From 2c4645de414e0ac07cb27bf0d6ed11812765a8c0 Mon Sep 17 00:00:00 2001 From: Tiago Kiel <186439951+TiagoCK@users.noreply.github.com> Date: Tue, 13 Oct 2026 08:52:03 -0400 Subject: [PATCH 01/34] fix(cos): check object-stream member numbers and offsets (#1815) Object-stream members were indexed with unchecked casts: `/First` and each offset went through `as usize` and an unchecked addition, which truncate or wrap on 32-bit targets such as the wasm32 build, and the member number went through `as u32` on every target, so a member numbered 2^32 + 3 was read as object 3. A negative offset was clamped to zero and read the stream's first member instead. Each member is now converted and added with checks, and kept as an unusable slot (preserving positions, which cross-reference entries refer to) when its number is not an object number or its start falls outside the decoded data. Looking such a member up is a MissingObject error, as for any other absent object. Co-Authored-By: Claude Opus 5.5 --- crates/cos/src/document.rs | 113 +++++++++++++++++++++++++++---------- 1 file changed, 84 insertions(+), 29 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index a0b1f56ee..97c832eb2 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -311,11 +311,22 @@ pub struct Revision { #[derive(Clone, Debug)] struct ObjStm { - /// (object number, byte offset within `data`). - index: Vec<(u32, usize)>, + /// One slot per member in stream order, since cross-reference entries name a member by its + /// position: (object number, byte offset within `data`), or `None` for a member whose number + /// or offset is unusable ([`objstm_member`]). + index: Vec>, data: Arc>, } +/// Where member `num` of an object stream starts: its `offset` past `/First` (`first`), when the +/// number is an object number and the start lies inside the decoded `len` bytes. Checked on every +/// target, so a 32-bit build neither truncates nor wraps a large `/First` or offset (#1815). +fn objstm_member(num: i64, offset: i64, first: Option, len: usize) -> Option<(u32, usize)> { + let num = u32::try_from(num).ok()?; + let start = first?.checked_add(usize::try_from(offset).ok()?)?; + (start < len).then_some((num, start)) +} + /// Where an overlay slot came from. #[derive(Clone, Debug, PartialEq)] enum Slot { @@ -395,7 +406,7 @@ impl ObjectReader { let bytes = streams.values().fold(0usize, |bytes, stream| { bytes .saturating_add(stream.data.capacity()) - .saturating_add(stream.index.capacity().saturating_mul(std::mem::size_of::<(u32, usize)>())) + .saturating_add(stream.index.capacity().saturating_mul(std::mem::size_of::>())) }); if bytes > Self::STREAM_BYTES || streams.len() >= Self::OBJECTS { // Retain one oversized active stream so its next object does not require @@ -890,9 +901,9 @@ impl Document { // Offsets may count from a header that does not start the file. usize::try_from(offset).is_ok_and(|off| at(off) || (self.header_offset != 0 && at(off.saturating_add(self.header_offset)))) } - Some(XrefEntry::InStream { stream, index }) => { - self.objstm(stream).is_ok_and(|stm| usize::try_from(index).ok().and_then(|i| stm.index.get(i)).is_some_and(|(n, _)| *n == num)) - } + Some(XrefEntry::InStream { stream, index }) => self + .objstm(stream) + .is_ok_and(|stm| usize::try_from(index).ok().and_then(|i| stm.index.get(i)).is_some_and(|m| m.is_some_and(|(n, _)| n == num))), } } @@ -1005,7 +1016,7 @@ impl Document { let bytes = streams.values().fold(0usize, |bytes, stream| { bytes .saturating_add(stream.data.capacity()) - .saturating_add(stream.index.capacity().saturating_mul(std::mem::size_of::<(u32, usize)>())) + .saturating_add(stream.index.capacity().saturating_mul(std::mem::size_of::>())) }); bytes > Self::EDITOR_OBJSTM_BYTES || streams.len() > Self::EDITOR_OBJSTM_OBJECTS }; @@ -1096,14 +1107,9 @@ impl Document { } Some(XrefEntry::InStream { stream, index }) => { let stm = self.objstm(stream)?; - let (n, off) = stm - .index - .get(index as usize) - .copied() - .or_else(|| stm.index.iter().find(|(n, _)| *n == num).copied()) - .ok_or(CosError::MissingObject(num))?; - let off = - if n == num { off } else { stm.index.iter().find(|(n, _)| *n == num).map(|(_, o)| *o).ok_or(CosError::MissingObject(num))? }; + // The member at the entry's position, else wherever the stream lists `num`. + let at = usize::try_from(index).ok().and_then(|i| stm.index.get(i)).copied().flatten().filter(|(n, _)| *n == num); + let (_, off) = at.or_else(|| stm.index.iter().flatten().find(|(n, _)| *n == num).copied()).ok_or(CosError::MissingObject(num))?; Lexer::new(&stm.data, off).object() } } @@ -1115,13 +1121,13 @@ impl Document { } let Object::Stream(s) = &*self.try_get(num)? else { return Err(CosError::MissingObject(num)) }; let n = s.dict.int(b"N").unwrap_or(0).clamp(0, 1_000_000) as usize; - let first = s.dict.int(b"First").unwrap_or(0).max(0) as usize; + let first = usize::try_from(s.dict.int(b"First").unwrap_or(0).max(0)).ok(); let data = s.decoded_within(self.stream_limit)?; let mut lx = Lexer::new(&data, 0); let mut index = Vec::with_capacity(n); for _ in 0..n { let (Some(Object::Int(onum)), Some(Object::Int(off))) = (lx.object().ok(), lx.object().ok()) else { break }; - index.push((onum.max(0) as u32, first + off.max(0) as usize)); + index.push(objstm_member(onum, off, first, data.len())); } let stm = Arc::new(ObjStm { index, data: Arc::new(data) }); // Same bound as the object cache: decoded object streams can be megabytes each. @@ -1848,8 +1854,10 @@ impl Document { && s.dict.name(b"Type") == Some(b"ObjStm") && let Ok(stm) = self.objstm(num) { - for (idx, (inner, _)) in stm.index.iter().enumerate() { - entries.entry(*inner).or_insert(XrefEntry::InStream { stream: num, index: idx as u32 }); + for (idx, member) in stm.index.iter().enumerate() { + if let Some((inner, _)) = member { + entries.entry(*inner).or_insert(XrefEntry::InStream { stream: num, index: idx as u32 }); + } } } } @@ -2362,7 +2370,7 @@ mod tests { assert!(reader.document.cache.lock().unwrap().len() < ObjectReader::OBJECTS); let decoded = reader.document.objstms.lock().unwrap(); assert!(decoded.len() < ObjectReader::OBJECTS); - let bytes: usize = decoded.values().map(|s| s.data.capacity() + s.index.capacity() * std::mem::size_of::<(u32, usize)>()).sum(); + let bytes: usize = decoded.values().map(|s| s.data.capacity() + s.index.capacity() * std::mem::size_of::>()).sum(); assert!(bytes <= ObjectReader::STREAM_BYTES || decoded.len() == 1, "only one oversized stream may remain"); } drop(reader); @@ -2483,6 +2491,16 @@ mod tests { /// says it lives in object stream 4. #[test] fn hybrid_reference_files_read_the_xref_stream_too() { + let doc = Document::open(Arc::new(object_stream_file("3 0 ", "<< /Hidden (yes) >>", 0))).unwrap(); + assert!(doc.repair_log().is_empty(), "{:?}", doc.repair_log()); + assert_compact_matches_materialized(&doc); + assert_eq!(text_of(&doc, 3, b"Hidden").as_deref(), Some("yes")); + } + + /// A hybrid-reference file whose classic table leaves object 3 out, and whose `/XRefStm` + /// stream puts it at position `member` of object stream 4, which holds `header` (the number + /// and offset pairs) and then `members`. + fn object_stream_file(header: &str, members: &str, member: u8) -> Vec { let mut out = b"%PDF-1.5\n".to_vec(); let mut offs = [0usize; 6]; let mut obj = |out: &mut Vec, n: usize, body: &[u8]| { @@ -2493,12 +2511,13 @@ mod tests { }; obj(&mut out, 1, b"<< /Type /Catalog /Pages 2 0 R >>"); obj(&mut out, 2, b"<< /Type /Pages /Kids [] /Count 0 /Extra 3 0 R >>"); - let inner = b"3 0 << /Hidden (yes) >>"; - let mut stm = format!("<< /Type /ObjStm /N 1 /First 4 /Length {} >>\nstream\n", inner.len()).into_bytes(); - stm.extend_from_slice(inner); + let count = header.split_whitespace().count() / 2; + let inner = format!("{header}{members}"); + let mut stm = format!("<< /Type /ObjStm /N {count} /First {} /Length {} >>\nstream\n", header.len(), inner.len()).into_bytes(); + stm.extend_from_slice(inner.as_bytes()); stm.extend_from_slice(b"\nendstream"); obj(&mut out, 4, &stm); - let data = [2u8, 0, 4, 0]; + let data = [2u8, 0, 4, member]; let mut xs = format!("<< /Type /XRef /Size 6 /W [1 2 1] /Index [3 1] /Length {} >>\nstream\n", data.len()).into_bytes(); xs.extend_from_slice(&data); xs.extend_from_slice(b"\nendstream"); @@ -2510,11 +2529,47 @@ mod tests { out.extend_from_slice(line.as_bytes()); } out.extend_from_slice(format!("trailer << /Size 6 /Root 1 0 R /XRefStm {} >>\nstartxref\n{table}\n%%EOF\n", offs[5]).as_bytes()); - let doc = Document::open(Arc::new(out)).unwrap(); - assert!(doc.repair_log().is_empty(), "{:?}", doc.repair_log()); - assert_compact_matches_materialized(&doc); - let hidden = doc.get(ObjRef::new(3, 0)); - assert_eq!(hidden.as_dict().and_then(|d| d.get(b"Hidden").and_then(|h| h.as_string().map(|s| s.to_text()))), Some("yes".into())); + out + } + + fn text_of(doc: &Document, num: u32, key: &[u8]) -> Option { + doc.get(ObjRef::new(num, 0)).as_dict().and_then(|d| d.get(key).and_then(|v| v.as_string().map(|s| s.to_text()))) + } + + /// An object stream member numbered 2^32 + 3 is not object 3 (#1815): the number used to be + /// cast to `u32`, which dropped the high bits. + #[test] + fn object_stream_numbers_beyond_u32_are_not_truncated() { + let control = Document::open(Arc::new(object_stream_file("3 0 ", "<< /Member (yes) >>", 0))).unwrap(); + assert_eq!(text_of(&control, 3, b"Member").as_deref(), Some("yes")); + let doc = Document::open(Arc::new(object_stream_file("4294967299 0 ", "<< /Member (yes) >>", 0))).unwrap(); + assert!(matches!(*doc.get(ObjRef::new(3, 0)), Object::Null), "{:?}", doc.get(ObjRef::new(3, 0))); + assert!(!doc.xref_entry_is_exact(3)); + } + + /// Member starts are checked rather than cast, so they cannot wrap on any target (#1815). + #[test] + fn object_stream_member_starts_are_checked() { + assert_eq!(objstm_member(3, 4, Some(10), 20), Some((3, 14))); + assert_eq!(objstm_member(3, 9, Some(10), 20), Some((3, 19)), "the last byte"); + assert_eq!(objstm_member(3, 10, Some(10), 20), None, "at the end of the data"); + assert_eq!(objstm_member(3, 1, Some(usize::MAX), 20), None, "/First plus the offset overflows"); + assert_eq!(objstm_member(3, i64::MAX, Some(10), 20), None); + assert_eq!(objstm_member(3, -1, Some(10), 20), None); + assert_eq!(objstm_member(3, 0, None, 20), None, "/First does not fit in usize"); + assert_eq!(objstm_member(i64::from(u32::MAX), 0, Some(0), 1), Some((u32::MAX, 0))); + assert_eq!(objstm_member(i64::from(u32::MAX) + 1, 0, Some(0), 1), None); + assert_eq!(objstm_member(-1, 0, Some(0), 1), None); + } + + /// A member at a negative offset is unusable, not the stream's first member (#1815): the + /// offset used to be clamped to zero, which read the object at `/First`. + #[test] + fn object_stream_negative_offsets_are_refused() { + let doc = Document::open(Arc::new(object_stream_file("7 0 3 -1 ", "<< /First (yes) >>", 1))).unwrap(); + assert!(matches!(*doc.get(ObjRef::new(3, 0)), Object::Null), "{:?}", doc.get(ObjRef::new(3, 0))); + let control = Document::open(Arc::new(object_stream_file("7 0 3 19 ", "<< /First (yes) >> << /Second (yes) >>", 1))).unwrap(); + assert_eq!(text_of(&control, 3, b"Second").as_deref(), Some("yes")); } #[test] From bf27cb259a26d0f1e326d4e7ddeb0f4b5e649f7d Mon Sep 17 00:00:00 2001 From: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> Date: Sun, 18 Oct 2026 13:07:15 +0530 Subject: [PATCH 02/34] Vendor and patch hayro-cmap for tolerant CMap parsing A /ToUnicode CMap with any unparsable part made the reading path discard the whole map and fall back to the font encoding, so search and redaction missed text the edit layer could read. Vendor hayro-cmap 0.1.0 and patch it to skip malformed bfchar/bfrange entries, unresolvable usecmap references and unparseable tails, keeping every readable entry, matching MuPDF/pdftotext and PdfCraft's own edit-layer parser. Fixes #3177. Signed-off-by: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> --- Cargo.lock | 2 - Cargo.toml | 1 + vendor/README.md | 1 + vendor/hayro-cmap/Cargo.lock | 53 + vendor/hayro-cmap/Cargo.toml | 87 ++ vendor/hayro-cmap/Cargo.toml.orig | 21 + vendor/hayro-cmap/LICENSE-APACHE | 177 +++ vendor/hayro-cmap/LICENSE-MIT | 7 + vendor/hayro-cmap/README.md | 25 + vendor/hayro-cmap/assets/LICENSE.txt | 28 + vendor/hayro-cmap/assets/README.md | 2 + vendor/hayro-cmap/assets/cmaps.brotli | Bin 0 -> 250740 bytes vendor/hayro-cmap/src/bcmap/embedded.rs | 129 ++ vendor/hayro-cmap/src/bcmap/huffman.rs | 180 +++ vendor/hayro-cmap/src/bcmap/mod.rs | 416 ++++++ vendor/hayro-cmap/src/bcmap/reader.rs | 123 ++ vendor/hayro-cmap/src/lib.rs | 1620 +++++++++++++++++++++++ vendor/hayro-cmap/src/parse.rs | 409 ++++++ 18 files changed, 3279 insertions(+), 2 deletions(-) create mode 100644 vendor/hayro-cmap/Cargo.lock create mode 100644 vendor/hayro-cmap/Cargo.toml create mode 100644 vendor/hayro-cmap/Cargo.toml.orig create mode 100644 vendor/hayro-cmap/LICENSE-APACHE create mode 100644 vendor/hayro-cmap/LICENSE-MIT create mode 100644 vendor/hayro-cmap/README.md create mode 100644 vendor/hayro-cmap/assets/LICENSE.txt create mode 100644 vendor/hayro-cmap/assets/README.md create mode 100644 vendor/hayro-cmap/assets/cmaps.brotli create mode 100644 vendor/hayro-cmap/src/bcmap/embedded.rs create mode 100644 vendor/hayro-cmap/src/bcmap/huffman.rs create mode 100644 vendor/hayro-cmap/src/bcmap/mod.rs create mode 100644 vendor/hayro-cmap/src/bcmap/reader.rs create mode 100644 vendor/hayro-cmap/src/lib.rs create mode 100644 vendor/hayro-cmap/src/parse.rs diff --git a/Cargo.lock b/Cargo.lock index 91ed3c490..5e801ad77 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -2544,8 +2544,6 @@ checksum = "9f4d0e94ddd48749f06bbe4e5389fb9799a0c45bcaf00495042076ef05e3241a" [[package]] name = "hayro-cmap" version = "0.1.0" -source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "24d285dc30731c8485de5fa732fbdf2b3affdf01e4da7c2135022ca6fa664bf6" dependencies = [ "brotli", "hayro-postscript", diff --git a/Cargo.toml b/Cargo.toml index 597dd316f..091296254 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -110,6 +110,7 @@ codegen-units = 1 hayro-interpret = { path = "vendor/hayro-interpret" } hayro = { path = "vendor/hayro" } hayro-syntax = { path = "vendor/hayro-syntax" } +hayro-cmap = { path = "vendor/hayro-cmap" } hayro-jbig2 = { path = "vendor/hayro-jbig2" } lopdf = { path = "vendor/lopdf" } winit = { path = "vendor/winit" } diff --git a/vendor/README.md b/vendor/README.md index 87f9d8682..394433986 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -27,6 +27,7 @@ A PdfCraft patch in `hayro-syntax::Page::base_dimensions` expresses the existing | winit | 0.30.13 | Apache-2.0 | (1) on Windows 11 (build 22000 and later), `WM_DPICHANGED` applies the rect Windows suggests. It used to nudge the window onto the monitor `MonitorFromWindow` names, which during a drag is still the monitor being left; the nudge sent another `WM_DPICHANGED`, so a window dragged to a monitor with another scale factor grew and stayed on the wrong monitor at the wrong scale (#324). Backport of winit master 488c036a ([winit#4341](https://github.com/rust-windowing/winit/pull/4341)), which no 0.30 release has; Windows 10 keeps the old adjustment, as on winit master. (2) on Wayland, a seat's text input is destroyed when the seat goes away, not whenever it loses a capability: losing any capability (a touch screen or tablet unplugged, a remote-desktop session ending) destroyed it, and only an added capability recreated it, so IME stopped working until restart. Port of [winit#4747](https://github.com/rust-windowing/winit/pull/4747); (3) on Windows, an IME composition reports its cursor from `GCS_CURSORPOS` only when `WM_IME_COMPOSITION` says it is set, and at the end of the composition otherwise: the Microsoft Korean IME never sets it, so the cursor read as 0 (egui then drew no caret while composing; it now draws it after the syllable). Port of [winit#4746](https://github.com/rust-windowing/winit/pull/4746); (4) on Windows, a hidden window is not maximized until it is shown, and a hidden window shown while maximized gets one `ShowWindow(SW_MAXIMIZE)`: `SW_MAXIMIZE` shows a hidden window, so PdfCraft, which eframe keeps hidden until its first frame is drawn, appeared unpainted (white) when it restored a maximized window, and could lose its focus. Port of [winit#4587](https://github.com/rust-windowing/winit/pull/4587) (the later, focus-safe form of [winit#4575](https://github.com/rust-windowing/winit/pull/4575)); (5) on X11, when allowing or disallowing IME replaces the focused window's input context (egui does it each time a text field gains or loses focus), the new context is focused: no focus event follows, so it stayed unfocused until the window lost and regained focus. Port of [winit#4727](https://github.com/rust-windowing/winit/pull/4727); (6) on macOS, `insertText` commits through IME when a composition was started (`pending_commit`, set by `setMarkedText`) rather than when marked text is left: input methods that end a composition with `unmarkText` before `insertText` lost the text, and Apple Korean dropped the key typed after a syllable (`한` then `5` or `;`); other `insertText` calls stay `KeyboardInput`; a commit outside a key press (a candidate clicked with the mouse, dictation) clears the marked text and leaves no key to swallow, as in [winit#4748](https://github.com/rust-windowing/winit/pull/4748) (without it the next key, such as Backspace, was lost). `validAttributesForMarkedText` lists the underline and marked-clause attributes, without which some input methods don't compose. Port of [winit#4650](https://github.com/rust-windowing/winit/pull/4650), the 0.30 backport of [winit#4651](https://github.com/rust-windowing/winit/pull/4651) (merged on winit master), with the underline attribute spelled as AppKit's `NSUnderlineStyleAttributeName` (`NSUnderline`); (7) on macOS, disallowing IME with a composition in progress (focus leaving a text field) tells the input method to discard its marked text (`discardMarkedText`): it kept composing it once IME was allowed again, so the abandoned syllable came back (`하`, then `가` typed in the next field, became `학`, then `하` + `가`; Japanese `か` then `さ` became `傘`). Port of [winit#4745](https://github.com/rust-windowing/winit/pull/4745); (8) on macOS, text inserted outside a key press without a composition (the emoji picker) is reported as `Ime::Commit`; it was dropped. Port of the emoji-picker part of [winit#4749](https://github.com/rust-windowing/winit/pull/4749) (0.30 backport of [winit#4748](https://github.com/rust-windowing/winit/pull/4748)); its `selectedRange` change for dictation is left out because it switches on press-and-hold for every held key (an accent menu instead of key repeat, and the chosen accent added after the repeated letters); (9) on macOS, a key whose text was committed through IME isn't also sent as `KeyboardInput` unless AppKit forwards it as a command (Enter): Apple Korean answers Space with no syllable in progress with `setMarkedText(" ")`, `insertText(" ")`, `setMarkedText("")`, and the last call reset winit's state, so the space arrived twice. Same goal as [winit#4478](https://github.com/rust-windowing/winit/pull/4478), whose changes don't reach this call order (macOS 14); to be proposed upstream there; (10) on macOS, when the first key in a fresh app reaches Apple Korean's `insertText` as a raw compatibility jamo instead of starting a composition, that key is interpreted once more, synchronously, before it is sent as `KeyboardInput` (g k s typed ㅎㅏㄴ instead of 한); only an unmodified, non-repeated first key whose one callback inserts that same jamo, with IME enabled and nothing composed, focus and input source unchanged, is retried (`ime_startup.rs`). Port of [winit#4744](https://github.com/rust-windowing/winit/pull/4744), with its policy tests.; (11) on X11 without `Xft.dpi`, the scale factor comes from the X screen's configured DPI (its size in pixels and millimetres) before the output's physical density from RandR: Xorg can deliberately run a dense panel at 96 DPI, and only winit applications then looked zoomed; `Xft.dpi` still wins, `WINIT_X11_SCALE_FACTOR` still overrides, and a screen reporting no size keeps the RandR fallback. PdfCraft patch (#1150, replacing the launcher pin proposed in #908). (12) on macOS, `WinitView` answers `mouseDownCanMoveWindow` with the window's `movable_by_window_background` attribute (NO unless asked): NSView answers YES for a view that isn't opaque, so with the transparent title bar and full-size content view PdfCraft uses, AppKit moved the window for any drag that began in the 28 pt title-bar band before the app saw it, and document tabs (which live in that band) could not be dragged to reorder them. The empty strip still moves the window through `ViewportCommand::StartDrag`. PdfCraft patch. The published crate's asset files (`docs/res/*.svg`, `examples/data/*.png`) are left out, so its `window` example doesn't build; (13) on Wayland, pixel-only wheel, wheel-tilt and continuous-source scroll events report `Moved` rather than starting an unterminated finger gesture, so egui follows modifier release instead of keeping Ctrl zoom latched (#991). Finger gestures and unknown sources retain their existing phase policy | (1) `pdfcraft` `tests::winit_carries_the_windows_11_monitor_scale_fix` keeps the patch in the build; the drag itself needs two monitors with different scale factors on Windows 11; (2) `pdfcraft` `tests/winit_wayland_text_input.rs`; the capability change itself needs a Wayland session and an input method; (3) `pdfcraft` `tests/winit_windows_ime_cursor.rs`; composing itself needs the Microsoft Korean IME; (4) `pdfcraft` `tests/winit_windows_hidden_maximized.rs`; the start-up itself needs Windows and a window that was closed maximized; (5) `pdfcraft` `tests/winit_x11_ime_focus.rs`; switching fields itself needs an X11 session and an input method (IBus, fcitx5); (6) `pdfcraft` `tests/winit_macos_ime_commit.rs`; composing itself needs macOS and an input method; (7) `pdfcraft` `tests/winit_macos_ime_discard.rs`; composing itself needs macOS and an input method; (8) `pdfcraft` `tests/winit_macos_emoji_picker.rs`; the picker itself needs macOS; (9) `pdfcraft` `tests/winit_macos_korean_space.rs`; composing itself needs macOS and Apple Korean; (10) `pdfcraft` `tests/winit_macos_korean_first_key.rs` and `ime_startup.rs`'s own tests (run them with `rustc --test`); typing itself needs macOS and Apple Korean; (11) `pdfcraft` `tests/winit_x11_dpi.rs` keeps the patch in the build, and `util::randr` tests the policy (run them in the vendored crate); the scale itself needs an X11 session; (12) `pdfcraft` `tests/winit_macos_title_bar_drag.rs` keeps the patch in the build; the drag itself needs macOS and a pointer; (13) `pdfcraft` `tests/winit_wayland_scroll.rs` compiles the production phase policy and exercises modifier release through egui; live COSMIC confirmation remains open | | egui-wgpu | 0.36.2 | MIT OR Apache-2.0 | (1) `winit::surface_fit`: a window's surface is configured within the device's `max_texture_dimension_2d`, both sides shrunk by one factor that egui then draws at (`pixels_per_point × render_scale`), so the whole window is drawn into the surface (on DX12 it is stretched over the window and lines up with the pointer; wgpu's GL backend copies it unscaled into a corner); the longer side becomes exactly the limit; the depth and MSAA textures follow, and a recreated surface starts again from the window's own size. egui-wgpu configured the window's size as is, and `Surface::configure` panics beyond the limit: a window restored at another monitor's scale (opening an Outlook attachment on a mixed-DPI laptop asked an 8192-pixel device for 8600 × 3423, #577) or stretched across monitors closed PdfCraft, before its window appeared when it happened at start-up ([emilk/egui#8361](https://github.com/emilk/egui/issues/8361)). The published crate leaves out egui's licence files; `LICENSE-APACHE` and `LICENSE-MIT` are from the 0.36.2 tag; (2) `winit::catch_errors`: a new window's surface is configured inside wgpu error scopes (validation, out of memory, internal), and a failure is returned as `WgpuError::ConfigureSurface` (for the root window at start-up eframe returns it as `Error::Wgpu`, which PdfCraft retries with OpenGL); nothing is kept for the window. wgpu's default handler panicked instead, and eframe configures the root surface before the app is created, so a driver that can't configure one closed PdfCraft at launch (a 2015 Intel GPU through wgpu's GL backend timed out waiting for the GPU to go idle, `GpuWaitTimeout`, #519). Unchanged, so still a panic: an error in a later reconfigure (resize, a lost surface), a device lost during the first configure (wgpu reports it only to a callback), and a window shown at zero size (first configured on resize) | (1) `pdfcraft` `tests::a_window_larger_than_the_gpu_limit_gets_a_surface_within_it`, `tests::surface_fit_never_empties_or_overflows_a_surface`, and `tests::egui_wgpu_carries_the_surface_size_fit`, which keeps the patch in the build and in the paths that size a surface; (2) `pdfcraft` `tests::a_gpu_error_while_a_window_is_set_up_is_returned_not_a_panic` (a real device: an error inside `catch_errors` is returned, not a panic) and `tests::egui_wgpu_returns_a_surface_it_cannot_configure_as_an_error`, which keeps it around `add_surface`'s configure | | vello_common | 0.0.8 | Apache-2.0 OR MIT | (1) `clip::intersect` (the clip step every clipped path passes through) starts its row walk at the first row both paths have strips in, not the earlier of the two first rows. The walk visited every row from the top of the clip for each path (on the A1 sample's full-page canvas, about half its strip rows per stroke), and rows where one path has no strips produce no output, so the strips are unchanged (vello_common 0.1 starts at the later row too). Its cost grew with the page, and a full-page render of the A1 sample was about 30% faster; the output is byte-identical | (1) `clip::tests::intersect_walks_only_rows_both_paths_have_strips` counts the rows the walk visits (`cargo test --manifest-path vendor/vello_common/Cargo.toml`) and fails without the patch; the render crate's pixel tests run through the patched crate | +| hayro-cmap | 0.1.0 | Apache-2.0 OR MIT | malformed parts of a CMap (bad bfchar/bfrange entries, unresolvable `usecmap`, unparseable tail) are skipped instead of discarding the whole map, so search and redaction find text the edit layer can read (#3177) | `hayro-cmap::tests::bad_bfchar_entry_is_skipped`, `unresolvable_usecmap_is_skipped` | Renderer startup errors: `egui-wgpu::RenderState::create` also wraps `Renderer::new` in `winit::catch_errors` when the winit feature is enabled. A rejected shader or render pipeline diff --git a/vendor/hayro-cmap/Cargo.lock b/vendor/hayro-cmap/Cargo.lock new file mode 100644 index 000000000..b19a6c70f --- /dev/null +++ b/vendor/hayro-cmap/Cargo.lock @@ -0,0 +1,53 @@ +# This file is automatically @generated by Cargo. +# It is not intended for manual editing. +version = 4 + +[[package]] +name = "alloc-no-stdlib" +version = "2.0.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "cc7bb162ec39d46ab1ca8c77bf72e890535becd1751bb45f64c597edb4c8c6b3" + +[[package]] +name = "alloc-stdlib" +version = "0.2.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "94fb8275041c72129eb51b7d0322c29b8387a0386127718b096429201a5d6ece" +dependencies = [ + "alloc-no-stdlib", +] + +[[package]] +name = "brotli" +version = "8.0.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "4bd8b9603c7aa97359dbd97ecf258968c95f3adddd6db2f7e7a5bef101c84560" +dependencies = [ + "alloc-no-stdlib", + "alloc-stdlib", + "brotli-decompressor", +] + +[[package]] +name = "brotli-decompressor" +version = "5.0.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "874bb8112abecc98cbd6d81ea4fa7e94fb9449648c93cc89aa40c81c24d7de03" +dependencies = [ + "alloc-no-stdlib", + "alloc-stdlib", +] + +[[package]] +name = "hayro-cmap" +version = "0.1.0" +dependencies = [ + "brotli", + "hayro-postscript", +] + +[[package]] +name = "hayro-postscript" +version = "0.1.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "885c5ef0654933139a9b9546fc2c69e18d37f38aa2520f079092b1be18f1fcaa" diff --git a/vendor/hayro-cmap/Cargo.toml b/vendor/hayro-cmap/Cargo.toml new file mode 100644 index 000000000..3818d177b --- /dev/null +++ b/vendor/hayro-cmap/Cargo.toml @@ -0,0 +1,87 @@ +# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO +# +# When uploading crates to the registry Cargo will automatically +# "normalize" Cargo.toml files for maximal compatibility +# with all versions of Cargo and also rewrite `path` dependencies +# to registry (e.g., crates.io) dependencies. +# +# If you are reading this file be aware that the original Cargo.toml +# will likely look very different (and much more reasonable). +# See Cargo.toml.orig for the original contents. + +[package] +edition = "2024" +name = "hayro-cmap" +version = "0.1.0" +authors = ["Laurenz Stampfl "] +build = false +autolib = false +autobins = false +autoexamples = false +autotests = false +autobenches = false +description = "A parser for CMap files." +readme = "README.md" +license = "Apache-2.0 OR MIT" +repository = "https://github.com/LaurenzV/hayro" +resolver = "2" + +[features] +default = ["embed-cmaps"] +embed-cmaps = [ + "std", + "dep:brotli", +] +std = [] + +[lib] +name = "hayro_cmap" +path = "src/lib.rs" + +[dependencies.brotli] +version = "8" +features = [ + "std", + "std", +] +optional = true +default-features = false + +[dependencies.hayro-postscript] +version = "0.1.0" +default-features = false + +[lints.clippy] +collection_is_never_read = "warn" +dbg_macro = "warn" +debug_assert_with_mut_call = "warn" +default_trait_access = "warn" +doc_markdown = "warn" +fn_to_numeric_cast_any = "warn" +infinite_loop = "warn" +large_stack_arrays = "warn" +mismatching_type_param_order = "warn" +negative_feature_names = "warn" +redundant_feature_names = "warn" +same_functions_in_if_condition = "warn" +semicolon_if_nothing_returned = "warn" +todo = "warn" +too_many_arguments = "allow" +unseparated_literal_suffix = "warn" +use_self = "warn" +wildcard_dependencies = "warn" + +[lints.rust] +elided_lifetimes_in_paths = "warn" +keyword_idents_2024 = "forbid" +missing_docs = "warn" +non_ascii_idents = "forbid" +non_local_definitions = "forbid" +trivial_numeric_casts = "warn" +unnameable_types = "warn" +unreachable_pub = "warn" +unsafe_op_in_unsafe_fn = "forbid" +unused_import_braces = "warn" +unused_lifetimes = "warn" +unused_macro_rules = "warn" +unused_qualifications = "warn" diff --git a/vendor/hayro-cmap/Cargo.toml.orig b/vendor/hayro-cmap/Cargo.toml.orig new file mode 100644 index 000000000..59f5c99d3 --- /dev/null +++ b/vendor/hayro-cmap/Cargo.toml.orig @@ -0,0 +1,21 @@ +[package] +name = "hayro-cmap" +version = "0.1.0" +description = "A parser for CMap files." +authors = { workspace = true } +edition = { workspace = true } +repository = { workspace = true } +license = { workspace = true } + +[features] +default = ["embed-cmaps"] +std = [] +# Embed the 61 predefined cmaps into the binary. Adds about ~250KB of brotli-encoded data. +embed-cmaps = ["std", "dep:brotli"] + +[dependencies] +hayro-postscript = { workspace = true } +brotli = { workspace = true, optional = true, features = ["std"] } + +[lints] +workspace = true diff --git a/vendor/hayro-cmap/LICENSE-APACHE b/vendor/hayro-cmap/LICENSE-APACHE new file mode 100644 index 000000000..4947287f7 --- /dev/null +++ b/vendor/hayro-cmap/LICENSE-APACHE @@ -0,0 +1,177 @@ + + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS \ No newline at end of file diff --git a/vendor/hayro-cmap/LICENSE-MIT b/vendor/hayro-cmap/LICENSE-MIT new file mode 100644 index 000000000..ae3742de4 --- /dev/null +++ b/vendor/hayro-cmap/LICENSE-MIT @@ -0,0 +1,7 @@ +Copyright (c) The Hayro Authors + +Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. \ No newline at end of file diff --git a/vendor/hayro-cmap/README.md b/vendor/hayro-cmap/README.md new file mode 100644 index 000000000..bc93f93da --- /dev/null +++ b/vendor/hayro-cmap/README.md @@ -0,0 +1,25 @@ +# hayro-cmap + +[![Crates.io](https://img.shields.io/crates/v/hayro-cmap.svg)](https://crates.io/crates/hayro-cmap) +[![Documentation](https://docs.rs/hayro-cmap/badge.svg)](https://docs.rs/hayro-cmap) + + + +A parser for cmap files, as they are found in PDFs. + +This crate provides a parser for CMap files and allows you to +- Map character codes from text-showing operators to CID identifiers. +- Map CIDs to Unicode characters or strings. + +### Safety +This crate forbids unsafe code via a crate-level attribute. + + + +## License +Licensed under either of + +- Apache License, Version 2.0 ([LICENSE-APACHE](LICENSE-APACHE) or ) +- MIT license ([LICENSE-MIT](LICENSE-MIT) or ) + +at your option. diff --git a/vendor/hayro-cmap/assets/LICENSE.txt b/vendor/hayro-cmap/assets/LICENSE.txt new file mode 100644 index 000000000..2552cc990 --- /dev/null +++ b/vendor/hayro-cmap/assets/LICENSE.txt @@ -0,0 +1,28 @@ +Copyright 1990-2023 Adobe. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + +Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. + +Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. + +Neither the name of Adobe nor the names of its contributors may be +used to endorse or promote products derived from this software without +specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/vendor/hayro-cmap/assets/README.md b/vendor/hayro-cmap/assets/README.md new file mode 100644 index 000000000..4cedb1ff6 --- /dev/null +++ b/vendor/hayro-cmap/assets/README.md @@ -0,0 +1,2 @@ +See https://github.com/LaurenzV/cmap-resources/tree/hayro for information on how this blob was generated and how +it can be parsed. \ No newline at end of file diff --git a/vendor/hayro-cmap/assets/cmaps.brotli b/vendor/hayro-cmap/assets/cmaps.brotli new file mode 100644 index 0000000000000000000000000000000000000000..ba74bd32fc71bb274e3e7e5c36bafaf89b9d6815 GIT binary patch literal 250740 zcmV)7K*zsZA|dw_tEYEsI=n=1V>avtPI&%cU3I@=`?{|B z_xsoV6&Nuoq|!zsr5$ix&o@CR9C4zK6H^dptrPi4q~khT06G;-OO+`sDlUQ2vT|90 zvXv609jK~SP+S9C1GT_abplCXcA?&j`wdVDvQdF-!p$wMxT=kzm5bYJ3X-DS8Z*gE zN+u1H9WrAO*Z6^pgal7;hM^O*lE{%cR-mxMLJAxb!e`tp1SsaNj&9W>W4+enTN;F`YtWShK;j* zp5x@N-n6h3Lau;KwE~n9;5w~32*hU8TdDPy>{dw4`B6MY%>$FOOjQU$WtxQLZK8dw zdr4sd+B{waBipU)$SmGDu?s6Gj7RIRLXTj!eHKf+9=7aWRo0Ilm1LHQJ+M}g6HM-9 zTxMk{l)^OblsSR>W)w(pcU!{a@% zC_^EHqC8sI-F#34Mnf9bfC43Z55W^p8YXtLU_!7sOpOT(yAP6lgk+vVPdy$_>R9Hn zf*Ll))Es~*h-i|Te46ny<&~Ueh82-r35Z2OL{+7P(T-6Q@s|7WERK|`{ zlQ>PDVk!enrGUv+wkn}8Bhe78Ps0q7Y)2O`xsfReOpT%E5@*8JqQn8bq{iYiz?3^C zsI$O!G!xOuv>KJO=$!ZHz}K@wHqGlnHxL~F};JHf0q@?PH;E@7#GpaYi&T?bAH=;EWh>*-A zTX=;$$>cI7R{bu-+yZ0qjqRpfX}r=WXW6zst9AZ@eUVdyu|93zv_~=#eziiDI2Uqvq)A> zt4ukL$KALh3@teaIYK}HOJM~w#n6Z!gC#Qh3@R&(oA!8oJKWYm0LQR^BNXR>3xZO8 z+(I0deXmm@WBYNdi4l>%Epr7c``@r>NqXW<59A4KCLlrx>&NKd+w@jNLS}---B{Vb z)#^a2czbUG2k!*+3OmRl*h(BR(LzegFERY#CLBwKZ-v8QywKNY-z6k@kiC9xeGQ6`@B+ma-mu(5=RADz1Er#! zZ_W)vgVT`CHKDSQp%^~wr3CK6Do2K`{^`CbiY$ZYBXLK^3r@gQ9n>DZgQcYf@DdgY z`Y!eMW;NX-F${Dwh0s zoo)jG{$ubKXo`U;2BsL8JZOCfN6PwPV^3s%TX_N(1RZ;>Vrk_glC3{MS5PlLD>CrO z7`{Uf(wZKkKeg}|qGd!%{B}X%yM=`+H&Ny|eNE0aR2hko%qW3{fKH2Z^CVMk`A3rB z3Bs&-j}rYP+b5S3V=S0xNe6EPDWtX5CHgG#z$2Mqtw6~`z|?y-oeob#8_x*B9c+NUnF0Ak%(V` zR|r=)+Wv3IGi>#xB5@qW;G;#SOVusHvF=;1v+0ddul_cvZEv>vZ;_LORg;9tLq*oV zB{t67Ek5@YWoXBG)ZVVXeb>Kxa-Qm^UCY&{s8Wc(f@-H_1&>NtAS7bAF{u?Q66@ih z3c-38M`c$B2pF?eV&5iwzXN6#6;Dx){paa@bT@Iic|`Ap%Y?Aswdt=rv+W{71+=lx zu_*!dqC7_l`?XE|wZ}q^Lzz4?j>_@f?`!?%i1gp+Kag|SwU9YaoQHi9apg+rRNMG) zXWO9As2jzfcF8Z9_yI9GKzu(d=@z=pVoCjU|I(Bq47^LReZEcGRqEc4Zj9JO@3w8# z`*FhGDYwVI1&!sfs9>o4sR88R5dmc$d?ewzZy};7A!4|-8MQF#KBr)B@q+h9G+HPS z3kN#ZLO&=JBG`_@HhyRI<5v$(j#2+^TbJc082uf}4#jsReYqC3c@U!5zMgW5rzT0} zFdA%KsCB=Azy60YCT0jJeM$(s7SCQ~9Lp6{rSMv!ZJ-EUhxDg3h zKK8-O!>bBb;(GOvuQS`lh~HyiTE~z?yOBYHoYtRy-fY<+_b@Ql?UHcv=k`_?d`LJJ zz99qr4oZ+rjTStaX{_5jgWR&Rt>su**hDftVc|piiDyW``0L1}`>MDYm6z)idKVG6 zB(t@nI}a8>>r+r(NECt=)C(1YI;8~ZG9lm_AsAv6#5}Z8ui3O^@e+ArLhXlZm(2~5 zEzFQ?VVh0&ez!>0BA-VFwqJ-6N!Olp{fiN!7KyTvx1kiyf~ja%5)?2|1Yw(ZUPQ&Y86 zY|Z+zov>rz5%t3_c(4^bi7x*weA7KUKJ=sGI1U&H;UFN{!avCZCYecQeAn|nTj9gl zh%v?($`h6=?2RhN*_p4d6OMf?_ef@vnPf)0mioTG-Ul5r#u)Pj;rrId^Tb8ppY~%u z=~#b`%x99>TA~UHb~_N_P$Hxi<@?sZ$`g~yb=^&>f^sbCy#4^?dhw!rFf!FX3u79z zfW;v24mmLe2Er5rQw&T$O#B^wp?5+_%7c0Z!BUoENJ$Wcu)7gbvX|-0@?sUj$bv~CoilDKT> z=ea~%F-{~OEs$Amuk1A@Z5*z+Z+s{(9RZzSl1XMmEfbrFvk40gAnr5K80C?|tMWkxw)f)C@U6A3$(QcSM>sTX3{Q9i^%A&_c3 zV$gWX~`rrnPBR_Adw)7 zyWKJgg%Hs-N~_do5WpP8R*ERL{5lR33;lzpQyx{a)7 zM{o_)jDr?o?gEK^ynsdCX`F{gCUI35b&<;`X<^K^Ppmd~^_I57M~#qn)xu4L>?N6A z&$d^Yl7=(*zv1cAgs@@WVzPadiK)df3Tgf&sz5f#;`T(y0qs0}3$*S3Q&;!DaqLz4 zIP#l*zeGEneDi^UgyVQfK?}ve6hl9?F408xlF`FGu;+0kV7h;ggqExB($0yO(R-R) z%;bJvY`NR2Tau0KZ;o$)J`Z$H&qkyyl#^_=lxnip6sm<0xQU!fxP^uNuUk#PKe>Hn z{-bDUh-A8tX4<1q(>~P_2oFlvYvfz^XqWb^-ymP#S$R2g3!9?jl*mK>@YWKSCOt~P zI?OLuaY!cIOT@dFO=HxFFA`|tHWn~93Ws1?g_A^4aefjdN7z#e z3b=VtL#=kOv||g&R2vEt!%qh;>Va7Ws)eIF!3W(x(211om*ABsp&TC)%wMpCu##K` zp8cN%R7qXy-g&Y0c3ai|RV7!Wpq}#uw19UM{3cV{ZXmqW>Dva{H|L=O&&rDfE}_%J zP4O@j)Ai6b0n@+^s1&4t1*&W3AuxI1_?^CPtJUAtqX+0v)sINp|NrTeOvLiL^ACiw zpuM7rj(V)ZBr|zikP&J1eISZlt{ZSLL}_ic*&TJNy2KRiUU-IhWq!)90kq9~_N<>tQNROX_xW~DYNcd66?an6 zlLO>tg+vq3=m{#8xVMS}5d;;DO3Lji0mFbAMNq*SM58FSq;I*zMHjHFAYk1h0VAMB z5wN7J8g<2zVv8lFYc_35vZekH&^lkn#J{eIdR`y9V@}kGi5z2Q6SZNYR?MPAE=A=` z=F8ru#tgs#0HAh6mtFwe1;mFN%{IT~r!x^t~8_f`BG*%(`m1=F$qHAIk{Y@f!gcX<@Rf0s0_1%G`Tel2j zFkFhAl0L?9|!e|V%F$$Bk z3T{meK}HN@L=SpV7rVQ~-HMSVV1c5NCbWu)^@@R6KZK|M7qZB@wVS&~d`Wm<8JshS zW6Bh5M4PlpVWv!6oPR3Dd;q5txWG9FZs-q(JL(%U*gRfIlSZ4QZIVjb)O^}0W8Ii? zMfB$?aE@J#n{mZ2K-C7@8G(wHaPPfvp;@{mRRaJ9FgCN6XI`)>2l28#=shOJy{5UQ zX(v5s?LmS@C&L5JVVow{|DgAQ=OH)!wstJ-m%z+BpA$MsK;FhE<_*^xPu1Hz$u8Lg@0h&zj5lzP^hBP zCwI@IKjE~w%9X~mv>$acH^kBB4fCr+|0A2*M%aW>okC#@U1CdxwItU1!E6!s<+xAxuCw zVB33kUXhuu;ga z#VV#wlZt7RDy6oybJu~l`A7@vyq#RZ_uHwglpb<3sC(`YS3!%xP(xCN*k;5fMc|5j>E!M8$zC*_~;+_pkn# zW48U(_5J@poqozKlKd=CgdOg^LL)~5N1qG_SAKy6c(M^OQ(E87vemk-%T~5Ci%sdf zeni{izxfla-ksgOIYH1rK}<}H#Q3TThv(Fw*}mymmUEW3 zxfk#APqG`p3h4`YAt>Rc1+eXZnXXy2MHY3%3+UK6&wwSl7?OgD7s4`D^wLALfw%pD zRg=wB6~Xe4+sX|rZxHZf@9OZ-Mu@B8e)p#jQSk(gB<6pbOPB3UAA1jP!z1LsV4*wo z?bC<_E`?k*u5I}Le(d{nV$=x3Zrtxb>`*vFO$qn8XTy50x}8r&MG^}|;Pa6=`@dCF zJzL$9{AyY)F%s-|P19mqgYX+LY>3rOi{B%Je@hk|nN5QnzKn6$lxv0&uO;mNMnbT7 z@qe#1EBtVlQ#yZy!)LSXO{`xCc|(L9u=6(()cebF0lnY7;%3%UsT|VGUJ8SVDNg{~ zgE&>Su_VhfNOF?eK^J3}C|=rOyY-B)9s3V}X#S~9|NpusZ|kyOhM!*~uTVfl)4y!@ z!{9&KEK|j9tVgb|Jt(}FpF<)O zQz?Qm=W>E z^X9(rrtp7>w*Vk396KQ4NT>=`jY3EDRJIEHJ?R1k{DtNJX?f6BOMd%fI{5=ZLU}mIQ}bLCw@JYwgu{4@ z;YZr^Z%$`)9Lcqv89Nq|m%OZp*XhK~zlo&3C=&P|2YOelXwB*cDaZsSKI4IYNb)T3 zull=B#mE7TfHzpUy1IA1g~%~u(jDyF5tQ;bXY?~yRM(W(6bVTy3JX3(5!6-8mxI+W zn&QLr{eJD9^RPyPwi@yKS3;uVi6K!MF?pgB3L)Xqr7gtQFVm@Lu2Y8Au6*Kk><_#ngbZrA3@O%(A``)c4aJ;01 z3F}0=Y~aa6>Df1P;rV~Y+(O!0H@O>PNv-)`Q_@R*CoQS0f{uSjgaLf(RcB#(ACgrj;poa0jLs zjA&I779fnk7?&%*mA?I9$}r{ab<8kk*fWi#`p?__zd@-kE$EpmoVEAbSA5pYm(l$) zQU#YuomLDd``S-8;kzFoWp6V(^PY<@qeKVnRD~s_5ml$E6u=m>HBvG2#2+x#v^Xag zUZQ)GYuJ=a`!D=iw&#^?v{en=jRc6h5j2SA5T`&~Lb5E|(nx;KmOY+#RBe!*LAJGF zzQ4J)y2e92><6tiQoT!vCh-yo72fO&^1(wT(Tn@>cv)(sptr|Ocm5&2esdBC|9-=7 zIm?t}O8#D&+s{~9e-!CaRVaipP(6PzFufOlofR#i_^1Da#;~|6-1#C||J@m|AlM93 zO&`WS)R^kZ%E}J@otz~N{CAh+1%{SD%>r39mva@j8n+ac!4lX&* z$Zm<))go!)ugHI2KFSRKSQ!=7ptd;!Vni}SF3*J;qArxo@=z;y*Hq0J}`yuWjLbDJF zolHxF{te#0-cQ`6id(n?+oXGG!7p%i-@B)U4cOl~-_yVZs0FHE3=3SMVhsb;QuN!)YF|~la136) z3%zUwxe&oHWuy9-a2(53?Vi5efBqDEx#ZSFv)c?b9nwq}ffPk?YVs3tsLHGf-97tA zkyG?*xi9qMcm(zb78csa!&HWUm_iB<6B|4pKKOh8U+|ws-ipNn^6oopX3pAaRhxyo zV>BZ9dtH_Pc&4V>(QxX{xttXa7jxw#{gKi#jB@)(CeAsj+jasgZlS40D2>p#e*ZSV?A^5;+8Iz)-FvIvfCk#q z;2)L-TVE~s8UsP>lsNR+$bczeMi2uICERgW`o5rQl$;yvvJ$wJyc4h`RAnps|HU{X0JKIpBiMRfg)D_qNS@g6z?Ma9 zSVWG6o*Y}22Jp1xgI)mdJg9Y2B@G&Y)w-=4{Qv*jmv4w|-3B#l+*qt1yrCU99l6Uz zc=~Mv3fN`E0y5> zn~2wcxWB7Nw;Bnlm^FM`nmzhQo!#x0S3>MWe`Q`KEBI&+&6#c}+>GD?rB5Zg2B=K= zq`>ePXNJO|oZSnX0yrq&U$DkRp~OVFl20MC=l!ei|J?|R;K=NO(XYGe)H`!~AgrN? z3KE$BSzE7+s!jABh+-balJy8|gPna76ZnArM1`$0mGmQI0fho|xqI)Dq9zNQpDgvHpvf}4+)ZZZe!};|fJIebh{lE?)Kjh~iY47+Hx~QGQXGYN zI-~*x3)pn`K0qJ*^FQX!n#Z*{W=%D@{reELOceEysFXm12cBz!;oZEk|JkZ(*3_wc zzNNUTBfkrJ0B;kiR5136d>xFR?e}l1Jra4%tv4(t1NU4gY6oBM>8208ScK=E*$*Pg z#Vz`gSdfdQ-&1iYV!wSTG)gkFt=K)eAvQwEU!lTfS@_lSDHCkxJ85TnKLtT#LNnr~ zpvKU+dliSj2^WaMLRFb6iQ_ATLVy%ysWK%g!LVXR559u`+euK-BB$lY*v#AsiwJae z-Nqt7mylDbXyb`~j5c#2YO0hXp^3GtTt@tTd>0kZlteQs#zvOC&?i;FYKUDB|N9v_ zS{gj{%i0t>DPDYRO>dknPw4)8;p51^<}?b0!kwa!l#j}w3=Zc)G{S#r)>O44*DW)% z2Rw|^5u|5-swBvX04lRpZA=3{WLS@7_(v&inlU!Qe|39!H`iv(WYWU^nGY05Q4*C( zmMJTi)2rC|I;|uKgD@yG$qDCk)~>;UT`nj%jFGHP!518 zaE1`1lVl4RO^_`xl+M7W-Eo~LuW$Mj7z6+Raj>4jl_m(J^YbKx1-1oy{$R5S7SP*$ zwVMA@HLnSHhhIT*_(!6Je5 z`T-za08#}6NKteHMQKYFC2ueCHWHw{B*54rKo^2r0c`{9&6=>CnDp+GwXNAsYfZAZ z)N)EYw<2t((yXoRlveNQmt#4Ve#^0xw&j>xX@_fWd&}XJa}LK4Q8&G8k1PQG#`rc7mYlH2ehGkfyw}b#xq3>FVk^ zx5bVfK@X6_Fsp%}Of4y6JHyrS?#f*4C5sAJs!ti=-29=$kJs zZzD4tW;IZH&}#ATORC01mzYRjjxEplZJ(r1@8+fK!la0Zh>@^K|4lxxl^L&4xGk=x zA;@zg7`9ko_7sA66bf-dzW7Yql%l=Qi8a*hi>p~5EHPm_+SFaQGO zPmzU#lX~_G<)6qiiIA05qR>Ft-cXPb$MOahX@H^KeIKK3m{vblxgv^_0Oq$J0WZiP zfOLSIq+tQ@3}m%hhT7AOIocU;?IK2B9=!4i;buR$vV_ zU<;if91+-o7$hJCdvE|pZ~<2c!wH=dy6pV}1Z90gD@Ud?$l(e_ViyUFfMLH($DCHIRGG`J zyWyr=?z-=xN1k}$qtAZ$>6hPX)cNDT1}$2(XB^JA0w^)!B*~Gtk18z&?kH2O-GFfu z=E6r|qG(~$gAIYSeAj-H*vKjs_@=|nvZK^#Cb*+nyHd|p=@ss^nqS@6&l&G-@t{XN z=5a53-8 zNZCjcg*I#U+bMx~dmStUY*rb{mOYgISH zubR2kw3+y;ORoM}Yu47uxE4xjzrqc<*k<=&c%#>$M{Rqlqx^rR+OlK0`nMZ${_?u= zy;kotrGkqgfrbcH_>f2aV|aAB8;m`3OsIk$0x!Y)@CAGa-($29*QM}Qs8!ZjONp`_ zcG_*f0}eUrq*K)B(qqVo1uK_bW5d=Bw>WYVAV>(RdmafFE6#JTq|1^mN1<{x8a1J5 z)#0NrI{m;f=&yN9Jp8axgcxx$R$60=eN<^NbkQZ&?A>&St3bgX3-?lzG@0_W88&S` z{xpd!ISN+UV?R&flBFnB>#HCBVPWIMokt>TjZMgu{YQ_9KqICtVZ$CFkS8Tnobb+j z9K49d0m*-S`Kd63M*xdhLWs1>Lgf3V#{_=Z3_?agMXL}ajT3Q6e3d}R{rIaA!3$e8 zAQ9FR?`p=D>c)metd{7;>eA#^y8271>$nE`8Vzeo&v?z*OLOyBE4R06qp2GDX;<#5 zmRgrVZD)z|4d-@GEJ%zrrQ-|SbTX{Qg**$Y0?X#<2-Z13m#u=qNW0NI&oWhzu@(yUd#5i^!-yY=9^-z@o4mwY*%x?!{&nd`IAdzsPX5 zXq~qZn}LMSCB5GV<--fJ0AT$y{Ky; zHyziE=l3ZDVb^LnqTk5_ZQF98Dci>x{BxbSLj`vj1`6un@v%+5#~K=pV*Qt>H3$JF z%Jw_|F)n-)M91{M=-LBhVgScYkDYAZcfalVK^P|xVNzC`Lwo$`oX)y=ES}1NkB z1jvo`01$7#`i@WTnB0Hu*pB!L0L~9$G5F!5teuyOfndNO&?Ng)@+;QV@Bh;E&(tq3 zOaXLfmbvl@g2TXdukI&|mzDno+)AnTxigz?gT1g{E&j8{5uC+e;q$p8C7pHeTUp5r ztV>oEaL?vd>hw+BK)D<0IX}}Nz&C!vteDZBe|6GnR_772dN0(qg5s9MpXLSnv66E= z^m)Ul9&7B=>0dd(w1G(ZQJWrtM&Z_>nPjpSON0X`Voee>|4JFShB8D^tMzS*NWdqU zB3$aZTw~{NtSPA@x-6xmUc{X&LAiZAdE1abcvBZC0jNLFanw?HSwE;9$O`Haj$n^L z<2iuE9UNrjO1oBr{2MjGU#`#mwVL8@*DB(<>$x`5&N@zuwN6I>J*ubNGW|DT?*U@V zL9slDhrfb&Ll)nT|3YFyihjglU~n(C%Od0wlcu=-Wy9}3r0)MaLdR-=djTIO=N^>- zRYCCxGM^PLyr_f#SU7Ps8HGeNmjH}}ftBN`lk*7>)cq(9r~`JXxi z|Fbsozi1P`KM&=j{8t_0zd607_@6q>|I!)$x6bl^yiErEi+b286(-f1mSfdPd7`k} zBQcYjh3qdz7>`@BI{h!e_wwGQlvovdKoi0X;Xd+>EV@?wH4A;0QvXBPeRXOVyG$=s zrFmJ#O#Aau^S=xI^5FhSwXJ4j%q`sguyNS2=JIv}2f$6{+#b}eu^$-s#b-7?*Fk%x z)I`+w-twa^JncR9`Hec;@lZ)3u3;6cgl=+ur3arKGIHnNhBBU-Y?A6&Y| zx9WRsYMj^xqGhnVc}}yn=_fdUBasIVx@CZqbTUrfH5N$LEe|ETh;PjsiltTP$RGTt z1Ecm0W9-oZPF@k+#COa4Mz!8L!eRNa^%cF?p0H!3n2=1PNl3zYLMlxXIXE0%h64a2 zt;+$I1Ry2e8O|x~Tbv`V+j>1l%}mSum=lbFME&mtYAl!NdE?5mFlpScc6?@c* z1}y06K*2L)KaUe(_sMnYbT}5MQ0T0iG%K$?Px>piYZ}yh*I?tN01ThtKQejp4a9G^ z^CmJsKO!>G=NQ?&cB)&KjHl<%}{2S@)trM-OF3RJ*Zro*Uf zKdL%l2e?+iv9iFK7RU-|l16EE2cx$d7};cz$z5-7-^nRKpJH^aj~NWder5DfVNg7VKr8D4pq z-r~W3e*k|sW2C@GCZWQFN9GJ8lCO$@tG5LAC<3C|00gY5yV2O29K7w@d-wGwHPK+e zr~RrF;i>i*e8nZoKY+Io1o!CR;VEAXFw(asBJ6tlUH4?O`h2~?$KTq4DmG>rW#3o( zItRqs2K=W@v*U!R#Dxe;(UbZw?-yDvkQdg%l4W8q(&|RQi_!vqp1BAeK-z2;qb7jN zCQ4wD{3WL~1NMe%yuyRWo105qPb2kiSWlh2!ANuU^+Of?#>h<1e$x0E(_^1(bC;i( z=?I&q%mumS_Gs`BdQ)$-96Y4eWv8xeXN={+aL_X4mYki!B^lgg^Y`8kl@=VJ=%$V4 zwL)k*MrnPhIwTD-0YEDhMGI|+n#!iTlI5^m8c}3X!f=3E_E+{Z!z?%dOo0P9cEK?n z1Bz)N?bGNKRmht6rg~6?kt$;IBjD3a2b?_%&bP9D;(q?lD@qLW!l$;kW(+GCwb`&} z-_sPJyAC&4bavKlxE*)kP6rSUmcTA#LEaR2TuspuZPGv6qrClR zHEZAQ`f@O5b1`@G1fTIOzY84p0I-Cpilw+A>T6L3kilhDZuT19RHkm;+ei3>)Y9t- zo}mB`Hj)(Db3dVyi6JiJguXBlW_9S&VeiOlwX<<1o{T9X#U)2dP1#jykW{H@tKC{!EuGxjwt<24 zXgl!O5C9+}7U{r32wR(?-bF5Tg_}J%P#bK{QZUm z3nvSIi(s%wxg(D{_Qc-U2f(ubalmmbu)GhfWL90Q0<4lx98cbz+yGWJz}m)IVeRRn zv0ewPYk^I|S;Cp}?3;54*eby`9wI^%m&}tt0MUaqb8!Le!eRFXh@Hg>u^Plt!Hq1oo``T3~-4IPe@C0C4yP91{@% zW$uk7jg=IzB7i+0n*sr$mHmJMl>MBlYwO zO6D(zJ(eP~fq17|dHTuHvME~clgCttLl319Lc#qiUMEX#Nah01km~CsfbB#aq~;MS z7I}kYFt>(6;sCV8k0}f4sC=X{Tz7S^*jTHyTL!)BBPl~5q#B9Q8HSWtCEgPSyj zLMoTOGp_qCrTcE$om3l$qsVbg&IpKxfMvK`r5z*R@_p9qbLRwQm); z_1=AU<*|9W8@%^yA2-;K3Sv7qeb&6(P4nbGL>Mfy$&{9#P{N3~$U;=EwBW0#DJI=en1nxB z7DPd+pV_45nVrO2?N52ucd~Grk<)?|L`a?+ZS1bhvPK@r_JxwvMmdmNuNqxyM>Z@} zq;WQZv(2YcdXrMbNe7Xy?AbtqbTe;4E7&))&os%`$&eNst3pSH=g2}&MmOzClW`-Z z99J00gu|`3XQA@wh5uUg-kN z)OPwr=qVah&X-Aob4Q=BL#xPnYsa@{IZpHEZRjj|(U2gyadJ z5`Oa~A804bT@0eDk>}0^NzQkfOI_#^hd6epW7n-JH%w+-gw1<3S2sJb*J2s7g9Q+0 z88}c~$BM7(0SPIAG(`xM{}+3HjhS=?sCThvQX+7B*n8q;{AN z{j9az`K-)S34b>hh)IJRYOcuq)==Fr$>kPK_msfG3pluRy!MkgrWEvR!wu?#vy8Yi z+4GKja0D0FT9q&X(AQd)zurwr zBo7E_%Opwl7^(6zWFf-ew>k>4%2) z+vUF^?e+p!(YgOM#CYnN1pGIY7Ba_JL$I-Wce|*)xxtwP&j_DRaplv(d*bE;9AF~< z*a6O_VrAc5%+fBy@rzRyq^@5XwZtXm0JH%xc59^+h~pJ0fmF~KU@<5Vp(=ox$7E|k zv}etD5UGPlV+R*pM_z@XxsMLy&m~4w)a;tLEg|DYd`zWrc0#3gc8*GNc1p#}mQ<={ zmr@xwy9{oxG*vPMJG&BYgO?&^(YbM-n7&ez%2TND6UZu5iEUN1$aQ491Sk!j6n#0U zMicNdaud}>uP_3*Ub05*Wl?!vTU%s?cL94zfNoNq(u1Nz&kaaJ6YI`!_c#oJurzfH zEHRCc)%jVDgme=lFPZ}F75+qJ1;RN5hp|@Gp`PTdoax|t4Vk|cruut(Zz7-@!Md0i zNS`W{K4?_(LynR7Kcib?=Bx#2>79JtL?VPsG&kJO4t%jm zk7Ol7(qY*M=ma0ZDa18`Kz;>S<nW~pkr9IY8(koCUr&Q)<$+{*ZqT6C9(V3n}J@mClTjpUgkL=Y6i4cDyg{PB@ zAZgJK>t(?!s`Cxv7NFI~lBu!O8yrD;*m)6fL2HW$uTmtBDKz#R&erq2{H4OIubb$G zy>cDGf>oS!AL8jM#qG9#AY%=Bk#I%AR(&R&Ab%@qgHDc;7AB3TP?5Dp?{5#w!_? zc^To95T>QUYmIsGzsEOQ3%NRwAFk^@@rr(~Es5WyiZ$B$8^zo+pp;v$Sc;4s}b+TR0cW z5uI{{3q7<)X+FvWXPiFCFbuPJSdstSN#l*J8F62Qbx&A|BBU10MgXdU;a_PMP?01B zQg}ZCPVqXHGyiMqzWLGHkM|KL`4@tg@pnA8QaOMytk(e@(@CABFp8lWT6<^y$HnlC zk4RhIA$h;*-#+$fCP0S*rhpGXweiXQM<*-%d>kVb01D@HxWJ_Eq$P}MpGfF`BH*ID zik|;TJO#jO-S%_fFB3lMSIrUO-xPqFr(63$6R|>~J_po&*FNs}H!Hw=4Hf{;00`zv zfCvCoIH&>uOE(g%!3`Ze4^X-jJV#g&G+B*J>}L|B01zL66p3dUgrlUufJeltLQY1) zDb+Y5eKwns0s+K`LLrsJq5#AK0u)tgdX$jKm>C9uH;_OI74a&iV1T&-9U;@2%YuL+ z1D^71h@sLsGfB|IHaqF#B#B*3x3w|7t86v`GNxIn?L^h7tjN|$8r#4_q}zQ}BA_y< z6T$BoeWOwh>S}||`yoK+Bq_Cddoj7Og1cg}t#3bQ2v><*5>@C}a&Bp#gcfK^j&U9THOIt|-*HYU$Pu)ZRFn=whQ$ zLx2hm2o7-v$#n$D{ee&7NWrVQq>baR&C z_~m$M$^}~(EF4DSuA3)tfYV0iHCQeN`HEtz!;%OPd7#xnc~(?_4*-p#1IRpMPKRxq zBfBR?GD}#xqpbj_N>o@$P1B`>&D!)FXktkoRZ<{SCqTAajE*SEnCEr&GDW;hX&a|J zDME@uo_YXKy4grF?FG-iQCXTMiGxd4FxfJz!189HBaCddCGjj&AVd$kU+XG>%?u{8 z#`C3?l_}{AZ{+3$(?Jx79E{>*)a=N90^l-kToGk_TCuPa=}r|}L_DMEEyYtlYO)CO zsZ6NRxqU3!CV%(HXD{wOcoE4GaO65*C6O&B&@3v2TpB@p8X&Atd^$t zpqOKJ%sTDPgWmc|vBVnyH$ce08E@X>++nva{&Ld3YbKz+DGBj?_eu9qOXs+EDW*4i zawP%B_$T4KCgt!v^);v?BqBULG^uS2)9Ey2!*p+qJ#)Th0-+_uK6D*>-~Iy2K08?b z&VZ0*iXhSd{b;ExCOw{vmx7v#80|V=9B&TgywZ3?8Vr(n=*-J5-CUE~qN6~$8$tpN z`sc8vMAWZjAbCr;IGh5tpj%8p1A>w5!f~9X>}Dw@GDIsky!G6EJU&MVa>bTB+rWM@ z+8IevWoMZ0?l~iNv2l^D7L5WPcL%e2ed6f6H=_N}{?XvcseV^9`gwD4tNGZW?&Bph z=>3Ah3%iiY>Fvc)-7~%BA)IfF6W$2r*GBGq{g(2Gy6a%HQI2rz)=MVlOPiK z!hJtp^r=5x>OjP+jC%vBYgUQosu}OX?gghGJ9o=He-%Y^kJGs)um5w#}{|!ZM$t^ZCT#>9!4;p1P(9v41`I=X=oNC`-8)~-3 zR47P61zL@jv0?sx{F{sxH{yrS>s)61u}nBymdPB4n|X^Expt=)NiEINCauxnPGTJ; z;DQexIFLZhJH>aYY3%>MzyKh5vlEhwje~R$uk@S?orsNKMlb7J#fkNuC(5%}pvw?h@DIa3_LxUJzC;1o z06z!?vSA@iz-7o!%z-V@C-Ct~fc3uaMJ1^-h62qBOPqruDG=ZGIB{m#R{{Wl2n+|n zKp3zBbpQ<1!#klVEW=--Y;-0TL`n2@012Q01b_n6fCC-^QJ@^40SagZa^PLa73d2K zqet?2mlt)0tUbdq{)>a z0>A++paL{NF3<=Dz%ZBvBTyJ#4@QmML-Z3$zl0;uo7%D zc*c@gYqL_elsgsIaE;LR_#j{|Km(`%1z-SfAPay10DK2Fz$Vy(qVPs|5Bvi`p#mi9 z67rnEZyvm5Y3o|YCT+srT=vib%I}M(5`N;3aAejK+@uBO+w(0MO{S9>Wfwd^7>F$C zs9k91+L`W5#pvM%hJkL35-6g?!NNnr3ra_6C~aUYpJa#(fpIWFWt~!-*89Xhh5tc* zR$vtPMRXuFurF{a@UGgfj;iOHtL}@$MvenZU@MRh6aodn{~(;hoemOu68cM=ta4B34+hH%<0-r=?BRGVHiqWx{5MNd&eO04PNUjC_Rbjw#505>j zgHkgTmbuCq4OOW_bz0YfCTJSipl zCnbv&7V{)<=6ikqEjX()5rY*Wco2rULPTKL2gjO+#{i@z0btVLqZyaxyDciK^Hk|L z58YJWnbvm(n1m7yzmg4FZi^!w?W;23WeZx~bQ8X1{jpHT=FCxck9yV3UYFjp9(KFW z-R@_%`j$0wN}g1(bGRK2=h1zk`VuJ_B}%H0N+m`@C6g@3Z!#f;Qno4rMWPs$6G}vB zS0a^WC0HpaDkW9PD$zxZ_0v$30c3>VG`3p<1 zpL*7H;{Lx^Q%AyZIsbK4O<$MEvJ%t=htnBO^1YnEKm{P zKd4Ueaa*oZBE_<;WbsUiU@xWDq;)8()~MMuvQ8W#0uyo|YjEP@XA?WZalVhQSyCqI zmj;$u4njIyjMUOf&x;68p%hHveC9t#l0+vKn$BF#F!?NFCzwErxtK+c2BsW8rOY!5 zwC(rOm0#e_SQJ?pRYhHau_To6VH)aLTUt@>Vl*vfBBfhPa}wuDEk#+99_%OM>is3O zL?10$@877r!&R?bDt}cf-;HTfo@rL@R4YIATJ7JjTqqd~1FDoKQz&MZ>FM$GdpbO0 z!q;coXIabJC(V^8-U5GMK#3XcdGTB}#EDHf@h;pZF0xi$N~%u^b8qAg{j3jlgc?I_;qhPS@3FsBI+K`Bc))ey6(R=Q0{$ZgQ@ZGb4^J@0dhCQ5 z1tJealns>N0KoZ|A~7P6)v2}<9+y4vNAkfux*usC z!_jc$Q93G(ilY7~6NQNBg7g=x!gu|?YG}RQT}^n`$E)e8yXgY`({`DUYyP{};Ta~X~Iw+w7=AQk$It@E&gS*BJamcXl#9oo*Qm63S z?p|`gtDSX~#yj9D6OOuI%h>&GyMc4&EZKCoOLoifp|i|B>%6l~BS?Lu`&iXv|FDr9 zBRFYelC|f$Bc4%j+?({qyjP~ueKU!PPJCMj4P0Zb;X_y3KW^KFC732c2qJ)w`<$HF zi)Ef^+YAoO9-cHL+FFzUskB3LG)M7_ALWoIbA&^k++mM=m{%prS*IN$k3&VJBTs3A z$@Sm~KvEtgP}mg1#iOga4L90|HOf!gu58c!-&c@w$k*q)$8=ei( z%T|ZulP?dxV`>#g61WTo}P-UdbO;c`_+~ zn~9g)a%f0?=ZZ6v+lK#4XIwNbB(Ugu8C+&vOa&zY2}bz-mF-Ex!4M$ac53CksBSTB zy7cMYidE5k&C;K1j%?ZV`AZY_B~#L{58f@5lJCyW*6bAnCZ5t2Basy4x6I3e>`Imx z1QCtEA`#+hc7$GE+LnX*3RbGdn(LsI+DcGQH599D_*q-7Lm*m3hiDR=qir;cM$v!V zUCGeOb6?!D>8n%Ta$p?elY8+O|9j}Jh10g2bV+lEgR4Z9=Wf5u(YM8YX{#x2@$b|` zjpU%s{+Cwze{xbC0j)93Rn+g_Xv=adO&+n-@C(T5aoRZ_+4 zru&oWw`NXLJUcBiB9Ta)T*MTNN>*S@1xt~~h(#!)2nJ0SB7%0YHlYepl%Ntm6eEW( z3>ud!!=7Zw-BulPmvbC8;vVz%HPSF(7i5nv0U^KDxI}fCV0f|8n986xVzZ257gkes zAH{@6C$+@BBp-^9RW_ML=~=hhzb((EFS}f{i$dc8eozF8LnKrNl|a>y1*(7qC;$PF z0n#Bq6of!%9B&V)FbQ|TACMK|M&3c2XalWc1PqTcF&`dwf3CVn-b75f@npi95@{TO z0!?WWL|w8pp#sOy14GeBo7XvIs4dCOOk7<*LI%0Kr3j{qG?bwVv4~_TfMmF&K^_`6 zX2c0o>s{DF*O(Ja8psQYDijx7EILp$)A6;#S#-Ahsin=m+0C|}z3uBt-3D|SHl#1( z(cl*q#2^otA|-JygM^Z^=B^{k7IvHKblixZUAP<-S}G9ZN=)p;T~Hw);~(4!5egDu zEbw5{NDcbnn6(~jwFc|3ww9#TS-Z7Zvt3(REiE*+Yt$*qCH3mhkE`yjJW9JN`MPZI zNqgN~&-j|2bP9>MZqY-mT+0 zxs#oDr+eU`_OLB&ajng5{;u=;{7InE!w^&a)9Wha7FaL@GQ=Q|JX?{%y=fj>v_t2t zxQZPVOB`#__&nlsX8G{T(D7$GA|@gT=~HLl8|1h zmWX&HDpm^QJ#|u!d!Gy3Bc1F9jC4K53U)Q>G3iAPlB{Ng3Hs@^Q5*DJSkC2N!IjzT z%f0N&clEC2?)268s(lr{DqndjR^B`DUXU@m7S1gU+H2Ibq_~g~36u6XEFvcCk2*Hn zxi_k|C@C!`7E zV93aaqq4LUhxtR9Wvqh%^;?QZs-$|OKsJMXwL+|%jM_9RvWs%{sn;UiYjs?hcAy<&?0bCMz4Fk|~fsH-v-DY3JKWQ$>9GIf+LAbF5}G zA4ZB&nk@2h4jaOc-*h zmpE_BV`ZCRKKHPgdwH-UJkWJ4XBqdi!iaGIkw0X?F$W%b&}ELg1hYWs?!X@&xI`jfXV z!h#+#d}e?0wnGBkH4uRVh{9(XfS|mrb3VxMfd>_|@InJTRZxIs9$A5IM63Wb&bC&yK+rqZ_aZ4;f04+9{zt={jmHI ztDzU|p$oXgE!JG%DrbQ;dUsCLgQODBjuS=0a%k6WWNKsFN$CF*Izogf9~CG#%<3F7q|vj^!eXaygIJ%=l838k=K+9 zJ^y+2ypXfm_SxaXKeSV2mz+BL{~o)ko4(ijD)AK+100MJl8?wIH3fA=Ww{NmDMvEJ zN{-V?6*W}y6jq-l9pveGiKof2#|-T>(ZG~qvsS{{%4Xh5C6ko5F(0y(FG7+!2>_cz)(aM|DYAWB95Q*4(8jr)eC1&LH=q`OHEd2PqUks z|MPphTGd>&rW+|=xhM9%{MyzWOKAkEo*N=b9j=BLUSI0j1$dp)N-3#=YASo{&b4#* z1jia<3iH)S`|={NZhf}i+IaK)-C2J|&&IPl%sy1LH`mz3di;P8I(vCN3Iwq6(gIQ6 z{6i)mwA|LsImhs+7M^R~S=YJb{`Pf5@@=kFgEnyn7cVv(JYg;Go4IMe4##!+F3{+c zFc3f$d>BU1;j{u;FsUd797k4}Fd>3dmUG8Ze}d|SxZ()Y1Y7D4|02bV66a3*JSmZ^VZ{63Qt=?E~ z2GS1zRPr6Kj!ystK*5eP7G*Vm_r^#ebdm?D-A=Vhp^-jXLvnMC~!c~!bgY^9yY9)kzzrD3iS5se$8S! z95Qbwi67a@1Y`J*7KYh{<5UsiAeU+?7UGe<+(V_WpGZ{7WuX6wOPYjbU8_YUEZe)4 z-|2ng7r~0+iCE%@pkOH~ptT)I`x!to1ROy#xt5z;uv%M)J~v-E-omF9no6xi^oMuo9-Fi+AR@W%rTr};s@M<1f;kt z!Y7e?^rD*WE!xjoC|9ab(ZVIz7zlBQ1{(?D?&?T^>#IH~%#K3E5S?}d#DFLe0iwbH zxDd9G=cjbWq1XfuY4PNUr@7`;ZfQ89AHfca&q&wNE1mF&~nSXk1$Gp9#}?dohl<(hQl<9?nQ4 zq(drHg|cu;`oS?}qXQf5ot@Uzg&XhBjrO;qjdZB)>IxMTu!NFLA#hSbGnDoap}Ad; zs#dI0(D6BVa+9AQM%nLIduJ|r?F*HbU;Cy+B_c1;?^0euhC;wV^(~zUHzluhM0mcC zgSgAL*}Z$Hp_<>fWoHk#awRILoTxup4)x-_I4{XFYC5yfVBkq{Gt%UlXFTN~;5*jr zdfsf$xNMdMb*$P>!hVbmScrkgO3W1|FE@QRS^?kaH>76FNK2C8XVc|l-EN=RBM)`E zdnQl&LkO?m`H{zg8_xLgQxZ9IUA$G(#p$nBQQLL%P`&-2ZykhnC zH!A}J|u)p=ye`v%rIg`iBV=m&9|gwX;GR}7!)dn zMDZe*s}|Phto7Cg>p3cvuBNlt=ayY;=}j&FKK8OOM~)Js9AXg*9Oge3?YFna1QAmu zZy1OzC|b9I;Ae5TxKX3TiJeOR=^<^hM$=0o z>NL!ouGfvbe|PPg7Fo;c*SZI9{o)(nCTwX<@0j|PyzXi>E|F=oMP6D~QBFd}Cj_ii zQ;X87X3LCjc z^jWGXr*a!tJa3z})+3CY!|!~!YnY(K`RUoo(Dk!AnKXxwGQ@bsGT6Sp39XiD zne^5sd?m(mYM#)0n9#s-ncGl}D%A2qFEW-=gJv#%Wk&* z;X)YBrVQ$jdXT`ZglSX$;LCGSgW6P~dh+-anUfXaRr$M;J7G~dM-3zR!6r6xvG@M4 zkMJ1eRTSY%cEpJgzR*p*tJoDD5gRlHUfLO}u^xZne@fo)d`un^%AhhTp(2WsBO~zm zVp~{LqG%DK#Cm66gB%p&&ukem)d7O$8NIhsud~Ytd{b!$aXkPv93xGWPnh(AtTFfH zt(@?Z#+72@Pv-yL{CioA7ItLW(j^Bj{^L;l8>*b}6rZgg@?oF&gFlf6ZAYE%Voy`a zqx#UC(s}KW=LZ+BL-2mK$X)$M96erF+lntxp9U0HkPURwLwB0HZcJk`)12t6AAAW* zJI)?B*3cNCvOZ+YsfZ<|!6Cj;XL1G__4+Le5&XjjOk($@#nMV>WuW0EI;p`%oa~fF z9(?qrwNBQ&r-7IFGtXwPm_y__eY6!!VEEXwMw+-ME<{RXq@*-CX~~%s-prl{eBY)2 zluC^c81FMaX8xjnn%lkx8DP8pd=9>ipCBE;OBUK?*9cP}C#p8TF4ItQ9S4 zx2CmRjjJ^CrP*vcrnL9jSK&Mm4KBKBuNvChCoSCWz@tzav8X_dM8j|LDuE2A?HZ_#{Kgh>NKEZcaA0^rnz2swc z`-tm;P4Y`V$t$^)r&84OdYKODcy*(BVk)ESiS>T_2pkJ?pa3{f0$C7uOJ3byg%E_+ zgqyGg*I_XZMB50{ToO!0>Av(#YU0E{=L^~Pk~J}RX`(bxBGlhl-)oMxt=Y@PLAfO! z*kksk{9g<|1r`L9@WI%m7@bONif=xpf|XnSt0x<42K8iH4jcWezAN~(1~gy|z7sE! z9xWtwQb{9;nloJ3DV2Q(~@$)vDPP+NU8HJi*&i_(7E6UyD40 zpPb~tN`DmHQP|L*Cv%H^9%A@>uO%dT$`Xoa&$y7;nQNVs5(sPnNcb$q!a=|o00YE< z7ULb&j`jiLU^rk2=s+^i1SY`*%s~EN8}w<@A{khwQ4zF%;c?I`-()B$KVSqr-~#M` zOK=Ew!4`r+keKc+k;>o`tMv6iVk^0OgS+8E`lu^gM^YvIa!0f?Rk>N&R9IQ7+@Vgp z%hwG*lPi4_2Y{$}9 zrRg%TOjTy7@>B!xb$Bhl7Ej^fv8y+l79=J>DMbIB^tb z<5*7QI1c9+&cX@8a&vEl*2r>N4$E!1EWhQk%!;963Zjq-s8lML$E#7kR{hA4e?(8h znNS%kQ|0Y?fG&~LeMQY$`7=aZa3fjP8n1cvNgX=*vI2X*|haubs zKa3oh;3b49WV#I1umxl%6;-fJ*kt06lo;ftJhmRYha@yynd$%2PU!@=5QGl_#5JIU z?ukJ}SX4OJ^x$!APC)v8H@bf898^xVh#<xE=Irr|5zHujq(3cxl{r=4Z~YwYl}<`qrM1#g>7aB|28^ysZ>5pa zLg~xq&f~4_^1SEVW;(@8$NQm-=dS%yQL>6;h!B;HW@V6!1Cj}aMe&(1l!9}+t(qn? zS5#d?byQt>%@tNrL)LIFUd8!cdsNv_h$Bi`Ix5EuL=c{TnEDjYfl9otGbyoHd$1%g zI6aY6V@`@dcj`#hxbFMsrf=gFFLJ{dzr;Bg-t}W{-O)~QjW>Lg z-v3^^|xh4POl>VL^NcWx76I~L+Jj$%utg>IRR z8#Up`ifxysojbH@&YF7~58UQkYFgY~#@=@>e_hvhz1MSnzcU@Copg|PhdIrOjasAH z;2Q0VPNElMr~_XH5nUL^6llPL_ky1S4+>hqr_c=kkQgHs$V*9gYMA<<-DrchXqYz2 zY&*8^aeE%D6VHjZ<==kosQlNzWopCNI+lzvWBQmoW{)Xj#uz_lj-_MPn7Vi|tjqbK z_k4F7!01Eo`pdum@taTl9ll7IwPHxM(NyiVmdi3Ne8|@U71#; zO=a(P;aZcdyV5Kle7&^?PP(ABG=ER)Teosb<@ImkJ#hMEhfhB1CI`>I@{~hYTr^4b zlp{{SJ>d7l4G|u#=1m&m(hl|AH}6b{@#!$_kWKzLW>G7f7Y3Yh*qqr&>sD>aVph5F zjqH(Cys5be5X_qJZplJY{MfSsgn2N`X!N)}Nz?X~wsj`(k_6oNUP8t6t^$ zx97UoxY8}akO4#yfe&7Mk!x~q~ZfQ695RXz$0KjzTL@UO(_*I z{w8*|wt`U;CIyx=i#{KE^fFPW7~5 zgDVaiGw>%z?Car!$BrMe?8xONP(n{M-|FVLs;i`apS}c9X3o~Ew!Z5+R_qBXvbtS1~Ljd1mUrArOA*YQ_^l4zRO?3iKA|i+!*U@1zCfo_aTC}(e4)_27n2H$W zxG-P>0`P!E3Q-_7Q4Pv8B+%z>bc+`0{ir|!euR++MG}A)S;#~wQYb_@;&2f_E<%X$ zshTOiG^yJxzv(x&H(cnNZuCCrOVP6O?S<}f8Ax9jQ{4O##pNT)qf7lsUP(^Hl9af# za_9GcV&rr6LqE@jw*JihvLATB5i%ts(v_BU#3U~Mfo$uOBN_i!BWN4Q*6DZ zBQs@^62o28(CEH$U82K`S&4>&EChioxZ|R$mV5)S94+ps$Zo@dtiahcQ-|jOv7$hP z85Y~I1&9Cf0DEyQ|KWaWaVyVr1(n#K2`)@_Sd+cCTa;3lI@GOnypIR@mK^2UG@mbh z$Zyf2(Y-$I56iZ~FSgb*KI(Vd=uThsYXqbVNmXN|&-q0%q&XWG=BK^x+q|SbJ292o zlq-9GaFtCH433RI$FE0F!^~DW#0@anHLJy#8xVc&i5f$rvT-x^(DR#9COU@6u0^iP zKSof_3dcgQlttwJta|}E#a%?{Bmkugy*VEJoX80viLwGrO#p_HYikj!-E;{bUqz#BrC z?GP|6Wh#%+F~!0XBNHM@6$yklwE$nPMJaN@MLx3er(KGrLj3Xx9}mhK+9X}#l9IP< zkxKbWR(z5x*-|B5`Nl(1FO;M-$afwVl{$G>yvDwWb_{i@@eZQ_6CFY`Ixy3981_ON z8D|BHSj8-cS;&04EN6<@OtOSE^fAhSHFoVG9(4C#;gwjft9(^%*nKF)mtQV=;d@S| z+>lI^$&_GWpuyc)zkvw@4*7?szOg@ObZO=+p&9zMO6|JA7St$fNv_oFlUv89WS>;Y z^2)6j@lmeK&VHtNs76Q&X&^HML3xk}f*~c82?ay_kO~q)GT0Y>4hW>EwoG_|4NrX_L^mJDGm_YQPA;-<{OT38=e}^2x-VLe9bsR(8MTFBtnu5 zASryRAbMa|Suk3+xtX0?TZ@JFwE8utzI$d7TI4ch$GGEXUDyXcY z&Xq_K2^G|#i}dbzr=M5k4G!|ulv8C7*YdFF8RTRx;iZ|B@+{;yyE(wK!)7s8GxFk4 zQ8I%gJTV7ym>ao*o4J&uERq7)8zSpoTi{6FJj!3qN=oMaXFJ}?(*=No=s@LUm+tz4HGVMa6TcA<({m0nx@&B zuP?Q|-mx3q>N+>M#kH<+;xa1+D_F)77G#?$4OFevZk?*knGDd&Aft@)bQbB;y96}| zZ>!B)$1;|;fsHI-Q>)t8(v~&HQr57zujMSy=K}rbhfbn^3QEakMZ+xZ``oFO(-C(9 zSj_;9*+Mh=EMpy+Ianh{Cn+o>CNfS^fkKG|ssYJ?QRTz*jzr}w9a2tF<~K9E1Hzo; ze3W5qTw=^ZN1UygVZMxtRa>l7vSyiDrEAwHUcHc~Yt-N~Tz7~a9B`a%>|r~5*I zIT4Jih$zaWXAjvHPUKWMDF>95L(;00q6*aV5%$=Xk?HHczN&YPY0n_h%&8_}fCiRy z^MjZcd=5hZ4Oz~wQS1F6AoauHu6wNRaJuG!TYO-41Pqxl>{n3>jk?iO16!;y!VrC& zWr|_uDb|^#X-zm!m-CO0i2*W<7Y8m7!ekgB#YQ3ktiw$kLvZZfcGKIN#dNWhO{}5A z3YN2orslh?DJGvH3b-yK0hH``4_eWqwoTj(lF5<;lEjD@nhJcfdsu?mp)}7C@GDEw z8(*SBM^B9hClWMH>;#f3O4w9+O5}7!r@})5;^%p`ij-8k+sQ6&*IpfC!f{3)ZITIA zi4|g_uo8T*2qO@LYDhu@3@nBRRp5XYn2+@^7cN%Bf5d2*hc%E40_o5I%OM5oARm@u z5Ef!S6hR5h!E~63LtvSjn1N}SirJWgyLCGMq&KO&mq@+~6i0kX1ERQ#%KF_58VWa2Ahr zvcnHO=)m#Mu9KdXr+V1jI<6V=oV`SgwOj?&s#;66N{#yFYnEDIwuu%RHo;Xs=0*dy zQfi<$D|LYE-QGqUDb;YXLcOM{f}Q79b~)t=9WJrQWiE1oEpBj)9eqkn`zRx?am+Q? zxvRFzR%2(c)-IV~#MRkVzK<~a=;tC8RJm(gjv-1D9vW#U@OaEFBtj2fvS_~1_;lATy%!A zpvauHiNiqv$>qT;MOcn7q!2})YsqEoa6eVGWf!;%LMex`p9YQ{HFB91xQtg2jr%Cb z2=Iu|s8tQvxwS?Gym4%kc_-FeWBU;#n@XcLxj!vxv#tJlAhf^gcQ2FSP0CT3N?rKwa^Kt^&e^LrK|Jfs0=wNWA$?G^0j zAp0auAcL9_6I9~ktz}#WxZi^Mw8k~66;^D8zPGPGU!ALV_k8fNorDjxJ8cZRGdi7g z(15e7==M=Y>Z^ZWwwa^d+U6n!B6jEHHzC?YKdaU5A2cod+M`!uCK7lu#emgk7O3n7WBn@CpQ%Lqic@NGF@&K<8cGZsTZaV~h#L8D!Ph z)G$voU1Ycq)F~!A-^PZ|FuT~paZYl96BKXHhg{9~>}D047_pv9%*%Ra*g@lzjoFQZ z2Th6YZ!6!Vw>;bhH?sI1?f4W3EC_%L30TmB2_r(VBa8&bpg}i=ki(ZKU@iDC0S1VE zdp5C2bDs`*%T3$wkAva8|s^*WhYw1xKOxAwKZ`|@9T*L_-%SuNJg zFYT`0Cb=^g^QHDCvb&Cw4K~p!%@v@W5KgFR3@OcFrv?|(6j5KdIeVbO@_CIDoZ+|< zq99MEtUO8+x@BMNLPMeI`YJ=;2!=SZ>Y_ZB3 zix_c$2qOqTqENyMr}uY24zFCcncmm~yp9|iTsQzQVIYGA3GqE55;QdHH*CNeA5PCB z<~x(9PQCK9!|7AQ2N0m7V`Bt^YDNg(v1Vrr)WjAn7_s8Uh4W@a?&{tlhA@I?!kDQM z%213()S%N*EGri!?m6n8ESDI`gP> zBFWqwtzv`QWIQ2iODTnQRscp4-|Y7--pu&8cUE&^iLZH0%i&cFsz{T5xpIA~l&R5XM4rr;F4ZLC7F{k8`QN)Z?uZl4;t-BtuX8wz zgMP!=tm?#q^7^;>wXH*1r?uMJG3{ctt)X4qR?7ctDi6ESgMH1gLHcW8)-4L*rUPkh+H=8|PDd#OvgZFl4rZstNacafXA>33g( z7&uVmQAPG@pPbOUixtH~o^dR+^B+EPtT$=OjNt0~ede^`6E5vLw?(V1z1AA@&C|lU zPBRUBwd5zhd5d$idN@ZBg=A1l4q3Odhl<wmJ_0m%%Z>G_D%q4|B1L*>#Yz zzV1{1@U*O#V4P3QG-6DJQlMU|@B1#rpcyO`(*gg(%KQx++KB~dg|z0d5HP1i9$U#J zP~o-`5}i37VBsoMg7`s%3DFJVf21H8Z}FLS-U)A#%Z5B6*F7!=9Cyi456fP;*<(B0 zHa;y^qI$_P)xf{+B_?T!>W!pfppl4xfe<1FOfM*55=4nl7(I0HC6{(zJ-U^Yn;)N6 zCaIJ8H-BG%yLEw$%8k%N3;gCepjKSsGsOf$Fq<{$Iwz8a)u@l2J5GkL<9)k_3%P`& zk|h~3=1$Ab_FZm5GEQOG`D`0JnQ&o~3>J{P^@5yhP`HtP+HCX9HC?Qwg2kI*t$l+< znGWp$!bX%J;m3}0iGh5hQR$S<@|@X<{wZb{u;j3%7G3M`h36fEK6)uqAWIM`aRxZb zulM-*+q~Uu#*h&U2K9^E`rRH&W}za);lwOLZn|M0Oopz|5Drf(WW}Fg2!WwK)P?3S z6#7CGveOH#!C_aJOsv3itii+V-Y;tvubag3a~!t!U5Miq}$8 z3!SL9s%HCF+xlOf71>CViDV+)^aczPAj#x2mQqHO;trOWW|0M*(!)Xu(4p! z@Q9<1JNgvI9Bq_&WTy7ZRV=*nm920kE4oA1xXLvzd!Zpfd zBs3UM6i9R${D6rFjYSDgSVY`sCSIy&foR?NMT$q|4<-u6VxrYZaEI@+C(e5AP)Q>_ zbWle#^|aGU7ghYa4Mv;DYM@wV>y)xtkFyc146j(c04_QM38R^pSYZS|c}h3XU`~0wh~KXwjkjvclh*v+wt2Em-%09QJHu z`L(kRI^QK=?}&5VLsHaxDUqYXCYDiU0c9Su2PVy&HfQz>_cv_RI-F^C=TKhU}v}u;X0GOle=+2D z{gID&gsV>(5YFDCapTuYCoIL1@oF4a?g%uneBR+ z<&;NmVRFczLw044XzsB`5ov*WViz%0(d zRAqL#$H%8!F;DZ}ed&wK*00sd)s%)byFylVE!6;o<)_X}=@G`u1QajXC?9Wgb(SM# znN1_}X=jL9Mw#Y8dKh4mm_~+~D#WCf4l1apn||gn!d$9IrU>Yvi9tY0A)T?Aryr)a zmJa5=dUP9vvG6dU!OGdj(I4lnK}U3ol&FHR9FM)oN1pZ#8YH1i4ozNqA`(eww8%k; zC7gIi>#0_6-R)usVLyFG15g7BL^WkA**a_SYb&~#8Acgql4-_y4}2=8voSZb8f&b= z5u~O7$K5PsDJ(G-aDmJgv?kMGc$vFLc5>4j-S}>l5{XI?E;nh^xMRK;5%>{A024+W zWZi5`Wh$~MCB3uR>eVV&tXicPYNkf)te$J!tf;VQNhuPO>_6B=;6gxW6M9jC02Gi$ z9yxr73{t4)iKQj7+>ET0jJ{TBzRFai4%Mk$ttyvMttxpyn;XYn zz^ijT$7S~E)T3KHYuUp1nuBJ0_9$61OY&0}L&@)OszbyMS}j1ic` zK}CxWiY;+G!$nj9QcWopnH7+9^e8dY%KQ{liW(+%GO;3sP9bjc^{RI(BN$E~E}VSg z0rr4{!GsYC0SN*NELI#~tLmOg-N-zP9HA^(Koyio=~P8N3eaiFrDQ6gAmtF#DN+j2 zAxfecrBR%+D38i1lSi~xrbNEcE>0ye-pQ01--;6Cp%7UpKnAigWx|+Cp2;ivVAnKH z4-|ri1g8rK1~COX2&D%PlL4x!;s^<%Okj}p=)f`?Fv1Xec`G!q%807MGgh4`BrG~} zmTQ$)pIqy}%45baJ8tZTX6PQDUca4uf8oq$J=0lM9y7|a&m&*=7qgHTcJ{RM5B7|+ z&%Fz7_=X&Ay3tzesK0z84OVQvLTwHIf1DNi{W$7W@?5h^l`?U%gdx#D&1w~^^nt1i zn>X%8<7d?@u79xsbg2bPDv3lxAhF}JQk2a@ zm|!6>V+E%HXCpNj69C&`At$jvcc@MMB z_#RZD(!CQr)|m35YjAb0QJ-j8OPE3g1{@j;iX?zE1PNfC4wC6dK}RLcm$1mFi1_>f zjtvcg0*V0XW2Ncl`}~;Cl;X?P@r~p>IjZGzMa%H1!X>Nc!?#5N4K%l@8-C@?l{(`e zjK4Gg?`Gp32l&@h|3sdVXZZQWcXgKefv>2u!H;vzoIHER)OosS<>luryNWKIzN*mJ zi`4PB-W&6nCNzti+1%zgyCHNLOFDG8+};1>4>#4IhBcr*{r>Wle=R93(aZmcPZ33Z zJKq1DwA*Dhgd009G>FbBfbeLoD=3cS6Mg~L*JR_sC5de#O$W!m-7(n+y|ix6)4{7 zRuyFt1}XwsEI1@`h`azmB1mKjy@d2Db6Q0JML@d0!FJbiq`Jr)VSW9}1al;VjMGmr z#V`%47-O1AW>}(_Wk1V|Dj_JTuaeVJZ~sY~H)hnV!>7QCZA1h zRCkMxHqXmyoROTEU6h`do0>Uy%A{!nlgA~z22gR}LgrHZ6AG?3-S~s=4pnvATO(H2 zS+De-j^f8Xwo#6E4C5Tt*he#gN$0}-ZsnDHi1U5|E4evGlgMN*uaT9wG}*`o*0YK- zuLXgu1R>9KXN#LqFfDDW2iV3Q4w@7&?>I+|I>J`Y`8OwA&05a5mb0$WtJkj|a&vC0 zp^H!5$b{LZxHkiJ#BYQCATm$reu zmN-x5Ucb_+mb9W}{nhs6+UrL%?)1|4PIs-vOIX^nB#2c;SuyO*i`a+X2}w@x(5`m$ z)_wCZ*1R!b&VOCVub!buW8Ik42j~2=Z4-qS51SoO;%v0@WTCHz zU<?ujt)CCYP!v%_2WUDm|cfZVcb$r4#m)+JLRQ=anH z3f^9N1F#??w9bOcOg=*81cp80BjO1`5j=PxXbyq{Umd%Lf4uD{2HSmGg}f`8%wG5! zTF$=f{AD9i1v_&iAhM@IegZC*7bh$4e>83c6o$fEA6|NK_47RvlAf5nT;xN?0hBCL z^Y9+5VLi9<0Ip{H*?(C4DrZyvIXq+S=v?c1wmP#WRjZ|;uF2(K-+o(Q7tZ|Q$=$Qw z$+z(L7XI&JbkzU6RSb(z9`HU|IU!+5e=I zZ}NO-^AF7iY}j${OJ^g)rr0fiq{k$g^8d7fcj^EgE%}3PRHNp!Y7*FSz? z{zv!ELZ8FSKib!+-yqv!;;!>vw&ySY3pDhi#A{O zCcpW?TW;pex>wsCIHYeJ-yweR93ST#@3nj7=YM-fQ}LTOH_9O@Jou(O@Zrz@$m4(2 zn=k+LL3;NcUwK}3^Y0nZqY0jm-hL2M|F4~2dkXZq^TlDmfe>P7M6+S@c0b(szmgkE z4vwC|h)-W8h-2C4x|6&isAnZ=?0LK^9|r4LY#M+Vf2=jvpbJN!KW( zWrKsDoG@vg#o~O%#VGuZPOboF8*m+EHrYy}(b8#GXnEjp2@|kv5u=aw!co@b6f1*a zraw}5Fi-k3nQn%Lqlmee4YVC%ujX$AN;t8CSg4FW1K7x%;y`~VGT~M8#Fk(45+~{tt#zVA7m}c1M7vv*75kycON@7}bM!|^E`H8t*%$$oCiSCu zU84!{oQ0tzMXcw`wSuxt=|URRxd~URYZ+TXUngcbp3X8gaHAK3)D$JK6U-W{RyBmP zi1{at1p0sFnL3jYJw*>)UJRv<@;8A_fX{Eq^mw5vy;XmcwE_yc!nMb`dYUji`0~N# z^Ve~d>&Rc?Y`q!!&oS?<|0?6pt|;$*KJsi*_Z-MQb?V;R*L>Gnee9)kgO~0tzn*K? zpZ*?v`8alE*o^>vmfp+{ZZxt*GkfP0*QI=DV^fB=`q~thyYAB+=R$5Vx^x1mG%=qX zW_dNpw1$(0j{@KjYnkgBn_lAWXIB*WCm$P(|bv4&9U?YSOs*%8td`{8@WP@Ur)sA zJat>^8tUrR5YQ{2^ZF#vWi^st4Ep!_jvy^#)@9q5jt=s~ybv1a`*}nTxzVL|pST_0 zLat`uGpK-m|@tBCaGD58!u(K*;4I|p zB1GTKL@-C*KWs)f=QT$*PiV?-x^|b*GB2yWb!yh0ma#3x#hoR8zKJWbL;6fWFI*{mnB5fQ@US4$F~V0oswD(}k6~G|Tk*fZ z4@1pS(3d^H<(YZTW6mSRN}|iu=6KB>sV~(npy#3u_L==wQ`8dj&5mN#o2`R$wX+l& z=@sKPVTYGT>VI0gEyI8($Oh7SzzaZwgHAFy)o~{?(xPpM0mvu5TA@k$oq^X}r`6?r z$?mwjQlKmh6sSQ0o{Ko28UT8j?npedxd#QcB@!l?&P&$hf3Z#>Uo<@jO6k$e2?Slk zqFyqRR7hfy6u`N}Pm*y(oU?}4NRpDmoMGNs;2z4U1dM=TF4`>VA+NnU##^hqmTxPV zR1xkkVaaDwXoUV;r@+@n7>zEvtfJBxgqX}0p^fpr=9OieDRZeYQM}x)3JXlnIvrCT()n!{+t)Q5ai=wQ()aSxc_%iQ zJ{g6JW0YSvbs%q&`LSKmO+Z!!ZOAp`g6@o0D6q#PRB2NoUc^8D^gcsY&nj3{@{q|- z^T5uHSw{*5MD@5Fd%12%RcJ^>v~e+jZYZKIzf8A~=WgH<4gqJzVH_6cDoJ!hsC97E z5xr^ZtC*HwJXD?x?vWQNb`|l4SlaWF4QAOvkuO~2JRtydQ>i8R;GPB49o#PA9e7EF z$OkOFnEoZ>62#_da{tw%{K)fwMb}Kic55LU{xY0sZ^SDgmzsf=aQ)Te$hPdK97+~u zFot8(t4+B1#@R_IQBx+jOlhH2P1VNsL$om6Q6~hmCY@gRT_8=bDV|!;iTzME!X;xu zmR`kyqK}bBiLxfb@EPsGG+>pq>yJN7q#=Rah8&Em@(S~|4rl)fcC!DWmd#3JV?+g-W>0)a)ln8IEWU$kmjO#o^}?E;8E-WLQ&pG#hZ zt>6n*jE zS|mGPbP+@BsMh4NN>p&8y!S8H?$Th(vR&4s-hZ)_@MKD~=)|`U+U8P>U^LeJ9@>&s zRT%m2pLqS{8!FWry_=gF+$-D%8arCsT8s07ejDz-M(6YOea}h+tvWBd@AcXE!l-p- z^)OaY^^$5$?yU-6=zi*I;;tRZ10+|6ebjjSnpGML0HzsD@5sn4oTXUbSx zSkad$oR}}*DmF(jTs%ruw?BMJ6dshpV5UM8nA;a||M%3==TvR0LYv|huua(-f9`3V zQ8=AHw&h+xR)20E(vdWlYtUue%C5G%htF)#%3T%7x-kEb9zEVvwWV*CxNS(fv#gst zhqe`Nj7`kuxEn)!?tQlr3dVc0pW{&?Z*MdE0K~Rm@E#i|AIXpKcN5N0Gz{aCB*ntiBw*JPXauGe?tHhWac^M z?U78vGh&DMrvJ-0Z;1T&(glK=mE#UDN0x85!N ztF?(IZkw;JD#aT6>qXD~b8Ecp{JclT>GcwyNo?wH%Z$Ifj-Tal)dO{Lp zX>K_`75;Oy=fXEu4aUb8q6AJe)$Dz&zwKh?WuxNX_*{o*Kfd~ZeQ)MyeI@!1ySY6d z`5%=2%qRO_=KtK5WWPV{^ZhNi$sND)$1hZIkBu8Tqxt2W=iJy+pSq=g=gj8{Hvh9J z;!gPdrn-cOT+yyCaXU*p^!Wenl#VMwFF>G0MT??d5$1)Etcr|Z21mG)rC9FmkasG_ z?d~dr@rhfKl?yu)l`Ax`pi_xe;z0&MV3t=C(4+sy56PqLnoIq+UDbxWU2 za@7~r>ovsRee~Qf{wv+{UsF8#!_R_)`lbreGuh5>_g~L1Qd(#+Fks^P620o4C!E2@ zT=3HWUw@B&FM!?a^L}X0*!t7GVn8#kRNT>4-O;iC{F(nbZSkvk^+SH}9PC}CXJJ2L zEjE;v`v5Hm7+MT*;ce1o#chWF$}0kt^fELd`eQXXr05m1B<)UUHxyfKj$JeQm)p7+ ze!}TUFcJUdyMvP-M180Q^U`c4-&a=)kLxf7=(D|gXNY?K;8hoXm|85`-GdF!=>HEp zu|kd!RN}NMG-iZ2GGEjxiN8|C`MT}9p)4a105#Y@M?^OcLzdE#Q2tJG#1M(1Pr>>> zLHWJi5bxp$e-=~OfBQ$I&6I98sudTt0%oCgMh7vEvTE2uCC5^1aSD7=qO>co zu$+35=JDA;!upUx0Ju#bz23KyXq*x2Z6@cp59$2v2BrHTJIgA{(}qlL3|RzK@2REL zR(i&Hq$V4{uj-vVca?&&a-{3^!8Fl|(8r9*HsxSg!5LM;T&cKby)CFBcFN zyhT&3obv8zX4NcLNu9nfzwX^}?G)g?dYZp+X*RabX+4v)hXD(-SPlIT*I&hgYBgD4 zk)@Vf=@~D0(d**UquJv^*?3(~9t>-H>{#Fuunnn3HX~m-3sG(8SQLi70dHVO!~+RJ zf|E>m1i_;;(n=YzLOK6m3?ycn1Ct-uL?$+wcbYrQ=Nj{k{U%tGJ(F8_t?N_%-?Xvf z#qyqdV(d?SHZGQXq0~`uylPCrmQv7q&NBPZ^GKb=H#j=7OPW;((oCzE*ehwYdp(9S z*g~KHRt7r(``^yKBn#dNN=NvLVj~_R`h#uoV~a%9phK6QLsv#3(R(lpmgYoifHcO5 z*@!uSIf}WC+@lzV1%@R{-I1FkPhjtfN5RVAV>ndgKN+xupNa{)+-XNUn-@18D){Ka zNk;%1Y7YTMVf7>oz8UCCWTFIsDS7+If^?sfJwu_;xA)vlRoLI+HAYfF*7W#)6aRIW zP>)8Gq}ciVkkv{2O|b+)hM+iB8S9OWI_YcfMN^F%!muAC^gf1oi>Wi!rsu)u z7mqb=p^M$DX^NsU?-6_Ih3GaT7x}1{W&dnPt_&J3$H$-XkbiI8KCZ@V=BDn3UVYsk z?S1_#lGjiA!0Sf+Sj$zEvhtI=#gFmU@q2kDqgq{Mc0hbZugetd?xK6Sts)D$Y}MzH zc!AZ+a{jly?XZ^yCD!omg4rwBv6GAIqhAz1PL?((JWbGW_Y}_-Gncxu=<8xh@nTE~ z-kk8A9N6ukvcu0e1xRBFIjWsM=Q#hqZoeM6#f$x_f5k8C_ixPKU~}0s@5UQ9e>3Me zb^SX};xgHA3mOZ+vx3Rs=yj0DPjwk!5m=_{(p@THgl2})^hiCgINgF)9Tn~H<;Si% z9(O&}z-MfEr}ws>y|WjA=`JOH7>ofDs>xjrWgnM_uG9FYnJnMSiF}{R?2H7K~?f7ilWt!!}R6=Pq z2o&j>;+o|OcY&ZH7sh3%3$NP%-C8$*HK4CeB;DYQ$nJd&r%WfOn#^fEUkW0gX&#`5 z>S1NdJ!j3ijd4xmn|zv{+#TLFzSFjOE>PA2ZT%-Gv~;%oxoidK*1fHrt^e(2AD12v zpF1yO*XdmguC7pFS^;Tjj{KOyEqpJTAhhDGL@qwTv|fS49R*Bvc>zapcY|lp`|)ri z-+O%zDK=y`+q>)?_M@;5i^s6W`2S!yhNr_gat&-c;77wss?a9DLxe{BcKp`n8K4ok z7v>@31z3ujyJjkcMV>Q+Xx)N*0SobV5hz%6d@Yla-rms3*k8QR9TXcDzeaAMH0xz_ z(y9%f3`uLm<5n)l%~;0IjQH{1sQB356raGy9K=w_2J105#*awG_b5h9hE2yRmU78r zS_dyzaC0q_p=XNV{Z9OS)Q!n)OdT{hDY6MYauMCYivLy|Be>Ond`bKTYm!_? z88j0tB{Ju?ZLR$_P2^AWb;MO@kr1kGOGV*MNzIZ?_+JaVN>|S zgnPGHYrNa5k9v35uk{@~&o;c?onS@M2GaB%0biV?#mCp)?KEXssCndi?`|_W`Ll(o z)K^;u(P!%*`?FpsQ_}!tDOqV*;4FL^9P-c-nmy7~lDb~GNX*9luBQ@b6%AZ#k{?ke zfJLMmbUWow)UL#-W~y6wZ!_wKi%>+$>8mUVfV?v~D|y7DKK z(4AsZ8x`u@arK|Jn#EKya6V((X^*R|cl7Q{o73pJ-Un8u4}O1jz@hn26`YNa za?`mp#Wvs9)U}3TK&A|qU$hDyJvc!Z47A+%Nl)q?r!56kvA78A8>psz{vE2?dSa<_ z+tl4tY9aKlZ{E^|0^zv_p!oYSU?R%6P(Z`p5~R`Yw1*|+9_tz&Hfyv7|4P4w0*aqM zTq5_3FX>v|)J9+Snn|08i=%E@a|*%J0dh3Dky^KIBd4_*%`XS&v{V*qX4aq7U zNG_;r*F4hUoJ%j&qa%d&(F+CeFf^4wa;mX3dEadagpD;=PUT2p)|)T4G*SReQ}7hi z3dPLA4+3p*pV=15B%Lo$nk|hTJbZxp&8dZi$fd3Pzp@GjzZ-0~ti7ptvt_FI?exY# zX%ur0FxmhlAnh^+zsKdTSKTAGnLmSv2wLTlQ@bI^1&oBRB${>k{lL#?W|H~Q}*|IAx#{v#9`T?sK=M(hJ1>4)vw zzq;^zvt%N|A3TvNt)()aZvWHN``7c&QGYAZKndDZ05sWbIwhcSp;$pTsxfof5)M* zz7a8VKW~&iaJqGqmi2aEy-R%#?by_}MoX*)K-J)0{;NxC5%3oYs{Wu3K?*&PM0|Wr zhWcqhiS5%k+bj}dv}_BRDV}JCCxFGt&TKNBOG~7KaDidfp=cwcJ&x^#brQAhEkg!I z6pr30WhzbOQ<}oBoBg_s?^sUBfXtcrV*^J*tF(pGw`VO&6fIcH57d!fW3J*-q!p>q zsqLggn#f&+Pn|?72CO2DxphQf21U#={Q;*)w=4%Aa>(@_2zJ6JTphf)JwHf-tJi3{ zam_eb7!Es^*y}{ww@?L*KZSUNGL-TAdvOt5u3G|xf&z>(+M^T3IMXu($x%L>!Wp?| zB`<9^nbAp%R9LT-$55YOV+LA=49+&=%x*=pc8ZM&WFv#AbXzK8jYDuPb7VWM?8XMp z{J*IUg?vsvI-%7GACRE~%H=dcuu$*mQ=Jx-L?s86_0rZ=ifawpH}mFv(O~4xPlb zPb?-mhya%xFi+6Qtc97$#NFCJEm-6V+d?$ZRn(1-8B5qeif6PfCB9BczcY?a63J?F zigXIjQ~khpKhr+S%y?%g3q59XeKRL+F1lzi={h~26DhhsvjL+n21u-ZNNn$O_mbn4W5iCtx*h(}1cO z#VQTh)C^Zk!+_xEq9{`VC)$q_LK`XZ07o$;wTWUmY!42O!Eg)$0k8(;B-?O=QS`qg;F&Ii(40ij1E}zdxA97=j6JZ<_oa?#KceB zE()iasi;#&R;^)fy}vRHPBQA_tWWn@h!Ey{0`MJ(^)0rXY!ZSiTCr6&s_9A1iaZg8 z!kKZUp2sP9JY{oA6q3YAwKXCmsx3ZHV$%-(yi~$w)(~+-D&nzzSQ~+~2k;zJQ?Nkr zcLp&{n&gZIhZA2bwxvvm`Z%2&T-17;ixlin5mg>u)g~MliQa&f$P#zeY1$Gz#GS77 zKLvQq4y%&eDK0)w3M!F!E%ZU9U$GQ{=$aj7-M4{w|58@LD><8T!JIA|W~Qii6K9jN zjtb9R|5yo=Q;Cof&oyFFluj*BmItW|8t48})Kjk&d~VtXE<;<|N^cU~=LzEHdP1T>$epb7Mj&zf zfT2-$X`t{l4M)KbeAim5e1T4KfkmUW@ChAOti0t4!p0FWDjeVVoBq(o>mgzbjt;NT9G(~LYOXk`tHbql#USFmX7pd(iA^ta1D1!7QlZep;OouL8Ru+Av2in zaNDzog5BBDAWdesPaA25EC!?J)<=VlOovh8CM#QZu0skikFm)aC3<=T%j=q+8WZ+B z+LSnv0RcNOE>wq4p9RwbWm=zkv<_b#{4s(+Ex2_&?^}8yBs$wSUXZLMQb?qSc>rqb zhM0JBn3%vRxhJZhRrf@velkwhDtHREdR!wQaLP;Xr?rYno+1&VKR9dB;ZAsfU4YE= zC5MdJpkTXpf~TRjDwF^_5xwOvWX5~V2Q@={nu2XJwq_@=@@ydyPmRzTw-Y4KU=D*8 z>|D=PA*fFapD_QDu;5+ObYM$o9u3QA@P{uZKGcG9O56Z(-N;;YDQ$IEZ+7+gVHP6UH6DjK1oJF#- z-cHp{WFUF`vh)8J$pIL8VS9@*48}cA$u4MOru=V6l7>^U2=gCis`Eq_rryJ9I$WLy@T8R@k!2_ft%bFDLxHx#*O}kee zHoL?98y8>-3Hv;YK(7`=3j>l6^AXD<@t#*=Dlk^6*YwjYNc)gUX2SVI*9fA*A+S?x zqXY|z&pdh~7YZA-wL(X|Xf>nPS+s9C3?Xo%_NQelS8g?8HN(J1NuH7spi+8)ZY#8U zX-=+@a&i(MO*@yynVUf?aKiPzqbLZvNegm2X^iL>p}3yZgezNfU^!1zcjeP#8Y_0- zpwH=A%pqYm3||e2?U~jjA#P4>F`W|@W>1p_%t|@0tcVnvIiC37f1`_vvprj!&$x6A zga|7lpPWgZ!F}leRY4W-Yeol3aBCu^p%5wn5;YPcL-P{_sQ3wjzeh>Thm&i8_b=rK>U7Qw zoi#3q88`ofQ(QbWX&6haos95F3w0 z?Tpw`d?F`K+~Y^#{d5<5onu=-DU$qRollV-gk{A!o*_YygNqk*Ts#rdEvVvn+S#2^ zY^oGad{(E!M_0+>1j>skiM0TOh~o_@e`7a&Nu1iruU&@z87=6WigO^ngx3+aUJkY&6x%h@pDbbLd8Y|EUr z)Cl*lHpHcgo-KrT+E~HLila1y{Sf(+sKFl0gVq&veZ7Zp9a;r)Cr>^S^;n7s_ zcV)TDOk+H#jn*ke&Nz{{3%x*rE+>Ns0mLHI^g$4~ z_~>>Oj}$wvw^3@RSbSGcB6CSEqt=`Y5Hp8<+~Ijfi#JmebDn!6vTak6R=!5LAf?FM zUU?C19!^-kirkLACms%0BOvFisEOq)PP9Qi56prgCht?DetP57NoEliL5MX(RY@&p zOf!V<^AxG zoiD1!?Y6>aR^;qiSKE}k?x)$7(a;PQs8Bj#gwJR)CpbpP%1s-qaDs>*j&wv*r7}Aa zauIWgxt?W!NRXY*d-=iNJE7%}W!ah}Pl9oEnye#G#_h8OdRFR%2BOy>-Y^bQ(GnB!QaS)Eqi zg<-1O2#0TQ`Xl%{VSr1+tWIOnrx_Bj;_1<34&DaBvjIPn_WK^IL=`>VvBI7^_@)Dn z9WtA+Iq|Y#O&8QUvoHkZ1p(T?gdwGJWoo!v-~J>lL`p zrbld~D#PFvH&MQVjI0F588WJAbo>;~ykT|kjC48!dV731 z!IOLQt+aHE*HqVvGHHYWkG#|)N}NI9nBUm*lv6lqOH2{M_6g{xdS^ z2#Tp5lYm5Cr)33`^hv0F0`oshv`@(!Vx!X$5+v_+S>NB+>I(|RUdgIBOtx#lOfArUECv>O-TS$FHI3w5L0iPi--WM+f)7f!@9#Ww=?;wdxb zBUY>3Zl}GbCKd>|*0+R5n>qh`P;?;AO@y@iCw<%+ts_^5CV{BPihl#E^|f$ z=9oGBxu>alo*UMMp}AL1hj>OOBWAmEu&~TD1t3{#F5Bn zJLN|YqnIrl*GIPkx4H|9)Bwf_bY_zlAfyOMgd(NiTJy6llgAFdj4K(O`7+wDIPH-u zP_kI3L%Pmd0HH*WGcy#H&6uO1yZLQU7Y6O2*%<(NAVz{<67r&S$fv$xIWXu)lnINi zDp9p`II*!}1LV0L_`nT{0MB_p`}33?=V(JZmND5_R7tEd+$gre+0{yR``AY8DEWW? zPN${h5Hx`LTO19x|A5W2GPTv~Q$5vK)vBF(fk?I|pFWl3%`d;a$V;JwFkzNO!VR*J!5gSaz5=AOv^ps44%sU1h_Ls?34$`P3APRJ~7-+ffy3M6OR{KtHfc2xt&r-;;1=- z?Kq!;a;M^~fed0yWPuLBaRW?)chqH5{EWI%!S9SbR2E!Au% zwvW|$@6!=h(B-s-Dy&bJ`RUV$r%_q`ROq@Xo%VV#VaaidedlaWYsBhQPAub-4~(9n zg57D@y|Q{|UV}~16UzQ%Qaj0|xetp{@f5gGju_hS=_Z0F5T}hVc==KXt+hQ9JkP?` zi8rw`tVy?jmMl<1wy%7$<#7MN>UV=-l^L;vy3y$^t6kp44DKyjtnA` zlg?<`?Tu`C7Rpa1HZLRI#wq`A%?OS}k{TEZC3UTOV%Kx8=ei#~VE@rvyXuc~QnDM+ zdOsQz$Av^2A;DwVvXjrFr9TmfJY*g=wfO>c`e-oV+83?e;0zJDN|@Ihj4HIerY>Vm z%8a%rMtuQd^H6r*gl@1dCJgYb*)?rG(eEAO|!F5-%yleta;;>IwM z-cNzmUZYB_Loz2=j>?g9I^(a{ij>K#-&r4O-?iL1scgxqchY@|tg_PxKNk9$ge%T7 zA8>=XZBcw?RvMA$0(VaQ>7I$VD{6O7t(SI6dK>@q(sYhbJD!>bnh&xbCs>094;^U5 z80(cH_i(PyWI~l;HwAe1zO7*r?Swc1j#jo zJS1PaT-g-7w7U$imfw-^7$s&1wu01kcSywqrM3``Fg*xQN|Aov5tx#^sVwJzO5i0L z-_RFrP#!wf%9HKPwmuSpwjqQnH7{Rdv20^dg5ju{Ml~k4Q|Kh@86!nLUeaCdy-oI? zke)MTzWJfpz;gHf4W#%rXnMpQ%@A7&V zNRz=pMnvnKm(SEd1>Up-Ee1OygLB%?}&0%=+To^%}a*$z>sp;yt z3jHONVw9#A_f4oV@(AFO-xg<0P}>$kDYd~rLs7Sh7aHw7$Gh;3^5_GXr#ZHAqPpZz zoCX5#uQ|}7v2+ZDecG5z)2#f}o5PqlzgJ%yX}`?ck;ZUNi^fKEM%;2vT0jJU{~f+J z6X)tLge18ZhdEY(Qhg}Yv&060F0_L`txZBGO7EI0t}qPc@nn5w@k*^x;er?t0010G zvIgL#EJ|aLXpf(0{>GvgEPMNx|7~T^V$P;kapDvOs4Z}g57G=$AMcw%Ajigp5OZgj z_|S^@YV*-*1B{SQgg^GKHcuEQ&3B*bgOcfbn0N2U78>`_G(*zpgq*E%9duxN5TsL@ zWy??QJ=P@T0&_;ioaIXi4j4Jv6XEABmrTaZkSFw%&?+xLp9DBsCyB5o?s*LESVe$? zgr^kGS8~9y!o-dW44S6wM=lq$+e4kZkm#jXriBegvvoNw0UpClgxXlRz)UG~Z=gXu zsS8i@O#R9?jrEGotdkWY7}a@m)yvB+sLe`QUd3eGyH)A-E|UKIrtTNI1mZ z9erar55r6;(Win)1_B7D@N<4`8~f^d zlbmLQYDzV})lO!ju9fS4-_Kg@PQf|i^&tgvJ`P6p^`>psH;TdMj? zoM@T;j=Xms=h0|JW1$^(S!mW7FLZO`eqy>_8`TEjiiqRCh-o;O6n2dQjo-@PD74Zl zE=+aqBVS{QS^o){Mw3>3j{ya}V6IcVMu&>t$^gXk;3>r&{!259H`%T7l$kY9_B`#5 z)zL-UO++r2Tjuq|ycdG?F60GzTAriTi_Y<9G0ZiyGtvQ%IF84XEO-!AAbX5t%LtjQ z-zHq%RkAH0K5t1E8&b5&q~Ntx7}$941ZCkB#$U$rASM@471Ld8%M~I!nvWqCF?97; zr$rtC!YU*OlrX5WK36U12Q+}Vq7>n8${I3Q;;u@a3W7`7*F%bk^fHcO!OGA|C2fSt z#tTwGHE}$P^Mi1*TOUIlM-yB#T;*tc1un^?9msJ;=8lv(hG5? zHh3EJEPuSYtK+0TiQ^KuwKDgz`RaTTYg+RQD$B53#aysL_y)64vl(~bL2BbwdvosRs&OWa_ zv#amLjSTxoB~GztY?+43UT%~#?D2jkjR6^_7Spk1*TTX>n`YXbHi+h`U zgWLtrw!GB)xL4aHOSRHXzX3hx1}l|+kdohkS)rV$^Lc>nG=?zrKT!JJ3paea(cSdw zdBbyQp*Lk&&uPqfNyBqZ1`7C4K?f78%BpOQY5g>9gZ)Q)knYlN$TFtNyPusO$-_7;ydiC*9rbpLV42+>LD#jdGDCfgu7sFlJ zz7#2X?zD{hayO(5DudCGWynn{PF(J%`-A?jf9$^ozyW-K9uNn@foh=71N_C^ue(g1 zfn&NK?wlP2$|~8=7KQ3)7)_%{;5&Ew!sLk}<%(nW+{{mzG@h%ls-_^Odz)<5CBlSv z>Ha2k96SC0(YlR>w;S5hfgaR=a4p;k55Qya5)8uxuwV`rKnUwl=SWAK=7i&PVot^> zI(4Vx0EarGqa5pO&dc|(GHY=uSFwe~*yT~D{uUGxY&M*}oQRGRYX z_$s&Y5PzBtyD!6cttpnQA9;}oh4!6cf4FNvHrTM@en9OR8}}VH0s1c<)ADyAWD#Eh zO%$&6uy46LD!b0tyyuh=e>ea6?|%B93Zy(MqOycmG!o54E73*t7lSLUa;m=Cs;A~_ ztIq1SKGFjrLJ=fj5hzhd5?zd79McG}5*}heU=Z9yNB{yOV^M)9hz*HLDO=!fB? zlSeC^OmW0D_q-GeLrFypT=@wUD^USTP=z|QL81r4n8hXr9w8}+fn8Wm9?s$8geJ>m z$uqx#OD&rf1RL4PK`wGNN}PbCX)=c7Em*un+2GJhDY=EQGIk?@QZWC5gRqDqv*&;p zaS#>EVF_#4#%at1$+PFmSE5XXu&Bh8{E|p~(&XtA^IgFBaHd%nwREN^7TI8*E8O78 z=o1W@Y4%|YFSgXMQBxKyUN$mLA!B?60~R(((iK!(Syk20tTwcxODqCXN;(z@R6tl- z28yLJINZVJFmYhSz)6uIi;ati6u4m^C`JXUQHKt6V*oyK4484_BTO8CECdWP8a5#* zR6$2AC9kPtRnnzE%B5D?q+8a=QLjaZxDiukMGRyUM@EdDIBn)?nyenpQKQgGZPIQX z9(AnorkG~7`4(GznH5J)oP{EL>{sK-g#~j6h?F>WUXfnq%l46vqM5*|! z+^XhTF9ruV%<<0Gs>g`MmfK{T{Q(stV-r)-^9oBt85)WoM6!~}AXZKV6gRhK)1E`8 zuHE@ULc<~g(=hT0N;~3|b8dNWzUpF>Om--AjsQkbjB=`|qZvaNgN!lD1~4l8QOuNWIEct46h`Q)8OcjOMl0Ke>uks9e2v{U(*=t+R)mhD!txLX->w z3OWVKw9tT75JMQlH0H4Z9|;azL`ecr1cyS#!lMyxk_Oi=p{D{O#ZDm}tJuO0MhqP0 zIOlm*M3DxHE}5XAqQyydxH!-XZA4LtW~#K^T}rD@BbwHnmb5Y2nBz@6aGDusnSHL| z3okc(?1X7E=d8NcNjx1A9fwNNSPDw&!lDwgiYi(rBw)BK)#|kBF=5J_MaKyyArqC5 zlGiZrnD_g16c`HxN`zEv)S=tBIg3&X%IbzT+JO@W6%b6Mz(9fu6M0l14f3D@mV*ld z5C#B&8fbxbSOpdUNT7fnPPpJj05L#7K|_EJ7Zs|BNEu|DGaJsVbcjG&04T*%Vp1^0EH<=NJ{n8LhaORn%M?y zF=Eti`y6%3S;HnRTC?Y7$cUi9LIgyKnL^5Ga1lG$&3?A5+?AHl!i^+Q)G@>yZ$inW z6e?z%`1zb4RToz&D^jdXjXL!tTMDIAD#R&1iAlY*OLu8yS6oF^)z(ydv3iO(QJ}?Q zB}!MUQoCV)EirXs$Wo)nO`o;=%4=wWmTILs)Ta@RYqd6O_sFA&*Np;f0o zv9j@&6hwdicUX`0+K^4y^=LQ*lr;2QHBB`38#IA#&QAi{>f%GzURQk$H`$ExRHH_< zszZK-ln`1MCrOIXtB>uPRM1EVeS}E>YweTGobur2s$0~-yX;nHM| zin5fe5;dw-eM%@Tw?e8^hq^T-k0Y8-DzwP_$i(#Wm7869^-{vD!%2WbN+_p}cE(Fq zh)PJ#t(pXJmH41MEY%P0I?v58xvQX*~AwbZgJjF>oe*2=@<>4{6)zVOo{&}mKof$(&;UtET@{c8O&|~WLm!jWLJRHmGsI$6vYA}~5DZIYsiVa< zf%NRcc9TY){oO07j)i25iO-%ptt7b1K{<*&jRKBhZH5EWe$*Gq^0VR|ZtD0Jt z322Z~)MvyxS_T+plDRBp9h=y}UJjcSv>+xauc~2Ho&;wifD9!_0*VX*4n;gRQN94~ zZxSjVe8kAX0}+4_L?Q|4$VCB)QGwmK3=W5ejZc+`$i5S4EWSu)s?zAB)U|XCn*iIk z=g_dr0Y_bS-CYkm6HuGbR%5Y?&x+Nd=UXIY1TRFP3|&~l75+$th!{C~oCHadr%soo zDP0sH#0(S?mJ?)bVz%t5K|1JRj0Ns+;fx$$4IQGZsG)0MZf}j)YNx#pJMN5|9=p@j zGqYAWo0^29;t&ua35|wDff9}7`{Of=Sg;}CBuJc$`ZcK;Eoxnx+R?E(nLH>oJPI~t z@>we{IWT_8j5#Z=va(eUoMGE6XB`{Z7@KIbhmBvbWO!W7QGtrOj%6E`i!18XZ7tkL zAcrExVIroO5)`qdq@iyrGEd7aDjpdfn_Z!LXy%CM#3H4r^fJgelgzQnKF4xRY)WRc z^#~|PA)vuWfe|Zif=X2aAk0WKp2S5LlUAl)q)VR}^R}GnGibzuZAZAUYHGyv{LD_| z7;zIM2}+SZ$>dW@E8R?GVS?}9LC&%*BYSc<=W;s_@;I-S(FDsbZSlsRjmll6qpH>G zHg4Xktg+@=>u9IC*tKr=u<{ni_Tdd}j!!bPat@bp8CP-h#x}dq;>)eHs&*1DS++(5 z%P3reRE3pLhXkZhS_P<}qHr;iWyx18YSD_BVMk4xHE+qf^DevL3S%~zG;h&P2OW1E z7ywN`vcHN5vXF}?D$s~_C!IB7(wtq_Q%o#)sDMaO~!U>$~Cy2 z&p!YbX*2NFhJlzCqxb*02cKa0tAOe z!6qOPXsR1DWyY+MMawp=6qgyk`WH8ATqbJz%*3FP<0nlG5dY4K^~}O0qvKL5R;eT` zDlRoozM2>&SYQJ7e~&y@D*U;|=Z5gL-H>07fK2Jq**WkcDcz;MmG zZcX}TKJw7%8dmV9BRnEeb0kuk;y`0MyH4e0sB~k5rGnRpwp|tYSEtef04fMZ^?H+9 zksjDXSekT}dBt{Wz^i`c-Soy732phOn*T}>oX;j4rLPBA>y4=V*7n^g{g&m~qG=Kn zC+*G`=MXTfq+!KrxQkLWB2V$ZbDhQE85%XY;Q|WO4 z)M20NCNAxHx|5G5CXzf!GLg%ZQm-YXQdouay=#A??=*I=ypD_cSgoC$kAVRo=*(2q)WhL-Ow`YQ;Ryg19TDG|);r$OhQ+{UVaneE4iI#t9^{(arPA^`3 z*)ev`Jifg0@*;6>nCD%ii^~pOK#BZ!_$FZWuI1AM={5HDL;3%$`gOrDYGl9op?HDA zH@77`UUH_U!SeS&&WoWZyp{`IRlJ99F0MUNUCV>}oAgO&s3G8UjZVBuxVqQd3c3XT zL0sbZ<@308wP)avuI=qwdqs~%6gQ8Lxs-kGt?4=55UBVqz)WW_Bku01YiGp&g<<;g z-oLpIm7P3kBf{qgVD&s#`fE!NRvtCD9+`NR>BBzDNH^pRJ*L#--}{!Ml}mnWUuU?10>3T(X#TY1S2gP=ETK4u@KsBhyx&(Ql2z2fvyUHORw9>w z>0N;3NB+-KX~)ID&9eM|d#C3j^2{gIc1CYH|6)JV`M~rTIbD1sVIYIY$+N=S1)GPT zT9~zS7vOLpa!9!^%67xG^npU2iAPWttx@#{y<)l$@GjN++gftdfe`F^@pz0 z`?N_zHi@DAzdpT_XB_&6UE1x~-(=rPcv}N+Y}bIopL-&=csC0x?cW;}b5JlWh@8s_ zw>PC4gu4bQ>dYY=C^3tM3iEcZZGA}+1?`!jIlz*0_rEuOzRmPX9Ve&p+)I<9*LoS7 zz~nwot=1@076?uKs%VKb%ed=!xM^{UsbF{-&rs*wL{Ew+q1UC>cT5J{B z2g53}{o_xMgjEesSK7}C@-+*ryY1D@@qG=1?-jU%)2kbHe5~2;zqH=gt`ZA8ixgho za`LSt!?ubuKIo8_rB-XD!mrkmxDy}H+yvGmRA`KHm@+keMrn(onmawf2=fW@;_%!q zuWNB0RLm6s?#A)v{-8kcM7hmAvIy7KyMAw9pWdTlO2Daw+t+mI7j+$H$#*u`J6+9E zd7DD@+I3qHo+}`BN{3R7+@9`biT%K`pD!E3TYE2SeB72Y?IZ zUpuQ-7S`s6Z3s2eQZF=1hgZ3x}Ym1@bKzwALTAejzKD>C}zWl$x8@?Ot&-(pHqd z(wn89trYaM+*#jg(49oz(2|BYXXcPVZn{>^xwsP?OMZ-{kio?DTba*#FU(;|XQXv| zOPkSSjFv`iX_$%@v5JniEYe(W9w-sDr6NY3Rgj zM8-`I_fklmlupRt=Jgu$RPD*k{NAD2QfQR@dqK-yznP&I5YLOdzw(N?Los|p+jF-~ z-u`N^8La7mdbs}c{kEk`k2XfH`5ukCA>>aG zUBbT&?NY!EI`+h-qN!2g%h=SHv%di)u}ddY_b&mg4jh=q*4{2^Xv?z<=QP++;A&cC zu54QG4acjN+0RVE41pUuriMbyka>~#XHII2V40QIb~b0*P5;;Z+)FaLqHf@J7*>yx z8(LNWcktw%MrZ4sZdz3O@QK!$O|$63P*8F>@EpnPNNZ*8*#j)E2~X+aBce=wx?8`xo^% zXz-rnTF`&8$ig8cE$0Pvgg=OLD@|^^AOV;o*P54cNd8{kzkS*@I`O#xT)fzT7?T;a zovLDgIb^JU`BNH5`-@5XVW-s70Plw%tqVOiQYqjXI(TvZ)-hWG3vlZR9O*x*P+;or zmc@l%D@Sxkt4uo*&yNQ0CLxy4I%r4m&JyA2Oa!rywcWIprHtW*zxgMB)#TW?)|>d6^ynpE+o`@C4yYnwCN6BImvy%3 zb``(o9kWS$?Mr3`?j_3j-GZ1Lh92Oy=wv|{@a)2FMf*zH*VhO}-ufuu#H#HV8Xc*& z))}MY#F*d0!*Y&hWd)f(B-GEbt0%AwC(sLpG->4xUG zly(pwk@+8Yq{cA(BXbGmyKC4&5Yw`eTO72aAUGb#7l!ffWD5_M6|ll8`OgZf`g==C z+YNt}oZ;E)iVo~i&g!>@@jex$G6w_eQl*@ocP6m{_&?u~Rxdd3pHe**+w+A|%2@sK z%M*-UK3U4S7CWs5_hR15un}0_Zu0d`VScej{3!6B@y`Nn6|hkSq}BDqOAqhRWNuOK z`FkT`>XQpzdOZ)r-*Vd6T{5-fo_dfS#|5}ot_eQ-mFZhD4xZbfu%Vs0D~vW&`_3+? zNmiSL^8R(FRi56UD*E)NsfeHMZmdI&dS1te0{wTK8QabinO>ZEDJJ%L)+eom1Y^8$ z?8|A3(+}FBkG~y7QqlS(Dki<%{HOAec)#iOm}At!(;5(~x1#hNV|H2%NxC*rgXqU) zbZ@^JpRS4BTH?JcB&`*su` zE=S#m!5|A8RW6+)iQ$80^;}D+V*JwtStaoQwzya|BDBY*1@ z(bK$kNTtNnbNV8K0<7|F5qNMGG*^HVd;oy{FV`}kJ6LCYb#{q!+Kz~6a|^le}`b?*s#!JwJKa?uLr zOV_Mr?dF%mN-mjr{vh<}u(3A@8n@8u+Ykn;=g)#+t2xZXI{PGNmbv%=>yd9f;D1Cu z@r!a59+9H@ew&(t^?~Tc@xV)|d9irYUk;E0Bn`CpUlUVY3$JlgCTY>9M58{-(Y~Bt z1`l>n<;TN*Jpj%i{K2ucEwX`mN(G2R%Fs(8bTaU%F>@F7>`c7utBm7lJ8s806Hg%7 zSub4^;=Evb>!TiwtO4(|tvAi<<-jNKwx~r;*mAwUOy%HMYNs&LGnBk?Pu?A%H=D*# z0{4w=7)^mg$O?XQuMqO-PcFz=kI^1X=ytWhH{LX}fK}Y;4j&(dszuZ<{#D*l!9p>G z*6VUma4>GnKgwA2>PXZ`Dnf3k0oP>X8z^bNJ{0fGd5xL(*halQtggAzwAJkJX({tc zU()X(e*y2%(IlE>ylK+-A-gmn4gWZ)6U*vUs5&DIW=bAD7?dNYL%@Fd=~!GKNDq@* zX`Bl*9ZVOc`8Opg>;GG6;lYk>Fcp$M>8IZPP6kHL6v1}t98?ufw{=f$LJzVCHDwRb zxa`=KwixJt3Mgl;w|%kNL?=ki2L-u%8c0u%tibCceRKM?54EII%^n;`>LWV-b>o)U z<|Dd?fT$+~SUV8LPJt%;u=~+K`8aY@k=32e=}$L{UUGkLSahr;fTe=303WMMd2!yc z?W{jS_m5#HZ$YQiE`LXt2ahH1=o|a)@=j<>f-T&#Om+;76q)Htc(lnwsft5xL3VJ^}G6bk1PTSlZnhe4NWFcvClGRrs<*YA83H()@^z;oQB&M zLFTPwE^Equ#BZ+OfQ4tmebrw%)xb zWKKp<;T+8>lyc^MxiD{j+og(yR@IbblQ$m~dGn2EpUm~V|0lO?Y{km;J$41W{5Cz1 zyRR*rau--+Wr;KSdhvpDSGs%q>V$Y#*_$K(Vek&IzrxZU$Pt6*V~n9;#~=gGka3$Q zpo`PMVenywTifd@aO8Rbe_<-F0v5efa&y-q{XwCA8Asd_f-k3Fe&;+Z$U2VUO zarW5}OcLdEjzM++3EytXV3%~nlh;g5l>36?7D-*ZL*E?Y#Fff|+-Wg%4JPm4<^W@4 zo+u9tyfe(Py)wc#IR7<^^h^2d`tIUXNn7XeM^B@JnMa)Bm@Z^LS`qZGFco%1jxHz% z&<`WwhmH9vq_%6b6mjX9xW0Dk#8$bU^am~p6lQ&^So7(+I9wBqwx($k%d7EM ztm~dmQ?XN(hwO6(=-Qn(cJ-*Mc1?XkcL}{#V(h`Qpz4 z^501NEw#6JRVbi5XgJaGSoB&Q)kMwN$gG{DxzIhXy=t)o5EB9)Z5GO(LIt@4Zv$m& zP-q;fl(l*`CPi$Lx&{R49vliDsZa8srx#C!aN;!-htg*_8)(}wkP6SwIF1g>{o38y z_(ad|D!3DVllPZG2TT#68$LMep4@Ik?%x`$&rH(Ai>5#&YFF&>->P6VpN`pU*^O4L>+)0Eo$%xoE&voZ-{?+sjBv-oz^ZeujG;2rc;`g8s5k zMcg=e%Ox@=3%Y}VIquX2$h z=fSR(Ehi>t4-9SxnBWfU%a4qJ@?^hy%gr~j$e29{Kr!i?;4Oue(?Cu5Mb3gwrp|z# zx|ev!0=wyGlsA*-!~U+TmUFI)vcMeubcfDY!u*bnd-yoOxHZO(U_!=EUZr`5FVPJ9 z<5>IYTfO(8UT^qqcx}P<_l|fRx@Jw4Z5TjUOq4CbZo|tD! z#Zq14lZRnh=D#ZAGD)3wpiu2huNL%L2;dq#$l*E0cG&PPAYY<AE9HdXd}Kz74YWXRI1<_+ltH$Pz0ot}HoE zN&=V7#9F|hYGAaAb$ad=@j&Fdr=))KfpCndMNM<($onpOZ;5{mfOfO^;w=+ zhmaZ5JEzx`p8v4Kadc#?5B^R5oHsD=xqk{^oAYhlQ+34olR$64^SA+Db4k|#js62Q z+|Y9}wt(#yH_UeCHti$8Di_;vJ#*lV(}7H3;=#r?XSKyfq_-B>a^inpVpxyYOETzN zJEv1*ua!Yg)uv|jO-RGVA|4280oWezP;FHZF5VrS>W7&xzX(ct(){Z~7m@07*EMTW zD0N1NbT2ejk>w+7Q>DQZcdriWA(NL69BwA%Jp&F{D#OTFfv&3QNfO@TysTBFo037N z-ljEfGjZ!d`~h$S?^GT8*X(${&2e!&-<^*AP5n|UYCxYs*sx-B{sQ#1i8E;gd@=i- z=0j7;haJHUf1eM1mK0t4iA(uwN-)kB_a~EesYISY)BgKfUdNGwh@Gcfh)Uxf=LD3C0u6m`~-WeUny>Q|7dldJX!4J-q69&N2a+gQ><= z*)nTM9bJmr&XoQkwr!s4LgdZl-%m;WT0R&3nbax)X0vUU1mqp;vRPBwH6G(zSJ=%< ze&icFUNyS<`m+N_SGfI(;l|w!+TrQ0g=@c6{o4jPOesEr)@TJ5`?3mj`krMUqCR zgkKcf*pNdk!?anW4vp&mKG^Mw8qdt=&3(WX%O=iD;|83#!r^mNDqmkCt$x7bopmtv zYA&dcLaA}hJtLPOEeBK%Ophp!QGv3q--k}jp<(Wz-tTgU!(e`A<36Gf3nQ7_RHqQ- z@?WuF>l;C5%MQRrVe`M1q=hf1oZuqNSzrmK12bR?zXvVPOrv1AI+G7K`u^88(HzKS zq=U)Izi*?gGbOAsosR_na~+JU1&(*ZnO9_(DWk@ik?Bm&ndZ(q#PdF?Qm9(WDQ>9n zSqBU(2Mwkz&xYvj_dSy{xgV?I>e3juf8YgaK{E0QSD+06Xk3O*7Le_LE*nMy`1qHc zw0!Jb%Q8zvxIB3p#?dPlGcMoVGUkA9J?>8@>1)$-h|)p=0j0ed_<@VM;`LjTs~h)) zn*lcYUTvEHzVi zMuAhgyCF| zonaBOYA%;rHe-LCW7+`U9^JnHSm8JKc_;08OV|WhNIx;Rz4rD<<$~i+#=3UT?Kkzr zf56rFut-#dgb5!sbRAb61F2Tp6VMEP15PI|PKP-Cmgt=byhtaeUqc*9NL(3E*w81z z2cyOqkPIA)it^8#I(WRZSOpq>hM9hD&#GIyeHpZOf5t50w}AilHoR6wzuS4QG*Ise zjL6KN?~DtVL66Utq|Lrl#FdqyZ+J%WuN#g%1wK!_P1q-=oN@Zq#f08bX!oH`pM1RX zIgC+unEuF-fQxywzzg&pMSl*N#_7?lpKCKXZy5N1aD*LicHVWq1PO z0!@BcosZv?35?PNzqs+;ef+M8a3hp^C?L0Nx?cb|)j#|+A(sJi7R&z>ng%EdsZQ`tryTc@hlt>{8v18!2Y8EGf}Tj4hchJ$;tL14$I z#gZE^N=LYmf%S1<$#T!(+;?e-pYmKwT$Ge_z@@2r#@m}6rw99Vd(Q`jIdK4Pp+$0EV-fgn(oI5Nt_pYdUXG>!|c;X+Ki_>qB8%L|Ox&PCrr z-?6dm_GCD9i@J%}T(9s?4c3KXlBxGad#pR60@jtI#=%>@JR$A%dUou`+rbluKH>8g z)-OV)mzjNOTWm&;kCl)Vs3!EH<5x+{V~^^%-C@kAUOt>{gs}Zgw%2ytp{mlj32-0O zxqqU{y2~EjFVBiO&Wrt*`4{ zMIrt32v$+=yP~`LjjEWQ_+xVSt?hM4q$kLqtoJX?47%`Ut(HHGxep_6-DyJOme)wG z#ULU_-&}+4(atyp=?h*9Bux;t9YBDIPyiUO|D!bEW{?3a{J~VTF*<;I>Mmd>Mg`*d z+>{&JDdXq+<4Kkt+FSVzO$2Npe`pHO0USVQM$Tp%6yrEBf@9H4G!HF6E4u>`D`N_N z#blTc^IdcpBFZBp5Cnolw1^qWAQZwPWNtq_70Dn11VhxYFv$d-1)9Mm7y=jI5L|=3 z^cHM2e=^K+_ZwpXe3Y;dgZ(cu7frNL#?!cShe|TN%bpWpQ;#LMOf9Y=U6}h9|N9*>9j)p072UDU4=Y)IdA&tuxs&p+StcLPttXVSyQ?bq8sZ)~1Go!bEK zDk^<1jB6}h{AkZIx(q35sKLgVWC^Nd*ak;vjU)BJe%Km2U{CCZ4GH4j(&W5fY%$pa zSDkd)DTl50#br;Ob8ewpEc~lrL5;#&*`F=_uu*abtP>& ztZM2=M|Q-R4X3W{dL?Iq*okd-j1J9NH{+Rh8kwz?S+LF1{Nb4Il$yM0zh53>OBY{9 z^q|{%(>7{9$&ndTcZrF#i~eoV^;TnR0wcy zXG4knUR-;W*3Su30hKf|#XkB-HnN@-oHm8F@Qgo`AMZh);#Y>~Th}OM??p9b#019Y zKYo}$D%bMC>eorw23l-B+wHRB^-q&GzXm%40vHpK{ezdl-?pUbNI~YV3G+HxMB2@w+Mx+})_ABrYQ z^?3Gy8d!ZyVou$MwsQxeV#Ih9x~kESx+(kI8q_2oQlx(c6jnWxs$*znlx$!#OIgdU ze&s0pxfwru-Qh6_Rhz02QP_|h_6AI;9u!!J%2ub62B6uE)zz!kHRB8zW!y<8pKGcG zXHnZ8J5evXjY|xkJa9_8l}U<@s3Wg&MrY)dTw;tRR(bNQ)J%Ov&i(KT?BGg=(P3DM{^^^ofM ziKm``P;};}>vT`)u=01i@9vI#xj*wQ%V|0TVU$BE9w-RJ`4^!qMn5mwuDtspWi2A} zJPiBK0?uK`1#_2xip3Q|nDOrq6c_I%Q#{I03S->br8z0oTHaY}gY7O|E>*G&Ions( z!uQMHikx{ft=s0>!$&&K3661lY=a#*;}x^s?hfCbTXUm_-Av-U{OapE@8)SU^MIQu zo9?bq1KdpGVc{%x#XU-;R8Dh!mbcSBL;jxJrPE`LnE!IHTPsU+0(T6O2Ee^krQNc`vbi*PmrvBiYf2S{;-+ znNnZ9o>XQTt&|#RlNNcS>Hfx4Ja!PPP`7OPdfRg8VRL3~R6R`+bwqs5kp^)1dL#Ux-z=EuQdnm>ISeV-?yVZp_Q3MB<1Kx_mhLYfo` z#E0=Ba0kdAF~RO1U>)2#F}88y^vs{wOsB&Jq(s8ucyH#Q`v;3ah~>Kr>}XVvs_-_ zvFCfP176`rc%N+d_>^DbKmPK$*Z;BQ$lP*FPJ&a*zduWmkG)|r7DFm{pb!5*7sQ26 zZ?7VUH}E@zpc!^UFV*xlE4r~7Q2^lZ2}bb_E;7Pzyaj${ib>>^{Mfk6e4frKlR%c^ zrp;_5henEM{=w5fn2v)YiYZv8-^*;LL<=pr*b>Xxf9A4~Mbr*?U;UT7u#3E?D^(f| z!6p3seA4@_@||(Q_#S^BuCcDEdmrsMzP8~JJJ~5;nowuW{K@(nsaXZ8Q>#*H)0Ec7 zn0|_dHZ>*V^1|kEd%T_iQijwZ1_W0enb~PW+A&Sk+*(e9G(xZy$Ch5>*0 z)BTBB<`=LpH^NB2It2L+*>Y6FOmyt{ht_`m6@2n{^mFmw;omkt1JBem_KZF=&)gG! zVtan_0aMJs$TrSA$7UwrK$avNBy!a3ix6@cfMFCqj^e?fam5fQqd}sj(OTJOJ+@{g z-hKPhTlZ}ZI-Hb>spoqJWwcR16Z05AaBgUkMxTnTzVM&B{6aX8=%W3>Q9%p1lZVLC z-A`)1;_~~Ct9RG5Gzds={FD90C{w`TI&^KoQKP0@_^D(^PD5K)fa*IV(zH#Cg)V+5 zJb(xF0sZCVpS6>7?oqu>5(~4!k(Ld%fhtu#?q%QmuUGg?WiOa_pFupAis`p3btWRT zp@$OChy2w^J3nW*Ygd^FZE7=E&zCNa4Xybx5I6458MY&!gJA@j1jw^ z@mgv-RIY%+P&fiZktZCcUpQB|SQrU|uvs|z{80Og@S%_MNRM^8?v1maM?C!pKF+bv zwA1hY;x#+Cl`Qfpp{8T_2cNzFCPz)1I%~PvE3FRB2N!|Mz!hKumV>QeBiO!oKLh{( zK>-WY@FJ1bKoCzPShA6#M~xgS8trL^CWq06hD}5lW1vKcfl-7aNri(^C}UFNcyxxD za4BOhXFZ$qmq$I%XXK2M2{T;=WtPm4Su=EralsQEmo#Ud#UhSb#U4(&WWUR9>9 zJ2Bg(MO|^N8+2p1)@&!8>#R~O`vCm#q6WDrN3nF5U8$vQG}mHuO+V;B%kj6n{oSiD zcCUZ>u#+8v2;4K!C}re8vRhxC15gK%nW_~im77Of&7MVX;qY}y~k5%DnUA?nzbY~el zA-X7`hZY`BmKYK$Ntm#;w-N;7EYrtTMp=`q2wk$axx__Yffox343ID(M}--T4yMS! zKw>(;C8(ZZ{U*(tF=tgzr>Hak4}LwMP7VouC_){MaEC7<2_=$5%4uZirstB}5I^?o zt+UtcGyCeL7t$;Bp?7`iTfh3V%WgZIaO3kmU-{@VLoAxWHESM^(E1m=Cx<|Lxen{t z$yAqNqAOorPJBXX$$O`MUG{ZFl|k9!V)Nw4otD3F&HRU;kgTxBF2}fE{5^3`$zD3y z6~{d;6J?x?k!eRB3ycMP_DR<`atr7^MiCqm0f{s+c`6Rho(dP0ek$=0Ach-1wQA~9 zXiT{wj_aDC&|w5aW`Ey!4Hgm_1q;u|CL=p|hJI36O?NrV@(7piVvAN>x<4{>G?W

#oUbX3rx}bd>WK`j@Vv$kaQ$C`BPM z5l6YjENWH3H+aHot=?LgBu)hAa>}LTQg*4hlwWEtaDMDy%R8$Wrw~x_u;iSu(FoA! z7?_Zx-_zI8@+cTu>l;~p8kd-nmXeuwJ5HE9XZGrkY(2;@Q5wZ0rjct3FG{A$^`Ck17uRGQdiaoQ@Tkl*4~LG6}kU+U&M&dq-5 zY^Ke$KQC3o;ug`^b>%fS(>u{tVYBbHw~MCTCAD;U@c7Za)x3X#HcK5W2h!7R?YM>L z(@VB~29v}jDBeta3WfhGO>xd)0*C8A2YoQhuQAzSr8|O#F`KxpeOFg zIjRTkS-i**5&Qz;C66s+v{rAG>a{r z%tl)sJzb;aT>4#@mLeBtO6hG=OBpwvIFh$qrI!zSre_cB)~AK2`rqT)+Ly53_Bh-i zKrsOX>NukAB=^m;Z_-(jN>EWy*&E9`$=t)Hr?)6dpy1)29KniKg!UGyJKHv+cKAeQ zl$4+h#%vJsb$_7QR)5!sboU34LfNsM7L>1p_wRUZZbX<3PQBwCaKVl z$?D8u@csT(ZM4TWtL;JVS+%lqoz3^n%sc}+ITjex70;@=su`t16xs1duBA?iYdBvX z2}!Xd4>3kI^ST=Bl^Y!6M!CXE;>TJ|%u&+1$M!%*AsJWhpuJqhi}l{-y0+@e&{lL> zw4iMEMX@w~43hKorpsfZlYZU$mIIY11O1j*8MN@_=%gPxrGCG#Mito<%$;_x;E(BvpLOwvW)oJQsDQ z?I>RZ!wDJ4fZ|PytbH&yr(SAzc`{{k&Oma}!dEv8C@QRSz`W82X`$?e>n@*YCwxvS zce&lm{C3Dy&KO@uBT!EZ;iBOe=EBB9&nT&XF;TT%yG?HDS{u;&P=g!AzPo%D?m&!C z?2G#l@r6!GkS*BrwSp`c;)y!IsGezHzhJhs&wxuyp&FxW!rayJE@Zf?SDtYl_s|w7xHY0=36NASsqRDT99J(>^1hzbXIMP0r5t<|U_YKv(xiDHz(R#H&?zyfx1=?#2P#LKUu{TkyhmiXJK6d;}BwE%q$W)a8;Rdr|E?0 zqklg?8mg|_);tioE0#L|>>Q(HSbW?}w(uBVZlM`c^JY}Z9T5)PPZc+u$E9=b7adW_ z$#0_%N_f&9u()_!9}sR;WCUhjq25z1l2v!Z4xiz$dM)6nkT!zGy=Q55!eWp6^J`bQ+Ai} zn`>}AwJPwSia1>Pbh^~zbe9?y+*#5e440XUmmUsPPTH?1%|g_Zp&Bi0E-JgTw&W$1 zVaabc8lFe&(ZRCz68W?1Z9Kx?d8BA|z|sEjLP|DPv?zLe74x7i!ke(cJp$I~wxAM3vvhu8$#)m68rF6B)_&UR?IW42&{%sc5_|g44ajgEAwxj5?O0xhO?PHh4Po0DSrwdoQX~C z=K9JPQBGa0TUNha^=}?+UXT$=O{N_#W$h^F+%Ob2ZE>eg?Ys+|o>O6iR8V+cye>jI zxY1}hs6j-XrgC=xWR#csG1WJv52V*hOf0e<^=NHG8I!fxCPOHsL%8*Yd{E@{V6C=U zNoT7GcfgIQJ65Pe@6yX`zpnU}#XTwZ!#XVX)v`RJ--!$K4xxEE<061P+hS<#4##Hq zLgaf3t8JMvd<`J@ zN&KM$Sw21(xCtgdKh`C$+gINohTzni%#wkyo@cFp96+{++SOH~Jsfra;xD+@o)9&h za~5&grpk*Rgg+f?)VXa4ZzkfPBw)E0tFeR1A%eGI2DzIaIcBObyEiJQmsxfukagzP zOmp*UpU|$hn7FQRLeBd)XTWa7!r1vtTG|8 z(PXa4nDGzB?Ix#8YE34MUkqYuKU=;@ntpQX?)#?&9E1e$p&RkFsy?HMSHL>iWUYz4 zg@KV@!RF@byq`u8Fi>?v`(rasj|rX-_~l|wh&8T{OULYp4`}d19AiTqU_8g%NL?2K zue2&S!!J6tjtm{me;)p9&D`3F<3}d!JdUoI<3rT)>GPP#ANzt8eYo?72;gIIaLe>s z{Ab8nY`W(dF!mMX14_7*656l~dlm4Q{~)ZHE2n2{Vf4S-iuk|ZnWTue_Yiyi4D`Yh zEh)f;HAlUP}xxJlW^s14M>atGByy&Wh} z=~Rg-kt%j>ttwC%-On;+a+&dvTk?{P$`CW`j6LJeggsfFiY1r7x1|k*e92w~O}`xG zf19v7;)jQ4JNr=SgnLP<1J`q3Lb+Sxv2Lxhs^bw_KkgoL&mtNpxOyFy1KCO9(1}C4 z+p}E7UOBhTInzP(dnf5#ZQc4kuIYMbU)ND1r;QyM791aRPw4M;&m8w>$&mSbe%_YZ zHtM%q6t4JtA>5(OZcIA#-Et*3lzDx-CC;I;Uvnqp(13JeE(q1Gp>LxI*k%gW{)q%l z<_1KDgikf42-Nwz%DT412JJ=!6d}00aX9%beAEzWG*r84>e@T@66bNEy;r`rhUB+c zw6B&_saT?HZSHGTXi$<*ULAhw_g?7DFXi?3@;mbBUiYJaN6mA1OZby>@1MW-vh)34 z4zSD@M{}h44M`z%wI`zntM3SpV&oKKMRyiW~F?8v@U}pgzNw= zLRpTy8@c2>4m`9e3zu19`kKQ&nK4Cv`Kf1Wz)2D*MMJoYdrdXg@#vShk1E#_7h2Ui z%X8!)5&Mybd=#O)lp!8kDn+9a_@QZ>D%I)*`Xh#A;ng9I{%`5Lbj-iexpg$%@%++! zYkh)_hnu*uJGt>&3B0q%C8mg|>mdM?0-_K=SDjvN|5R0kgPyfBuz;j;HRj&d1j z&~TM5w8BjfGtG&zyN!)5QoYS;79-si9z zXS>_x*~P7!)E13@D&tFh0>1fawX{(AouD1zk$vZWedPXmH?0|WYl}nryFFs-c7tx_=vBZbq5I90WC^&^kr zCyn>un3vzGz-Dt7?Wjc(Rgigt-JrR~%X`v=I;ORG5`ZL4enLP-gDG|nEPkiL54*-u zGJ^_zTvpT+Wo6{lE!pR4O&c1;L>%rT`B#0Le>q|CO{CyNd>(<;)RJ)~pKy)p?_L{d zPZX=YX+z4;)FM1|b0OIKn>x0p(sR&V7k8>pOzantd%nK-Tdloh->wr<6$BG4;tONc z;PT?=`^Lgo8EcI9s_zx@T=%u_Tj!Zr^6;7OEM9x8opWdM!q^|UyzVy`F81~9ztv~; z!?7j3?=;x;{~y+`s(BryfBVs!KoD%K90U|`$^|cGe;qYQZtsySOXPpE+1zY3NaKYN zgc1Y6P5s-0^+8X06CsJFGn}fMcCqdfF5}b{UD?gJ4FK2!4uiAc3b+S8Wy{0y>J(Ex z3;`ezgAvSN1v@yw4J8l*ji3nHK^=^P8L$HmfnBf%8n6#Kkic|b4Q8zLLm{H!(*co7 zU1nGe66(tZU5BME2h^z3ZPAK2LIQzN!V^w1o+dS;IW?{8h)(K^F4lvQ297#dl1T+f zE``vs<0ed;EWAcz%~q(~VoROHCI>jgSuW?w(#xxyh6-a-1qu}}S-MgiJdY($$#e3G zJXE3+7%@mi9n?cJv_xxkMtAhWkV%{1gtd{Ux}E5B&B&aX^f@~h=ElrLoI}VM#xaRL z0fZ4p3hCrhjtW#l6%?Zu8lx$iqYDNSOenD=l1vVz)X_i_eGD_o8XIhLz&C{=IKYOW zga{KYO(wFDhg!6tAB$MV36Pl;#?1>Ckb$DGcp^;&aZqV)1r}36$z@WogWVkAI43#H zIUXiRpR0UyQnpdG&eU>}GnG^d0;r>j9%gWtAybynx?*zxQ9!Q0*E)5{IT>YwRqpa& z#L1>uaH+vVCQO<+M>}1VpfE5~#Y^CjUr|YAb=B9hR&|U@4A+k$n{hh<00=ZXCM$56 zE>J{F`*c;+oi0WkX_6M{ls*}fEKPxcN3>ziDg#WXshk^b%>)J<5U$o^c%F!2FJ3S%R5FFUj-n51TM%x z4IY9H&<_T|6j%hyU=3`6A8&ilMEJ}SG;RG(JL95juDjy{t0JYQdEp3_k&2ZpT_!#; zIW?~nUF~UKa~G|twz{*1G|*6!&DJTg*91@`;HaY#5d!+L0v#q?M8Ig!VaARRKS9DI zf#47+(?Ao3n8$<+A2X#?SivUt&^gE{u7w;yOtHriEbjOcOgxEXk`I((h;R`iMM*gw zF&Zpwq|wHks93xCPPKOHv~C-*F`KlDD=1KLrG=`ip{CmEZLHY}S(8gevwc&)WYWnX zuX>7PiZ!7%)h5kY_OPtV!A2mX;u4Y2k|0N!K^9rZ>~Y98aDc-cI|L7o5*9%WEHpqj2q3`-3nIva z#XK=eQsk&`uJN1oglXa+4O-exITch?ANsKc`$Z80a#{A^Rzw-)R1&F*8csZM(Muk4 z;;E-!pRKgjw$|3$zH2$@L8m&ySq{JaaVMW~;nm)E3${!;W5aoSlukL6PsQXS5BX@o zO!F+a%3AlCvtq}gYyB-;jD$iZ8!396%;GBUizr}pL4wAWK;lUxn?m4VVIoCKxmZh; zs?=^p^2H+|iAc57O0%>|e+_l^Z0pC4Ew;pnshg^-b=s)yx_QW)!Lpo21T(y$1)R5_E&xPpOItYIBDkVGC) z`j{q75fmN~6`w9oqFH8kL)k?YRdpS8+W!XG4Fjp}G;i6uhXS5Rq5>SXMsLKlS@X_) zvdz%a(KBGP!b*fF77PMqw*ThV0K4F2W3lCjkRr8q75;m|6J8Lad%)Y?7#!Z|yf88bX_rg^Z z(#w^v*bMb+i=6T(Ls@E6yPh={zg;GTU~|+crj#1$F?b0QV-t3qxGAQRYFZd%7!k>s zWoz(=u&@-KNMk4}nW1Xcs$L<*v{?>iDMy7WQ@xtiranz5QMsxz2&q)GSW0#+@+qd2 z3LJXr$4eSJK60p{pSf}o8Cf+;QgV$ffg;MO$XAdsm{Lg~3CU{LC7CvB#lCZwUMuP~ zWZWSTUM@OkY{;Z3(`K)*zEuvLb=Ny zE>xo#Z5RoTP0ndvD2d`JgE8Uxgw#<@ElrFwpADQ*45~JvqN)p7Q)@&F)fg=mnb!iN zXosv=7N7tQV2sS%bV1Epr!Cr}qmt5O7RSuwh2y5J948IREQ?4vrOGX@LP{x1Q6;M! zby=)Jh$D8VrCrh9_Dx_j6rE=;SaQKm?Ht+<; zIL;Y7=&>E;cu(kbc*X0yT_=CYqn_EVZgRou>uzn!w`vV*S!P|EW)CNEDyMS>7jP|i zZF-rtcR`v=p~6HcsgkNHQKd*xVni#WdO0nLl0s{eFL9}mDyfww=@uzL)56NFIBwN7 z)7IULWTb*(Xk86V zaL}lTrM5o$z)7c^ah0n%=!}P-^@P*Tf6*IMj}{VAD<3>ObLxx*hYVe6x#d^fjB%6y z|8Uf>ynB0~;;%7Z)3gfx`uENJ-Ky$4UUCoLr=I_R@7!Db=@x=6iZyLdAN;8F`)bLm zuErY#tq=B~EzuY?KkAO%`&t^xm<9iWh;3iHuRZQZQ%BtC?eudNob1zn+P<%ai+}AN zv-mf4-`DLNrm7qJ4UQ%N9lBR<$wpQ<_B-jPAAb3>gZJnOnW-NmNcx0pi|j2!d|ls^fe1hSwx8jk73l0~ZcE{> zn2-r_Xov5Pyydiyyz5kt{L{%ODaUSB#@orq_DHcG!G**-GU-K9v;ThM^34_Q>V^}( z`*ra}yN}gAb~I1Brf6~zaJ$(4gPpIp-%t8KBKItA3F;5S3$bB=#U#4p8ElC|2q z9ycvKw`1pkGp-G26+SZUCzafy<<;}yAG8HI{GXwZCxWW9{l|eq-4r4o(21?>KRd(A zFSb8a32&hKrfPXc{-STA#5#X1x618&(_LUI9nAWYsIF|zoQD5L^*It5i;~IY)?YJ( z&$LG$>nQjt-haz;b)VKW_sq0f^5)xH7=P+*rRbTqk~Ld@CBG+D_G~}XWI`^dfHfcV zY}efra1U?OJ#(B`T;df*WGwPo=fn`8nx(e4*L0SMGvs?59}X-l2sV~&T%rEoeCgT) zi&ZIpFTxf5`OH6#z4839e^kxczURo**bnvZv+ncp>*nhWWb}V|^mXv(GOJFiRvh;S`(i-q&%A;2wE{5rHsPC0 z@}-%*xIlR5zAAjQ-z!fu}9?v&1z9XG$1!uqUNCfi4(W`w$N+O9a`?=FIQT8PDJj-5+atnm()E@_heLnMWGc zYpbHgre&X1UXH2KUY%&74eGu~?tZu$)i|co48n-_G zTw^qu!#^?^t+bqRTaIkTs<#%E6gM5uGRhZpKl;<_zs}Jr z8N_hm;#4nH8x@aaP~ZUJ8VY~_hfiHvMkodTbFFW2Jy6~MmS^d2Ii5;e4}AVtu}y(r zyctt)f1RV)@wip+MJ1G}zVw^(;JWbC^-bOtf3EA_q3H~mbbGw(F9f_cMG4xFEjg7m zX*&G}lN@$H>w*einL%RvPk{H{AOD+pa@9Y7oBsM7;R{=-Jx*q0X*iN4PfgFOL8&IG zHU70Y)1Q3(IB&{Cjl8Iw{N$e;kf4=c{|zVvgun^6W&ilm92QGw%>MC&bhO}JRhCTJ zIF)Fm9IZzHNQ$}#^l*iSQUN6ztE8~M{;?r;RST?|M4UI8U7YO(19#*5~!t8!rF$?#EFd z7JdwmAV+~>WuZl|?di}l+Ed=HBlsM!&K#4rU|EPBmNxl=S(?Z~v2Pp!LmM?<>93nTse%Xhx}{-h`FsU9*TIyqg~Z`zd=w#SvDcP9#Q@XuoofO5D* zz@rR(38WmZ#$>}aN!J99!x=a@Br;N{LVGitJHiVFalhBZ_f`E=zt>;%_frB*4?U$} znof15jk%C*l-JVf!ZctSH7&^NZZf|;J?qe%K6Ex62X^)MlR!A8z>fi~I|74RG@}*$ z7;&cI9(ax9fD48%RC3j?B+8>|Z4*jMhEMX&-Z>uTrw8%(?IcxuerzTorq-w%q~A^Y zODefkcFGqSzK-EL7{2Q~sERSZAI@G}fQ$MEmCY9afmZ3DRi8?{Cq|D6nM&H3}= z3_r{8^9=u5qf-f1lfNa569mu$0k{d_Q%%4@32=fVZYO?i{#E37{AfWHYg!k5ri9cG z;&k!If>qsz-W%cFw1TKb@_(URPLn+yye!et_2KJ-mUiwj zD}|F2|*`)s6D3$vPXuudPjM|KqD9G(_$G z%#}d|iNAn)fKMa(+MZ3ryDNwf>_-^?;r*k&@;*^7wH>?pmm`iS=3yz&pPR2Y2!*7M z{~Z?vjgiLRug@ELxyWOrzYY*1B3i~#mFN#eG1-hEW5BC+9#D|2ufU=7IGBb3dOSjx zZQ&PgEcHVpAyI773~!l5xg#8i4pc{ES=EQfc@w+}-$Jkv?1Zmk7|9EibojI!`pv*y zD2W~0SczOmr{gvPw_v;ARq8Iy^#0iYw?Cda%aIKOr0LR3DOz^ERHu}xrq1m&@e5zg z2~AHOds(h~+Wg!McIftPruaixK0Du3xV31co>Gh~CKg{cNle3%`=^{TPT6ZGrFtIv z3sY-PM`)3c49gA6yc|DF})DRtKC6`Dwm#DD|-MhW1dh4Ln z%bCvtGCFmOH@NAudXKsv_7Rs>zuZJj{5ONWg$^qnE)SCW`FWnoseQkZYvgFmK_7Cx z%UUU8GyWn_=6JMW47J70XftD6X~~_gAMd1fj7(HqxWY32*BS$0Chb9JK5dBBO+e2F z;EjJvRhcG9;zSe>h6p<_bO($Wmj_rQHhu3IwtWrqfP;D+;{&MBJM3l?3AqVnfNIry z1j``{n5eD8FqdP{3vE}%L0*!1xcV` zW177QJ<1lAaS5cRk`A$^S#QIuQ#+_z0!svu$3Iq!tY#q>f~2Yn%ptV~V#^dEL1kb$ z5D`_%fG|dSrsI-;3#utLy)=Z4Erx)RY=o1AffVYGa~Fr4PFk7bQl;L^EE4ML{rn45 zTP8|%p?2Wx0F*^Z7p~$=E|MzaR$59SeP;RP>5W1NH8E7;;!9Coqng$L6PKhku1prU zL&~sJv#D`ZS&(QpN2*KmN+m2Xx=Ze$n>TyLN~jLP6b+h~ zb2xtE-kX&p)3D>T7+T8N=EC+L?FPV_o}!!Ye_G!a!|1P5Iifd#q^(sZLj5bYVO61$ z;;UH{-aMF4sMOjnM{p^EN<|@}z;D165Z(JXgRgq`TZH;9Yzo9vHx~Lu^8*TCI6GiY zMFJ*hU9JHCftUO5g^8cF)E-ljTWT<6UJBj7LBL3Br}7{euB9br-B(flLErJ({!p)Q zFnjy8O$|Bn;T)-R%ezCujEipt|LPtN2w*E}3GoY(2!bmN`H`E}(Akpoa@m9@ADwF| zWR7h%LrID=bFPkcZ}|R~^q#s>1oa=Kg!P^iuE}gUhE(s~&h^J)3~+YNe(e=DDEfNf z7tGNu3uouF4?G~N{G4rYF7wC>@qP6;bo>{9_+P=&tPj#yvxp)UPhKAm%|26u9#l2z z{t@+2ne)p?5}Wx`94p5VjUP)3dd~d3s{nu_i3~ehH`kDp)lYX+%N#y7H}(g39PSC(g0P9g1Ws zTEXxgEfDGM(1$Mo&5>A!)i3bD&%UBGX9I`iw{*QbUnQa>1Zu zMhXb}fg;8Yr%8uq2W3p50%I($BBWz;1r3?0ECL2EePdFWX6Mr0QM`(xf*N`)UWSTL zad9I7X-ZaJYHyGPKzF8=zlMvz*C#^clZw!9%;N44)b}@VR725qaE)$Ugp)rS#f(5F z6KcGG$AXc1M6rfG1<{=_qR3ayn&7xPviu<46k$^19dc&bdoENUx-Ybg#lNH3G{;8iL;B^%eToCo1 z((NSxGy~j}S^$Y|Y}AXfoU+0gBB>{s;piovQlvyeI1hRdD%{}wKA3|*l{T{w`e9a& zYHoD)FTNAHMFgiDBbK2+QQvpW3A#V$KSPzqSd$KMXu}~A@1*`mp785c**&(u!_((k%83!6U2-YN7hbuA_L0BLuBk{;Zmf)aJkWq^oA(A%n z#iCKjqyQR9^*WU4^}3r>B`fh3$JG{8D^r}qUv&Uk8qGRpco$xVd=#Ms)u>lLDnAvh zh1FKUe08`C$Kmn>%ok93y3@mNj;UvZaKgBG;Q~S;Vv^FbuZvLDr{I{{-cV+i*2Vnp_1<+H@6v@AoJLLxQsL*j7 zD8Rs16x^yu1g7T|Y0``}Q}P%~SRnvOL(P}TmnJU~C1QE9F%zgthi2QXy311jE!5v9 zVPTi+H_Tc9-}(6g3)`lU-wv*#Teb?!F8)bG>f9d8jAy&arj<@17e(~pT}SU!F8)HF zk~6E#Z#c_VP^a^*?-WVk_?OHsuPgvnNORwUw;aS9ebA&! zr|6#WKV5D7{WxsES;o)^q>4ott3fJIU>X>ku*UU00Fb`iIJQhvG;KHP!JfyS%n#&m z1NHMsIWcLm5#b@Wk-vFp{H%m~I)~oEAZI^7COrd8-6W^~jL7B~GmrrH0qf2!o`z=r z&)(w!y!|NgmoCrvYSu<rkWrl;HtA)7>)%If zW%*%Vl(r^BiB`Z;(=>@e79k#^zL^wuCI5x>yJsYE-Wh96@B0~l4>-9A-!{{+<0>L^ zYQa)?PVv(3gr72%wW12hgBBl@Z2Xnx0rwm!6)HqFnRcZr%|4NC>$+(y zfPRf>VL=#FlllBqy;j@Hh zqSiXgvT$V7st}U9VX1?fdb{CV7;ozi>D_#>^nV>Wf@6b?GHj&2Ob(Zwc1MpR==Md2 zo{5Au_9}WjYxD|qn-mioGK4LX2OX(=Uv$xmdgX6_qJPp@AJGg0OeFVf$Q^-4i%$rF z2yoLcoZOp_z2fqypibn~Lz5bIV-MWZCE9};jQdh#>@@mu2K>M*oK5r~D71mjweMM$ zd9{3|#1J~-zV1YYnFm;F%b0zEzwa1HZ+pu2O95Y?teD(0<8=3a{{ekYd0as^|Iv=v z;_h-_8*_z0qCrJ1c+;)M#2T8mQ;TbJi9qt*j6%Zz90Il4VZE7E23&$`)XJ6#c#t3x z6`h$foP)!ThJItP>5?o?T%(Mr3DiX0k3M1FQ7&iQIimR9v^TQSrxpaG44wJv$#6Nr zM}~DiPm91gZVQs2JQ3iMqbuBv2LwQ=@G#j)n79oj!1gxaN(*#6SyKNS_CS&Td$QNl z0_HdYcvfv^I?YWU{@v7d@FeF#+@H2ODPUIv3?K>|;SlkZz@gO9(@j!GQ7U3W$~QT1 z;}+Q$w`%`7GvS9+8-14CZ~c)hkpmkuA~fA^?3n1KqHt9bwlyUIp_$V z!z$q)KQ?uWZd(2d8d@1FhKoG@ zPXWL&Xc<1mTv>6n?}=n7>h4&S zP1@xJ3-aLEy@MU@ROo*R*bmj<^kYeDJ38dd8X71~d>=9uiWpGS-0Vt>#%VN1=a@J) zz)I{+zQ$(n*obcX2G@;+S){1_?quRy6GxMBcC0QXJqaP8F(fMgBUq1k;W@P9cEN^o z*2`e`Wls?5kQXs8Hx}{7p@wzf?Hvg4#&hzfa12?2j1AjOlO@o_jQ(Xcu|LSOkF%mo zeYHKl^&6CpeLNf-Ggc}fXs&zYFCk+hqB9t7e0tDI&>^&zN~Z6Mu%iT@*!Qm!&Mb!j zL@oe8*Wwew$9PFzR7|XY>ecj0!r_QY`(19@TkX`TB_k)}iT-es1M88&h#ij;^Q;>| z3ZMcfVF|1PD_+nWF?0@-WiN6n{a>lNrp167Y7wr|&2z0aL*J&|Dr3O2P^*o6Ha>2m`a)re+)Yd-3 zpJw{%tkBXmKg!i@%kU5{vbH_wKs$_1FaAhsrUID zg{~?uf9+2{NeSVqyDG~x^gp;+ZDcq4U4W2Q$$NZownAz1)3l?=YC+2QUn9^FG`}+! zabL4@@cqx+`l^V&f9(kk=(z$+%^yS)^9vR)#?Bsb;axC~(`!_}7pooDWM|v}^np~v z)tbHad+m;XvrK22yA%|KzBVCIxA@GT>AO08dz0-jCX7P1EBoYaS!a7eX@jmp_cDQZ3A`xMfou2- zv`KYjeq1IA{&w2?eKu7PZ_^XPIxQrE?}r|Nd&Dvs0&am7j*Ih0R^bo)G&`PeOV{$W zJSqPZiDV5mLZ>#?OqG{HxIp@84-)6Ap8etS>^~y*= z%{=O0(s$g8M&|78*SC*iY~c+6#*gB=%($3TN*IMJ-F zjn|Y2?x?gdh|*qV2MyiM8hHSx>kG7LcwMj}F7pznU02g>;r0;E2^K^*xk=mO?kqs^ z7c$iY^>incs7b!FyVdZG1o;7l2DK|=(fNomv>{##bAZd7JiG~APj_;bbcG4JuFZ@S zcfHO=!Hhl`Q|VW`Yu&`WHrt%$9fE(Il+iz^DfQbWt!g6ACYjonXoojfyk0k&IYYQaK{ z0E?r)&cqB0=jgHp{({9(X(dOi@G7E;rH4=;_1JLWl3b3vZ^!Tn;z$?9rFGq;g#CV} z^@^4CsB+i1NW;5kO|~}Ik5qsg-SF_*H2g<%utjXk<91`3=(Z6pIV`uz|AzRG2daQR zn1#OxOH@JXgk}=X0mcIefCtp!ZY0w5hQuPxC`3AkIlBC#hU!M`2|Xsl=bnP1@u$>_ z1V-`Y^RIsE!v<;Ip+%z{AKRW6NTOOBuKG>Av-zxmlJ zU;VRSE;@tOwO;u9j(K;i)k{hhrJ%iEjT$lAa5n`9T|grOmR!xhs?Y_N%Da<84`X9e zO}_JaoU!T7d8ACShQrjM`r)k|bIC}ZHhIlQzWyYK(fsdgCT7ojf1LSZOP1_}8 zMBIL($HFEuTaPSrTgGPuKgMIKmCwO55}MaD|Ht0Cv!v1Wr7 z;o1Z%eK4g%QY3fqrQ$#L*VocMF5-1%*s@j&pK8j)~M%z4M(=>ZgTz=iJq6zS!KG{Wc45r0X59KAZ4hJaq|@t?V59G#;mDz zRj>J2Yre}u=?w*1JYgnea`Y0GDgqzKSTYrJXo6YA5~^Ivd?3ow%M)u)wXjZq=t@B; zo3B>>Mubu$wZRBw7Fp8V68B&EG1?pFv=u8FK8IOg;W)~C|F#W|@qv7vW)~?b#a~YVNmH@$)GsW$ZK!Q`Ky@Z=`X9s&XNG)Q z-~K#KlT@c7gMn?DiV=m;&oo$rUHZ%!1BvPnCSNObA{V!cxkCtHbJ-+ix4j)!0>E7G zq-4^MNw4TgAKAsxS{}p!)!*5|J9S%k0sJRH{%F=2N~ zIS$L~?_&FgRnW4Juv;9zIbru4y753Jw{$W?*2WGdr}?}>)6t|<0@UmAqXnn{0f>Zj zq@;O}dB}W+l0_}hQ>hB-yMb@UaHsh$uU^>|st+q86S{**92^19d=czM)PeWE;0C8l z4MBIEDMO<3ZkFn|`6r{(mfz-zo{mIlA+wC%tEy7n>hC^YpMUOAJ|)_bPlF4Bes{IL z@&3&tNb2|=S|yN@w`)e}ojrw6&~GfuzjMQds;EMfwd}&&z#~B|cn{o;0A>>;6|)&P z5nbdrEtIGMqrJrboPD5N4>uxvwa*jbece#jKdRlM6;UjT#6IX8kp%~1i9_-& zgmXY#G37V}jNmVr*iAS0%8+c(vuIU-k$_Y}v09W|uhfM#Z&MN(5X(J1F)Z}1VPvPT%F3?Om8VJ`*4(Pn7L`oBVE(=@0 z85Zyix%1-=e9KXE`*Uo>liG;*;oulEd)-LI(r0aNTd%jd%>edN{{m^#XY(Sa#NFJ% zj=w6i+K<*KNNN7DFC@xV(J134C|`$sIopP1A(qrLd-dg%&-yUgwUY#Q$sKrgoKd>a zRiP~PI`MOnWS2N%yf*E>bGyf<^8vP!B@=!)P*U{B6t!?C={%b%eWP1q+RQdH1bx%q zjR{h)9(!MHE~Ne4Hin{8^~r~vKkA05<~W+p!187;yg886(QSB5Aq);&f0WYokm;4+ zWyb5i#XXcQhy+WaCv~tTjgu&)E5%GsE7$h>h6MyD{3a)-Gvw_CvEJ~mS{1u8>$xWAho+*1$9%i!}^I!Wn1 z-E;7nq_a^c$>jc=wSpYth8#Lmn6rrP9jcn82Sgv03|~iMv@nX6+9mv#MRt5j=?9)Q ztele>dxz$8gB>*ZzgdjdOVfHx5j`0>(xE#oKiGjkjkD_AydCZ(+%0+kIRFbb!~>an zVMd&Q71RhlhnYp(grMkBUD~oO*-k;wE7U{Y@HY5`=R(|qMJW4hWpYqvnqeq$Vn3Du zedV4V;v|WQG7j0(JE{HBb#;6?Q8CgE3;SxI6IJCYpfk1X+e{xVaL--L2ibfs+&e|K zi#5g`b=dcZR(Q$>n9RAtx_KSNQ$Bh$9RXzTkpw%%*E5{{mIh0;&~{|`HVmTR*QDa{ zAFo|#ACjfKh+H!Ox!nmpy2=CFsf>)_OPml{NA4*HUJ3R z2O)@?sS@8oIT3!hMVdz!yd{_vUp!U`iXjVlK72`)Ml9k9K4~~~eiiGOci3vAEvb1> zjFfhK&knOfS!%Y>vv5qBGkn9ta%^cC`tgT~`C0cHa;LbGX2~7GrN61Gru(_tzEbW# zPZxL?e1<1Tr?iGugB6Le!54usmXRyvUh{y=J4>^Eh#vcUn@W}wZI3u3sodLJlpTba z8AFt*6#?3miPnUve(=)|@`qu_wF`gCBv^_~5`4w6<^J_O$}oKquTovtVJK@q+V#iZ zei)gb3E(?t?197(!dLz@!!Dn?rqpaBlWX(a=JJjjA(u&jaGx7k6y}Neg!cf(FrW`V zfzKb=!nrDGb%aCc#Q2pDHQ;7HmUQF({ZQvOj&%m;O{W8n96hH9&qCvPS}CPcnQg3) zzfB?OR%EU=`?eK~wL-)DFatY+bzood$T`+?CWrt%fw9k#=uUz_*HS5XfvBU-@%QYn z<)FLcOF2FYp`$jhMz?9|lVHk1oGCtP2SST|C}c{+^`j*}u0ytE!D?*s#PAai459V) z@*f@3>?1ly{656)40A=H72rj6Dqw=|L8nee|nm?4#OFI zggt!oK3U|(LC$}h9*8umyIuYMafN0`j%_YTlOFsYOF6@a3x|(7KpjvC-2;ShM!9n` z?E>|J&$<)}?v2cE&nJ8E9pr<*iCuM**z9;WPmMDqQoXTW$eDK+0TnJbxtoV#kDJ%4 zTjLu(^?+|VHZB@Ii&R_dnfX~$*l}8pSa3V>EroifXK{iamSd5LsMvG$G*p>1rP=oQ zmzCKvnwW+c>T{oXH;HU+^6pb6w#BlWI%a@Sbl*pHy2qtP+P~Y|5m51yRbg*1wu7CC z;uNH|g=;je5YrSD>TEuwg4Y&l!N4bw3F!K#Kx@g%rbWoG$Q!6GLl%pM%aj{!7!$< zi|tqskZX7sjvOT@AV>;sDJyJN#>vF^$00cd_Rw^x9X3gF@J4SHk(AV=Bx4fFN~My- zbcz}ilU31C(zRg{BjH2>C!SRDu~SST(>13B?OFV1j?$HBeo@NXp@to9_|ZibG)Pk%h~-B$U0?kGX^OmSWPoihg<`(I`+5TDI;X8bRZ*eZ}3j(&KBFd6P7O@0U zfbqCu9Qx>p?&yF?cw+!=7(xPJj8jAhrEoc~YH%OXhS%Zp@Nx zhEt&e5`*1>Ta8EUXq0NOgc~56TC6i_VNe#wiJ{4mR-mRr><`ukpBb`p)Y`$tmsmd) zBN|{Iw`zas7 z@!9MBgk7AMC7nUjWC|%>B4mB7O2Qv|!RyTFHD-Ww-OpV}la0!|Z_1g6Z+Ii~&bFkj zNQL`?f;MIJouvP4(2)n)WGPzmL37b4tN-+nF9L=}Go9?M)$(kfd>=@|eOQ}>! zgEUL4bd)Z)S_UMj(`($MX%P#`s*18ITKcATr_32$a@5#WHCZz>SFB1^8<$0_ zIaP`**$T$VQ=n|wiosz~S%sO^5*Aua2&PU3sT3ow!YM7Zk;9k`=T2p@i2((97zslp zum!qWVh~R<+y*wn+Rby$)QSuJT(A?OfbU$lZ?>C811xD%@ljG zeCR5Io8yyom8uPzGAXLtWd`REpNsfKaEmzi#}$?!zeD8|2#!6jcoR!9=?vB8FpT9! zy;{_INxjx~U^TLld#!wPS zIdI}hW+bIV66D)2JI(q`#OvPqM~vjYwcwvhn&{-FFa$-rU8}k82d&&%t;K4r-FmIZ zMs50>-%M6z&Gj_iM5h{Srf{he4elx0d5QY#o~o0V*rWm=C5HUDVrC>h`;U+Y2{g~hS=F_20|p%h1uaw*o_D5 z`MjVFc_ADa>cgk43-7l!yxFGkeC(KsbNMpLgdvzo_~h=UCC-#hImOQ0E$7_Z{ae%N zaH0;*sj+9Jz73>gBYbT0ZIAO{(D2*5y48DzZ03${^{!pdKD<+#-}n|baoV@w)ZH8B z#VN zEZ~7A%LI+XL`}9?M2csUNr*&>em&9m5ylzeCMc8t!-UB3PnKzNR>ZH-cc-*M{yck( z(f{lpf2t$?@V!pMG|a(bbYMBU(1!u+6G2RgB@`%)>~hK>R2~JC5~!j$*|JpD`@iXw zH5Wbl(zno7*q}AG=a?tGrjuR0joVtg={T3T;BMR)htQI_f6sBD!zETMAkOr5zb*K0 zfqA>mIcC$kdmZq@wAsIslbXpv>n!Y~U-9Ya(&dp(TlFTVY@2n)x?#sX<9n8k+O(T} zvXN&bmVMaAiAiT2PtE>Zz#~)RB^lvy`9C(f934}{zKXP68H+s$fgNoJ)7>rd=!W>) zOsAplY-?2h=z)2)$2OdQJCo2iNp~B&U0m`@2S`tAWx1>?x3S(Tm#Z#ZO?^%E&yEzW zT|_gbd?(O8k=7^W1Y=q+&9r#;)8Ru$PFZQ%rfEobjyL-bZ|l56XWPiYF(br|>3vwL ziDd;Aa5=vtfpILNgj+eA3nGjmmh@#Pp>!r4B)Rahqoo=el)O&;Vr458?$AT zowT!UYD;+qs;a)eI1~MEBrJw%fG8UL)jXpgd<57OY zTG|;Qf`SX{kFNA+9b&t)%sWe6F49WJBIH#=jnOV3Vj&1^&b$bL|0t0;|;aUcgKCB++3&x zEjTMDqk-%|sHCca@k-99fh}y~c5kj>XMgp7u9OT%Hx12{Yz=omSL1o|HuKBN-c`L? z~%yEuX^jE8T4IlBFxv?VPsZlQH>ZZFsl$c(ZGL z+zhMjvdMOLm=F|k+yN&AP+OHGZ5msnN4rU`iV2{+q=F_2XrPQ7GVC=E7ex{Mzmckk zVhftoB35a!BSdOXwK^qpWFKvzrP{4YK|jCDy+CbjYaBivawG5I9t*M znr*&li)E_TsgH<(0-^>26Bdi96mG}0+?8wSV~;BwaKbs>iWntK$;MvNP=BrL7?`{f^o#-`$xCi$4o)VI-Wp`F>5lPVG$v^CZFfOYJk zEwvqXAF6NcE_L=8oGpke(rwJW=puH5nkXSknC7tTt5MyN91AFFL51>JIfX7!x0Y2& zH0#s~qHJ?Dt41w~t5+VSl+-#s#qRpopzZPE1jNF^Y;og$kEo`KI(mq)77IpfSaA`e zkOFckqls2}XrzAGCXDEGOc&nujkyTqudVXO*7{XTh%9e}_@}P2m7BUvn?HWXiARbg2=GhYxYI@1 zlk`0<_or{0sHH%@SJ*Ght(mkn{JU6MPurG-IpvkyV>qB==R2}l70oS&q~!MERvBEa zv^i}m+|pj|#-2tyf>F2CV)w+^D3cgAZGDQrkbhzBvJL| zU$Pu=))AAoT=uwS2U*EO=uX}JrZNzmh`-9JK#3+9{p(wF%h~f%*i=&;FhD9|3=vD@ z&{5)!Kk=l_Y71fPCJ}uhGYA9Ewp+=e^=ujBKP3@U+3vXi<_cD)VJkhDjl}fg` zB{n8-x!mmMk$q!d+bcE_ySUii?b~S+X%3kN%bD|=+!&cJ(%|N{wC6T~k?+mpZ4YxKG19No!4;I)l|ZL8p@fghFBt!t&8C zuB^nz7B!ubvrddk6ZUfH_d}#*1n8xpL!U01GBWR(AEy|w&_GxW}#nMt5z~KO@ z-;O)-rIn|He=@@=N5Y#z&JLXHgqur=kbcdA4k_ML zW5v(toJ$Lt7E>Y$E8@6fjXRx~V!1MQAVgs;{Ajhw+L~*y?pDS&lpzmzcx-H=9_t9l zHNhd$rv+x%VB`zQ5Vc zEXp7=qE}dDb-#mHd8!tuDg{?Ok+F)FC{2o5#0d+eKyoE2HdVUG3YF@$ zqMHh}s?|00A+0yU^^|F;P-hDfFo500Uf49)z&c7OhJl$An4sZS-8%yZGEVkd)atF7 zJW%=L-K9dC3=)ioibWg}3+Z2-5gWf#!S};!j5DcZCaLM zv+sna0g>IJGd{hdN+YQjnz7&2`9DUru>qcw55`2PCyRd7Mja(u5qTG^Io?(|dRegHS zjEed<46m<4f{=H?lUnA7!CBiZN02!;eG%~gW4JbY!w11Hu|IV9+d!_td0$lwuTRL? zQtDl{(_jp|RDO54e{0D+9@1nV^-yWN9fOsOB-9g2upL#B{gUte#RG9zL+WHKY+}=^ z{?}%M7ie<1RkyPNH{Em$h#)Dw5cHpLYVn@ACuFv|}&7-Kx?AK%QSgIN{ zpWZZjMkjbLa1#v&7d~LTp31aO-kc!a=u>W5$0PnuE9eIO#^cab_7l_x-c_k(m7bP+ zOBn7-=ErHa-$5_RexnXh|2bXawdBR`mYN1@ro?EgY6~;zmls#-yfV7&U;ba8(r@qS z{B>`&DrDPQ*=P73hVNte0frxXB_y7+fQzjpJYeR7uuNnlC~LT)Z8hn~PkkfF=w7}V zZ*Rx1BXz;(=UCVT7lN^m6Q1>?`JR9}+v_IxWIaP0&jzz6M>Ew>r0d3Vzh4$)jgS1`S zv_}U!Mz7l1#0 z0AYalA|_VAFc5}#k75)hE;LqW&KQ;aw5Z1Uv4pBY{D7KFy2VZ2#802?Z$lKg2ONSQ z4uJFJI5{p7mG zye{oMSKGwrTc*)lp*D4BOjiw>bK#x2&F_}+GkzL_|M>gK@z?&v@an%m7`_nW+@I@A zrl~a@=1a=L?zo$82X4Fl&>jBV+xe;cTEWw_^yR#!7aAc#~H)r3tx84yyNbZja`EozJSeB?PMo9>1T_SL$Ml=VPt|H=i-c#ly!hC@d-=C8JZ1k^z&ZloUf7HkMt_}JO;0IAVy);}C%?P) z&2*2*z9-jx04mT8yr4UHg9fyq10V1W$8km6DO6GA*o41+N^^|i8>?TWV`E?4R)F|m zLA6tmW1a;Xe!T^N{0YA!0QK+QX*Vv%o@EBS!pO3*6%$jd&Otc>glK}I7aizCH~KK- zbfb(h+J&rISm$LU7U{nSwEA3!h0k5Q$>iZC zZKw;sG`XuZwX4@{@p|W`>I$fH^{u{-vIV{B&+na?4d0!>g)}QRRPx)5DXJCVVz**} z_47^TO%b44Kyf!nl)+c9lac{C{%v?v?RXur9JT#VFQ+#0YNK{pJHo#7btuRruw%X@6wVU^UOGJcYgX>R9tQd#aE z{(DLGS1Se7eK{B~&G@4p@cvP*$h~ODEjLW@U-it|shf(FJ2kOTrt7 zb<)om1z*-8t;H!uJ9e;IQ};3|VhJnQ#a<5cGHH&oF&S{VzCwvAmxnoACC}SbP78z2 znPq*sH{;q@7&3B(0R<{+q`rQ^7&m|pqDW2)`6xy$`mh&07EnfQ(PR{@N{_7E?c=9x znO16#J{`PDqh+cwnJ#WjkR+&5d?s!h5JE}D*+mgkbqh=un^mETrPkW!Y<=b%v)l$| zM>Gi;In7ld!;Ecmb22ODaz93#ed{i6$FHl&m?x^{@)YP7YjXqjqaCHk>PA&n_G`SdeCK4iDq2HR@;Wf8Lf^Z`nbWB$4BvuT%c!1IBD*$4ukr0wkAaQL~k+cdmFPDDASk%(gHOm{GuJt_I$v#Ct3xtrN|dmKP<2>E)73uhj2UyvhGr70VrHfgHw|1I zAmVuAPcCHJOUg@BtkZ@>r6yC^kUOd_+7#Uc0jRVXXOj-6T;i43>4UIeL&uD&meaJP z1D(?aeZ{bSJ(q=ATkb3febq8nvYLsVx7Jw;<_DnkaFLR=)UH7VwL^YobAVFX*fah| zrX$I>H!ciD%g8HQEWg4!sY=zcI<$NP6p_Y6J4-c^LXsh2B1n2bbhX-c;@S%rOvtpQ zOsZst=*1!~DU`RtHz{+Bz%G=dqoE$UM6HVp7qXsXmIwtnSIR22(@Ezm1$BYA_}_Xs zOfbQN0zMqMKvUYPYg>vm5{RVx797=V*!2chamrz5IFK$$^3lTuwy%Qji&!JNaz3XAq@A5n;17w4;kbXe^qw z>uuBI)fAm#-3Ivc7yZqkiz9>l{c{Vn$CQ>F_CKPLw?5YF(4-j?&{>Q?hg)(H`0eLj$nL zF~-H93=$6e5@tIFNrIzcQ3u1cOqtcPDk`tXU?b{OS$vLSC2C20d)XXcD)rtX8^a%?&tz%b?} z!sHA~FJy$I57-lqP?zYDWrOpIn6I&n6d3g^2mdM)$$pt%8xrx4)$*JuT z)*suCX3eSs_jKQmQ-+bO+`y<_@n17dxR%sa?Q<9N#uTO~C6Gd7W@E09hbKc-`m9q| z15uX4{2+$Rg~CRdC|B|07%|TyaYZ2F=ZOJhgqom3RjC%b)iRU;Va5Z8fh*J`ovK>3 z(xd8z1xr6LI9_-^&T*B2#$Y}Q*C-2v$pTcp=7p%}GHe4&ea6j*6{sq$=74B6dQ6|~ zo+r@_dro`VXfnr|PPc4yK~hRcw@vItY~kSj(plIcK&i04UV}Aq+|oU|qX@+*iQ*Qn z$bP31d%Q1S^10crIGu<8 zD{b4Vn|H@q&MLHHgoMw?rd_s!bzS{rcT6ibSSb#sNniebt@HZ>KJcj+(wEdc6lTe| z29IS5gUr#!=kSVp&0WhzBeE?Y@iczS6MQv=?$OKk+^F$Wrms)u&A)-IT>HRM@C5iH z*bVl8@#C!HPTnglxzxV=tEdF6!jrP9PYV~jSL}{EdnG== zj>{gWEjD?;OipGmUre-1I!6N1A!`aX&~$E>%jyY2#B&mn9t z?Ap-r!yB^w$PPZvlsVVY~AYhl>6s+H1N;mePVi@197OZ1>bzmR{)wBpALrWRR4+g$(D7Zo*NTwjjE z1GIEHPR$uO{PAuphq#Cj22TCu1*T1S7jV8RhG>6J556l&68!lY%t5>YrJy*i6^ zR@a1nyrAN!V?AC;xnYscVuNE0xVFs|SF^+QRNuMl*`O`9vx}`~aT~L)y*=aCoCN`5 zmLG~dG(sCprGPT(Xs3%|Zn)=>-*XWR`Bk_5;qbP9JaFF={%zwv*o?adIJd=2@U^PW zP|CKrHLJ9@V;%4O_AQOhRYBFYHjyo~>2mByIjWTwN_(ZZGHa|D*T0}3)0lnG}nw>q%3k5Rg1>O@M3mxWKdh!emc>-)0q}ac*FYKhv&P_J)ZEa z_k8ZVsap?c5D7qJ1-h^bIjTN81oyoRje<{us=89cA<2;nsZShrDWJu6v9B8Y4;$Bj z?#ukKOS?tdpvy=5n%wtOO+V{g%hnZ&4==yabJ>mT@+a&U$l$sMY~v+o`qSet1YW)DWsC~a!`)T+f~o1X|E0!goa5K z#i&IG`f(Kkf{YTzIZU|7naduIae-UmMwuS^S?CrLS+Pgi>tibps@A$STgGN>y|g;( zs<+{0i@bgtWuD%`7PIUcYEW(Bzunmu810<#-Vn86jyF)&0f~?A)|=e1dpK^6d*bo< z;$QLSp7%3%>V|LXt}BsLRJUNshF5sCw|HCDkE=62=<~iRN8pVCtFIN)?^&-H->tT1 zTgQA*13xTHfj0pxYk^IVqHs3^C-$oz1PSHGXQ-fo8pesU%zo2i3QegAo7ImMuA-Si zriidWhD~wGd6|H6_r>LCu^wBnTQb=cR9Wt|opSos->`LAvVBK=>A8W#bRK=i zIc~9ct7xW+o=1LCT4koW7Tau_F~@wcUv$o#Yl#o>`Y3A|~yITzr0f@ene1c;HN#YdPVAOac<3+z;Id+|Qa(q|F~PmvN~ zwVL$l7q{1vmNQ=I3_j-Ng_|d0CKh5b+R%y&J16Y$`~-IQE<8vjpJJG3aT6fXLnq$L z{V#`!p@{xE8PTjq7!&?=FAotWL_4$S-K_D7HQ$_$Yt6)!IxeMD(rB_~?ar!b=vYy( z8^zA(onHunH=AiKw5~06bYAZ>VOG`Zq?5nEu`7OwTd}zrZ?A*xwZ4om_l@0bvuuH_vbDC; z4qxh}-T9|TsAzZ~p@zaz=Cm%(+|P0aCAIl~!0awh;UM`mFlGqH+L*!!0!2;TO6rL6 zRjw*^YXVAb&{t#{dq~7=+0j5k?wPV}4vcyw_P7J<_Ci`2DH;})oINM?ce#gr_EA^7 z{@e8G&ba%ytrVgL7;_5TK_HX?E;aRWvWpITd zL59)Zs8lqk(a3~C#Y!i<_dD{?mhjf{w9LGMus*y|?k{E-EGjN3O$Dv=GsqV918@YY zVwGB;AT>RZU2b&MVxbueRu79tG-I%6P-u8k>cjf8G%nwm&X6iGCE)U=iP^4Jy&Bbm ztAD3a=W;j_bIa!}S~mK|&DtNw+}p+;4h&lHIe&t&LBn( zew%%~RHhq&?=o(7HBx ziNCOym!-r3bx>uZ?R;xmg+YW`rl0w$X1#bW+KPC9q3sR6F|YS&s?xj`dsZns>R$Oq^+B8F{?poBOmszNJu*ex3GwJoIR% zcg~+#rx1d~ZkWxnFW25GYKbypPyR(~rB)iA{FiiU}u95|ujDm>;(C-_t)~`y<(ZUYYv@Z=2_>MI?q$vby03Q@8-}_Xyup*HhlyB9 z;o8}vvgK>G*tzU*f}6|eSd9jX4g+IEjE`wC6XwOHsEUTs1iA&)I~l##qm4KxIslnC zQju6rRb+kH99R5c4VVE(vA@r8EYqCuQ9ww!R%31(~oj<6EEOqLM^T9CUHNsBLdL# z&vONDVQy;MxA>+uw+}>p%U|1F*-N;rYfc=;jq&HZm5>;jIbmJc5Yo4)ZJ0n~nny4C zFp^1Vsiu|5&T*d#_O%IaN-`KDG0!qB=5Kgq^iWxBmL6+kqimc_vIFcn_JMU|omv+J zgn$!9!by0E7uU`ebN9%L&Bu-`7EM&42_u-m5>}xxmt{fw7{il|88_ujXOiQ*FqWWM zj9-V6mD+ESU0h+YhMH-vg>rjVrt)Y_Ttv6a%*_guf-#swEL4Wt(AI0)dwj2VZ`DbG z9y)TE2N5^{Gwj%pfK5$>iJ5A1Udux(uP}G&q`n`1Xd*me2_YNMMz zlvG38%I&-|f!8YwwtBX4&iZx71Omw$css`$m2AxdYx(|CxAVZlCYQ}<(hoQQ8vy@yl~pXe)n_#^4_yO5GttO;j)$M zv{Y3oEKo9bZ>d2W1xPQuJVq*8s6+$jJKyVk)UX{!oz==KW+2Iw;~|SN6b~m6vu#yd z*?jAmwrCR6)nG3$nFDwS77%6#fl#=HuCkskspB7%8O5J4~mg7Wuv;^VsWw0xFr#a@zG-x9gKR2c0lgLu8cv+C^4aib}NNACT2-;&d)STwJQZZ@>f*ba0{zxkLk+ zr^Q;Q{W{6CEMiM5m!d|rW05L5Cuf`#FOOX@nG1=GhZql`6nk*hXV_%P3D-%Ol*vg$ zW^B8<=5C9p5hQ9ti6xQRu#741q6CU9pK}3VGH?KMjyE6TXy<*0oO9I=TT_)P2w2on z83u%e#eo-{AdL$3Yeb7$*1GP(pSGSc0~2O0zuL+M&#;GZxqZqGN$AmU*gQf=p^fpF zt)J3ZbCZhHMAFYD$8tQUv>?>Gtzmbi8GA0%Gs)VsV)U_Qe3FwP57klssM9#&DL+Ho zz!7fZ&P^}B@~YZds&aBQQ+`F2R9cS8X%Q*G*^WA8%B)qda=>vXy%ix8p#}ry?WH?K zk#eu=Xj^$BHQQILRKH>XMw%*c5X^806dRL14A@EGL-;{7pdS+~NMoATHOgvlL2he{ z_ML|1JM9kW%lzxi*gLdRLy0*QGaVBe`?c?bAA}e| z;HdjF$J6FC8`5V-148yjP*kG{Zkokl$@S=M#=?3}DaN93YA`#M&tvILJJ1)RYMRCR z#K`#|S9_wTJ6vA-_*hWrFq)My4gO_^nl3dhM+^shFgSe^!SiHbnUa_Y!|GdC*?`px zqd~z&D~JplWd;ygOpXc_NQ_p^5Ktshkg#wm(7=eshhc=4iG&j1T9xMSc#=ZKrbLk@ zCVc1wn=8UNHO`!Vsd=5)f<`vFg^g)tvtRmY{pi;}qp4U|SPx@3{~GfwFW{kaZC`~s z3=r>yvBF7lk@SFM>E`&J$Z=4D$j~28$6YFh4q|XCC!9Nqp7=BTZKvN4GpuMJ|ApVy z3bv(d0U z;lu(5jgl}~drGyz3fo*+Ql-^aU!=h%nrOaA=_=K0)^4TO4ByrqKcDOsA60uTtJ|8@ zO>b|9JAM}~YjgT@fY_K&k};tY3Q-SD&>SrhElA{kFP|w6cw-LA%~Ig%x6km8!7^tk0I22bwUCnAio4Udxq-TZjh>G!q zVwsx=pbLS3h9!$aA3Vr{gKn&c4}hfiGqfr4G!-l-j?~FgSqijj7dK$kta&{vvnQ9C z7Du1LOl5lh(5m9>qs)l+5W;^i_Cgl-(74bsRU77!su zB_nB;?pO+|E>$#irE)g7u?g=0PCOCv30ZijE zq@0z=TZs}UMUEon2osw@*0G7*OcC4&JF>u0M;Aka$p%WbL}@csn{UnP&8pG4U{@BZ zy1Lqmwzs8c0Ah+128*jO#Z;dKiFqWbR5g8TxC6YlFP294J9eL6fZqOE4v`u#K zl!u)2+Ae?4X_OR=DUm51Cp3#FMw=|M$3c%;v|`VdU}0&K%D%z*|3T9Gf`q?YI3P!46|55 z0M{c8lGc!!ZlUqgmD{jMi&)WMSIVyee`3*j913XCW2=0Fr(wa-hK9m~l@u9@FsRfa z28wRfExAgy*=m7$;$GKu&MOf$VL&+ENlP`%zzUS06q^X895^o{4M&Ajxr&>pWE2pd zTGfV4RH|D~i`8QJ46RG?!bLl+Yanjgx@UT>7dvse_t;>Ioenzew6N7P_H&w+O|_QH zqb3p}$#4;4rj7^_NWlTIiA%x_Kk^u2Pa#bBh*1&*hE{8CLHWw1mm7%3>*ltil9VloOEHZ`YZ?WpQ`BKj?eXl(KLkY$#g zxM^0f9>SzD?aTp}CHqLi3Fsw=7x0U#A9?R*$A-!Cavf@@qzavSJN#si2K08}X1yDUD2&%c@$^2zhF15CAYENmRW- z!$!^eW7MQYv%mzQ;xF8`x#bfeX(rim_Fa$}6k8lr2Yr z3dlks%211TbfO=l!I5T6Uh8!jPl_2k7izHF+BL3?Ist$p&_uF87lh^RD%;v<>UfMI zQ&p7+7AZzvg%wpw#j6apOWl*EI?_0>sN`t^f?$Gxg~x>tod|JS5Q+qppbXbBaHzUt z`e^`5tj0uC)OAfdw9&BJJ_ku8rkCAx*Han86`i+GIH740QBsVT^3;S+I5};!KfAQl zX*HLqwRpuQnr8rGxRb{UHeaTCy@n;sqdU1XMwtK;XA(?K-7sK~AP4+QZeSx@+5T=q zRlOFS5eUNIkyw_|F*=zfG%f)dE7!L2(qkvhn7!)22^+AXO|VG=R~S5L^Y&PFsiKOl z$YY+Uif}zS`Nd1dC1q8t78sgZv^4)^h-nr$k*o7+v;e5*RhO9}$0&f99Y!yWB(7kk=`vJoHH9M|&9GRtXOQL;5%F2&rEd`NPL45t|@ju`iV&5k;S z(*UYbk0x{(aLxteW-NOql_Zjk9J3M8JsfBDx&ElBRg*;`8A5iB(^l~k?bCpUG^%l} zfQR%%qXS9;3LTRc7*=*!k(F4y@sV$`?&MR*bo*~p1rs!6^pU}42m%%-A(Fw5e-!D0 z2R?+6L>cYq!2%6Exr7;6U26hTMm;a8nq0nOjfArNkPaiPVjZVsb@k)tUieN>RB_d* zS!1OFLkliBxM0ZQBd1PVapkpU95vT`i!8o28?|Yx3o;L0RhI?O+87%JbRDpV#|@qm zPADn618bGo$gP?(Vtzu{F$_dgeBgJ_G`jorXEGL;=2B3_FpG$P5R`^~GBHH%%NdCP z-*Oh4ILCKPn!~`K2VT!R&mVi8-oRZDuDBY1Ega}l9z))tz^Su%+sC#HWGtF-aylCE zY1KRw6lK7VSjdzxYBCm4`JSRqxa9iZp5+YH`s5FP`Ns?N+!$SEJYnc>UfH8eI7agZ zSGwLgf4gI7HW_g2i3_%bZ@c>2^he5d^78%yKvY4D-VqLSwPo95b%Kuo4$n|MfFjpZ z9Jh^m2j&ErrQdS)?dAAYam5HIKsg%U|}i@_`u*B+~~ zO^po2Yc=1^u~sNcQ~bhFhEdvS(0Xo2t~7gnY-Q#HTzCx5ooN-P!+O z_*@N{_nu+#31@@C;Z%KTh^?0O9~n5Y#i%VPS*;cE^mpBbeGQV9%ug)D04@M`N zhD$ae=}VlZW|s<=+-Z1)aLOYZP9VPHwvz>&kX{n-q?psd3H!%%_-94%#IZDi$;@u= zk0Hgp8ka2XZ9?-+*6NkKkXp!c|1}No@P98Wg~e=vXG!4s4Iwf5IOuDHRaO|L%MAbj zFtr|?@yOX7n85VZbmxf9yWNGiy5@(Q+T^dh-sewu*XjO?xA2^iNsj2vz)UZ(UQ)9^ z&OGjs+YCcuF>3e286qMgV&o;bM7K62A|hsvRfCzCMMOkIM2s=U7-Ni4RaI40wboi| zt+m!U=bUqHJ&a2}F8whyTS#t9F{xFztC+HtwvF_(UTQ7=0FvYJ&j0>Mv8Sx#z_L96 z0001(;UV=kU+`*fKmGPfV@q?3ZA)VIZRNCYlE8E%W^V6(8EIxIv&zEH7H0J@N(~?Q zGq+}56AaTVlC@r5K;aKWwVZIchFL9ZA5Jjpq{tS}$zUrN6MRICfjGeuoW| z{fDX421FsKZ*8dBwT{Zq@3;G}-Y?a`q2DTTPN^CmO&GywWm9cyfgkIZI3(hb*0tOl zT5wdm0vLNip$(I5gV6VbXF9b_i{=}hzQWAPKP$S*Co zM)3}sJl5V8l_dq1x^-mb0v?TbQSU89fzif99=4Ww^1fdFr&)s}iuP`;upu+p>UtY* zW{ZqgXU}%jyVK@8@;E{H+i-LB7k>&Iwy$pRt1D~6&7pe_R!d;+{BbDI*Pf&09xv6C zL8!Qv-@n`NwnW7u$c&h<@+EudF&mXoYSgNSg{%KyI}DYM1&=@ZzHRjPHC23odo8;! zhOX3n?Z(i@{|m#$usbn&TP;{EZtbShJED0{9RKjfr?^L!I8-<|M`ZnFI4{D>uAD>#^@sT5-DlOj+#Im$EO4-h0NME`4a*hu5_wTa@+DYTtv|vS}L|N z*`|LY0iudGE}XEQU&n^Xw-H5bPgeG7(sr;9iJJpE6*@78l~Ge#MDf6gLUKr=JqL~1 z;o&yEq^o7$@T4a5%tV!v8$if+W0488rS1fB;{7W_g+4DUhbXjOA}zk0Q!Z3P@|RPHotq zGH3<|J61DVg;Z3Ph)PZ~Yj{)URH4geEq9fpMeT~6aGb|&O&X+QVsxlsiV+(O<-!r8 z!@-voLtxhiQY|TGC2$3{YP+&~5gM;{`k`mfNT=UC5#<&F?wcrQ=qSAd7#C$dZ6V+;M>*v;i zy_5totP^2)P~wO%napyMT{k%IZA8E^--Ztxk-2tE^TSfXXta`0=i7tBJh!py#TZLY zXYRJ45``oqTxNUQoFnvL@u{^9l7wFGD72Pz~+ znN(&@Pr29rr!G?C1Z`xy8@Kj1K_f!Tj_9TdM z7`5fz2t}Y?LDzF_($IQcC^1n(?Ol6nWhhui$ zPj-thhguty1~$QC;Qw>dDka{9Lqn6Tsm84O3jlo>^(DRo26(Y%@Zc3_hjtlfZ&GZhY2aMmB>!% z;?lTevbNhs8~bZ7-jN1f)i(8M<{rzJY!yXwMlhfnCA0vbQCHDGQ)641QCzlpHyxSAx@Di+8DasI(Ef18Kp zRdC+Iw1(Ctatpr@Zsv3nmy?_@1TI|ukMnQ}zM-m@T3KdhsTA$@MTJ6~%_YS32?m*9 zed+b+!0DOkn_e^jkCJyKtxTLZKwgZp=@jnthTDQ6+&2OsgJG))&djV;{;4@@-)3k_ zLxekhJjk|cgAHWhYZ7cnesU2H%G8HMcgPj5_ZgFtqJ5N&*2bCE5z_Eou`9qYDBa6($mC8Ju5p@M>vBlp&X3likO znONyL+Z3C@5T?zFY|yS6H>05jc6z7kEhao18m^}b-h^4}3zOF>kBMm4gbB;0r{?vPsCM12BT)Az94lWnR}Zt%C$fgj+;;6 z9w9sJO_c-PBB`U}0feiIZ0d`3wIuW0mJ>0)*WGfLz)aKk6*N4+2_BFjf(#X_OX^H{ z%QiT)_p?+oXE#lu0AM0J$63rdA(SdWd+Sp%QA^jb)CAKC)+y7?HnKCH>jg-`2Gugz z+Q*jBE7vZw_X(+MIlJnT@3_($UBzapb>g@((fb^frM96`5AJaUd0TwQ4v_0Amym|K zgm6&O-6=^K#1f3FF`bc5H$#uJSm?%nwgO$wKWm^ulUlBNdD7twd=Hp&GRs@ zb1{qSQ*_ReLoc*+>88&p^9!4z-`2#XeL916Q@Ni;TtTh520nW3zFRnw+?3K=T<1o+ zDc4jyrUrUJm=+Bgr)?DbEsw-r&ygXoOtGum2KSbJ&gG=jWcvP|ZmMo?o z=yz)wajTVQHbL|ohUsf<>*U3oPK*oMROL&=?mZ4D!PscO=n~eNtVaRn{=l;8?~Ya| zf7}ysl7kCX;pPPNIxAK+t>MHE#0MbcU-V!p?(8Z^NW)H3h{j2trx>E=!BjO|>*TU^4e4ey)we`DGne*;h7emVXUF7eCE#Sfc`5+z` zBO3h~^O^pepY9JPlzt8@U@V<%sH|6Px_XP5RQsovowMt}+Q+&awxU(~WtMETjkY8UfQNE?y`zKh}#}$;JxJ9PfTGWrNDK_P->-7rsZ5tRaEs= zRW0dqafdTHYNwOnS;h;gP6T;b-#^c) zpXHAoi@&9@(TJL5EyO>&!I+2 z#!|jeU*hemDRgfz?>qgU?VU+XFe}C7F)}bLd246l-JRl}4 zuo5%~B^ZT-EVn6YU;#M&KGhM=m0hD`WF&1Yd_#)3? z0#ex)LGZa#KbN4MfmjLwKy(X7!EKvJUK5D3z=o;oA!%q5M?1SoJ4FaV=Wz`#AX`vv zUD6Zl;DY^@FLwtOpHRp_2fx9Rw_4#b@DeJxw$PR#kfq?(@Ebz0=8{8To^U`m0*-9p zT^W?%AZ1D<;H)|e#*$&#;pdQVKr)WKwC38 zP*e68Fh8TR=W8jy`(YD66^6H~8*|fd@&AF%&lr=mnwD_<#vEE-xT$z{>us*CuA_sk zXa76NjN*`pBO#G7_7~xjGwPPY%F$N`{&Fte?snrz2SHp+7%U% z<5f$C`pj(FX^d%7thPR$^MkctFY4w#hvJ1^w=v=xXRtw)3AU+`8TXt|;bY21z(#q! zq+wGB1@AqID1~T{#eO>wT{y756NW>KIK}khTLrX_b!B2*GmnA)L zlT>q&Yjg%&kUxq(kV#*3)i2ZByJ2`h^s_WG>jeQXDKGP9pJ(+M%dI=1{k^NASt9^M zW$(@)(q{J3bs8pTO&@tfY6lsAnnJ%>YTNI+VxM4!ZV zY#t}2I7S&1{z&&u5o}293V;?MLph_>(6={ZJ(Izh-_4qca-2?ubolWCJC}oZ)g}Vu zDb!8iGZW^t@9rSK)&<^ZwkUrneuij(7_GN0bWy9E%janR#wx-q|H)F7L$y$Kmt?f1 zri_!1)qLY_=~Z||R$LCdZC-YBa9at0j-k$oE|4Cto}Tv{1CBMQV#AX?f}nDGH*Mr1ZeTSeT zDej#0V<1B(xi{e&>m7}s&9dq&C^q>5kXVe`xP&_Z0000WA|fIp5)ly*5iv6}Gc${b zh=_<7V~jDz7^AAHs;X+Owboi|t#i&f=bVc>+gnc%5{pqgCYayG00000L_|bHL?R+0 zA|hsHW@ZkTf8=#B{K&?0k&eXpjXNyR$coD-rV-h7UCmOQ>&^WZXtt|p-#<@KQ|Q4m zf-+&BLDzB?EQ_AkUUgbk!GYkk*{7vru76H=#aOVP5j%YR8A3 z{_P*E4$1~LK7cfNip^wwFaXx7U8kP9_0|UiN?rJu9iwUp z^BB!z_Jv)JI$;sQ=bQ(!X4wtvsJ6}K^?;v}jYU6rYQ`nsUthwYxdTZt*OPgIdE5$8 zff-vD!34l69r3U;{L;u!!7_oy_|hf0flK@fqGpil*ND*HeF;BX>^>pE7aWjM;`VZq zbMK8r?S7W*d(d+C9T{!+(xr1w@z8Ua=fzZ>N@ld>rAMJhM|6PNu-)hn9H9N*7s4<0 zje9D>j0X_;q)cLDDZI^b+D9G46$cUmdmJD#Q&6^=AYzVJmh6WD6$UhZXK2rZdD#`@ zzj6q@a#4p`bDVA0dyZ5cI>tM~#K>*~EIs%E$fSDOL3Hm_dt-9|L+y#UBI{iL3Hi6f zd9{#_sbeIRHpcAgcrr+8+nAqnKQ=FCpE(E?}rben$N;5Pc{0*y0o0=R{7I*akPYNM!_ zD8*Iags)8D^XL|q%{4!ZSkrr2a16wFW}?KpN0=aPyTBUua+6j|>5KrLWQC_(jGIHP zv*J1jtwe%#JDx&MR@2tPB4=}pET+S>JmGWX=8LLQ=FGdXg~@)Aw(_<5-L>9d(;YHG zJWX%TG&0Krb!*18y{+l$-pp6rRtoBGK3Or)W9D6A1qmf1ii<~7^(I^=y36o}*=(Vj zIhcCMA)ZN#7vgirnd4p7%DYt^Md5XZEti$-7%}o7DV*~vT$-u&Dl$=F0kfZmZv~@U zml$hA{O}lw9DN^7a25D?Ci>NUXm<-^YrX}VC1{rmj z<4>*W<-ksycuk|%5@sYuL278&OB_l_`ZD+JklP7@eHUr!KLN!6NTbU8*~LrZtZXd! z@Yl7sHcI@=UUn=-sy7BuMS#(Dr|A%tOd4n}NV!fm)eigyYv>W~fU5pj$V?_;83COW zLyodU5HVN{HMl4Z>%sD)p% zLZ7v8v4UH=SfLdNeSczMl4esbe;sD_($eMcq=bruoJci0&KvbLch@cz|DYDPB(;Jg z4;|c>X}!v1zaUADtnx>`W6LJ(N|-}_Yru?YIteUfi<^K{9sm5BkbJLe&4y}4%cS2TWLKl zybs$EEX!EYWRz%fi3&jBXHoCiIansQP|w9u4WMo5gbx?r4DDS`82!WERIPKH64)IMY)3W}kNea{INjm#s$Y0lFo?GDM4l8Ty%}i!JQD2>{4$apxr$S0kL&fQS}% z=%`zrtpg#s)(+OJ?Abfl{YTU%wuT8&ivW*Oh4}X#K5D&cLL~)WGI{!$&AH`{h&DAl zPAcCl&1+12hbvrJ;YY*75q0kCX9SkcY7klyBeZvpD}c?e*VI z@P4@-(#-FI9ypKr5zqvaxn5^IXSLAoB^&~7>B|0-SM!z=Zm<lyaEe{ z;Ul;VU%(6nuVsU;iy#50(}Z9s-bRvvQ=nFH`?+)seqix4L;_+6PnWG_j2ZSh;ws?* zkNHo_#9euamtId`7Jq5w1&>gP^<%8 zp&N8W1zn2DQ>;>(vIEy|b?8y2oPPf8nazfbunYd}%Dr*_?>rtW#Yh23!tzp(9K@O^ z3q!PoB|H%dA&zHz`8Rp)TbYvuX-JH8M*ULg(RhF_vUOBjs;*9`hMMTFTFMfCjQ-s{ zv@pW|2qB=cL=zoMaZNQwtOSWevdTPkl;}BH!)!Zm$53pJjB!XgicQ6(_k85QulU3_ z{)AMob}XpVyUknmr!WZ1aQ!v@=Ewif=hMjgO@@rxa^k@JMXPpXjevxl35y(Z%loK` zM%ny$~O%h zxnq0mjUWAzTCA5s5_&ZrzfONo0fJ`9g;zuXP^mJ;fl$#?aby5aK(W7zIyMO!*y8Ic zHeR+b+J0O{affm(8FH%1?Cr^`>HUu<`fdpHa3DvqfXXFPV8r~)7MeW4omPwYj;(Gi zJ>sLC@6*fDm;8OF5c8?sGg<%)_` z;qc*C`YZi4BtYq3bsd!Yky2%i9OO}+=4G@viPL9& za+DaAYvrrf?GWCob#+}Qy4r)D^kc+ieYPd_uPa9%Ql_577spIv*X;8&p2fiA`}=XL%}%! z*a(rJzTWqEM&Kh^oLt{-g(^9_<`)~viR4{J2&Gx%$F|URnajXdk&wA7` z=VjDJeKbY^Iy=iGgI8R2CvCXt(p~Iy(PbB|y5^=^Z-+CugmB3e`d@iGp&K zs?bgyU87rcLVXJ9R7jRmHR>dEe&QKdzVhmuAAaodC!TfTC0E=f+hAiBGR0;G=U#A^ z7OXh5XvMk%7jC>UX~w)2JHFx@yvY09;bRUt;jv|d4ck65ip@Y9m$aR7P#99p$J~s-{|Mq~>a;KfC4b@bn`4vG(;` z=XFWfbWe|?KX?526Azws4^V-D0eoOgA*7K*34P45!Vd2Up@g*TM30j!W$LuS!iJ9+ zD?ySJIk%$KtYZUnZEU_Nn_0?oRrkssqn53E7FNPWFhd;Lp%*4$9roc~^v1V0Km7FZ z&wl;do9}%3<#(|X%}Aq+IzEc$Po7L%UK>AU-omAuZ5Y1Qq^UEPu3pCs^W3IElWn?0 z)I7x+Tb%PWsAw?TRJ3$WrFIUT%DNWn7+Sis)26C2WXe?_D6YBkDygQv##(Eqli1CQ z9F^w5gvmwBw1#*h$Ye_u8aEVZGGzlCxS~vp5p#B&xEY$Z>uS9@TdcQdvCMFC`_hbM z2hKh7){-^HuHE@}>jrp)NDwh$aE`|iJhrIuMo&CV*+?;anSu-|Fi4;V4r_2&g_d4d ztk7Hla`RFUZ zi*lP4u4)L2j7iGOX;+_2R?3jn~aFP>76uL;R zP+OQ;SkYJ%YFUop=D zzfmP2+_&V|&D=}{uX1k{b zwj~>PzblA?Y}a8X|3zlP`K3_p<^w}2m8T%7_u1nIix){C_F2wAPvdxEC`OU^uI+*0 zZ?6{D=RAMVD(PIU3U*@K&Xu7n8>P)`-u+Ar>>B-ICcM#2oWtKfP+sj5{Gy(-R1%Y! zXW2D6Z0qxdyp`VFG5Zg)36Jv1#;%nX^ZdZY%Y}j7H~#kh;>YW(qb+q;I4AHtW_w^Y z{9v?*@1>X2eNq2p42!JFClWPrcrNhpb+Qs5dDPyR%Kvc9dGtIa`qZMVcK_1Cl;5Y) zE5O*^z`dRNkS`%qqKO#jI>rXT7tJvuQbN9y$Rco&ecJgVo7I>*)pJAH>Nkj1yPwA( z9a~cEe|xaa@rgLV$D;g0h&WR5@mVatDX{j3PMWPH;2Vlg8u`WDSC;>C?7#e4`=rHa3g0~1Z(I5P#<}vt6sU;4H?8toUq1**+at0_7(vE5 z8U>pMCgqgd9k^ZYAW$WZ!=r%e?J{d(sQ@0X6*ue01A|iT!;Zs_#2)Va zdGk!FQ6Sb>XY87kZcu4xwD>9P_*BA_cCju);_N}Pyvw>u(Du>EgefqJ6}D)ehyb3E z$tZ1=G$7%9fDEzrpxJt@ONe-hZL^vBPjQ%! z^#q8lhi4lgEO|bU&jB4Q+AmDd>7Yk8675*3xMfi`7>K?o(26FE$J`>o%4oo7Ou}GP z5itf+3klsZ0R7Piy-{BqHzABwn;&oBgbQ{()GJuwg&%HsV1WZZ*gi-pGCfC0G47^x zIL%>HTB{P0sc3v8fJ&dRM2)`LoC|cbfzaQE)&O)R@961$eRoO#C~WhgqsL8HymI$g z2VNqffN@|{^bGt5SPe|2Nsuf}u863XOd+7*fXMhX9JQ=--Pao_vJ@#(r3)>y2P2oB zJCl%>Rn^w_=ogk8nR{>HO(?x{OOh-oO^&{JoAFF%Bl~%h7g1v;%66EXX65R1)hr~e zWqe|7$Gg;(?)PRN^hy8JYt|Ym6gUFC3621$v;;_!rbdeyFEr`0;K~z1u=+=_7NyS? zg4HWa4qdo%=iV2aww*Y0?NOYGYZU0H;w*HS*uX`M9LKN_mP0!ng-cNAVDJ$lMurL; z7v+~wa%trks*}Z)Tx!|Xf;=!m57=M?Ai#ivsH9R#FQ>99mG7dE*3KnOWNnV33l#!Q)W;7md$Sv}IkG!P%Qxz+m4<3 z4L|saw4hqKi`w-Y_7L)6Jw%}!#$gwpufIL`;FCwcdgJXczK!MBh$M>mFkT-%o;@2k zdBKuZt9Nt8TBe!f7K^N~!4@4JVV+esIpK`Q#oC&ViBk(JdnW>IJp(IS<+Y+UZK$q> zEI9%?AgZXUwmQ05h@o?8BESJ*Q_R2&P5>A-a(5`vV8@;p>NFWLW6P1Nk*c|<1>Nbzr8$M{#!LbTEXbgM3R<`* zs4CR$(PFhX-dyXAkc2d3APZ$^K-=`$xCv`lJ!sb1)<3c{YNfVnhhM-E5RtH=8U`C8 zc0B0x(}EyaV8A-IafJswU%1{bT{*b(;KeVVx$W+IA9};m? zs~!{uIzq7*l~oNHv)&cU){QC0qN<=%fJa6HKsbwlj!8^TtBlswmMu?-F1pnR2~?mn zP%xB8T{@MJmJ{e3+u8QW*wl#+zW6t%Ig!eaMt2&_$gb*;e0s4H28Kl>$|9e^wT=Cg z(ar6Xg_RHFsnDQBj}bnCDSaj^*zh@ahe#;+aO_|bkno-~4O$Et10V=aAVi4?j>P1m zZcXMuL7<|eXJzl28}c^SHB}1aKwZe zm)!DNQWjcXzq&R!oX#FTADx_ASl>K;f9AsF&6~HkAMH)&FV)ddra+NgQ!RCT1CRiH z7iS~W_(BBL8=IoZR3@7**0FiE%=XxU^3Sv7oj7Nx47!L4(D#BX_T3zEF{reOGUslHOj4xTY2uMRVX;)eyJ8)n{oLOVt|PcHE8xv^{jTd6c}P z7FuWP?Xo`rdVS^EX5&@&0GvyrfUR;!AZU*NLA|(N zhT&ihGKl2VRk~82cm0cpe@ws>5IW+mWo$FWj6}viKi}9ECaKxO_JCWheUMSGbfzx% z>kWvCf)ArJq)q+UJqS6HEUhV9+bX2~yl68-zAqUH7@?H;i1`bX$|5HV{=yGIc3XVT z*ogW94?^8SjiH>Fs1;aWI#iCFea1>!`>Ko!zVGys7k32`Kzb?w4&<<78SqurxSLy| ztp@)Qx+DK%0rVSuD#RIQh_0sli7Q$-)n)yV4M$OWN_yo_2MY3Q!3`5~4S4uO(pAtu zqUsO6YTV+<)HUjl2C6}6`<4;3*!}Sb5F%<+1*o3T_gqCyKRwo8Iqnlwf$XugfGjhg z0yY`24R;T&0Kg^zuIb@g;~CkPQ7X%+D-Z|z!0x6|fi`8PEfBf}f4`xDKQQ!Im#3bf_Zc5b3dD2TyyK& z<6%i8$j3kK%)HkwE}uy|-tqC{Z;6MUV*aQLx7xD-PS6G(FzhqUk;V-*;Yr~I-om)_4g>(x5nQo06C5%?$ zll5P@Q+b9Ti8vfU>`HjZd-f%;D-a4&wpJ<;0rIo{QUZ2Hz*)geidmnX&+sUrQCtZB zxqS-kUAUg)QOtmCBn>>YR{*CnoK-GG*&+L>;@C%j4pdPZ!_vHU7AvV!?{CKU&aHwh zp^XLesAXZhF}LcvYPvPk>FIqvgwTiDZyk`3{K}9EXTOz}$=bS|xJ4BdA2u)9%=T8k z{!b7TB1QpH!k1zv2v0m#rtN`aRNHqBO@J0(x5A13+IHJS@jU0+Zo|T_Ooh_1+P@Ra zJ=ABgM^u<~l2MQw&$K%iX23pZG_;JJtKgj4*2St(3pBtE64eH8N3Vhq^{Zva6+5G= zt@y8|df2gulUf7E;52ueJJ{?|b{;&p*3?%ca)VzxXW=ZE2+R2!84GnlrcR3XJZizI z6SoT|NUn4@wS1YYcu_R}RJ&G-(RD*svSSp(rbR4$`H|2~dKpsk?xJvI$0_10{n0YQ z*0NfT^*obOgz|g@Jb|5b77Rcuj5qttOHWsRb2vS|Eqg5ndHS0>!P|v>1B8+%IBowJp=-7G4-Ev>>I|zgBDR)hEthYnd2x{tx_+R8>y!#dF(Z3KRIp>YxRt#Tq zcl15`1D?$hpI~WvqtHX2{T6@&dL|2V17Y9}J?7^SJ(|kRyL(Uy&>9UhuwFnZ4zO?v=me7*S8^<_U|SIQ@d5@%qOXK<=1vmNfJpz^e|_QUR|(%K}$$)~ak zzXOH~c5s{IabrL4TX{z9FTV=q06GU)KBj1Enr3=S&?ebJ{7(I^O;2d>dkt6O2m6t( z>b`B4BPzq>z!)rRx%U)J??f^LxHLcF!z$G5D<7_K3 z0^Em-rFmUbl9^R!ruU)?Q6_9+W>FDcE_qBHlsnNU_2wrdPCOd-3r4w7go|OF`60Hi za-&ePGt2!03{%yprn$mh$Te@0r36P-1aQg5kB7HDAX+9qxVes%QGVc35YNswhR*qT z2x&48RuH8URv`oOn9ggdiuk*A|0H8oOhHJwTG;T8; zLBIn$L(`Sb_!3HX$YeyAJ8?ZVV9&cO)Y-Kvqi)tQn4qzn+O<{3-XKt9d|53#_BCZu zKE5W*-V5DCX=a)ZWLkj_z?X;=kX+YGX+#7kaL?q0dL=$ZNt?78#C!H%QE4HV&(cde zYtnQzy-!;0wK@D6f_Y`17 z?;xBKT~&IiHDJpgu?tSn1i%VNbqtmSyvsNxuYy9rZo=VPH4E7C?6r-R&D z2F9(>=t9zHhso8n8P#4wlObHAHH&kkFu2-5M`eWVD}g$Rdlo;94DNxQBt^JNypu7V z(k&a)J>YbLLO3b&IR|Pgo%e`-i1@J1(Q8ZZdxceEeQk;4C$d8!Y~>SU%xF(^Ft63f zL6uK^ini>52dg<3|BWKdjmwQyy1&St32WW#tWf}y z2ZF97K?TKWE*Bl~Mw4cJSTR~pHoI5yjNl3pQaRrwM&-@5lLhtq$jcq~eOplPFRdLL z+0j97qTu;ejDU5BPP&A)(TnDrD7a0~P`iLTD>Y5)&;U2^f-Ff6x6SSEB){t-|C*lq zaB~JiMA6dyna9~)?w$`t+z>Z*mym$u46zV8aRMw<*v`5hItp6;flcKu^`a7R#D?y- zX2>(xV7O!DxX^!mKj+X9m_SM5nBh#^MPF8Ns)*y6pLa?*=HmD;)xeGTfzqZY8vfiX z8YP8p!*h`+8Bls+d}{?oDgAI@E}qk$s9#(+Tq_?d~J|a-csELD?0Q+;H(bm zPiHLa0(L??_*r}|;8XVE7HWt#%MP5ia+T`s{xd;Mb$xqn9yd~=vVGJr3@v<)1jxzifzF{vQQ1_J`Rpk=S(jNWN(Ho+%8%MK`be_xl&lxrtIZUrpyz+S`EPE# zF^tcgoHhunbGgNoLe0Om9tx$-fMI6bWe%R{kWMe6%C^o{zKh~I6DEbK9**^ce{U*tZQ*oG4L12HC}GGE6O>}^ppX!b zM@(y9|L|PIU+=#Z;DvX^@Aayz7AiQnxf9m;2D&2ONUVI#@=Uh+6eE~~9hh<`q>QL1 z7&BYp0BKXT_)Gngmh*&x1{ZiDQAchLPTP;+(v!)G#Z}FM{CWyZXL+v(Iw7 z0^Hi9j2W6h0Jq^PU=}(GoYaX8=$5l%I5#lf&gmsZ;&GhJt#(8N_^fS)OY1d*WY$$4 z>!ifnQ;=C)=>l`Nh}(mfx|H*Q#o|Fhu3Gn^DQyR{C6>= zM380VWu4RlB(Ox9Z?MgG*c=b%${+|$7^YD+%n*u8;!|+RudgSM}Gmgf~ z6Mo6T^eUavND$vMudFoiB!r8; ztVd}<=g!&W=|u^-LzN?Is1XLDDJ(oMZg7R97o;y4PIXjs4Wow?al4&%8t$!s<*LFF z_cO|h0dLwnY0bL4y}4-5qbg)p{;!(+MHv{i;GMuaSVyNp4EFw8FxPtV#XC8{H+qJo zRMK$! zOf`X`(wJGA;H{V+Z0utX$cL`SXtmkeK<#6`vQX{A6;MR_3KXm80Mq!FJG+rQ$B%R6 z?d8EW_rWw)$21NF=H!R@(@76&t6FKjt+$F~z$JSo880l89ZKJem=0*r)bDX`Ca`v2R;PY_XvG zlLR}NHf4S)bMO=X_ee}M(Md2MK;MWZ$jLwmnuQ?nq5I5i-N!Ph%0I$o_DLR|fUFg0 z48>ttb&#A+5I_nziJRn`&_h_Uz&zWMjLgsAT+3AD#LQ=}J8HsZQDEP$c)1UHI@dKm zdNf`?3YM5`U*IKaCF|SgYus!(Uk%qg3{_P$2)hvT2x9x8XipVcw$}%s zbqnCYD)^g>rrB6eW~7BY^d0<^=w(_t4Zo+HYDL_gCE7CC_Rrl}Nq8^&1(AkGRpMZR zN(~bXmi=xALW__lmeSMx67r(ZB{BVpG-#xB^R+OEToK2mNbbps-Q(y|VHekWf_{$r zf@X9sA@t}&#VP7XAP%HK0eBXkXrh}3C7kG0Nh~ZYNcfU!`?~$Roab60+!3yej~;q6 zkmS}$4}ArRV`sxF!$QApmC^u=wp47OL7sPSZ%+nDOW8)5r=$_hXDkI&r_`p9PweUqc1F#;(OEo3=Wv8SAqB8t_dqp_XGHNo;2nNdUAjmzZ9*u=Lz_1puG(3AL#bk5iWr?aSUanEK#*fROari-c! zq8jkTria(@In9!Ple@+mJH92-9*&^wx-x%KxsVkEP@37v*1zqw4;qXXiNR+>WhOw< zCi`s}Yngg{HvmhgKFaL!h4y4%8`{i_lrK9i9Q2d*w{hsB5+Zz}E5gsiwU|uzC4-r; zms7(OX2rc=`F%h*1q&*pi3(#jvWq*jOHaulLWYq+U>HXN6XuJ~X;jp{BW}>w0l%di zISRjP3Px_)%fqiSrsziCB(fw7nHPaR#3fXyihxVZBN^UiUB{D5Z@90{|Gj8k`5pBL zikVJfqMMVFBEeD-rGs3W_XuU8P2%HuO557?W)R$kc~e~h6W_#))1X_3j(bUG0nu}} z@lEY979|b>@FF1q@6uDb(^8zn<($asTvq93Mpg7i&&9V+=lZ9iE;$di;r&V(HWzge z9WQ+_T_7=aGL!EgwvL7ZmR z9K#uMl&VU7YG!jAJ>;NY85=ix0sn0UL%~rf2DE5(G#{;qw)OE%;b?InsahK)tCw7; z@i)=Z$@*z5IH4jLs%O9!Ikp|>+)##mfg;^IlFpc}d+IID{HRfCOR5h5!$B%gLEcru z%JEl~ENYi`R~pE)+FUB(d=XB#ZemFK-?wl!%Kpka^`-D4!jJSk(~?4;()YLEs-LitV+ro0-qaHX|y#^3~PQDE9gp4yySjTTrbytM;7cev7J}) zewR|^RYCh&Eo=#&Y6zOt7#SPrL=v7 zy1A%9j>byd#~Ch9uB6ju<|G?xe?Q#J=^DCDDBhpgw6j*u@>zXGGfKe(S-rAeFuUP@S)SNxeWY)>y3#`W=uxqYkSUUKw|LI_9v9&;m-vW|rYp)= zpSB@43`HW9P3UIC7|P*@W~2VC?PpsGUWb5iOq@{e)o)kRSrlry3huXSn8H5t&4~yR zB3I&g_Aff<19or{!%uR)&QE=7WqmHv*S9;9`((F3EXI4K+*8nDWE7A_e-^4u8$|*7 zLmSWrU)l@voeS4X8jxybj>0It?zPqFhP0Lpo8pskOd~B{32r~XuEHd-bx4U z-9`U8XuxdaTEd_3kt3-P^rL*WnA9;nIsa=#pF|k8be?p`z4B&)+g83bQ+$urBnL+ojuoYJVkzD77i+`3g;DEcz z`y#LK-)lx%q&_HClhEZ2$oPw|etafFU#aYCo&j}q{4E}qJlRW5P2}!N5=2nP@r20L zuyDZxSO%G0;^>KETsF7J(H@P&Oc`29*d(XzG)>V5Mi2G0>4P$$To{)8eW4XyYrSl* z4#&e+CQACcRcqcuYmvK|hB_AY7NkP1vV6HIWO@1Il)7$~1Uo3-fS`$IfdMRR=)qTM zyUS5F8GdCck@U<|vkZRjBOrJDhvX0wren;9@H7jP3EOnbddQwGWeWMo@ z*j0DTsWoW|gRd0uVuFnYeEy{VdW!WJ)6$Khk~U zj%KGybUvQ!yPEIxGUV{*tz89-pi40r*GfF4sft|X`=@TmUw304*`duz!)RQCHKD}d zEY&HKQtkuF1~W?q4NNRe14*K$n3e5o*IN2K9^7osu~}ZxjckW5XM1J9WA^NgpKsFk zgxQUv;Mn%jt@2>n=K9@)mYN#b*E(p`=pQnbLn)Q(lt^rgx0mq7P=l^I;7VL#!-bJh z4jV7xKf_7-_JIEBRw?dnpHcwxXLkOISqMO9r#)~JzK7GcPrgw}x>~sFgS?@oltE0g z{bEy^4F~sFA@>P7OCk#O?Dl&9H|I*UP&E}(G1lm5TLyF(t~ApGcco8F*Q&Mu8uX`d zIMX-G-Rk+dtMIer`XLIpUMq_p=`M!Ln=HB4FxPE!KSBAV1G!01m3f*xy7Q+i+*e?-S2R6}2HdoEFl&?ZF{gIIExTn% z@_S44*)KjOK1>i(Fzyxk(oPnYQ0HXHFaS8LJ3>aT!pxT&d|$cgyep;Kp|A8!;dU#C zRt;rXI0+jYO0Gd;6f((&9;WDshKPwxQiSSEEft+?n zEi5V?br5vZqlMjLS}-lbM^#u??6C@dC2A>-e3x}<+erIQpcXm|{)Q&!dG&O>e=<|` z6A9)pbCfVqE*(#j5^W#W0qzIK$#;u0?cG%3x}`AtdZo^_Vc;0Eg{uiDD1}(qL6aKy zA{@j)_Nwo$20gbwe%i;-3H$*C>4&Zz^o#Z#Jeuwyf!Q*7HLx3-I;6)sV@eZ_2FtOza-T@v9HeeucEAE z<0Ww}ar?w>DZyVyk~%jVj`$na>4v|}a0gfgFIInTzR**mHgd2j5sEg3$@}kFJDE}wv>Bp!Q%`y0C2!K3$qk>W|4+go!^L&oYrQ(&u^gY zHhPv!%x|ngAgf;d&;U+vN7NRxibS#ZNNyn~Y1k?^8k+R;{5v<&d$8%x0QcEPnY!dk zrVFz=>9{QqWL{vz;d2-wWE(%8i*3|^ILXj&3BamneB?7)pCI53#IP0VqG20WCkT2C z?#*(Hp~jQ)jRL8|SedMj+vMtr->);ZqHs~fV%XOdId&7Wh)%65C3KtDWT#wj688vC zY4y{BaLSp!p#kB%K~W!j>`5sbuB0mBQ9duHASMwVu`J}8YpB1af-!LH%TxvS(eu3g3zaY^{6nN_(TzIDQK zG|X-FRQRrC73MMxJXe`C!8kc@k`*F3*)NhE?d5H<7?UZ*jUydbt~*`s zUaz;k)z|&W)$*%g_o`@MYeylj@a$8hIT?FD- zj$eKKuajkoX1nS;r2{OEDRf8w@>e~K{-cKp6fWU@Fk^`_%_dszTE+IPWv_5CJOUyT zAm^xQz2V4-Kl39QxX1SI_}Xft+Z(&ng}cK7t7JUP919$C$^&oq8k-VX@pr`MYO9bmRlMaH9qO3AmljJSorz~}5-b=@zOJB-N6?!1$X znerYyd^57~;k{e4!z2cM*96Uk1l@veOSh-Lq$hpFh&&JV%o-5SRfzyrH7eDO(Q1ln zsjee^VE$dC6NYQW{uiEnJbdw~_WN0=#n}i$G7VSbRP&MW8~c7V7kXXPGO;`l$&T{C zmspHi_c3;_cy0S9SL@F%rmJqc>!GJ!dh4UFetx#2FgOBffkI=jmN>i>NFr0HG&+OH zvSD+$Jib6E5=*2qxuTSdYF8Sj-0xJ_I`6-Nw-idOhQ6VG-hHL64&N`g#aA34fD-qm zkcXwQjAbonc`I1aN>;WCav)F`9D%ezp)puX9Nr2fkttLfoxx<;usK{FUmz5TB~qDO zQOb(F3J-q0MXA4O*@QsX1I=J!L$T6)$RkG0UMM$TxkaX;;+D7)=Z zGA725OfikOIk^!Tdnx^QLMuQElGp?i-D5jZkQ|qio zjw2U)O-&Se@bxCr6^jRC~?f8S`bo5QO9V=4@G&~Q7 zG|l|#g_x4Inm(<$0{>*3#x_mN_b|QGbmY|N#Ky!4uQ(tOdF3|6E)R&iZ8)0tp!kr! zo9@i<4n}rT-tnLZ>-ynG=e$F~0{~7;AZy$%kaEDb5n!yN`;SmT+X|+a-Mll3xKZ|x+;IK=;pQ_uxp#(%=z}??Nlv!8*$OI%X zK?zPs!Xz|dt!M{!!jd{Tz`J-3*Ixe*n*bNJ@GU5Xd{GI3@Q1B{c- z*Wdqh;d|o$XVFuFmH7W##~`l;*k`QNFX_C0@7F+2N9J$DNZS7`~a&JRG8C zrA;oZDT|0-%ff>axmb!s%DB;`#m+Tm#aOy$+Q0$Ma{Nv*Vm8cboe8sHm_UpCDrq~r z?(*2=C=@=VJ�Nq0O~MlJVKxP2gkvDzvHwa#p? z`yA+P0I86sDIJg>S&)g2pFD`;X9X33=-83uIo2O&xU1++g*0o051Av98 zSa`0M0qi{^EA)kprhj;&12_4c3nHhe4jUZC4%UN4heklOXq{RNI-=__yC2oBDF(>L zQiF2V7p!2xffnW_C60$NXp(J(FM}rdASecxQj)vOa=M=0MPI5AQk{M#W+|tH^)6qDBiRy_dST_>cQj6e)`N2ldrYoH`6FkJ=yVS2!2TIn=2Q^a z%)=wR1N;gx+|jwFbF@`F9StjB4Qzldumko$W_IvakUJK~Zx5$tQI8lFOEu*hqnNSS zgNsp(=IUm9wb#?^l@fa>qSs%Ej?jES}(SBW!_t3{bp^j4V0IIF2w(y&`sN@_x+ zQkg!AOFr=nJsK1ielNQ3NM{>;bd%CzHElaN(SL(6(EvB_5@+&y+Jy|VIG1*5SIz^{ zlemC4Xh?f=A&{QIQf^E8@(#KONUsq`0@_m3G%H;pQ2?|^04?8uRAL)=z&|9;QS{zo z-Kn7Pq^qyZfenT7ZB7cc!w}SUDG)3LeC$6p*rGGY_R}BG19O--=6X%Qm_}ZlMR@hqs(hDK!Pi=?ZCY8|Of0S&LmDPx$v%P=*4SW+ z9rnolDS&UA5sw20em>}a_JmKt?nFL-pGYD=-*@EJ}# z>)P20;MzTl`CwIhmb*;zM?!l@xZkH57Iu+Z{9hFJ2E;oqM&{b@@2Xhyw{^=zb5tg^ z(95TAuW~-*+`SWU>Y>wh*ZM4w=Aya}ZSj;Vqk&m`tnw^Mo5NeeS|@v^+5H0MebRT7 zl?fr|BzOKfIZ^y@j9R@Gi|AbOej=Tb&WaUgnSyETez;EI`o2y;L?GPw5eKNeA5gdP z3Y}%^Z{D5BCvV!xeI5oEKvx3ZbZ;I>O5-$uira7%7c3Y8iY@BWh3H2NMh^w3n1>G% z3SA9ek?75*xT>X!I4)18bX|z3dW`3{s^By#>}a z7PDRWINX>nYmZ?@i(uRs-}??5y{ws8ddT(bMsp1Y-ZL{FjAr>Q9Jp!4c!}zL=WvIL z{P%o-9Sk8}26~FG2e(v16oKwj`}ygj>VHLCouOrpIEC$9ID|fb5?EtnGMzOCwg>XzvC=^HjZX>!dhIj6I| zAV%k)(VbHPr6S(^hV$3Su!Lvk%lh#b0`*0?<(KY^UD>Aap4|1G zrQ9eezL=X*5Ixkg4`=qX(e=DyT`+9xqI&kkYt0{)?L{N4yK)#1j_%chYx}wBh1`LF zH>zmuV}K$OHUfDn6{?x)__-2hX<4I|sc~PBd97P-6LKHX`$tn7_R@LglDA>BQbj}M z0XQO27h0v&6O>6SJqsY?v;AOtqo&R1CLNBMSf!!wCfKGkc#+xtl#zwU<*c7tjGR2P zhuk&?7KLo29u1z*2v9(URME@`m8&K)KwX0%4(eC)cUi`;KMum7I67y+jWa}HdB`pR zG9{~Wzd~d(&g#^^1(1DwVg(t^`YK3Asx)_jPb5mwiA-5BP^uS-OdA5N<1Jf)lQu?U9(S$T6a>`R_$6dvXLCf$Z`h-E{_0U`JzZ?<}ejm&|r;J zsB4OGBg`&&6#!NK%mmW>()ul}wGnd_@xg!;WkIkZ+Fh2iOe}w4f6N-&S?KVJ;=FKV ztcYd76shES!$Lh+Ju*5QBYM16P32x%2t1We+Wx@3AvSrX26DZcPIfk8-mlWYSIh0CqeiuL9z0(w?2r+p;&!4__A!dGj zjXQI>$Vil1H!QGtcLZ|aKAEigr4uVgdMv9owT`u__ z+u`FTQpzV2o;aGVWMc|d#vW%sJBIfR|COy#mB3Fjb@r1fLUCxeJB?zxa~5xQ^*oG; zB4!}j1jK%=k>w~Bz15(1q*WodEDQa6skCZEzorIXu8jW>kSy?B?m-E=Y*Be^K%VkMShASQBo zp2aDy6vXx8tIsXHcNt$t#H)K)Gs2aaGbO$;kwcwTb&~ddf+L09swj}6%+c3seM32~ zm*z^pR=AGjwZ<%84LyaI()r+{&>7?~RDR|GP^vonwm0l@*TN-!zm~J-ex*li_AHIz zPTWJ!G#Cm*Ro*C=8M$Sh3JMz;%yzfLrLMJ$WY@ZK;eDjyTJ%4r5bUcPhb3?Ku=-}c zQ8!Ptl1*t+;+&-_Edk#i)nL58q31RIO*j4mF_|2fRAIfXqM9wCCr_`1NB6ta1$8^9 za-&nOAy4E2^x-?rEqln2DJ0yW2429`{547NFn_U_6;KqMdsm!P*y z&;16v><^4`X1;*alDvlEU>*eR;0M2fC(uD?DDa#y32ITsX7)vZ%zG*da0guNFM?KR zb*At@Cei|waU$SgOTTU#hq2$w{#_rGbMoSOBd04KVdv)hg?w!^A=JF@^ zT{2JqlfHp<1PADbzVQQ<5q#Z-ZgmGc#rL<1FRXx|p@5$%Td5aT2xuu=4F$eyJLQ2- zm7#n90>ShEgq%;^@7%hw)S3@a?M$f7P~PRr(jTXa1AiEbtH$nUeq+axGxdso&>I(* z`VRz<2Y8{DFoX2cE%XoD=pXJu8UMm;fin~ktmOWFV9CV6mu`sw6Mx@-F;6h@DpRy# zx0@FFx#s=@Mim5NUIjJd$-b>CfCojHxORBp$6NGjy~?gUeB_N<;PRg=Xh#V_?ZW3! zxADc)$%tV2W|9Ku_rS%xUZmZkyx-$UYiP9f0X3a+-5!?U@{6-ji!b$xid;#3`>SAy z+3WV_=@`&-d&t?J+vl;qgkT=OCbnBkAo@$rgb%mFitYF2voL&7J`gmCm>TI%m4puY z!FT_Sq3G0*U#+r=PnBu(_x;Ff8mj&OAnhUozH9s6O9>`Sy}$nl@3X{5S5$`G`hm<- zdn)qYIDg^Qlpq=k8-(xt*ICQubnUT^2+FTYCif0MMtIAlmF4NL+Z&LH-vy|oQB3TD zY=>Ne;8S>{IVG5Q89P>xlml?hSx_Al>@qOa8w#O}WB~8hIW63E$+d;GR1lS4?i(*1B}_)6uOdxUlhwqK1mBiu0O` z3%#)yGMK2@A{^#5wQ!f8Gfr zUK5V~Uz=_Qjxgc{Mu$6<>%TgA>x26R9o-vV>g|>i{460)%Hwx=97o^l8u|a%$A9_x zU0q>gea40aQ?V7=PvS*i1gCK{N)9=uJ9~r~BbN;Z^`g+S9cOblsIsJS zd&3fBZO79`&i-422VaSV=~_%)G_(fHFYGWq{p*QX$+6MR|GA+5Uu(uM`m(dn6n#

NQ$eQjBQREZ%+5FK#G>$FIIa0dIkwZeuZ*<)xUY*C|a=2GE3Byc{^i<@`Nx=ZxHFgD}raQ>@WLr zmlA&7-pqG7_g~aHhy-ta{A0geDzykc(C$j@wG~!$`QL3eJ8Q!=D@(tr+Whk6t}{KS ze|^rgU;jV*9;6vOeDo^KeR{u5xLNmB=k>2FRmQ(|ZmZZjs6 z_(u2btaxtbSTtb|+IwH|TJ8+i7kfshm(q{{+g{(HBo<$Ds6LVd6*mVth|`{ zy2u_P*YqavX?rz&f){OHX-856x9+WOuxf~4$Ak)NiX7K>yU~5Xwy+2tNp^YdFyvGT z8>wbFIXa$wHOA0gDROl3Pj7U32EA*cd;f8{Z=!p#jh*aCcuGmJ*{PLt>E)00UUgj~ z`pP*>+1Ze)p92Oyj%OJ~w!In$4OUQU?$^-dpJ{J++pRxu%Ubs}cMhkNeVpJsUO0NA z_~c6F30rcj(Al5V=#$(^smtnB`($T2GpyV}kD)ZMvGf6_8%*zaN^s z6;vC8cOj(`U!kBA5UQH?vnlPG)bQ1(;R3sJ@C^u8)RzXB#tY=1A3(XAC5?=jGI8pg zueaI732M$78mSJOuyIoQQ-vua%yi3*>!F8UTEjiK=-rM|cTLpS zr_cJ6SdW@Ho>B3>^qCRzHi@252E>?JX48bx3Ve$SHYRkLI>e-2cMKy}@-F0hR)SnZ z8BI%MLL_WB+ZXYPm{Zvf;395r=%Xt4I*~en|9A)#o@{KbNY*YNrvbY)pmv-eM6}o4 z1O8#D%t*C0CD+vvyEtx%U28zGNMb)*{h%|fF=UPr<(jglWeU3FX4h)tIz=P3$PHv4 zxk@#P!&3Cd{yo{MavR*2W&ahA=vU;4U-9!tn!S(jDfbwi>N{QIGf#|(X$<`%(l3g< zQg-=r?OK?5@>-H$j%$MR+o2q;jTa)vqZt5h=ha!}zB(gc|jtAlW0d%{DS@d@gO? zkE&C+cp3%xIyn;y;Z7asUPdTbpg$&A+g@dlzvD6`7>->}s);Sdrd zl||(R&(LYFPuU%A!b~h)N*|wTi5D7kPr8C{@Jqo^DFAb z=;vK5&*MwViZ{)g{alr(`w2Zaus)hRj|ZkYXfofx`-rW|W7aH>WorWhtgzWO@d5{w zV5jo57ItSV?}eGWz*V%Y-sgAR%btHGC0R)#ZY*MS$qlH$O3`3li1&q;;1E95PgYJ5 zw;~QaOvOwm%ue~X%GI>}E{45>BJ?fh{HH>w2*AYhBpnE;PZ+>5La1;-+ zIL#%9P@?ac6dOLB($G@=AqqCqN(cQvNhdRkA@xui*KzDQkBK$}H$o5n+hEa58g)3e zXxv8XMy}0SAFkeL32J&|#WqAr-AX~Q8&&}nWQem0&xWj43*21z8hXidrD&p}b7gmt zyCBl@$#=YOGGcJ6VBu-tO)l?li_gs89N;(@`WA)S=oeONC4GS!o_0Q|fuphfgY`dgAn5e#U^g=c@^s z=$D=`$p5}Q6u+q@aD+m~EJ|wmT4A}|JI+-&3~D_)$tO=4k|0ZNuIbOu33*dLxo(I* z=yP}I9I$AhF(^JK_;20X=37p`?S?x_`6Rma@E1^xx}?GLl7^FyV!0<$b~r*0K-79 zC&St733&CqG-_*hw&;LueJqtM8fwUbQS>-n0~2aa%<9f|HHZpkoqyg-v>^v;#?-+ZjPy}bF$kqv8} z!}9ItQ!p?6h_Ntytv_G3FCs5H2z>uhx!jDnGOqDq8)Fy0;fWIK?Q>g~b=lFnN}YC& z@qj|`o)+`ob{bPA_?S3@b#X5Ws=0n`NiWRCe z9z9w*6Ll1BAX%SEM6|N(vn!}O*p&9jB+MI8T=Z=RSSRFwy&?s@eQuM3|CRvi8NPL{ITLH#1E@ z!zB%?$3Ks)TJR}7KG>&^Y7fH_nO^F0q6>^q0&ruRa7{T9El|B|9}>Q}7yMn*zPzJp_IdcPY2wyEhNx^p+?8hll+?$% zKgV)vRN#+mVd}s0KI0xt{S9)tv0LH9m0v&k@YHYXR|(F6Cv@X4O5T1O=IhVJel$JX z5p34EX@elQHCEq30;oFt1if=(8Y>3Y5#FC2w>S^70Z606b8naAVUlm|0eOQJ54pVW zl2#Rx9_TUHJ;ip!qSn0mAz>U}N-jc%QikI0pP#&~QU5xxq-c5G(ME4puxW|oq*b56 zKf;>0F>k??OF?vCc?!)dHycjkFWF+MjLg7?H*Gh^&111hh3y$4>qTp1q^K=F!3@T? zK)*0wI?%Q-7B@Cr=41 zW1e9l9)A4Y=6@8m}8#}HDzhRYsE1jB32AHOED;jN!|0@t{}xfW<{WF%9##C5K6 z(N2SuT<}d!(6|LVIyLdrmg(|7A8a(vxRsNh>8{>qvnTdZdEx+PCk{alW#AXE(s18bXv@P zq-}er^}Oh`zJ0Ma_^Dnk`Uy+D>rKD<(AOC+{dq+K%BbOH#>F3K;WX6a^vomZVF?Js zKV!vDEy~#Ge3s$jEP!=er*uNW%+FGZui&FTV$|7CYe$RX9ZQT+loHS0&`}YJQ>l|tO6*uO5bWs z#UY|nMlhINAlKtaYkd{svnv|8xiwj(&$v@$m{OED4xexi$ZKS3w5(WW=fI;?#^+#S zPw>$1=g>vda~P!6IZV;w92N`awL%TXUgf0Qh(iwmDm#&2JiD_|R4*(Y?)6L1@ zM*4Ums{%B~1s}-+7cU!EAdv3q61#43Q5&J@WUcAp(HYV*Gbjo@$vNa<8IX1tyV5(u z(lLY6;o?wQUK~#Uw9h~i&~NdQzUi5M=@rLOo70!DXccG%GPRPr8Ij=`m5~|k!GQNy zQJ({}&B2(u6cXn-q#bRC9z}F?*(p|#Z)R?+U0Ac#$w-rku-?lb5_D-MfdsO25bg(v zJPtOwVBqkbbn|S_KVZkx2OO~9TeJ>PeefGy=@@W+BtU$>PHB=y+OrY1MJ1?VS&Mcz z*dF8uAU2zV=N!~)>!2Z&b*U4q9(3wV`zEi=K#4o18(nKdcP;KE5(omG3itJhUIya*Lj>*H}NFN$m#Pu4+|9|{whpFRKyz;-v=ymwfNLZB)@{WX1qi$44!70mS5^{-x764{gInF; z2!cLtf_36y-fZ_Qng2)nvW|g>&pV-BSMv-SKexb}x{igwO??%8x7fO@yWmo$)M?$$ z3j%U}F-%)=z1mNZD0AlH^*q1E&H3GWncqlkdUF2Az2|omH}#K{af!r%5_a*N6^T)v z#AHqqrTz&BsefjyJK9GQk}qgXG8q}L09j7`ds4`uvN82vOquVqZ~n^|WwYZ+UYyi) zLs;O%2^~)-EQNWHN?Lk#F&F=mF*}ZvRF=aX6B3o(frbyUb74RdMaCy5JC^Kh^5VYC zMg7@W6JT@bCx>j|T^BhdCmCT_vdZyvM>0cI>@5f~i=045B@b>PN}kW$cL zcPcEVw3L*uw@_S6O*wYCZ7QW=sxFo@5@>SO_)YQ%}`1Rs*%d#t@fyViG+nwZdJrR|ykAoVzEyqDn zq`e=Rfld~Jl;z)^RU?thsM|R;4A0}T!yTtQMqTa&4QoVhcFU)I(hTeVPNQes$*TvCz=w!L6kp^0?knbuJ(U#; z{e);q5?WE6yD3O+g>EESNtnUiYZ7%N@Dny+S2{@13&g<#MXKb*E7HW6d5$>uo{czO z6j}?)FfCQ_WgbzNdsPL4GBCCz<%##a8MuP!R|yk|S5QBy+?yvRqKF!u(13DuOG|2E zr@y2ehW9R8YDnID?m~W1TeS{4w)uWC@@gk+NQMvTU3+y{t7d)wdTm=|ZQg%ML~g|E zwOjjsfB$=JI3=kQZcJ|p;Ab-*NaKxU%mZc8PXS|nS<}j!+k+8)Sl!jf-Rds;T>jV_3~rZ+l6!He#j)k!77%Yvc}%wmbJ^|@OIIx3hAJy zwX;!vO7~$l9n~K)W@YO)jk~J%$aPxmDd=V%ZuV(Rt z@k5b~jWLAu_vlOd3oOoOll&Y5Nq*QwzA?Z?^LrEkUGhf6dzZgs1j(<_Lh?Um3!=!P z(+FqUu9G@%B5-kRPL%noBRbazUMS&`I$waO=n~yfDvPp5ZJE;>J?X##EHV0^AE{oj zV=*j;SIiXxsa8;ad`u`jN->NFqs93Vc%y#IDjGC6KOHe+PSHCbM^E9R&G{L^7UTSE zv0x}tjKCm7c5d8(t#K!uiVq6LO(i^T#`?J5c+iN9lwjHUqY^1%GO@TYeLPjX&Y#C1 zfXfx_7>_ZSJYMoy$H3eYAMdeYoGY=Cv8+-m2Fxx+n;JjHl6Q*{@sEkuqPm zLG6Z%rh@mNmWFYv$MzH7YFT#fK87@Xf?8WC>_5HaglwCHI8-W(4|)U-+@$1@FR7 zwo<4;WHZkasUA?Vg$g!|iU!+EHxC+IcqtOEII%LV={4P!)bzKBgyZ7Z1=2Llw|TM| zCbw)3X}98X%VARxt5NK1xw782Ig{VEz_AY15oPq`ZuAEWo~jALZBODfeBG~Rd^kF{ z2?+EpINM>(cEF-^Ep8iaARy{tEp0bA73o~d=&O|w*{-blcy@qoRm<2mHjkX^MBNsb z3){1~XfMj%jy6B#ZmBBVPTC{!ihi#mNE;io(RJGeQ{KK-F@d+Q>|7<0ZYSgr%RCFi zPABH7OnJM1LL*jmb_xXDRPEQQajW=jW}O`wudQmsmO>7MrMo*M6qlyX(ggS0JBkeC zOA;cA-@p7*qIZ9uGqEU#|IOP`$BGaII(1Z*~OGfqur}k0+(jojG8NCnpZ1m z7Doc_;?Y8a=|^(JXEkxOj3`lQ)=Z94I49&Ze1#}b<@E}oraN*|$Vse%it$Ak>sBmV zt7kQaucB<)sh2@M+b*jFc3zbXhEA)7-T>4lCS!W_#?k{8%$U|N^WkyVH0#}Ck38{^ zx*ImN$gX&_>VrmoddAjsPkUAz>yN+n7Y~3oTrCdck&^ME4cMaoFt-A@1MIx2dA2dni(CtGFVov`VV1lB=9VxT-2DtO~2@ zp)Tpfifa)quf88vqcM1SSRLSFa=kz7+Ql(yCT*ME)Uv6Jo@cZ8gm>t}+qO(?^V+_+ zDekP3bgA4~*rRXRbM|{apgYIT{z_)m?Hqd*{noz&yEFWTsNmp@rlRWb7w1&&w}s|s zzm5ui8M#_E}Pex@Lz zQS(RLuW{9{ISliyld4N-buY-8(r2Y|2ui`G?47_Qv+6q+LB!k^U|7 zH+QNi*)uZ(L9$2vJ*0%!)-g{~YHI0KpH7V(iM!Z|iny9lg_FxR%%YAIGb-q1#B`Jc zk-@#WZENl4KH3As9R4{!2tMKf6V`L6ZRET=Y2TN(-tG^HVbf6?3BGW5Zu!o<=6Oq} z%x=|zdyD4bA1_;2Gr_p5zvgz$0arw_E>1lDQL*~_ z+&N9|$GUef{CN6^N}u)nPVF0C%bt3oFgADnyZ~$6v%f8DS=WX=-ZJl*3yoi2H4UQj)WU=KN*`8<5g-(Jis|x>u z{@dGv9W8Gk$#`k`(QPe>itTBRbH)NXOIe?_NYe|4zg)OIePN5YvZ?-Ve=jiufWr6s zuFMCRyqWtN2V4vGrv}WT4E4jpptiH7Y>zTeIF->o#Bv!lt-DX12Qs>cRPJ3WuX{)p zo=nwr_pxBsY$stP<1dR|$k=CT=;odO_3`dlDCA~xN>cyJ!o+T~m&^gS2!vrD+st-@ zjBpep(LflJP|~D2{4ECrSF*>O(_j6)rhl}*y2+dKKzH-YS01N3htjMCg29{rOe=n) z#_5y9`Tq!g$~mM?6qwZK?gOLgivs2_BUQGU@8+8PxGR`ZA{z{bG1;^%`1r|`|5?6L zo+p1(ULfBpFO--37C_qn>z=>cYsP-FXYAZZc^qes1-ywm_u8CCoO;Z_8=pCf8wfyp zHEw68*=u%~Jz^i>87`~ARCnhQRoLFASO!mGJj@p`ZU`qbfJ215iOX8K_oY& zrq_dYYzz1Z?PHta@6d5}Soj%(v+qDsM;rz*FV#rTrJc-ewzdZBfrsHp+Bh@{hs9&U z2DY2s7T&{{b#x^-Bc441AY^u27!{^=F(FnP)U$`%DML_f7oK49S!KF~b}|OmOY#N7 zZAgOSkbOJViC8zYW~>=#V$E4UWPBqf#$4+T=GV!^DXTre zDJwGo9GUFFkW&u(R+~dx7Jh?^z8z6fy{DCr@yJmEyUebK+QXZGo|uPeb1KgQu|Bmz%=f6Nt@!<`puD*L-zO@sH+x3+WAf zW|qo!RbjRj0(Acg|1>-zVIX3+oX%gbMnr?zF(!PmU5k{&hpMk)z^8(%&2r#PG9o`dS9OUkNj`n z-&_7Oyh^t#S0Pp*VTD8!tixKY#@5&tyI>paB+(grV0Y|_eO2#>e_W} ze$Mw+z7PPz?TF7qQd`zzkTi?S8dH}=sTuN=#pb2T)(`iZ{*CI~y!(ZsFjb8???7~P zm|V)@GK{HPTQ1J>YcQib?8Ux zuNC!De#T>`f=(o}W%;eM$R;ed4im1aQQ$WDDg=ve#2#vU@FcD~RsYK&9xc|<{7nw# z{~Yc($m;&Kh)yx|4=jIhIt;e)hL{Imj-uX06Eg4D=w{(fd=i%Dhkc*oWx4hujcWrr z2ynsGmh5w5Ve6Av<@?R=Kfi5{_;L}Qpk~ik7@zuzv^X?77ys*f@P*qx8}Z>MZ?&EN zKl9!TFO}bZ{`2Sm?0+BbiMBC<0uKPHJqM6)+smu&vUa9@NCL3h@9e$JUhyXImTOH2 z<^gx=O@*^M)#jR|sapJ3&N;*%aT2-haLeqT_o)w>N^3S<_2-@4|MoK-=P!-SJbM2G z+%g(Xk>*W`)ooJs@a0=uHN1fjFwNPv-HzoH&v0(!XNTT2WN4e{@!N`Q;qVNe4B+eO zrvnb#aH=chv4)8abt=n#mKL6GAG%|LgpjwRq=T&=8b;rznj{o%O5O&jgC}!QP1(qV z$+i&g?Vyy%xK!sWA$Uepdq8hxo#98RE8;^8MERKY&07TYKNGfrPK@97ibG)0G`pVw z1}&2Nvqh|GeWHM$gbNE7DHK{yjX&FD-iQH|w4GY%E?&)E>)8PN@a$pc_&hp1xajQA z?A%5%x{(%&4Z{p0&PNLl6uW~LK#PVm7dAGL3bWYaMT5830IaFLEe2d@7Xpi~SFq(O zDIIh;OBBKQUf=g-aajHV+iU^2Fz5@RC2BZM1GKRdqcg&GJf>@LhnX)4ELEnG>IMz) z5g^ThF3L*~R+C4K@Zj`R#BA^3JR1wuhfWX2Kwul!=qNZphI*6{;;qVi~)@X^C zwG&C}sH{t%_kE{R_yWBYUh;C1zF6{6gSv{BLo%xxsI^>mHQWciK))!wtSJ||+L5W2 z*DJqn7osL!Sj{e#DolfI6!eYoq2@HA4)!cG1c?47W3iEjFQ&Wen zZ+AtwWH3I8?!2W3g6V2rMFZ)uwi; zrRwzY;pi?)vpaU*9%B|2a&-Q@*u(t=#MQ#yf{x7uPfD8uA+`*n^ZEmlY%Ap0KB%!n z&`@>RDI>uU)Q0v5Sg>n030sz7TsX0}wvGVy-S~zhRO19rj8F%21~t;r^-5~2lW0RK zQjk$h98^nCUNH?S)mTqe=lErJ3iL*q>Z~%{OE_?_75~N?WqE;XJb8onU;fkI%JA!- z58%Xd@PPVoIgr-U^nvragP`B{O*2K-e>qWplVf^v-Z-2xo+@Wj8rs6s(9_b>RzY-$ zBkkF}m-7F*aECk$>frT|zmV|TXN6yO4fC*gA6qzg^PQw4|8}E8cWNU{bHt1kqlMAI z7@+Bmy{E57%~emEQyv^V+i=5{ERGVNIm$6DYCwC3|OF+J8zqXp%S?I)=2-$ z_&SFpj>=Q*(ze)kk9M>9sDgRcaj;m>cyFswa|xKs*r@bsCJ14=)#6k1z^`N7by@>Y zD2=s(7k!4P_a5C6FL44ZY&a~{;|Bc%^V0;SwQA3jAZkU1a%K z^arU9ZZ$z`t`ArhK$W!~C}T z$;plTtZsR5-^)gIu6WhZ!uX)NkxZ+8tzu7F)J54guUYGv6c|fxKV*!1a<23v`6E)N z6VWPV=JYMqmOSRGGZycMPe9WpJAefX;2`V*8<=5cfJrh7X%!`nKv^kylAF*Z&yz;* zTBVx`w;JNwBcxe??rxi$bPr7>nt6{@Y>&<+7uX#+s&<0$i0+DZgPEQ=smcWhnb!_B zdsA*vf)=MTVarpUM!Rxts34b4X(d;7=lBv8hXZbp2%tgM5ubeSA`O$3Ur}(mgVlh_COB=eiky$IdC~Iw( z5jWz;9PRR~tzALfG_bhEVRtOx5y6&co$Um1Y%7Z679#Fhs3s2LsKjaPTev+ms&{Cd z@36RxU-Q-MQuEYA{cbmJ@pku^k9fCxzYn9` zf{XoZ4=fS@@slDNqRCCHM4G1V_>@$wU2}Z1{r~LpoIcc8=xy)wSSJ{t zik9erh_RxyS9}F9?7k$mz)|fXj@kRgd}^1Nu?pHr3`vuvyT@Hk<@qXZC+kL^`9ydG z`ebCjyA)HyE)~?g!$TXAV4~e=nqJ+J0v&b(!m_%=oenW~u<_>VXo9zAP&j5>r$vy? z8pFdm>5u+O-o*DQ=laNG$`KP1mBq>6bPR4=pszOh!{TxIf2q^l9O7hYY4K$>Bd0wlVAf?_d!L&|>2}Jby--`hU^QQTy}m`Q(nj z{PJhQ&hg)Z_U=e_{>`x{+ih?=Z4Qm@JLL`EDY{DWi>-gSS7+!{p`5C^hmRQk@H(tY z2D@*+@>JE&hd+Dh;gAn8yA1txo2nUM{lowNbx~0 zz2~v!{$$KJ=sw^8`Se(spQwfk+i(ym0zWV#21oEA=kGOuUofgRQh+X>=VyG8x>^Nr zDSGUWZQp!Y^YylLdrbMR9;;KZ^n)|r_Ohu&Oc&Jk`4J3fnQ5-E?F3T}N}dQ|P6zgC zBfA&RBUqql|EIh73L|^Tq!2XKki^jPLryqIXi>ctEBi??#c_Ktt&vrH7@Xl9i`3>KiH`wt5Ag!mA7-~rp zS49_?Q;^!i6_pcz9V4gmlhmpl(W*)xnwg-E-v78Fg`>lHysT*t0ml>>8u0mgT&O}1xr^qO=KPjAN;>LSAYfD4JL>k02Hn~?%Lx&OzVY$VW$2`AwT zm82+N@G-Q^Uval;yyMlee{Np|2;b*@H8{`sTCMP<5#Hpwuu_ML(|h-+dNV%A2^s(`i}Cb zY!JA*LE;*y4*W6}*GGLc{ODy|A)qhXgMdkbAi8{j-9>lPK@apoPxLl@^f7~=$3x5* zgG2>`K^Vl$3_~$ofGQ-L6NsZoUE$)!91}4aQ!N_PEC#bM=TkoBX(>*E3QjwOSnZJb z4CrB8vyJ^M@--fUIv(Sx_u?5|;FS;K4QTi>K73rqcWJb1(nU^ zAl?ZgmIiy+`R~A@&4Oe-~Rgye4^$bu+!YYK3+H*kl&BdNMBGpyD@@Oqsij80YPyBv5q)y{owXhs#p~5Jtd~Qe_#i<-?|=rG=VCf3;yZq=&?>GZ<@hJ40Z1h z%Bi$@re$N~#H-m?>e14x$dt3$7sz1?5K~hFRj^8wUneV`H%H+2ec-zvbbC+TyUq7A z6go%osF!vFT>JeS4{$(ud697hx(NyE1>m9}W*JW9#8TDXS>4jI15dG70kHAeRXbN8n#~5+z5VGBF^Lxb( zaf2NS$C%ZGP+qu(8e{sOrRldu#`zQXS3V)|OiRa)w~dul;KRD4YC{A|T*&?7i)-Q^ z{_R%@=a(&`MTf$}ol)RwkY!~v4B}Fk_AV`MgLUG`M41$A)%ho`8WR4(oeL!YTbx&Z z1VT#CN6a)z;$Vhpl8agwJ_@b{8;LY;CRLW8uI6B7vGSv@mS6aoGVfoJl)0kvN7jSq z6+X_V@l_-_T{8KrCl7u?L*Qd_+YlZ6B#PE7iId>lo zCncA~>ELHT?wk@Mk#Wfb_rt-@;@yap_=t{5-dekZpW^^SC21M@Ww&3lOSP&>SIaO0 zQP;dAv=zVHujw`OTd4s6$hv@T8LHI_*>zFAqiQtZX9u^~Vc4SFTLC3ePhqzwif1a| zpYS*O_io`)&>KtdeZy+My??2`s*t5`iW?V&-ZfG6TfhfDtT zw{4Y&c~8E3nqZ`dJ^;cZ&_k$pB-)iO^M1M$=hrf<#su7g?XN;@y1=*^eNcRKv1`p= zKDssMvyC&$$_(88UX4yNsOEy(TQ2_P82Z-#U;1+IqH2*ZNSDxM;!J-B8WZE>{$#l? zfD92vBEg70=%NX3B#LM%?kLo9MgF=n!AU=~!AVpI!bT1TEw%g#MN%-dju3Aaw%6s5 zRVQUpX4%dur^!woCMWM%ij2lny*c6oOSl1^e6e)NY&p)*yPLrX#m-)(e7(h8M+txy zN73FhW5n8JElS}1vT6{WD^NE zCUR~Bf7^^6Wc*0Ghk56R9`VjEP%y5RV(gHTolvA}B>gI>gAlOCVxLXwe0g zxIMrKr>?4?5!xCSAQ05)jfBI%kLZN^#J-57WpMTiX`h0&;>Z*_z%7FZ6?x^eX1P0s zlW9k~#PaekEJJjP5Z+_5L)>dYYnZFT*`Jgbaupyvjju46sJba2gp3JS*$a@hJ z=@K0enjo<)ptPF+sXbXsve4A9xXd=#uPXEaA{WpiZHs#KN%4;|IyOx);c#|uo;HDR zH10H%*D%$#Fe%(Z>c&YHa{5*C6Gm!Gx->VSyWog`lF1ZWTW1m$#-NjmY_FII8d0_0Mw}Vy>r#ww=01l9lMxjb z?UAVl`waKv{X5Lu6R~6$A#K2gS_;N^Q$dq7k2xMmCDDjLoxN`|O1+o>pQ%FR4$^XQ z_L6ix0dmzddFZx4Ed_stBMn&as!H8D7O$11wvGeG6VNU^JcOGM4kvryB7>JQTc!Ek zG3ID6!74hw>917@q{1}?{80aBZl2QMv8an1jH77$eofDd)zVjC0 zdJ-!RKl4eW^o54q4bUzv{~bm4@RqJ7V`p8#hbyZPiD9~X(w&^oKl zRIj`wosN{3mK*l;Ic_-!#o?8uJGyMGTS})|ROMq$)xJ@Wqbb;|uWfe;91YrcE@QF@ zobQk7JA;+$fNJ5H=UE3j4Ta}A%88T5CZHKZoeQKubHDdqpOf#Ci&zU-3{%wgT zRDay)T21wRZr4a>(RgZVl1?4|Nz_vx`h*rSVN?6%U+^#Mu5bMONOYP{c8@*4^%Cxf zzcsk|U-o$of{3dtGxl6+M(>O&Mjgljs*hvAeHLR(SE;iTR1gEqq`9~W5D8Ef9@f;v8lK|4^s?H71!%;Lt;S+w zCV#8ExJDjSuuARDvkoIgWY%HHr-A3!3+>p@c_mr46D>%tzco6Ui!A|*;-(_;UV0R1e+>!dHp2tr3rYBFezA6%utj(gt!1}+mBgQ2 z0kS#s5%x#dB|>M5<=$|$j!3&t2x1NDMF;O*m%qWgn!qD0%)jGHMOpGkcggT`4fW0+ z-i|3raAtpSSh0Q-H=M%=L`?F9SOx_Nrd!3Kx(40EMO|8EgaClul!nAA{ zEen}(`1FJ;>d{5mY+-Ea5if~-lvJNiV(P&x+|5=!BbSk+)(!UU`Bq(}V}bv`S2PsL zN8SaF$TzS|ef-(`vlDHbfnQFv$5C6^olBXS8AH;qx7y#bA0Fz>fu5|s_}q9r>6?}C zbz^vbFku;K`FYWIC+vAcXWHZ6{MqpRTyPJXFM%mXA}rz__uv_AxO_?vx2Xs3Idj{I z{aU8Oe&mYQy9Y~XgA&wBUc*)ST>(4Y9{3%8JW|;`s$ucG41d^{s=K;}DT{u(J> zwTOo95Ap7#ADM)Wk%hq1->rYwlnWEX1R|2ZH!i?-xjGVv49JUJkX_L2W+rw&_aaaJ z-35|A;NOE(VHDZ~8ru8iGt#W>pWt0J3~l!X-y5^w?k;{t=s47UWC&wiUI-f+j(*2u zo|cDBw92h@s7D6@!Y^oWSKW!l3ys7kbXfMN;W^FY0cY&HItIz$me_#527sV4$;XUbT>_v;6nE^ zV&-d^K#nWy1}84tZv~1;@4A;3nsPGJfWwZR+lR(N}vkP$NhPK zslUQsle(O0I$cAp;q1me95TvC=ME`yfLM+e^LMUMjOow$~Q`mC&ZyR)nz2R_*r4&l1Y$~S?TB2E6 zqIJSpgt?I)3P!1@CaQ}Xpcbe->VmqX-l!iM!ZDo8d0fsN+|B(w$>Ti9EN1f(?{QDH z09TZ4a!3}Xlv7GqmXN}WBD(mJNiDtdRH!h;DN$+4RFUe{tTv75mj*PfDXnW?$09|u z;UX;5LMN=G1_>pR#3Yg=kTg<@)Fll`Ytn&?l6099noJ8v1hS^kfm~UsFAbpK)I)6= zOOt3itwY<>!5XTGnxa{nt7TfJeLARpO3-1YC{5YAV`4^UXjXywus{~c1XC=QCCY2L zn+JKeclOb~DEn@|6aW@L02**W)__n0I*p-`f)~|DwN&j?XZFrE5MW#En8InCNsV+Y zy|NnXjJj@C%rq)0YiNNU%wriF0uQ~4i8mvbw&U}N*bWH{S8l`y(+LIkSkfBnuCFyJ zpaWTy5fx0Zc!&Uaw1^isP7=K*T9a$8tSu`n@^Dc8NTGyG1R1iv)cNPlb4sUq4n{%g z-wk)%GRTY>o_syK_6`L1lydCEDlm~$2 zzaxPT7biV~MzF0=jsQhrF?y5u!da{-BM;6l&aDHgUL$#@Z`e9)bcMHVjHLzND5yQz zRB6fB%K^^VKXEeeXu)RPn=bJ=;g_t_lry{XrKa)?wwg{%hst}YG@>jb*4NbzktNMq z%i+4XI9=eB`u%oaoZ0OhEp66HUUFMs-EsFptf@r%q7_=w@pg~a#UNWYmlpqR9wIL+Cz!ffXPP%J5V^nbmePUFpUIFRM zog0*FzBP<2^Gx}y8ab~>y`q6%W-+{4+ZEZP$bpLRNn7~@au^w~p1Y3~k2#L6JMVP^ zjh7uqd)=>j64}LY+O)MXyub9d?(GfRF8|!5k~Ptg*h3H=%T7oilF22T+Ew|DIW<@k zU#B}9Kcy9A-iii05TS6V*AxPTybMO%>WrcD`{Ln1xl^~e&*KgY)X&MK)*U@tw=b(X z)tv1$UunqIsICM4Xa(Ja(apkXgWV34qPT)wje= zkD>tMb}>BJISt0Z56mbUOGELl*~O@YS1vnl0jkagV+Hg~v1&BDvKEq_04BM=>r_un ztJ_+ggL@S2rysqrt$5aZGE3g@_fCibW*wtz*ByA!&GDsYZ-~9qkx9Q%Y384<4Ht4F zYSI$n$(lRl9P7Vzoifdqu6xuUmK6pu^(fFHK&8Rm!G8jS;tuQ>5U|v(cY~FlX#rE; z3V=vZ(!;G|H)syJC?P9J0`P2k5~Y4!V~gETt|*Nv2_>mB&+G>oJzOA*eZ%1gQ#u;}W1_ zG`$i}mDbg}Q*(Ut5N{yBrDmMht>t|kNA`5Gx%oXM3pX_T$H&>xda`fzdUn@!&(Vn# zFxek#)j9M%y;kk1Fv3U;O0YTanbyKk%1F^uLlFF{H+yQ$dH>FBgUe&un_0cNM~>us zv7?V7#68za;eL5D1fZT1JWZ{at542&_Kh*t4zW66R{f53k*qlS$Z5vvU80pPUIOQ! z?38jxoKTaIKT=}JV%s$s%ofCLfJU8M&6|rY;a9XJG;agI{eo3mByd>Tlz`9pXbzxjWqvY&9Rwgg@Kom$1;H< zK;)JOODOX(D6q}{zAa)v61D{B-*hOdu~u+^!j}IIhR$&f0QvT=HY7k5%fyF-m;g`( zTGxi!cQs?gd%VI@NLaOrq84}JhP`&@((YFtm>7ClXCX=~%^kXqsMeX}!1`J`eWDYU zOxkeD{I1#T=wGDT-&?fH`um|nGK@{N?Ez#1jxeQ6k17S)wHvF*6ov_d>iQVC> zems_5xeYQ=;uTTDR@r8M(Q!m;X!Up#X0j3$HHR?dpm}mHB6oK_`|) z&;-xFtT<2p+$~*rPP_taUPcLpU*j(%P$pNNUr7Jeo03XWK zV!3G)0dPA7)IrpO@d>d`$?)kyj8!Uu?~QY!ra27)T=+t>RwBd2g@X$*l1Mk| z;GE@V;c8%_z$y}n25&J>5_01uTz7Z>TdAej)?BF3tp?hevh_tj!QcLS)4}-+Go=6@ z)Qi1<8B3j8L~XtXmpfV`Ioh|>FtvR#bpOfGlNZ-ORSf3m25sG@N{s8?a%X#u24Tvn zl6=`1kr{N^+i)czF^E8C96*c&%UmYlu^>61L74#LNsSR`6u=+_z>zL;1%nQ_3Mt?{ zg8(eQm_90(!4Ci4_kS94>&NZq-uT&uOnDAHRQ$bojIXXPlUH3 zSJCUnroTJ04ohG2JtpJ1@p;yA+$qOITM0t1jO5xHl3EpsiF6ceoQ#1%fr_Cn?tk_R z=PC-XtW7L$?=E0Yv+Z>%yn#rI<9m$r4l#5!5sUf7Ryx?xP2=E0`;~1r<%W`mWTY{C zzWd6rEPj#~@PU?13a3Tf*C5DuNE74<3Cl!60jP-4IR;LCLPbReNcbIADJ)iLCqwNf zX-dE>f)!D$Y{Y6Fcl_xw-@A!E`4P?XXLDWa?AX1f; z693O#-eZh>^J2F*L)>G_9$y>gFl(fVS(zB z!YKJ#Ca#BJ1d^N*AGRKK;+W(#xeOcbw!YA?qo*+I$fa;j*a5%v00sIC?FK_TwPT+d z-P50wN{v~qme3Y#$lHH)S#>wSHADashr0c^>^5`M#~Uq=)XIhXsC&{xM|12o{x6G@ zD-X?9Vp~L*L@}D}0qRPNq4fb&s68yq);$zvL!%n*S#+FT8&3>K<>A=+fy!Bcy{zd6 zZ3AE%eZzKpbg86D0rtr}6ZufbIuElS99_SD%g&prlasrIwR#d%iP0y5{R&Xs#}p`m zMmk7BW#4|_Z~39#H>I~mR5sa$&LM(4Lc{w*bu7ph=^4ZO_ zj)scYA8q{$vED}aM6y1JOVs)i=)|qRQT6?-&&lhLu)fI^YQ2+48|!y+`~KF~WV4P1 z$Vi9?@MQb7%EF0q&1%(XMbo5Fl{(eYp*$MYWBTKtztg(xiRD}&0?GP-ufP2AANgOQ zFTrv%kOE|Q)ISv%X(}UJJQ-a@TTT-t9KTKF-_{$C8`RvI(n4*tGg~vVIP&Vw%20nX z@cwO3o$D&>bg+4!ufizp8>>X?(vB==+Ov-U_P>FF8OBeK^u~u2k~&RCx?e9Dr|Je)9%;FB=##KADqmZVGO%UNMD2-+HK8_ zsTW9Y{d`Waz9=YlVTqRfK?;Z-d}72mJ^g-QO0RxpiQa&yYy zPc)IEDd(iw(K10MAuQ8ms!WHq`JocUMC*9iWE; zecFNbq#K$gWH-90Quki6}#WM12779@{Ma+v%m!(=NzY4OZA zAU@K(e3M=+Vk`N*_2oyAjkc-MbGT9SyR^+7+0HgV_t}__#SN1C!D`(I6`p@XZs@Yw zI9H}`V;jE2>Kbsq{4=T@UEd$;Q^zh#%8;sd1-M>_$>1E=1a=kRv8#atRha06gO7VR zNwov)U^~<(2$yS=bn4wWtFiL*<`xa)wBXYn+tQWok}h_NJhE)XATUSz zpc|7DD`clAeAeZx_Q;Nj?BZh9Qs%B~Bw!A;?>eg5jY$ZP2!A*27CjF`wvvu_J3a0C z#OJFN%|1xKG^#`Ki7Mf-EJS_m!zzPP*xeiU*3K|DRjZGFYfm~z?N=dc>6NUW`sZHg zYXcocim$x`a*YEUzy{a2RDuxrAOe9nflHFa1DwEvBk&PEz-M@W{r7riY1^*ndjMnI z3dEa#Fin!7UX*mD)gj~wdeCe)g0#)y`$0=wKnYEtkQ8VI`Md=~uAi|| zK$r(NfHQb;4{A|EAu-diTNA)M3dozU^Tsv-V1OhQHh@9W#Sb>GIq(y;v{rD@;8I;D z%ixrJE2``1wsskhMN*HANR$@oz$UpcufS#`VTE)fSuV}5%CHPbFIHkMmSL^5VuOsz zkaS{=v|&{nfPPFG+j!1na#lCsG=x8IY>32hh_3w@-LdXL3-mXfoQO2Y5M>CnUn04? zi|8K0Bu2H`_nJk_CR)PK5JLb?MtJInV`(@}PfUbMRK!5o8j4eEw80;z)IdWp!D}ka zE@;st&;oPa8_H|7j*0$M)@V!uU4}vWMN@2WgnqZjF%!@UZBUG%{$Q&0^G<%6nOM-k zL6za9H5Xq{1FZclG-o>v<@PlL zRj3Bqj^>}gunS7)SwEDsI-%r7N872?15^7%t8TLP>Twb+<*?(?N6AB1m`6} zKMT=VC|}p}?z~tSQF}WhmxOAf+WP?})Xi#1&6S#&BXkO(N?M7V)T2i7XblMUv=LC| z3I!?=o3`QxaWtfHLWRNMa#Wzu0Zn^m&<@RnT8B1KsK91jT{SL=dTN{`=ps@*(f1lF z$>tZM5r-QB>dROEYZ=~w+@5%m-C_|NU8nid!p4p#)%w?)$jQ5+2yg{a4+t9bhF zLEb*ehP4*Cr1eHFP1Y(Gps(?@O^Y>Hz1$ASYwc5Qoz`xRa!!56uRG?ZnB#oF?%-B9 znGpP!P`Agqc;ZaF+|F&>0;dDQh4{IXTe{=%Xzyj8eoRS^bf<~+e5_{ZmJb zsY93}%nkO=E63?4aL7Np6?Z;%KHBf;pPyL*Z3=4G4n|zAsZT~K(vd<>5Hqo|D5r4h z)t)eL7=ZwW%{G=BT zH(q^N5+78Nu`!(&p1c5(9m_ zwZKloV$=}QhGc&bKfh#gY>AI^WTiD@e2gnN6Cdxa&E1mwx;xQcqMM7(66p=*`ib}RZ2BCdH-qsX6Mq-{p#Q_ zzH_-Vkfx-J2jVjXjl}!oHn*n={54OX^NI8E89e{1frVpr<#LzYy7_-H*(Q@967&N@ zhEO=eAXIbv@M7)|UhoY1T(Ejl2sVO*$FL}6Yd58A<*vE24t4)y|NH*8&N7u0~~yK zIU>nla z9)9a&49af-g42-_Nf00LkQ9j#2dNMjF_8kvkqohs$bagQvA_XRBZ2=ZenlOs&>$UH zj4}ifMg?k6k8)I^8jU{VUc$;>#Z_cwLP!-Q2&EC7NS9p5*VRSE2uhhMO(+#48Zju9 z@cyLk#C2D$1XBUR5hftEN>DmcREiLkt-_S0vIHmeYH?Lg@2aE<8J#hVa5e^Gv@yUK zY>XmY)=Z73*)n5b;umc#n!(%iNdDO<31gPCkhw-mSa1$&jm+pxm>Yr7m$2HtV1L6i z`Vp?IFlxfxXY>C?NLaT=S33TkIjbn!K=kj%cjjHbb-W{8wskEUM`0W%Si&*3aexIw zoDaN{F?(%{iEX9d`8CG2RgS=t6G8V!aIuITY&r&W*uWlE4&P5*!vVw5+qHiCTEO%} zvE%(fP6;x2hyF>YEAP=SaQe`9yDm@Y87S`tLk|Wei??qs$3EFX(u;wC@j=ibRXC_Z zU@+Z+NZPcG;G}NX$6{*RdQWhUR~WF}ShwXIsdc%eB++>i=_1K>M!DtCU6Nf|d8vEq z)0YjByoAKx1>#x0YQVBU3U^g$l}N19%PRkLgA^qy1xYVoNLKZd>ukyDdKZgtx=>e- z?et?)@{Gmr$ZxJvC(h(y6yVn?$o-(HtvHfgV{|^l3aR91`AXuu)W0sv!6QA`KM~H? zLDD`KeQHc~XMMiWJw4Dv-O-~fmfJt;8JYIFxWQ{AX20>Ii!Qa#{qMeO?LlSo;*rFr z>G&wxQytm58_()lL_{KD(OP#RA$7%X5m&b&D%8;bH(%PY>rp%`eu>A2pn8jecv4S` zKlNnY)rdK2XWps4rG$eJW_OVQXeZ9xM56>fJOI!tA0y=WJRF8wbet#Z6(hauuZE)y8f1|h6S;(;@w}(F)G-Q^i1vQ!L z@BNd%^~|76Ux_RTaxy>Y(n~nlpNnh$n7^Bgt*WZsgc2+xc{Ax5o3wH*b;|Yb#r?;s zxH4QCOWoAx)auU+8HdoQoN_Psa1z`xNgTaY&x2bc!QwB)63|Sawuf7*@8}f5zy*Y+ zZ2lm0V5|d*QG^n^;GV9>PVC1%_rh<{TjX=Gj-5Qac5Rb53`ZKQL9=vWg><&*BwH0! zo|we(S5?kmA$MP4n=zQbGzs7MQ>BXoIF53d1Kz%uoEnv8_^OOb^mg^DNXOk z#SIK|COl0HizVCy4^%T$bHUP2G2K^aTKGMaGtCiVNc$=So1)6i3y-#QKRo435qru%R#S*)!0QF%i* z1-7w)7WcgJY`-%sN=^0AaV6X#Gg_(0-Rj=<B43v2s_oWfj{oh&X!X&XNK5 zCcWyaf$Ecza_?nQh3cplY40}BPEs%Hm90jqM;bDaQ8m?&RLel7Gyh({DVh|!jAddb zyQX!>M3fz5s{y9OWyUZWlV5^>p2tjcGZp`vq0F(PY`4QE!PU%QA#<5()czp&yBvaY zB{uxs1+H;Ewq1rj&`7xo`=dRdxXxckU0vsIE;q`xu6C&_CiCSF7vZmWx!6Cj9 zreK%fChVr1X-8eGX3QWV5)&L*)Qn946PTEA#xr@46wj9L7&p!`_NZT7JB;Uezs`&DN395Ahw`HU>$OJXzx!v3VGFHV!K^BT|!tZs|4Tr=$x|fS^w#?K4BGWO#a>HebEQ` z9z*>ph$xLQXzbC)6)zbW0O?PzlcpwJpp zpGLN1F^%y|Cy!Z7wI2{e@5j-|YS~#G6SB!=G83xj6MtDCBd+T$6{uLHhxQ<@H)c8Y zn#xq7B2}myQrLyv7(e>ICirH43rlQ?L2OtM7@Pc+5I)wZu(3H$mB}J-j3ddSKJWugP4fLu`rg# z68B}7PVLGjE8S8Vo2cM|6Y=NgC|c06}DsmVyRL zut2ZCKK|NXEhzmCEfqlATtL6#F^Z8UB_xZ1ay1xq|!gha`cTujTp95wSW2CAi6 ztF2m5rE&x1sXbL_vrSZ<_uC{idoigqPS@dji zvg|3Dj^87mFBkD$@~Kb9QzSdAe(hC{x`cYAs&B?24d{S6wNHbqePe7H{WH4bvDYLE zJDRNQh_a5Y-~Y?WzxXSy=}1GuNu;eXq2pQM)#ucwo7-IN^A5{+oF&Yoh~*TpfVtLF z$@7Zgc$Y;KHZtt>(e5;R)BvnLP>rh9s%9waQ`M?TEvoO*gLcQ>h}ct2#m?9t!#EJ* zu`Bk)!N}%2Pki&=sjsRda_|hO-L>>*ywsY-H_|UaWm< zuk=#wUc1)b8BgGa+B4$|yqNK7635dSx5gck5aJ4PhBOXw=q8*mTm;A-Sjz3hdnBR> z=LI#J&?rtdw4tFqBbX-UtI?Ts2GSPJNjpD&^-T8=&%~FJPo5)5y~wp?z#PgQ(N+2F zA~v=au-jWMyYO8#rcu zH%=V_&*7=j&6M=*IsorTE6gMC6wv>~UWjf=G|j*aO( zC+fOb8KBn}b~xGPVboZUxWYx6In=EQvB|5we5hWlu|cD}BurlG{#fNTuH?_DQxVHu zs?o7R6S@N{cMkA_2}TWO26H3q;H#Am;>(q?a$lF@!dx#778uhA%*DkQgIT+<(xk0e zX``olB)*%KIid4S6wYlH!70F6cgvzkvB&*w(N z4I9%Slb}Dk4nx*9LMw)DH&KcuFHwIOXcMML^edXyIFPwVp)<{yW`nKeT0}8bYns&v zEO0aqd!J@tf@H12Quj2hAukj@z|8T14hTBgR>aLb=H}3!P7Hdvn+`#XBW1xW%b-`3 z(a>{Kmm70?)6x|?+%)-ax)0yai=)Z@V7jHCET(hrICyfXoR%4xmZ`|$5sPlr4M>ps zjya>8VRmRYOb~L$?2H&w?~SZllnUH}*Kv%#5o|bj@NiL(Y#kmBkB6tjlRCX7(sG@# z<)Ns`G||Rd4YxKQAsxD<65qfSyem%< z$7^yE$Y%-5BVLs|d6n7q(|9F02~kotY9gCvl3opDRYJ97RwG%she}kpma3A}YbqU5 zwd%=0CPl9XsFCc1sftk2l7<{BViB?)WRB$-$-}&A4Q4QtW$}V#n}+8s!?Jn6a(O?` zSsM2TOO#8EjSuk)W;4Yyd5o#dniRFZt_FvoWs=6G83g99liG3d0B6jZ7>!4-~ihVyAo zJqalaHg#!8(~SS6Sd_L_Ez0jeG&D<&iJ!l;rvCnr=7}xe%>~n&=~Ykbj76d5>$y3d za71Qk_-w%}rBbU@-q*TQB-PQmv5N&&^0H;@9qd@UK+XT>)m@Vxm|x?~!Z$gmBF!I?K6qMha}hbt5_rt^%gHieLtf}gv~xl12xxYq?0a{AHv zcBCKN)`oDJu(@=D=dyCXKmWhlY;pP}HK5ZEWf-lLUgbnpr%$=WlF1PY}sI>8+iK^{_XGh2QXMZwk}0>b2fdCc-2V{pN^>91WvLC~INm*8^#t z-M(tF3qC!wxY<)?C%&e47PdXqMP;)rq(|8(^Zk*q>@V~uhvHfm4{+Q2f+@%}rZK>j z-j|Xf$)|0rR!EYhLa-#>uGDozhgZ6)D~Lu@P?kB-8_j7b;{$wvlky>$Y6;gI&DJc<WRu_3xKfmqUe#zQ2t!RDJx?GrIa6+!Qd41(UK)@ITsMw=>Ohsha%)^_r=VxIowP4UI|d!~EUzusp{?|Bb?0(J z5oO-$baI>?r#PM-aIvgTUD z!Dqb7d&ADB<91y(9Tfb;0!{}fC(PjS;nuCtQ~fP*e{R&hzUvGZ*1?@s2>Go#`Y3+s z$nNs~L-lzx($%gH)pan4ENcpvm;MxrOnr4L&wq1-!Y(9nODJg`&Gh%R9-eR?&Of~0 z$5QR08D?QRK39Ol$H-m@%C`_g3neU0K85j;7;7l0C0a}gtk&jQX`y%P6Xq z6sJ~7EzO!c&BgHiuMxm>>&*qo(5zZ^F*;@I9;>$1f@9@ZA3C z=ccDs$A!sL9`@_xhRL6>GRjBNAugS8J5%RF83>o>>xIl^`XL_p70N?d%w)#G$ixKf z^FqJ|Z^FqCKNw#jaYBYceumC(&K#F`nZa{8uJbAE&@#d|#^~qW_~oV3$-f&0M(*V} zL-Z4zup0<5#XhF#rDWKyBVm{9uk}_M`^N@AS9iwa%rV9+T^wMB9=mEoNZal5`*@gx z_Si;|wP*GdZYEd$EL}DNF9RGU+kM#Wvh^4Pfwqr&zuYR*_bb*et^Qv679@2`hdRjl zns|k%)e33`m7o=gbmYj;NdiPFA6%A35MganWZD9(bh}?y!Q^QIk}X38O2in1RG#Or&uUffyY@Y9rnK{iU5zP7qvM* zf6UbB3w_3uB!qrd1L$yTTea1wqM2~XG!2*x6hae~)vfjTn|hFewtOom6Ad9E9Xsb= zKe=;ewY!oybVx7}yn9Gj>B!*f3BWjgz4ro9gh5#43I6)=k_(=kF#bgdnS{dUB|YwC zpWG$vL zNcAKlp11OBkk>~Nyp#2<=FLkAo#2{8qVb-eg2*k4Y;nKHfF*p~(pOBaF!RNl?=zdn z%$xT_%%iJZtM;G(kkU>bGs;Y8^9g#kH_aL2NPQ1R1G#}9DKXK=5NiSrW|I7cDZIsb!o3y1&ndo`ZE{L|HI z5sF^1=Xs?iKl-ICp^x?K?eWdY?$gQ(IZWasgom?bzLesW0)n|mwpe=o8wf{|Epl>&yr1}!VcdZqu$iJlHDiA-HTKo)AbV+J-xWET>GzJ6zqQDpt z@UOUdw+m-4$NK)RAi&?#APRLr`?Q)$+Cg2~r1gGE+o@OEXh{2MWGf|~Yr`d+-#a=; z3}QW8X3vRCT$<8m>Yy3zwx`-ceZ_RDQd?t*%$8Dh(7MAl;ISAJqKf+1);th%9 zbAxx>7h2vF8sQUWUf;|2Ne&G-*ArMmD7DB##LoA0m&s{DGj|N9|tP~>U z&+EfFK4x_?#B_`CBi55wCVNg?800Ome%6<4xiSZDFg1oX6J{kKh3IkM{HU${C&Q>T zux#C`=Sa8Q2Cebu$DOT!!at=M7d^iR10bnhZ}i`&81qwKuy+`VH%;aTl0;& zCYTjoP&Ggj7<-0ZZ>Xm8{vGNvPL@xvmSj+A#PDjSwa}%^W%7+L{~Hd-wh~(m?$b?J zcf8({ZDRrtVbG%7fx~9lAEP-S3>bFT?4Ju$%P3|b)&|CDT@Vi>iCovc#dlE8&10o$ zeK%-k4Qw2&EXOQiaW#!UfEF$}`fOajl`Zds@2onFIw{?@on~VcQS2xs)KI}0C!P8B zU2hmVMI)!}A=@#h$+tiqg^ti*7cJ%-vX>@W)MXom7OAIcEq47khuw0CTxY1H`t!^^ z$}IWIao%axDZA;=n#*oeU>|QaPZeh!;*I)--rT#cyXxljDZQve9P}{GDo?ofZLsfExr-7@}C2 z!DRx1LW>t`Z+6!FLN~mkum}-1sh1q=#e><|V(T6ID}AMJI(>t?tJW!@puRHe>vILC z!Fmf=6FU5HJ2ijeMh*Af+4wNgDje69>hWZc*Y@8E<3`RTX(>hP{vCAw;3!JBYskFJ#bz6&V#tic1%SaJ91#;TC37+z583}34wOn!?fWy0 zlz=#Fpp<|7xwc@4(Lls_LJ+VhSp@%yWh*5!VhYjNQi`w_D)xvO3$4jr64CW{)^an; zv9LeOwcwR%v6U7H^P4DZI%9R`%m2uWy_R*+?>knwxz}h*yU@8#mn{;tlHbg>!hADc zDOQQFWC1X_CLlgt&+&sMd+zCQv&4;SnnW*0(Eged0x2_z8Dg}iV=2qMf(`nbr8&EK z&yop?-YmUTz+%k@7HZzIeDjVan%?Gy#lbGM(hLbw1+fq%tq?c6L50*v3(Lq3h><3} zm%1CJo*^!LiFojnPV1?j>$TqLtv>1`;=nihpfR1XtygT(xi`r;;W0NjP6jbnahioK z6XBeQj>uTZA{jNz)I}qohhCkfsloc;45$E9C5s;B+&`u*1sT&kpA$@eLm~N zzqjUe{`xj6*S7SxhgkK;)7^X_F{7q0_3zp7kX*n1J$GCh8bDM2rW#+$ls%Y=-H}`N zV=ev88k;25xi0FinD15XXHrqgU?U(UCZt-5yRuX3niSsYswdte__}p|75u^4eoE)} zlY|mE6G}UKf$1O#5)cD0K!RwiBh-dy=jWR~zyS(npp>_Wd|n?uf#T+Jc#h3cIeMs- zwsFk9!KI&Ga5Rn+T2J3_tdMuPksIR(p>^~H$My6mDb48LQPy{TMYF9HSZA$2?69S) zeP*+mZpgbl4aE^j*95k8^9%ar5ycOt;-3KyO0^nBa6tU~+N9RPG7QDmAP=Bg*ZjBS zsKe1z#=xJWiHuN=#?O=v%}w5$KY5Z{^NJ(qAH(hK%ZYY`+VebPqP3n;);P7evE2Bj z6z&W=!be(lvjp=a!uH&YUBP_|pHQe?w1jG&U6!$2e zD_^|0fDAOCb}K1Hp89arvKCu6yKcKk353r$Y*H54@ zz;SCro9RzZ!O1x(m&3_8C4c_7>)ZW+CwRR2r`%O-7XM$m)q1pm`}Lchc}Cs-dXkrh zFD5qv48{PH#@OUrM#<#qmEEVkcf#hM=HK(knY7<;$ZuPJ!t(~_?haHQ9Y=8zhaD<8 zpzF9HDg{95vCeubu!=YfJL(NyvhkaD@s>$%^&ZIoAb zW#zWpGldR(xn_1!6hTD;ADbi<4}6jqCmk#Bp`E6BDU=EYzAyULU0Iwan@=KVVQ;<*otOUd3<5?zOC1W{2K7v3OZH-nfV-cw8&6A@%CMcw_72np2* zp_TdHWl8f;;fJ!Zegvr9htw;92xK65H+2tr*FVB?(HTZ%j`{7W0Joon7|C&$qo;&_OUiv;ia@_yxM!Qp82LGiPVMx3w=z}R( z)urMsL3h>0tuk3T6`!d;qiv)iehm*sOPMHBmD8!K?E4^2k$0B4n!n`x3kla9rcQSa z+<|W{jL)m*InYm5aO3f!&NlfmwX#3`d!(H}5Iudj5-V~YSvNW!`W7^yv40Fw%b(M0 z&cxp=#jlR|gG__JrS{CJY_7pjP~9lFwRFrK1?FF7^|U{R75K`W?$*)?Us@Qq&ZD>e z!_z#m_S$;e>xUxvR{G#O6z(sq+e{|&GgBZ&eAKs=Xz{78!XKWz0>CnGzm`@_rum=O z)?M>UUG{V*xJ@Lwb^QoLaz})A_^x+usM9hnHLaC%p49w*<5DRL(k4ogaK~5~x3FDZ zET5Eil?g5zpY2N^ZhRa^`{GD}Sn)9&i=%v?_UE|IeL#RsjbR3Wv!DVUXaVl-5B!q0 zOfK+aU)Eg@{AxD>zZoF#L)tV23Q#ot2qNMmCDSKLqf`$djwpK2Ek3&sT^K}6q|gUS zDw9~ZvioNs3lt#}IFLc#h2H*o$ObOR1DiXV$V4F`5vWPpq=iV*PK^?29so^1vcKm} zkk&HgC;Ygd@)XbTldTB9u07PoQqMo|)MZ|s5${n~i1(|jS`$Gpc{9<&n^wEgwe(q? zL^t?Hbe4HCBhs9bI?voDFF_WRZeES1L3?G$L#Amfm&fb*Up94Q)5MkdW)^OszQ*&x zhQ9XBnX>;A?X}Hq^MU&nu8_Q;EOg3&ttJuJ;*Am|Mp+G*vKGD7->>Jx+sZQs`Kqbk zo634`6uQ?B#h2=z$}5s!4+C{A)Iha-?Mo^f9Gkp}Yqy_8t?=j6KfSrpZG#+bSMi(L zBjOl!wY*0CTYA11c{#Xaex8b5S~oyFgH};D0S4-znMHLY%WCxDXPk;NaXL=J zSvUnJdwb3cXO!PTUDCWhkNqp$AXXsZekMJ!zp99g%q62?Z{`;H6rxGEHOtLJOkJ9m29E;=e zL_FTH@N}oHW24nrczF|RvwEwuT5GCXM`Rx`tKDSZ3`B_x7vg*$LXR;Ek>GM%iA!;T z4=Fjg*q0$HA4U{Dn23CE%<9+vpt!a9GhA~YMn1Zs&bbdf;Jf}uc7T`P(asHR&i$+K z`VSQ!_{VNgF;ugKKEF;SQ(%9B`md=|)xZw}s08|hg@+80*L~34=KcRKA_Yf~G@t3d zmcH+<=)S^>;`0P?97Q@#BDsGEC4g!4o!$mY1{i>Vz#t1s0#E=0D3Iyhh|KZ=bfH9$ zQIs?sdDx*&;>oD&XkH3ET5btJG8vM75)hXF;>#Fi&R2QA6I@10P5D{7fj{hdUa<2w z+ke88-{<+Q7=P?NQr%X_R~~=Ogl=^;caOSRd|6!)p9#c(S2E(WVK5NZp%^Gj;D7PH ztdteXRb7zsiIfJ?P#VmKHLL}Y0&NOXhyWy`eP2Fk8+lCXX-+$pF~p+GQ!I|)nwguJiX*@^zA)!a@aIFab4;4q>abJW~w z+oO9M3m}BDkfmC1(pv=Og`KbL3_D+z6vt*SrPc`?Z+f%9#+{Di7|ciN9BZv0?{EWs z%5UmDZg`sL@6($w4P!9O;T_!%U8R$EfyVXqAve-TFanb>1>-OTBw*eG$g7>b+q=k7 zvnX=VpXe3}7xX2%IXIf=_TVU@JAz|i!C+xv!2y^`gMLJJbszCl+xCUKa342tPt^L6 zZO6hbqIvZ`{T+|-H~z)L#+d#k(0+Q`8F~&>$gxon8iD~C zgajI(14$lg$*Qti|%@^D*G78M1Lz3F_-0ftLT~xVV zV^94v?dCwlR+)%;Ng)h5u&FAN9OfXb8Ua$31eX8i(0x#4#{FNHCT$-$?fX|}ti>Ix;2L+=ecV|OLqCkb5O>vscn@=AD+`x( zPRD}9>$FbIf>0$Jn7^!N5vWEzYNXXEoQ4IW80E*OumluI!yw`HYc@)gZ7f$eEcW7=R6o6_JSxgLJ_99m+0vgZ&2 zKQ_h`Hp4`=qZzGefgm=*G^vJQ1k?u>C@$a-fipOzKe&!#!da|jSBF~%6i(xt1a%w= zfU$3at2Ra+1D7WzcifhObRWzVY5N{(!o`a+Hi7dlTxfB@;yMFs3(RGGTzy0`$Vn*T z*_z+$=)VB@_{bsxGxbLa)u@u8L4lfjsw46=DN!d)Sxt)SsEtuki#DzC z3iKu7caqKFm`(ka(fVl6KXKe|?=c(MOq$1^VgFCJvHxXCyfytHP-DJ;O>=8P+)NFS z)QHdKym!vW1X5n=^4%m>dJ;3u>gqI}7>S-(X-Ze7(R>A9=C6VqtEnAnCa;rsRJtD zY6redAntnnu!$uG5ZujrOOBsrI?E z9+x^6D|;QnV)QRuaizbwm;U1*(-&x2^3M7c=)geL;^waNl7wZkovuXI_T zwolhr!hPB_#F(^uz$X31H&KjcJ~yR5NC)=7Buv26{0ajsQcxC$Z}EM*%x}JOJS|Ov z=(&qx;F67Z;!iD55CuV++&Ti!&BwO0er@LqkMf-#Gto7&c)?ggyTWhT`X-^@!VJMV zEfu&+lLjik?(=$iv&;6Z6YCTWt_&hwfbbBU31`c-`sxUx{E{ys@0ig&fyn>Fj_AMR z(O;`W>d=6S3vU7?uK;O~(BhYp@?Q#wA|!9(e+#G{5eK$=NQDOM1?b^jPHJ~y zYhW;l{?;I21Ue9?VJx*3xq(9?wLjsM+8GR7lG=*w!t>VKGZ>3GjKQEd?M}ZS*M)HK zg&X~`07ec4DF-#KSjd5(0tf3KMtlrLfq`g1zyST_Il1JJ=&#*J=Ewr*FV2B=Re#mu zA!Cq=)31FvSI*1a%zG(#zdlv46q?79XP!&RQoIzI`}^N|Binde-o~168{USRU>n`Y zA8yeg5mv=AQ^<_W{v<@i)wmE>;^DZS8IEglY2qgA;v!aXgyz*0S9DPa#~oR?BQX|Cm zu{|X$Z6OC8xj%_jj(t&YNJoJkm9;_`@>q23PuCz*F*k(NIX4*Fr%;19hf z?7~AG&BMJitPcAk*0&Kgh8Pg`4&(0kSdB*_#y}p8M7#yWv0>M+4-ycMH;%DWt;@p@ zZ`H6SOtz0N+WvF89~Ni{FBVyf*B`}CFPIh!hDAh43f`YAV6E9g`FO4J35dNA$m+#P z2wCU1BK%BlSyW^{RC{gd@EW3Ik!A|#k1@G%wdk`h2UGQ~fUc@e1A0!~$OcX$m?jX7 z>qRhAtw}=>OBZT2mFp~DNtGz2s~9fMzZkhz@Kd#ZgVpIqM>AWt0@D~qkt$RKmrP~F z=N;>7quZ`6UDD~#06s8KAWj-0Ps0Nvs1OALLZh!p92{Vk2Fc&(!34olcO!y;=(iFs z1(1U*^)VKyr;+kqMjM@PG*S;Kl$T)xKm0ivt{7b0Ot1X5YcY}PZ4 z9cKv1hkU&`cu{|LaUs@#U!A{L=|`x>=V$se_s&AA^(;kbWtmcE)Aj~~Qe-^2dwHXZ zF<7Jmr#gPMYtmTuc%qoe8mvyT9Ba-v;mF&h(@Wo%iHZ8T`?ee#p=|L!xAZSo!#kJh zXQW`m`0_B1^GY)~YkMh7rmaY)Ym-?<| zb^c8&QP0F2hW)2(rZQu)*&uZPYnfiMriwAy{=)f{_4UuOM z{6h#I)NlqnJeIb<+|I*98Y^KZ73c%sEjC^Xm^88zl^M%aYY#lQz(A386cUmlgC&7? z#*F}LZ}T3cB*d$eLIyC>p+E!|P&^Ga@$t$afBXP(6>sGoe3(a^@iwgEBtF1np2pg) z;62#LaGu8|PKnnrly~tAFJiru2HwpJypK0y!%|(!TX>06-p&VkmglgkLD={_6Kj>B z-3ZL(yEuo_IhV`1nyY|_ROXwwh6_3K>W>>aiwn4vGdQ110H1FIT)rK!Nd+jtH$uC7 z3M~mqDW|DbK}Ev~mj0&m#4ja@uW%XK zO0t@)>=;92(nrHIDr?dhjmTOV?^4Of=)A99ed8DQW)#*NvHaT~li!^)^lxPO7oT@v z`OWC@apfVu`UMQLzeO8NYQI-{O$NtE44^_lv|JeBT#fX-*#OdYQc{Y}UHCWhNz{aH)Lpw%)1t?wc!cwL5l9m{` zts5`k@7KRKUgZe}d403a_gdjr5nlT$gbKs>FGq^C>Egp5}jgMjfvo>UAZgjpm4^s2=Gk?LoAds4J?&q z7?Pu#HBNwQ+E8%wa;(jjVQeZWU|2xIydFnbCr3Hwhy(W0pva#gR8Fnh9%meP&@n11 z7o4TajMSLP;h91la@vWb@16T)l$5fB>V?STg$?eh~W`73OA za8V(Ig9r^Q)F_ga#3Yjg^VWL*)*NhO46pGCK4)}2P*iA|aeR@1Pd_=TUTZRPbxR*zPHCK(x%`J?pTtzk9)X(MBqoclZQ!$4+5EO@*~{K?cBok zT*pJ)%uPVT&yTRgk5uQUcyOtMK7}BM6@gq zN(=$oM_1I93n!cYTed%U%c1d~`w{6v(O;w1Gw_QPu0FT2F%iEC&r{qmVu$i-NM2Uh zlX@?2-4mR6%RNBlJX}(|Zhti1Cj)DDH(>DpS=a8V-UipkH7J=Tl{0bb?|qRp6r>3Q zX~f9X#EqdTiU(cOG;LEgog*A+8X8@JshhItnU*P;iZr0#chg~bz$vmw<0L6$a+-|O zbZ#A8N!-LEB~1Bx_#Iru8}$Rf_bXDq#`XFtR&4mx|Mp=+58q%0399RS=sv$;Ne}qh z9hqNaWtuH6%V$v9y}q1zJNa)5-m8?|rp24IyY8xcG<%0@#GN#N7n49yj$o$Xh;U0s zkCFVO21+3U5-}xBxUee4R9%ILYDAudHp(HyJDkrSIXumyJibriF;eC;`!pC*3@8ib zAq~or>o5odiU~vl&D=YUNtL4W9iA?57$wr%-h`{cjUfz1zlzg1c?7etnSKTVoo*Xd z{{kR)sz8PKmm_fz7yY~EQ{{9A&mC_N-`CHb>;f-AQ*|%0FXAuKCC`r8(dhD8W?bx4 zH|sZA!V!V+ummbWMX6}k7)lp;i#KP96Om|763Cd^g`A$xaFSibb7#AtUDz&k7k&v= zIDHqqiySF^1$`J&5iAJ?x)7j*$`Wa|Pf#auJ$UegWmlz)6^I*=>CqC+(G^Y6eLW7J z8LiP7?a?^n4NA{q$;V1IsZ0d}=5gbh($7Wnk&T$k1{;Fi_vY7d1|HYCW+y>1TuE7c-IJ|*9NFYS4qNpp3DD=5vsRTozP*3$%7l>n8Bt=xjD|$>< z4AmFf#+N&d_R>u0XE* z=E>Lqe6#9gHD8)+MN}9dwR$NvC~mI8nj~*P?BiQN7iFeU(U4N$4^k$z<7?eEW*Ehg z5puFttSR74Y3fhw+sWNn_w|O^iT*rEcw|kd&hCm*TZj2fj z`K{WP-(psa5u8cGd>zkyK(rUi116_mvP*9v?G}xt3E%^z$ISSub7SpLds1IkW@LWxnHUJQt|%&rs6p#|H$R5pyCda+;P!-7%?W{SAi=+YS2-yQQGb!)>() z)$hfyw|@3Xc#dD(g&4Vtt3l27rnOaF&sfyoaN2O(JlW)IvbWV^()WqW>BA!O{WO5K zAPJtAcGaj-wW?Ot`wUxnJOtHmDSsb#mSme?H%ao=9i$2W9YT~3$1rc1Qc2AyYV zY{JO7DQd!-&?baW`GU_!d42=*XN2d^tdX&Zt4{@50FmyP&x^S?Mq`Wfmv6HI2 z8`PjfQoVJ(mIubtf8XRn2r)!!vYODDMOdo;g_n*GE!Z3#P}ifgbO5ui4y5ZAB~cz# zQ5@Ax&jjA>PHWKWjhRC0wfbXb^TO_Zy>KbwNNt-vA~qr_BF2(n^|NWq;ep;Z#Dlm) zciD%F?8W^?Yw$4FesJJ)0~|~wU2o($sII3&;&1SBkfaYCN-PM786mnpFgk>;mxjhV zC5x@mth5!PLMdXi(!`Faiw32PwNW2!N);bXW41ZbxDi*l(e|Hua>H)SwOz~2ahV%( z&355dZMl zN`3+lF?p0zKxqQPWXrSoioIG>0So_{ zpLqrtiQ?Y_6aV38B{isg@?5zfmi-r4P#*GfiG-!aUt`G&==ZmWx%ayzn!TqWDJL&T*54~kRhGspwnL*? zEvdzDAL;4p2T+;ks$lDy02+}>q!5;&kNV;=&gPfn zw#189d-26)&P0-}oo)DhDJM8c{$woj=g3S3vXY4yqGOZC%i-azK1+hM*K6(w8s6^< zBv5;&&k#TDjT-H9+S`3md$-1}8GC0T(xw5`6LcXbXVjn}RjXC)8YWJ<)u%4?6DotM zA;9rafq35a@epG&h#;lDe^H!nj8+Uqmgka(Hu@xou<`M@@$V& zOy$Vg?j$x+&{~&cWtD^=eD}DB;^=SuWSOM|Ckgm9!Q=O1Ou~6Qa%sB6q_397OY{<% zzDzSx=uP{0%I_1wFhq_6$MOD<2v5;t?-Xxy>PTxPz9CVFtU(smKJ06RXdl#KUnK@2 zq+_I6xQvl#5%)mDM~DoCd6!cPMLOZ9Ar&SOsv8+PA&W4fAIoiix9^?UBfkxA1IIFB`8IUDYtebb ze<@@0Z`*WRd`EIeVi(VPe1$c~GoS-9d|JzWBPo0M0$2r0mI=~3QA_H?Z@h_`u!_pU zq8l3q-MAyjk5o_7RVyW)aHOU21(Kg_=MUN2#qB~%>+{D4m+DHnR8*2JKew{ z*LBw}T_*)o00lMK6xL)>2!&H91vZ%ifW-gfJpqQM@K}H#6CR<0WuEp;%IDEF`-u>X zjZu^tta`C&fOe*59cnZ;Z+_Y?wW2sTe*kjIFTVN8Pyay;@*?JxN40NjwKf~5eOYV& zDR^PhS#r+6q|@r7K<3GmIlfwY1$!$~>^S)oL!{2gbSel>O@jl zMZ9k65#e0AX-*Gs<~NaaaTjwAkp`y|BkF7-R;2dR{@~(&m^>~*qZ_ei-+1i=;FIQ6 zZH<5LpS;)9SMQ&8Lysv#tuYD&Z7jjtY-QAr^&1jaF4Y$AL*cn#!OH&BU`teE7MI2& za~H|js;!SX3T=F(cSXdlVQYIC@+wA*t*kO>@3=L)>-j-Sv9hMOxhO%7fZL(_KKE`b z((Tl{DauyM6>_WDdc`Zk?II!Ft#m6~-@UqbS8iSv5X*OwX3vz5X1RTT zf4#QQMZRGit)p6=)4G_mdmWqgQ#jf54l?m;2Ga2fR-hICuF>{iWt+BsU`D9P?Z0Z(B?1LRJYctjeycm6!^Ww+YC*aH?AHPSunNA5=++uV^HFeljwW35{ZGMHahzBut&CsMVuJ_H1ydwkXxe8$IT zQ@`XBYeh&z#=Oee9Fmcl$6uUtbsb%+UcK~S`mM19tDcuCd)4Sf0oHXu$b5BF_)Jl{ zh~T5kJfFGAL#E4avXP4%WSp+f96ljuHstBH zN7|(V(;2~t`1brBKMn3dC%hOQ)^Dw%{ZL1(s(n`nAJPpa61GMTUKhI51?OID9iie~t<7n((9BPsRBhE!ecN_@`FCyc<3__qU4^Bh zEduYi(=tmN=kx2i)nc|;rEK|YR9p?)T^UtnHW%s`9}TZPmyLgKvqS>W4;iEkC-Q-%1N*fv)!wuAL}vhA$U zkgZMzG7lJ(Z<+Eni@@3sC)7-><7CrBElo2uQRC7q6rv_+E=!>wd{Rs0K_9A(C$ZQW zoomjdp04%#%gluPG7IjF>dcHk4l8s1@L-l&*e11;Hv4f z?|IEb!}HqDllFgZaXNl4eK4(hOWFS4s`_ z+5PNNbwl?~UHfmkLS583t~M>yE)1lFtPnt2%5uR46E_H4 z%3-e=D&2Et=M6kv(Rh+s@xVfFwK}5^4o?hQM(HjC<7U%JJ=85MLJBPu`EeX`7WFJ@ zPZJ?L+dJJCnI_pYD%%I!+wL0*1weknGa+|I?X&-RolTS4NeyCDQ)}wfI}aVL+>3P%BjhU#-S89!YnN=UD11zKmygw5F7?>R1h<|0D)c z>eum>U-E}krYhS-)f^u0g)8wD&{Z9LQ<>^j<5A%;hl>o@J{)Wl85-%w^Q9|ZDR?zTU*=H=4% zwW`<-90hfDnXMcuS-DykE626IE~xX4@9Nwhpdr8er`J88`EL&tI7ic9NEd0AW@wTo znvS($|2q=GkCYJ+(q)Md;q3nlgzs9X2Kcdbo4=XoNpgbaArT2y;A?p9Ty(c{%87n2 zBPG6?<>Kq`*X+Na{tMtf2Wjq1+C_YQTS4y|xJvVPd%y@u^Azn*`gPNS&Nc8FjQcff zfcsU+-8wqhtQEI8tz@3>8@&D2R247?3U%SUREmSN?YkGU3pM;%2wsM zBvqPnvvR7`S0!~(56LJ-hI{m58OK;gn5%jaMw@GBdHg3jt6J*B}kJe$p>3+(J>? ziY>|_W<1)#iYzdpo|U4ml{>e}oQ5S%I}hFOLb%-pJb~_YKDx(w)w{m$d%lgAzTrE5 z=m&Ta@;NwI2y^GDcqp^$WUKX_KP2m%Z1+XMfnL zt2gzq@!|ahbLVwB?e=gY12|e?u@;?N7FEg3P+(n19PkRe`;DZH;^h`u*nF|XwM&sdIkFG5iX9*TR z!KxHb^Oty?Ijknb%A`LkRoy1C%QW@L1gA9>(CTyk**_gDBN}PV0x&G$ug}k?r|nNG z|H@om(UHgtQV<{n0r00FddMYlm!JDt=ENfxhikCD(WS3kma}rZ_Llo9u-uf3a$c^> zRTa<^u(B+bRbgq7Fr(T343}j+8g2!??lD-^*mpNF(W&u+J-)fO$7AJL<@iu~NoR{1 zy`K33-_ix$&=p07s1*zZ!BV<7f3UWXUqB%~qRgdPw zpPtfOErfqfTH7>5leIw8HC6L84Ss>zU)Suzx@N3b5=*u;u$g8A)8q=vvJ6YI+=WXn zop5Lu^UPdwCF)!Af+yF%Hg9bM&%EJF`=Kj2%#pa~^unb)8+R{6(b;(GPagEgIY<5C zvmUti!%u$i&lvJ6OZPj!J~;Zj1aCq;h(QTLP-g(P3d{_&Qn6j0VWnZqARkUdYv#gJ z(Grc(98K{O$w&-O#kLN}+M|ii93XR4ZZ+m3*S|HvJ=QwTu>IRotz!GP#0yp;B~@li zG?lj{*E6Gh7oz%ac`bN#a2D&s%|ktfxo`IA7coBzn1_2RR+Dj7Ba+!KH~04>=7H2x zawGpv2@2mam+U8slG9r6W%?tsxkkS|Ub*o#2uA{4IhMJybV z2uo-nKS)Ic3LF;rs77r{D8TCLfttl09c#!9Wi3a#Gqtjwyd)*7r}mbP|lvdV=)b#pH6 zbrTmlANRSQV;qeKT-UMdpljLQ<0tq5?suJOscP?dXP-jk2>#PcUEp*l`Y)$EiL+A% z7D5mDF*w8j&hdnk@f-Ek=_n0xFoTn4VxN%tzxYo`LLU5JgeZg|jhq;cJ%S$WpshnO z7#%SX?a>uO(G%U#8U3d;`R}JF`5)6N{44OFf*jH{)xy9mGQ?Xwvv4UM#IoqbIOgLO zLl+%Szn)t|jJw@$1^-!OhyMO$UXXiuPl_G;iUHGVxQL*Xf+G`C`~rf-1LP^|uy(h` zgA=n=9FbpwNWOPXP^2jx0xUDjzm){Vw4ajECv zp#V$V%Xs+T@T}Esoqpb+2l8TY{1#H;rwk#6hFvo}g2Sa)P)V z54)qwLK|l;!y?9TmNR15QggFt)o9U*M%19zlh%hO*6B_$3Tavs^Kne*LpP;8>@&dW zy&X_EcAc4|;_LuWWYP^D9_BqS9(vlHBkwhBhVmtVVZ<0ay%5XG*kMC9h^BTD%_lhW zT;u8qfdjFCehO6>b7BS$!!)kdTBV!aJg%E2B{13*kvSQh_Mw-A)Le z#BL%YgD+7z>Xp5IZ_u0croF&h>a8KCLOSJ)I<0m9rE`Q@y;CfuPgj|FJi<$sV8D=03nu5T#~#ma9xs4qYO5CfzD9l!#J z0+N6MX1+SycP#YNb_Xmel<$tq*lEbLu~l5d~2bE%Cw{Fa(ClHN*mSXhIhz$aUlZyTER+ z9~=P3W(ze?3yl$x$iyHP@o1lgn3CzTm61_#R8nqdTvT1|WDev(UKE3xpiXEIM{+io zay2(_Q+9J35Azhy^9o~3G=j!~E8}W75(6x-B@Z+!Ji4W{+*Xkzt+n;H6=_F$kY1!O=}!hqu!Ksu#7L@?NV!x?y|hV>3`ma*$e>V!DL-XZ zc!DcQNt9X{X$4w|R;P>#T8}oSt!Pg=P@^?YE458Kv`0sjtSp^ZzAo#kG8HrFoHs=l z#t_pPXVEN`)s&ZV%N*sMJT_-dDS00&8`F-kd)a6A!F~pGZ2)*Q0Ty`5Pg6=M_Q`Us zfVMlo&>F0UmU+`vStoNhz$Vx}sIa4?-O7_lWM}B}>fMZ8f;qcl6j-sFMu#<4&9G@> z55=At&z7bn-4{*~G=0dG6C&8ki4o>t&cfJ@>ONg^3f)f+&_mRu!WGmM=g={6H;Mv% zDJUCCQ3o+(7{vspbp5C5JZHDMAn&?6&4kF_m2Pvu?xzp*P@H(C7kH7EdJWDCuJGs$ zcrCQuY#H|8@9^W_4#xqH&_^muwQz4eki zZgE@PtO3@xfDc4bT$HqS1gW(%Sgc*aMaE7z88@#i(yRkvCR(YB@UV_V&>C4e>joXW zHi6Zt%tu}8PA@tR_U=jd+MRb7BGx3$kWcgx!y+X@rAbB8f&UZ9_)J}Y5(qHK67yQE za5^9&_&C$*DR~}B_t?n54s*=*LbWPm1GuJqN2sY7X0BOgmhC3DTJ)S&|0*3 zmS!_7bVgeF^;a1v&;FY3u*Dvk)UB~(2zNnhTYN*`PoZNXW;|w=HRekRD?qX14RdTZ zW=`Kh8AV@DV%|EW+>1i~fY)CAM+u)&%rvAnqh8+@$4-D9+?GQ3`KZ5OKpI~sFlcN( ztK#Jsj=lBwmyWOgeg4}2MwCUf#xU8Hj}ID1cEb!#@()|}UV$MY3ZO7&JPBm{1{m$aL3Uf==lK}Z3r{s0D;am_F}q5IyYKkzlkYJcwTfm zbazbYZBHz3=Bv;2=gJXo3!DW79at@gFzAm8H=Sb4E|?0Qfb{-;~wy(cBuR@(Rs=nlTNEUQJ+b>J|)hL zP)fS~wu5Kvawyc9HWhaoT#BVVl6f$-nPOLgY9yQ~(OBayLHKk;bt_>Ciy>)ue7ROd znhU3Mr4b;lG|__(dT8vP?+$(rH#6aW?;>=kxJ4qL!n@S4GTyVCZ%%j%q|SaTsgIG~^_Kum)gELIETEbTiHZ8@$5 z$Ojv2fHVw5Pj_wf3lvmE3=*8za)HIDPAyn33`;pS?je>tag(41qdEly!owpdUDaZ0 zMjiE7HIi=gu^GME&;}%ZPUMXmLr|;fbVc4e5W$mb@1HxHke&$OyLa{DD__p|Id#H}JspIUEk^7geH^0Xt$b%;iuf1FaP!$!74G z61C($=`1lTWPqfw+~o0nHYT%x3;1ikMwGFQT?u7jbM5#PhEa5pOxXzNN?n#-!}4x0 zt#U9V5L(qHs;LcEvVl-xyzVOd1!x{!1gOit+BKm6bdZ;Mz)NwOuy+23E#Z_+_Smwu zMJsC$=)D-be3mMKT~~;AiXv6vzB+OCh@r}^FzeM>%1{1w?z<-B>@(cYPencQ@l4~^ zd|7Qm*5@c>K)Srthv~7k^Xw{}?%y z1IdIG-2NcPJJhzoHnVpBP%{hlDy#fjM7w^Z8bA){*Fs0NQ5UGI)JI0L2! zxwDH4XAZZ4vll0>`Q>B%b zWwqBh;8fn5w>%?1b7kR=VcsD3CfHqF(Rk%V!TP4sp6%G)y^n=;wivclwi`c&2K4EP zkMoL;E5+AXA14c+(v6z(W@ueO^+-foa(Ke)Xs1=8d{Jv`%awq>0LeO0={dQ!^t0N2 zp6DrK!3JB^x=C$6E7iCp-`3t(Mk$?Wjacm*o$@)H{<;kskg@h^mB(ovcKh{g`IKEBI_UDZFgt^FT& z#y)tP)om3@?6|U~(|XjxuciC@1HS#|l{E!)z2MPAsNo!Fn;LYNXpZ82Ca#CXN)St z1wwAM{iZBFPn%2$88Skc${i&3gf)9i=)FQwb+qwUFHK^py6PF%PQl9FQf?fXZ|py8 z^t79M3RqM3cD5VKZ(Nj>`R!+-Ms-THMjq$tmQT_}(f_g`SD614btm?%-}Fa9#;6XR zvl~k}C}QjvFd;x!ip1Ls!JB*{;Sd~iZonQOIMY-&jH>9u=5XJi-4?%z@!QeF{=01&jIpD9l{P-G zq2*i{ZD1MDsO|&;W$)5TJ5!zk>6qL^u$4GN&}*wxZ?n2suo}1cJ=HsFC+_VJB}K6< zd|Uz~sB$@gx0;;t_-*wVqX6hz(TEj8zyJzQgEP&;P}I2)=G8mN@bMc~gyM2Qq$Kl( zay3ij6TdWbAP)6F$vD+Z2D7QceZ}>y{kQ4m8Gr_Fs|mmFN$HoIR4MWJSf#dzWhnQq z)kWHwIQH_{qZ4H;E-bF@)@#LaJTb)yrymSjys-Z+%>nsA&Z5Ye@!ojU=CWoc@=G01 z%JRSWZ8IPtRdhrw?V?6rkWdhDTuNaN0XD_xV@ELPG;_VLwm~iUn?5cr0!;&Od?W=VP^9i1TCO#k6!IE3kBW524y|;ZcGw@$ zkgG9ps}n_x!lGy`?fM#mtdX>r3j-&Clv;Sh3MY+D@^(9*3yCjni0FR{6lubcC@}re zwI(i<{G=&*e*3FG5&|T!ltS0tU|3p&h7ZV(eeVKAw;Q@On0`Hl229vG__PWlfOw$l z#2l3LCD7_kHB&&r>9lvERK%qdL-p3pCl9sv&lA-|Xo~!UkI{N*Bs+Kfh|3iLiRtgh!0RmZBWKufUpRlnI>1)Ky*U^*1O(84Mu|z`` zbV2Hn%0#!sP-Bg*Z+pF1pGuK?F1Z>jA4T1s=Ks_fTr|NE_h>*s<}5=fTQpcX!6k#z zH_=lPeMU|=7x-*l_G0`JCh&hb%pUV&a9zxF1e6e$-;AJ1RST?uMweRQ7vu!rM|i#6 ztGaeKXn4OwpDm7)Jm@Y9tiz)AmRTssA!C>ya<_crF1|L88&l0&6ksz*8G<P~q=TcW`07I{_|dgyUIA<1w)o^cASD#O&^p zRbS-0Wb&l!9#LoGgi8SAsC21*g2+;U%O$4tECqC%w*{09xktnCV+G(DU)a7xz9#Y9 zUq1iz?_PQ7XDhn}y3{>f4!58Wb{v2$oqXsnZB-Ium|)ct+Xkx7RcZSD>7rLyjaakQ zREzg>3sHZ@nXqpl+-CMGM>q!KK+cR?@wXmEs?XmpJYSu;m zyedjSVehRdfCJ3|X;(AAK^0tL;mM*D(BY2?6`gA_4Gz@X$cXf17fettQU+eCh1)Kw zM}_o9dOzAki5NPk*o>r0^q8@L$)x2pSxEm0Fu+NL+AaN60EA8e+^Q&UE{6c)xml~} zQpO&en}otx^Fcn?n$mRkLIS5TdbQwwFj^=L8}P+GDF>vl+^$sruy$Cp@z*B>sfl10#vi}~$}}t~O=vf9)(kKb?k5FsOM&Y=tpQX~&9Z=Mgi50V zE@c2v{tRe`g9s=}0BAlhnm|%ok|+S|%YY(eb!H?U0_p#QC%?Gm#{)Lp`{VszU{QoX zkkIK&4@|v*7k_`Z`Wj|MPi@}cpXtNyj&L?wnRmUA%ytl78)DIjjTE_=flM81CpVO6 z4fUP`wjqQiW!L=XEUi2_w34|d6Xz7ZE-Uope@R1asJ9%Y!VB+>8nIO1%il=?nJa!G9W7m!q-?zv8C*k5*>Cvra+1k!1dUTE>H>=>-$QS^fa>L>fxqtoX z3oi~ROlV%)>Mj{|h0ZsXtpK|MUP0wZ_V8cH9WmKUGxtBoeh9~#YMA;~!&F@Aj3gs7 zc+Q5DR{G(6f{A%}Ho1V?RTZJrkNA}z7=|1>^nIEE1gvo!EaqqYxA~!|VK}*{iGtGy z;yZdIM(d0zB!1Fg6-&~Yra)bJ5_&U+T$TC#J!rNgK&$l=G~SEqUthXtZ^_BdWnV8^ zXIZ+l>{{hXZEmOMb+qK5Vdv|QJewaO8g#UA}Qwq_OykeeQnyiubbz3j|VWgkf*W@RFvl%r8Rcer&rft6Ps6jZoe& z9!A!=N;2?it+D&d470Pjd+L7u@|(6G|K9S0&PC-v(jP{yd@qZ z4r;gvc)8P?(n=FG=Bs>Ig6Y1UXBHa&px4Lxpa>2s1MAZUlIUPuXIx<^j;zL*_ZG+4 zO1MI-T7o&MATHY3uPh|LRbr(SAn2A4=6hOIda*J9wBR3qgylZ4|1-<-9ia1sO!nhM4=Q4la<ijtj>wpr4E(Mm(c~31PlaRTmoR^LvRQ`_k>8$*RLAHZ~)fU1)wif~keAHIalVI2h~7=A7vW<*S40Um<^8#;|K0&IruVv(jx&}v?E zW)luwh7zq1(gn=YLITkuQe1&Tdk~}iL`4I2p%WZ*l^0hDfKIHh9K{TwE?qfMtM$_o zBLylL>T>d7igp?7!S01MV{{!dERm3I=-~2!K;Xd=HUt7)QwU4WHJ{P7NaSOaUIK&x zDB@T~5J|KFppj7}R*^7?`{4h5Ce=EX0R7u)y_fE%it z(fZ0@aoc@wXt60NTf|8B1i^`IDkVWN613SE`hp=!NF9c7o49p=Ord9ySS0~H^hSaD zi5qtaOUGE^E|Rp!8)YZ}qvI@=Bvb`atu6$N1_0T^AZ-FjPpB>8B-I&*S`eWbIdzMm z24Z9YBFjs!yum^%m7voULvvI?wK6D`oN`g@Gjv!6>m|hspzKAI-Vqt53t~hfYKd3x z30e=FbPbE7Vuk|&*!zx5kH=O+gNK5!rI2_8A&}m9pilDiPx{ICF)hhI<|8zcY3S&e zO@RH~(9>t2fCJ;$7#*T4Sd0-)bRd7ZQWZV`ML@d0YaX;rubW&-fYmx_wMrQ>e+I}? z$1y<3DHL zLyDftpz;N8yoHZ`k;6J3&GePo>a!bZ|>X^tzp_ORmzEl42IV?dMGz|IytMF7rb@E3%wA{ zgOZz^LA=G63vlldLV?dqiPxC3Tc-tiBhdnd_Fpx+b}t`(eTDDxqCoVpdvq={bX~E~ z(QrwTM?hzmkB?~LP;NZqHcRY+ic_!w8KfKp7+$U&M$_~=#GpGeor2$s z43|+MalJ>tlj?w#id;notUf{l z*@p+FypB9@g#)A8dn~k_vQi7CDW?{D$UcGZDz@HM{ef&X_@c@KT<(B#&@rD|Dv(Qe zy5>KSRVA1|1XK1C<_*H2@U;wke?T%wF6SR@VsdlBCT%Cak}|3DJ7*d2ELMO#6(P6Q z_Xc7zvIP88c~qgmV+HG3CSrcCB13*pyjF)OpaZ1c)dO#vh|`jLlmIMCYIzZH<~aD| z75Ad1!C}kG%It54iw~4pKFO&UG({TZf6g0eK9D9T{wVOs`%5>c1^(ZN!Qh;W+pX z+sOEpsM?dH;d|!9G#RJzN0rK`IQ=k0qi!0FBXS>C;xBw4P=w&zNuo#?EpBodEKm+< z$RQ7|c9;axhYs;lFOB>dr8SKfeM(C!sWCjLN)8R3^W{#ZYvTvc`BH z>iPwz*hq{knx%UrBE)R>DwMD?#vIRb%WG-f@f3G=0Q^RYUUVm0c`f9E>6n6hAV~BJ z4jJT!h}+!-LFGO7STNR=FanK?V|I+#o6?rHhsf8{;(YkErBt}OgkJy&1O;UVo_vtg z7TjZgA5(d0qd!}fd0X&|IX&aur1;+vW`4jN3HYVoK|&G?fUmO5INAbTXB7Ud(JH$I zeJIe!*cZMKB_9D%lUg&KfO)Arl>Vv zoTHT1;}4ksvbgMIc6@gPXft@s<%2USCd})MkU)46DvU7PdxU_evHPM|B1!hSie--g z^$C%do~zflcKU%T6zX3bfEL2J(&`S=)YxlI>TS?>9qDetvee06*uee zA>#^YZ)ZT8JE#7XXLy=f`ye#w-VWUbc~L_;b=cY!y^lI9&$eN!-p9G81X>!N_7moW z0^27B=vnTH9sk)KogB-Ry(b?x&Yi1g`=VrFPr4k@%C)?&+QEI(?dj=#*Lkg-!i3&L z^o-m0{yCsS#{N)xy7la*;?FUp+<4%l1bpB^=Z*&fni_VewMDw(-jPj`jUNc6$7JTKt5rQ3<=Ig4DY2 zZ#{9%tR=ko#Qw==z0|?Hc0H^mjhDZTufa99hHs;xEvj2*_V4Nr0+}}vl-BkO4)G2y z_;f*K2H@)TQ~e+6CcazkfGBHiv08suXCkUUSm-@{ zoXR-#AhW_MPqZjGuFR5*N-E!wRe6stpJHhAjZ2~7&V+_)YG2U$^KUeLY5mu4lLw7Z zDzFuBM9}^9lEy&5_w$<6uLk%&<#+Q!hiRj;NrU*9Q@T6O%b~bG2FB6`*BzNZ77Wq$ zP6-bu0XW}?9{gZ8`(jRPJx_nOgW#wp070Y0GGHpkhy24q>O8hM5_F^I`_o8txc)YZ zS}>4NA6d+)L54`RF4fA0Mj!Ke{O1^{J4)wC4mY=Id1EER2%P=sEq+ZW^+pMAuBaZ1 zgb6%b`$~Pb!N?|n0-Pu_6HsuFv$G4E+Z(M>0vA)ocwtrC+K~>fqh@oian%L^(HF}MgQaZ8W>sj6fcvib=u%?-r zKbj7&C%3MRUWseb0?W>B&9PloQa+*;4R=#{3+u7Dos_RN7F3$%paS%TO;-ppPSXi4FjSxzzxMnK=x{o2EozAVDp-Vy0&B26SD~8V zvi39!k)|TlkN$rGf!TU7Vn-TSabq#ui?Rf48oQ>TIdHNmgo-#RxC6Xg4iX-mq~Lh_ z&^-A8wRh)CEIeFxU4!ovz%GP`UMlW9a&Wda4`WwzGM}_pcv5r0lT8~Q0GC178F2Hf zYx__+_e~1V9t~oN8}b{?fR$iBYW<}PPJjnpoj+73{;rK3 z9(T``#F2)t9@_;pWjx}nQ=qHsrP2HvBVg@AOCJ;Mn_<(MczU=h&i^+ldHd2dv!0_a zjyF-#`AaT8Jn%D^-o^V(X56R9%b}%>Y%@zdKpoH7D2>}Gm}i)Suv&Hn^}KlCYtV} z-cBUrl)hCDes(4q8~K$rrpOdNF%66dQfh*@1k>HBW-C0>#FRy+Lku5>II6`8^kvpH zo1@o|;K%GCooTA%j&HN8a*VCqFs+$|+T$e}%@!Q7SmCE!=c&zAS5Z^H(DIs&iXqvW zwR+J=ki}=cJEkFL4zIf2uOc>~^1f}uh}@beJlHHkx3|?pNRD(|_Uz!WlcPc%jH2{- zjvrLB4@#vwFNjIX+vym0g0<*Ep)_Z<$Kis;b`G$!|F(IaQR+-rO1i*uSq*=4p+uqe z_=ILox8{g&j4r;YCXwbWo;-Ta3n=j8)0QuuDI7d=rww03-!83%vSpQRrETPQ9bpG~ zt)jG1A&uUtN}kLH;xpG{N^5GJz_$T6G7{L~PC+y2H-WjO@dr#uKGafg+dNqC_!WIX z`l_I@X_w@>c3$R4Ad7O=mk3M1SO^BkBd*RNDAOUfe4w)cZ!j0p=AD4W$=pe}Q@tif zSNorj+MjVrr3oMl14m_y0jT4F><*!-3GYDD9dcOhQmQr{fF`J=s-q!gRgVx+y?`&R1;6Kx|><`;Y`=HdI`87#aSmt?vblSOhj*WU2Jh# z;w8cfSNCs|{03jQqAR&N>L=9RX((#b`Kt?NpqJwddeeEL=wMP%wJ?+Qx+KB->u7_a zH_-df(EualWquQ9)e`QuYjIMo*caT<)TGJ>R=~!3Enw0{we$Y3saWJ<*a;iVhl5ab z4xF6)66ATcMh>6qs=DxaeJ!rwYF?eoaMLV+9<9zO%Br5484oYgi}VGX>TDKuQ#b#` z%*$I3C$zQqJb05d8V{aVGJK4RESHV*Hx0Th+&v5Ih~U1~;669=c1 zQy!0=c+3ZWHa(-N@!sUzykm9ZBQH3sPr$YQA|a~31M%ZYKvQVCXNl9$SddY9^`-H1 zn3>hV3$<|%T@-RT?hh^R4g$d~`cbW5H#Duuo4)3)0er!mFpabNsR3w`(SiR5ImPwL zF!MzTvkV=1ft=jj0W3aFNi>cw5g-8ri%=h>^RP6w z*s_O?`D$+?21|#9ElDWiwFvW)QeuW=NID{l_QF-ItgDl#>|XFuNUAYKDFfE6c?TH6 z=5#XX4|}qDw1^sS`Y^)$PqHM*-G2$NlI+q&@0N2bs5fK|rJ8`d)k@+pMk7TgN5A-_ zw~@M<*xHcj8ftP;%DM|Qy^IJ_(~438TMrR0kU`qkKb=EnP|6abr_AHA$Z{$*r^b>A z=;0ZG95Pk&R-{7>A-Rd%VLRBHZ5@3KPY=PLg1qkI4v-tEocx`AS&d_upN&`$ZB0Su`8@7&TLbe{(W1?TXnv=PzXoCrfW2Re@#+6t&^NprKF_HxLnGfZ7}! zRw4M;SE(({kXCRzH~{`rT{qe}s#n$45r9@sL7TOrJ!nyGbTIfSb`(@x?ebywQoEOr z{!7oGa{9GP)Ow1Z;y^LCemm=HD$}FNRB4D>HJ$BDYz;uwjr)8B`JqbGc6%Wo>OuY} zP#vU~o~g_rAD)3k)@!K47qu0EsQxi&?FPePFDBjm#7I0}owuU6x|s?oDhnn$iy^mq zJGSRyX*Vv0}0MPJ!x;(6aTPa2HBL5wi4xJmSj@HmH+l1NO>Ug)T-HejRXkyH`nAcsL+v*s%Z9)*yLGrPvsKvA zO@CbU^u*@i#zv#vUQss6VWvH3qi;rx{-~`hf9)JQ#)cOOI<}VON3n95STe^WTg@U- zW|l7%rHX83Lu^j7Dl229#`%_JchC_%LkzPX2fAuX)Pmwz_$v9pD_ap2qVhV;&zevn zN6{P!6>d(PPl-UPW9c$Gwky_rr$b1b(zl!$E}JuBFf)^iQi3B39_1!A^-G;f_l(FMzS3Nw7pngUH?vau7{nq1V$d`NT2thHoGH}RFqei zi$xDsAInrZ9k&UK&o!u4w`H4v=Q4hW!qL^OOr>Cj8IWyC%J2n;qqG6MlW1*%+-vA5 zEmqvbPF?ie=KT&0Zr5n|iVbJyCmVNG_0t=ml>T}&R7E*6eW5AkzEPl<;XTb*NR`-> z7+)NTG_orY8ETzQuSH1PrCbZM+q%3k8@+?YrA+Jke_Y z+AUUZ`+;7gE;r)QaPI zjkA~%tck4(3ib3Pe8`k*OboNKl}OJm`VMDo&mbRa=}vhf{0&hn7@e@iexzhn$2Xa^Hxb}(HQze8tP zU8JK~*ffq=O@oze`0m#DO~eMMc2nAF3XL36hR&H!tYpcP!&Z3Bc$NCuf>xTU z#I3mXYRI5|_wXT&w5DR0yJMTUVG&HUO~l%^IteNEV@L5tfW+Q>;)(011xAL$kzrr1 z<+1Uinqc*X(Aktb)=sAo#I4~RX?DkYoLl5GJ@A<~C|2<{@*uUjSWiv(0(!tyTsKk+*};HB=_P(tmC zBjbDo0%^oy zb+uqbFakj_b|M0mON%d0eOb0h8e^`*Pl*c?DrH!g59W{Ls)t%NI#0-^ZXM^kLLBTV zk5s?^)}l7OdEYWBDZ>h${*M*WW6sPSA{>(JtUGP&>_bLBEfP&})bY}MhDiU7JEkTW z#_jp)=J?G&{3VpR?d3M9k@CMhWzIy(9hE%S+f$k^65lvm$h~ClR4M(?^_tk5-Xoyu zI>#*CBJUs;k>fke>;U z3arCEFXQQ;FZXv+Jnh&3YsxaKGCC}*pcwZyiWf8Un_pn<`h+skzyS`(NQT>2>dq7I zFA0ZfIQx@-y*l~-0{Hr!J~>DT2&Vl)MV&PrZloxgOsj3vUt*In`A2wblP)U|uRJ zaf~&|M3;A2L?6+DL`tez@m9zBE1aVveS+T@R!_or#kyXySK-kP%;F@$rR;`O_FY4V z68!v|UBqi2nY$8)k@7&38qMpGHCEeMpj6%$r+Mjx$*6TBd@RtR%TAzoS9&Ug=pL93zC=1K=Uh8w^Vk8S+*1?HZbmH%DbRi!;VmR>#)(xyL$F}DAdVRUf zOtSmP%gj)BZ?bw77{J`1@XX{#$3U6f%0LW$=fYivUWlE2ec`xMu;Q&zT*gC*bC6dY z8}RxhPnrk;Rssqe6RCTAj^zKb%CGZ}RD&4S9Q3X~N2pm&c}9o!r#On75W*KmSL2c*pg;W1JyYDXe<4cj``EL0wKWZJ_gL+kEta_ z%9LEFwHHar*N)f4uu~`x^ko=~3 zPH8vcjv5cZ=zs2HR~v~v8Q7~$IyLlG6C?&%f&HhL4-p^Ug0q$ZeG^ONHF@P25oVA& zG7wdnl70qA-eopOkX12xZi^a`pAgU6)QDVqFY`=V7omNAeA8B)_fSp&11le$NUjF zm5*Z3R5-w@Q(@ZN0WPLVl~O}q3lI)z8x*rWwdV+P3q|wI*&8&+t|20ON`i0&lsRf8 zynxCe$HGk-pbt-SuS~%f+Z*7hRUSC%OSCjPMD~A3BAT4l@Vq+I@TUVOWS$m=loPzy zz2a5=skqE-i_#S6dYx9YT-K#OpR(+Bqdoky_albbc#*jR$h7d+aPHUJIC}9#&&Quc z)@&d6)&IOtRE^i>*zN zKHhMD<%x6!?LcV^4Yy6CQZj^wxYvdDUKtxE;U9knem2Ykz|{nb*3>gZW&odC)w zb+R(gT}Uy1(vBv=@_;q9@+lqb%F0+3>nZ41zXlSUHW6$;%;)sx)%;p;3u{q;U6t(F z4Ncv?0_>62lib+66vUnd?4@)?q+tKrRtH<%rlcYrTZ^&{tO~JEB~V%9rgoy(&Y^OZ zWl)1%koo8<8rU^En0&f!*H3QU45IPfs@r!LsK#_}sHXPd{s^^#YFn!7ZJ_DVZp66{ zfSGyGtIh_tUVd2F=HwfDwTA_)9b$c*V@b>#m=dOQAfvfy)R2YTGNRKa#@IqKuAH5H z2Dq;zkwK~^)fOkDwo^N*)yjovx5kKYAc8H4_ZfY5kLb}oap49%Ph}@q@17=m^aV*9 zou;Jb_oXSdJ%{!^+;D2a`Lt$kJE=4AmryNzBf46nR894!lEh^vVOx`=_R%1X$IYj1 zCbWM;KBreAh~~#Q#RWsFFf|tjml!sGktc)^>omF*}(XlC4HEBVr9C=(q9d%vg zVy#G8tIZ?AtS%?*UaIEY@0h_97>cx9A4gX7aKyS8J8K4D&E8xpy_HzB9ED{*`ew~Z zR>xX9#@m=pHFs#s4ih`GYkRNWn6=)k+ZKV2|2XMNV!2%9S_`lH94q`iPxAV-O}%$r5Aw~u#F z!SzN=(>M3cd-IRCIa>q(@AW_`^xjOkpMyO_yvV1wfo&LgU-GI1|8$!a{{WDmj`>t# zyi}oKYtR=55`;un=sApsHLcY2;Xb_hsc`7|BN}K{r_*d3?c=M_g_Hq~zdVS;UgiSH zjIo`njaQMAKr7UWfN2%OX}z=>ZOGK+wDF?W4kwoAq^b7z+?03+V@buHp!3r0Q3EhNU0;_?91gn0r{ns4qw6LxT`i_Fj`iU~ z>~MYfC?(IAs^sO;l)PEGLG2TBAaOizVdc$Xbb)BuSfCclAlm>HQV%J6d2aO)% zK2{d-^E6E?O&R#)VwhjTHO{Akoob}OwBbGTSzso~?+vq@|DlDKv28d{0`YSBuM*=bp7+EeM@h}McropbZ`G$e)-Pe`3!ipYjuIg+Rka0pSULx$x;g< zRyI%Xzxgu(@Gb`upnXIK`;=Yx?&Y!w-e*5m6uq*Cp-i@$Ogq4su<@q(A=vTTEmi2y zVS}q_UJw|p3ho-L4few|ZN5H8Tz_tp>)iY{?`Q{GK$42FGS-(>FTh?Ed_{lgDm?fJ zD%R-+zbET`rY=DlD;udX??<_&z`0rVgnsxqZ{@gO`##>} zhUn%O21mF{;BWSkaha?5Ax;d<14 z7U`}jr+rCbnkg4roG5@Z2sVczfGq}zIDid&IHO+b*`0b&&+1h;Taj)E;A~$-2*{=D zhE1<>Ew^%?B8g~Ziy6D1__!#?>{cSBKuVoEP48-1a2;+y%AJ&V_Ddba-MAOE*#t89 z36Sixg@sYIqOOvQsOC#^^cSCcIG5HzZKN@=^yjSF5e9NmFxS`#JKruQ|68oCNeudE z8+~wGb1Z+?`3bvijy1C<$JVJO=XkJA1$0iN{%YGWt~BnDk4IZA-b8p9o;1_5$GY>5 zfca#9`pT^BS+Kf`>Ya=9!ZD@fc$==hid9uX4U`fB^*8Cc18Ad9XV48L$EoW=#RS8L z%+GRhZ?K%JQ-dE$$_by#8CRRh&$VO(T}pzb2ysZ#j~5u~M(9HNC^|t_Ou|=7tTQj{ zU7vQ?qHu#MMe9WpYyL-KfTN&hCNUWY9oVf&U=hJ=Q3TCwJr2+dR)iC?rNTsUCaov6 zjwaf*{iB+*Xhhg>EU`9t!1&O#Gw-Mfndv~d4I=Sn@zr6mq$qczrjq*3vv?mOd`RIU zX=sz^asGm8?|%9P!|4w18H@&yG#C27Jxx{Sgb0CqA@^X|lX|W3W&m*iC=-JsE8LGd z=o*Z!rnk(fnK1Lslv!++nT^1aiJ1U8W<;Kpd+>7^#yZeH{6MblE|lzQ;Gyv${(zt1 z@8yXmSU3+u8RvGWf?HGo9}+wy$Z0#dus|xOx2<`r%vBh;RSt=T=Cs~)yy5hzfVyBSmN3dzuo-`M$ z&DX86I+!g6vqqXiw0@7%ItivxL~wSUFGcI}HOlW8fO2k{XzzUeimDZtSS0Ir>tnHE z>g;XYkgcQKp0V{v)F_c~7&yWixL5@AB5kHFR-|gg4r;Vbmq=1ST!0ryKBfIgkc0BB z6_b7*Ef@ffpqF46RJ@;nRq!Cd3SRJmUm|MAOa{}SbN=q53>&ei606uMu}W{@9tO@1 z%Tqmad*=3SEi!jP?rioNmJXQ!e%H?|D*PDg2Q-nx(N@ox31sZB26jAv68<@&TeBGz z@~-H@jEkcodo0Ln=U-wDSy8jnEY=&0H5QSza-nJ@U%-yT zcsFCbZi_R_!5KC?h<1gN>LvZ9mIunb%{67IAg>f1fz6}lE>l(ZY_yc(jgE4RZ1j{f z8x@SdGdPUvg4u(8382B6lI8?cN|R0n$DHGG2t_WT9M$0ClgM!ePV?YUAtS=-;v=)A z%o!vGmBFns41`Ny4$GoDo60sDashzT#2@UEz8o3I7vv{GkDMcyD1cnm{Jb%OCWJVe zj54^d=q3=dZZ71$y(y~kyK{fHYt-CZDe?F3Jm27?|69YBniDk6)VAF6w!<5#-}Jkn zX5o?^+k2(y9~(A1y5qH@K_y#GE}5)7DSj8;BJ9?zZaq46C8Aoz(dW}0BwNmX5ewfE zo`atzvs<-n6pr}7TS+jJWEZtOrx;tT;FX#RAIJh;3*=?95v;iRALBfjsc=+uUiH$2jO~}HBt>x zvdJsN%xla@s?Y9z@)?sbFuddYd?fM7LQ6%hyyz#K_gH#eK!t3raRUjJV~XZd%?>c9 z_jv$5rhwi+A(h$vNtegkLu5V;k>zJ*LBxf!1i#A7z%F$IOT5V5xx^udT1!tIW~V!I5iwOho3Keiu)5Th$sAaBavJ?jma}MI z{i{}%Iq>ul>0KNQ9K@<^_{^i&myWAz+uoqRDY{SQW6xHstf!w$ua3E10d!^2ZD0*h zt*WfwiPpd<2@0jyxG^_^*{5}en%+^JCoYa?0{itxztM~QC--*~7nmn@d zM$3)QTH-^_+C;B10dp@rT5oE`XOugw8XRwcYjH6p_EQBmVQSqfQ+2v(jvFPIXL;EQ ztE#UJO@(5;O^GX(B-HE_H)A2BuJ_I?DzW{^fru8{hL|r}FA^uHJx+NwQ@y$G*xxvE z+PZdbN>g1Ek+{~^N@QcML1#2TW8pB$`7N&$Q;qqyhr!sOApw+|hm&h9(a_`g*3eTw zx6-49JmhuI{fvps8`Xo1GC>bhgZ9(gz2W(IYkB^9J$u^>7Bmqx?yFd@Z-oc~qin7W z8l0n(TIe*;3t`T1JJQ9?&IuElmfh}z(YIn7BVD778{(DQHHT_f?ieoyj$WqQ9A1xB zi_Q7Xl4)d18zM=9n<7x_H&KMWk@2}3{A$|KHh@~e;4M&Da()jVeN%`=d(7DMYH)@!ymfo6u#b(Q;6Q_qvN193-7Nz6y z`t&b48?qZZ3`;BHVZOgKk=Qtr-$-1K=Tihf%m(22#2YfxFGegPz2r6^Mg+OstV_B` zuf^#T?QTs1BBm-^?}{(+IjH>m%gIA#UQ?+XiWKrHeQ_85>EYe-cRL*mEwQH*QM1AV zHrO}qSgz6{)Slxhjs$O*=Z)Ns(L;fv@TxC#;?sEV#m;hs*T;v_`bwN(!7P-?Nyngx z?z?|`2DZMVE`96Wzu*7gxdxd{?*3WmlRf7uno{@FG$pz3h5En#Ao9$rGX7LTHSMaa zd~P~79JH29I|ZA)pG|CiozXZ}Iirw36(x1DvY0a8)@`SRl)?Ayvnup+W+Uj{1xLoBYBjO-E zoX-l>yI-Sj*q1$bG)#@2Gzl9BEcRtbwPHzQPVQ)K-`zCGt8@4A`gAXX+n zBpOnXb(_aB9Clj2Q?c&&YJq5pL58-{De7bUdC>66Ej z4-0;(n>g&0q7F1Eaf>>`;vEGVf^elF7y%U+0{gOu3j-Sl!cw(+EjIEbOh?jf#Y~W= z^tE`1rZMy5stmW7(+Rh(Lu7+0r7&{P3H{c25mZO3cMR0KxxR`7#qCu2sZ$wb-hEh? zgjes+mS5jItwba+t&_cuiBE`0?|vyQZ;Pt+g-t!3R&|~ll&?r7B;i;I<$v_CoO|?n zL&BN3(4~~D!=|+uy+pf;G`l^d=9~_pdwr;iEhSxtA*ST74fAf{*}5_0>$`X5#SXzl zlG2u(rnbSj5u5R{JkBq)nR&Pk)1Hb|`{__HFcXt$Av2bJrK6!kx@qiM(yX=U2L015 z=f-?jf=GGW+cju-WGZ8aP~2C(IAW?vg88yi(!4ewOEqmM;}})~{eYMk_k`~5*~=DaU+@R9%rl;Hd@ZF->mU7V;MKnR~2&tfy>mA^r*Jf|J$KoFB@XL+A~x!q=r{@ zFo}&s2A78$&5S0ppg(Ej0hHY66^mIGo2`IqKn??;6J(z9365sV_e;uL*q2D38QTH4 zmg&>S>!ynUd2_iT=qk52Kj!7_&b;}Tlg;(LjD!hB+42)VmXJJLequt8eEr5hnLpY) zGQObjC9knyr_OhM**bI~zRZ^&`@QA9=5&)Hi`0%vJo>&IOdxtvnWoV`Su!>|lQvwQ zSYm!G+U7VlMQ9E}LYKaUJd&J1aUpa=@IYqtWn^b$24we=nw{Ytt_bC4>(EvDj5)Z~$yqv?hQjEB;^NtFNu=J*&<~z1Z5)o? zJn}C?$5XQScg~&M-4kl>M2W%Z%MkSJ&r^Q%(TOj&h!5i8+|gb=f({M%W&uJm6Lm#&GgUv=Gh%9g^%`HT)dFe+(0R7Pv4EQ%eJ7y zUgfJT)*?y!@vtK6(Et3X`2phnv*G`uym83g;K{HPxPcsm0!*cXYQl6;Pc6+f;x6s9 z5~G|DRn*Wx3qk5=qKyau0N4Nrq<0>yKk_a;p^e{Yri9T^6=S1n#yK7hQ#O)@Iuwln zoJ{!PnhRE2VRphe`R=HHg}g*6(vgO|N8TYHkq!BTqHhgUU)`a;GXfzXqKBgIhv*}Q z0n-nCz9HkMBexJ6#164VTv7Cs5O4K6Lr#bemgniO6gMl~$Ph+_xWfh6!r}xi!bsf< zQ2u)8X7xo+ZJmknz*cmXTcA&ui+cn!&Md~d2am;sAM5#p+&0gCi#ai-o`^)LC*ibM zl}rjPo$YP04u=6kJeufI&-TP~Za)Ua?ET!Pa^N>-7@S!h$QxChgO ztA9~M*2YkbmN@`!F*ABD!8)Rp&Yck@^^486oFz4DGOxJZ`a{T))*z{Ia4#=@7cNU? zeW)(yM>vnbV(%cn^O4=?OFSch&M6MKwaHw~U|x|BVoQFXWqHmef)dp#VO9 zJlzm#sxJ!tkndWmw4T(W|5EF6L92Dm56zYvE1Nik5+zR!Maj}czLA~XjpC)%~d;s+d$dUg=+8U<+O6?W3Q#3^_Kxf5`#>le0V1G7Yo&jwrLd}1fN2=Zk2wbOD#qD zvXnu@far(XGJqQADwwG?z$}|rFUwY>Pp!|SsSUYn+0EVPph=1;Q&iXfplBV-b>T=I zkkI88Xgt^`>NyrFXh8#7lbB_r3XjVJ8kfgZ7P<{H@XU*#YxzJY2GG6yMt@d7Z);rv zy^pOxL_kbfftfw)Vd3EwR6A(}(~if;D(Z7DJqVDi$t{&u>9Zoq%Ui4~SJB{s!Bs*` zUr+1brBs!j%&R%1!!%+}pcBVbI$UBf&v4etDCVx}V!nf6vp_mG)`XQ=%tP<0EoOx$ zjC-?M$t5c%(YhJ26y)_*;274{d++APC7%U}H4fjs2Ft6}>`z&%Tp-d-c$2BGN?7-u ztEyW7EnvBKVhck_Xx70JCNSO}uBNSJHB$rZcp2=SGqA9}!hV_p6E>??QSnYVIC3zPzxl!%>rH+|Jxr;JSWn2&Hu zCI=d>^!#=VJS<`n3nLO=Dgr0u~@38fPZ)3TJJfQzMN?<~@)MJT=PUm=U1IG3{>tfeTmN&^9ntZ+2s5E!ymh)zvs*%e?uA zm~9S?92U_EK5I2Qt1MzS=fGorjkuaF!&5-SR1$a{x=QTc+k~ZlO^GQhoNe(=a|=bh zkb9p|9unGUPS+^CCi$SbMT_hf!(?)--Wd1c2BT$p0gX*xvvWN3JI znkg2C(J5RH<>9Ouv93{MiZJ4j_f9n9@+8@;ZJ+T4Hvgs;T9%>(k4%hXPDvAvLnH-n9{z`V5byS|BM~zIOo$HS?CCJZI;{0SaHpfKYl$B|68$WtvItC{M zW);xF8g?_8jb!xl`cxlfPU$E@A$_qc)eDC~RWZHJqVeJ0HH!tg$v=xv0f8ZB^ti*@ z{%nf$NO`^xS>Ybvj$YRQNQK__&OsjlO+61C%>fAYyQaw_M+)y7{lp58RG$V6 z*mv!Upo(MmQpR3E8@e)!oX4edh*5pPKX!}&4gEaHuc`9<)<5LPa0Up_*K=5RpTp%4 zz|d}}`kf7pU z;vcGC;Y{#4lbIf z*KD~sNe3U$s=$#r;qNyfJm*FL)JdIgJ4~_E>iw0_Z`IL$hd<^K-Q0DyS5Id$ro(6e_mB-&0JviwSRTWIxdKP{Z*edK_Fm`0kn=C7+OZHcn!y2Dp1qiLtqMNFi~>Ia zMf)F9P;1`@z=G1UeAC|0bltST3ulnm0W>%!e>oFh*Redad8#JP}idH$>oJ# z*EiG@BJN|0D2yKMVFCra3-clcjeTLUcKWa^^3W~{eyb?D4urr47DWk;M^6~Qx~RkH z=mbMI3Gbo{A9(DoiV@rfLAW1%q<4z$NHh+uW7;vI{9q_o7{d!yeT&nB41)05f#q+k zm|nZz#dVTAI=Ti6mbVK5B_crKA4>FdE(k|l)YQ$!JG?&v25ysbJv<*ZqkfJOn zkjV2+jwhYnl>#l7)joc~7x~gFP)uXoq--Gs2Nojk09|x?S=wmTJiaJM(nL$mpw~~( zY6v+^wnZG?QNM_IJqDH=6fRi{fo!2 zt^$u#;4;IDAJhE2e#%TNba3WR7+Jy#+Wi>@;j;OW6kJ)NSx8JK@E8hi9VNldC5d$f z3kQ1MYeIIOEx`bqzZ7kW!u#7frY+<}GeURg=h?rYq}#A9HE_2GvT57$J|t&ZVvgx$ z#2qc8If?BZB+ne-SeXlhBn#ScVKAb^`HHMB8v)5= zS( zn2V$V@sL)bc^4V>43Xe1yIxTw|KKnmRH%X@E=q@Hn!P;FXo1UoF2BYm`($%5N%Uop z(y4Sh`N>9!YZ{WkQg^bM;K;Ed{$!hK#N9y3?FOrU+{rCGU)PGCc!__BFnZz}1tdrU z!^G)H{FRW1CF8TQYq2;oxfYZ@36`i+?m!#%zw?%Gi9D6~o=9Qx(OX;YF;Tgg|_CHabn68@LeuUWw8zcdr+cOAbc27U3` zBpu#lL=N9GF%y0AiIh*C6SXv8Z3!z<;~1?;_SFTe8Qf;A$uI|r)!HL%Yp+R~PuJ!& z7&X@6D@&o{x~wDlQRp>U>*z3&?Js)Qu{>O|ZJ{uRmx^@w z)1)1d)SmUhDx^mSVP`dBFg_}*wf+{4@LJ(2kX$(YyY@B6xwvlX-0ZOam^T@B{pHdL zVX87a8T3egRaKiO0R>?kvfHUX34(8bl0+aCM*mecep&e>wEX$Y_9tQF$21-=ZUG;>ej>*tZDSiAtavZ0=glrl^-ywDj3%d^oEsMr-&! zo1zQNBMT0npCw@Wx1sI)9DVH|I?m2I?CgSGbf6mpX(Z6#B!)+qvyV`nC&XPVB|%^@ zpO>2UJa^v!Vh@x5Z3_V(JZFdN2P|DFrZ9Of10G}N@{aO2FWF-P^9|1Jh;(b3OhoxGVw%bP2xxf3yU07*W z7KC~p?9k494Q22`7qdK~(*#$U7Ia6RL|`AXmH90El(A2?pK9hIOYL1d=w$xu&iR7V zZ>igtw7~LK%=fkBbGJLN$~St*zG(|wk^ZN*VHO#U-`Kw&KkOfN4~PA%rtar?VZZ3P zJPPwA^NhTGIKB5t%zV}D-;e+7KMpX^{WrQiOdiG$(|>)s%6iu_4-oPG1620$AGOK> zf#02~he=O8pmiUp!$RKZK)@?Ci@$kftVah5{nP{;sJh24c|CFC(P2Gp?cYbM-FkLl z6Khk%dae#A?*xx5U24yEG; zyg~cFwb|| z|Fn+_z~U%$_eT-NiZMqSW=DAgWLLYz0Ty*l(tV;x=P*6$6_ck1Qhiz{d5dyBHB+QOswP31Gs5E zYE1+8W62%nqL%v6iHDKM>7pw7exH1z!7<$u*FOA>!tbb^#u+yobKKz)Qt(RJT&jN? z{SylAy%{1~{IA@EUmQ!4(~+ zQ?t#J4Nm-4_rx^+1avQ+7}04zBJmC0d^_M|XKMuC_ge5zVs+UQ%WwY^o7sBm6X;bF zlFMl)Rx^a40jhw!e3sbLGpK&Q;jOeJ^qKMgS_rDoDYi?O zjv2h%tJ{)s&DAw9h=!N>YI@IE8&a0;eAqF#9=A0;lM8OQHaT@$rsDqLtT(-sdOZ*{ zm3Q3|bq9)mq?((}01lsX)E1?Tr6-uhm~2?Q;5s#;vnBPEFHNr%C+O-omfm)3;Huw@ z@Amb%GoDS1Zcpe=^MF|T&fVK{ij9=TN=H|G7@+kpZ=2^q4Oj;JKRnOl!|}G5M5DR) zI8S?4C`m>dJXn}zMQa%xx#xw0to|}bl)Uq584glu><%KdBtY@K{&wQ&;&-`QM{uqW zYVzoduD=1rjpen4ByDZkGBnaWmGRM6Pxm}7ZfVpB8)Hk87NPAW@t$7Yl|+*(XzSJojliJW zNs2kfS(NlOg<#3d)bA67aCJhmBqesFyu@-(>TIXF8H6{(ZZBV2*X56p?H@8{wO^5i z7qd9b+a_lruq@~CQNx*#D=Y1H<+A2#tCe8bq|n3BrmZ07>xDJjgumU!v#wK7)x4s4 z&I!jobLr&c9sl5*t#sDdP z0q>};NP*AQhk*l^b{LpTPjl9Cm7a-1D-g(6X=qK>ovw&?+|2Nmp}1KK1nSa93k1v} zch}$5@eININp9#CI*mp;J3wp(wWkyg^pQ0=PZzsWsa}_vTy)31?vu41Zg1lAwR}A< zrCqn)m!bNgM&J3UAI?Y-jh8g^(SkVeRh7<{DI!j%n=^Pp%Pm~`;nzQ{`_tz5sbjv& z*~}E$-O@kbApgWR@^2up#cY1RN&l#c7Xf$Y^%>zM@xRx;Lc$jwJ_C5a;9EqdR7`=@aG4Ft!M~NMK zg9PI5Jg%PSiSER_3`M|$)Z#g}0LluX#-SY`nRyIJy($h` z2Dd|c+K%M-Ie!*)VV4`+`@=e^FtxXp8!k$`d*mbE$z*Zi#bfcsP~#Lir)Cd22^8T4 zhM1O%v1Bve)iCv9u3LT0#kyO$$eT~*pKyloQs{>bjFd`=7j+kNd7iG#lWZB_jsEuP zVw}+UU>Q3(*$M(v|D}owi>WRzPEKB`2-F}*2rdJI#ZlMwk&D(zGvDbZ9P}5oT*yTc!y(b|l%a`m(&tTy>OKeU;e`o3U?#hNb2O zAnN88RQ1%_>0tecgEkeu2*}Y^ymF%)K!IAfazA!iELUjh@fy%QyKcn|5fVL^nIk7mL+591pbOfdU6!vX75mC#1;Hp(aj;o7`%@ znBh%#59_emU+Wy2ZlwI-9D1A{Fr28&tQF+YwxW2WUlu)lAWn}?jQJ;hb)a>qXLU0p zwH4Ib9o)SI4dy%SJz=2trOEvA1m3hCdO7rZ(h_C6ATi>K?f=!^E(A{^!<7Zu zYDlMmd@G*<*kDyVO7}adYai<@^tg_~(vpqVmJS$>LhIev8WATHdUuP%;JD|DYq*0q6~z2`cPQ)<;9`%Mog)M|9D;E{wl9R`gNF2Fqiq z{+13cb*ubIyFa4Z-QP~Tb2bf51v-&#WS$n6z|m9lHLDirgB0I7JI-nI(An8BEeQ@U zLvF!?aN36p?k$51tr{7gB01|mXRKZT7;;cX%6WgD_o?euhfx-oC(|{0YD^w>FxBmU zQ*-*7kAwHrGB8{_IA4bz�-*q-TrP(eg;ng_LIn(|@Ar`2fSwokH8^l?>Th+m+$? z<3<=NAvcY13>tLtoudpdi;7svgVKO7IrR|`{b@?u!O#k}%Y|E`4QyOrEc9cDGx zm#40hu>D)(2>rs|F-kVcL63J0$LZf08=BIjA|3_SzO~y<>H}=c?(*`$WIt(7TDP_@ z4^B?qH=DsazIDG?rRes(iPnFgojAP*Pll>LPfZ>@M6?YraAY!?Oy={KyL$pfM|(Bx zVUuYO&x1#{nEYOPzpqN8COh0;-09We94TrTbY%g!_Bn^fq%e;<2>CtzyKYrve_aK* zj(f8D7OJiqs#|Hs(nn>du2{KO3nkQCt%X{BMtKGbU;4y{>+{VukjfpL|41CTMk!8y zK@bCU-Ze(8>rM#ub2J3V@ZGe!C~MoBuDLNgT-?QUK?jsR$<8@JdtYmDpE`P|x1zMu zV%3w!bx2#v)Am`i{wN8L|LbX(p;O^v`9XhO^y2GM)8u??ujpu@>1eRQW7oCv`HV8s zap1}81xsFPg!MXZGZSOXvu_hn_<$>5f+@?Bu3$l6;M4RA0Uyx`gBWkIJg?k7v;M@V20UhuBq!drp|@3Ea;U06Wd9G3ct)ABErlfB2V$XuJt1y}91 z3p!7i%%PdPQ^Cq+#cm|^6>nm?DY{m9`ixWd#f&2p*ZahhlOK9Lz?g}xBNTEm*xV6E z?jJlL+Lq&b6?hWzUHi58b9rZiRTALcIPTLD%z0s?<$Wr9+v9DH8vOWJUN@*NN-Wpk z%qmgh>>4lfT~P1Mk{Tbs?@UR&)c|g#(zQxeQyvB%yF3u$X0xtYMKGkwtas7N1cmZ?#uGz~$}u6*qqk<6W04J_yV81K9q}-gpfIAk|HL-uFLUMgK@w?*82Q z+xq?zyYHfkK<`uhlu_<_Kfm*QzRDHv-#kWUbY1>*@`09e9Lc@ve2dFJj zkQPZ%sH{P=kDAlp4v5hyKmB`V=wOQB0z2=q< z;xZmLGk^9;j1E+TUQYV8>_GTu3pLe%A zSArY1sCGI*Kmr!MB--|wf*A>Wgn}~dT~LGXAO#(wHiW}|`AG9@g2*D`;eb_8U)GCa z`DOX6aeXJC<@j`f+W#6~UwhH#s`HU4ML9hnj!*nf0+DzY2Z_w8^_&sYv;~=u)e@Xb zQ=H{Imo=(KGTFoRLTbFZkm2TBN|OD^LS?+F|0}fr;A(Z9R~)gIDC6yYvC7qbi;?DI zSk`)rm?$_n`!DK!fjs^&XNwopfVLm&vzCXZM#93Aa^-rn)T&i(E@ph_wusxlbciB7 z!ehQ@(>VP|WUOA|)X@?@Xq?u#yX4r1BhI;U>|YAGngY3Ze|K74O41q^xWl8Rh8+75 zR)zAgdZ2Kgux%$QPXdVcF&BB>PCouD<2+r~dB&$ZZP8i|ZTfx7h;XwlBCjIk?Y7v` z_%bn>JMsFEf0gy%%$s{|5m#uCP=xI-y)R$m+h&c>s(D_@1=7Fm+<(!;(j2wbhyB(8 z0_nXgn^1T^|0LH;w~yu#eHxl(h0}1jTqTXkDn*f1+Muq@4TkQdi6L2Y{T~5}IaW*H zsROL@>wu}3DB1t<=4wjm)r>L?#|#SJFYaq>B$ih{QN=^E4xRp&cO7c5sYCG6S9S#W zisLm|HLvN|QF7}a8X&o@*(zm7$1eW$#ATD?$B*^jiG7=Hx;J!f#NQW# zk!t{i1{vo{|_ggSJfC;c*RVl8m4hnffo)cV)>PgcWC;Gye!;v?M+ z^tYF_0e4fVg&NnLw^fia@C}^{pbE9G0BA9|$NAS2ClkoxpZhr-p>(|c*%%4VHEEu$s9C;@~$5G z5ji)9W?cv6CcQ>E^`y6~eLSKlE+$J(SX2tWMi;ev2~cQtF>HLP^k0&v$6>8>ZD z%holk#wxz{ofuwAG(Dd<_as!@lTqv3h%}maPpOIZF%Ho#kB?83MG8o zHj^8QJ?iAzA*#E=X14IA=bwN;{p+r`F5i25acU**G6=H0D0Vgx;cKy+7RKkLBdVTYSN% zx}fjIzPu{E3pg)aekaq|*ks~+@h@3BLl%T(Hono>4MbB~Mz&;SSgh9B z!buU%N<4OC2;x=QX`h1av9tMgeY=MGrg=1OA&6XJFLEX~a*DmJWrX<Z-lnL&sLb;yAT{3;ybWaCI&g2v#?E_C+1R(*pnld><7r zVJj}Pr=`=%tQtRN^VW38^ZQWdZvxn`er<<}e<2=;no>vS;du`I>jDTWMz^x`F}2Ra z`P+JIN?eam&SCUvYACzAr&5hg@p?xC&4QJY5VJp0SfaGLJ%%nZ zl`Vwk9l;9iYiN8WL$fsDQY=UZ}eJChoxFQS@v33r$4O%`di{Q4}DdAk}5>d-OIrkfWhW2 z^~4Yi3}hH;@=op!Lfg?h%k5|z?S$L$MQysd_4VbeaJXhV?NqoGc_$y>-p*#f)q+KO z3OHJ6YMUv4G;iCkD=OQv2FtxcA9KG!Wi;3%r{2F`?eD|ay&c37O*Yp7AYqp}9^`nP zPDCJguo{zQJ#{Ye686zeX5I&x^{wByi1F({bZY`Gp?=unjVnts-frod`r zyDDSa+c3biGthRIL*HEyb_rX8JLb`6C5%{SUIcc2>EV~^Y8xTc;hav`)r=!8TKHR8Dy4}&ElumP*?__eB^a~DUd$R zVB=LUeADG!h(ORi$oSm}#(#ASc^KorFZ`9l@G$=g43WgA0wC{Mw-_MYRTbVn z7r?~7>SW?e0SYW#YWN^AVn`c^F9BhF6(G@&jze7^PRQRXJ@s1m%kgAzM+&{Zt|Mnf zVkXJ!Zn)y^W-+0Y%|nCBR+P6)-uSofi{s=l@n;3TqvLmuE45!hy7tS!V0;}uPao9& z>|Z-v?f;-qV^YSCe?6+O-sQNf;g@qiAMP{PTK8J!( zWSPY`)}bA2I<^f8V>qDAjf!!ST2uthZ4t#jNgWB=FMV;lzMOv$mtkY}u<4Tn%>bEW zi)h$V9aPoh^9t?QCUX;8?rC+)<8R4@{L4F|Uvg=8G|X54`{Zid#ERrP?t%-PdB*m` zP$W0=e^{B_TOD_>ZIpXqo7@%Dw!4$qb`ST#D-wiF^0583;X)H@xnT62f1C04xb@qf zFrPO9Ba^2HaUZuBW7&rB-7?8H2xc>Dl6Pa!_MT-*-sjswmV6Z47=0yge9wYC#w8#p zKh+JUkm^}+jIy2E(afQNYZ)o0WYI8moRc4XP}0=4Q5hc^1Vl=vWFVZIS+Nx&+RSdQF7bTr#H@dvixL$m{Ng-eve<5d45wOx(#Ek9?~+t6+{K4%Gn;KT78n+4o&o0+a?`3^Ud( zY;l~lM>l&o`RMU2k2bg4*zne4)MTX%Vx+zjto z=kt0KI$kN+kEI)X@W58Z#TmZbT@L3~>wEh7>{pImNnCCx3$8bEk;K4_Uz+x&;N7y! zX#URX)Qw&{uL@4rY_+4Csm*wohfR>iU1|NlcbpxWLa*avBDuLakEW#oRIpQeg28py z3i#VjKjZIKc38UoY0agM<5svMR`PyYpCnc-waV+ChG)g~ig(`AemLG^}4{kzJ#3Yl3>LI+p-h{jGier80UMWRkedXrUO_BuDakH`rrj#of&Uy_s zDmx>??U_(@+gq=~OW4$B*DA&-sd7ZRl!9~N3J&b-Q}4K1HE=4cCfs%p7X4mzA&jWi zI1l%AH7QWNJ+P8}Yj1vgbbEMv+%!#1Ez5SQ+T!f-pgM4==n|QjuG-O-q;d-2MbjN& zcNvDEzLTaah-Tz%*QdM_1n|b)TvL@6cB{J;lU4`GGVS_qt%h*U`U%>Ite(4Ee-GIF zX*0Nv=HJF@x7mzhN^`2;F`BwNpX-1)Q*p1?4-+c)`#WZ@#LhHy$J!rVkQph$YBb%e zy0!$V7ahHI)tnkoI3(EDYu?S)$_9p((?KR{m4V%~B1)UEh_q?PM2x>F>uGP!HC~4u@0Shk3@hT4JCn;A_Xt866lNI)QiN zm1x-0946J$=^Eu3&d+krhJAA+@RHQG7cr*31C>4IzDtT4KDRVVw;mb#)OVp`_=@dvPM?#TH;lGL zp6|BU7kz>uRBl84_cXm7@tx7|J<@Vv~IOt-YO z?|bR}o=Bh6Z!d1>7GkHzBN_C6XV{xa#N=y4Ihl@?uT&pr70=%|9BetdXy@2fpTBTE z1)tY(I!vCcGQ9a)R?a9cLpfLc=DOUCx#x0z&y9Rv0#s{*(s?&$obkhrvA5aGN21h^>IHk8@>g1UIO8}}NdEKCxpz3ny8s5ogZ&fF#k6y$P% zi@tOuMt}_FiM~2{0*86N_0DYWi}N>LkKqKMG>LwQ(^pPlL_f~>c1vZJ3_~FkprbI% zM{z{IG^2oueg#hS2V@9PIEbm|q43oevZV17pT;kg#&3erlG@!BU42c9sgb3PznIXC zPIC#O#(!62o@oi1v1#rU?=_IwOnG#nS4%ZB>RajyLNS;+fp83?klF)=)QAy4Bxc~n z6g3ul-}OC=(_arpH)P zt%9;JOOT^j{Qvp|w3kd#IsIBP)B5=grg;I;(n&I?1N;yyZyECxYbH%7tMhoo%) z&0ZjR3f0V#|B>v_7iang2(}4SkVPp(Lm!;d$QF$IYu{X0z;Mt{Z0L zmLSk}h5l{=zd1b~*aZTnI`Z+cS8f~NVXxwyMR6HwQW?1mQt}bkgO0TOcsfO^&|Wl> z*q2~?O%S?RWFJ>UX9b5-o#T>!ETB02DVCHM-IGm0eS20UvN(by7bpA+a+D+81@71` zHx9#!jmqN+hwYpx11P6uco{-Dp@~mF8KXq>X5x698BxYjPNEhw>Lwt)6r|yaS1n7B ziwk$#?kI6VDML1FRKPh!p^YjG1LWrnGD|bwD3E0#_#5 z%Lx0Flp*mUW&0$bro#YR2p%;sQQ-1SpfaC2MiN}D9<@ujq8y-V`8l;K&t>KEJT*se zf!j~?;MhGwM}2=MF3UE zG5t@jNQd}d`%r&xF59g^nhuu-y8?*pbNCP%rK3TNR%n8jXo1#HXofbKlX}Vj`}hfDnjZ2Y1Rx6~ zs5D*O7ibFQRBOukbm~!SVM6-hYq9jd{AG!3HUqY(_JMgk8j5kGi)Ewv( zB*8l92FtAIiH1{49H-&WomQ%vK#MC6Dfn*=T+69nI(+8a#mS7yOn^~C`5_DQr@1^x zk$K?>_MAf=`>MI?pdmT}BSSD9stDhs@=ybQnjK>QcXWe3ufYX3wnSJC(_(O=Nnv<^ z%`b#aj4WXld2~IHQ-prrA2`_pwK+fuMU4fEywB%XoW;!S{NF%dTS!7LlVAv7>AW-G|WAO_`_y zc`i9Grg>#dhCoxP zWg!BVW=|mu&DM|!$?9diLzFtJGF< zN0H@wkGIHiBMrbplkDTQVa1R$U?=XP#B$;mu#4Q)&matmC^(2pdxnZMvCODGgb^GN zL2}qa=O!Ec%z*~S%~ulznQkJKMIc=vedAkFvle#(N@)Hue%qWFx7&=>rSRC9-HU8A zbMN1sGv!gt81E4iL1zhK`gSN=B^H8_uKF2st{VAm#eVM)bE?hw*WCK?^B79{wlN!y z!1Lzsq4-=@DZq54lM-ZpUbs+#JfO&k%tQw##tycV_=%pU?MgXQUM5B*K04OM4!#xWz`>AIJ_J`^S%KMjhO8)q^&Dhu>=Dk;1hW~6N&+X{N;bn0RFxcRWjhRuY2$uSx%scoOr9eec)dBMxS~XH8%x|)u*oaq%HW}-3nX+W(cyyv4Lz0A!;mRy ziCq|=3C9P*5P~b*3tdGeQ?hfvF~K3_KP#{>PJuUn@8j(33Po0}zF~-pSNQiTZk=uw zZ#3^Ee7=f05y-kx%Zk>Kei-976Pe-T;F~6F_IPk>^VUYvyLB_N8{WG7P6E+->%n{v z;MU)iq71r#D(Fk zH27Ebr4||7AMd#q;DUQ!W*|3mFhKNVW zI#mxuujfW@-t)lq8Sh13_jMfHpyQLs>d&rEG_hz}-u@D7d-aydqm(||gkNFmn~I_i z6s7y!T@RUrFh$?rK;s4{SNDffdDsTTBC(Ph86$Kf-`PoQGbCazB^*9XPbgTap zQvEAA?7@|wcqK}IU8{z*6mp2)!}%uf5BA84R`E2>y2{=>53=}8R)mAuWM*ORj}rry zrB6SzvO%BJ&N8uXk6tEE58dwq;>p*%vtP}D?c{*U)Qx5TV2AL`P^Mk{``XVH#M2L` zGILwy+(wzvySir~Sb8i-nYDTF?0&>P>%UbM^HUC>NOafwABy;N9dKm!Y#@4$7!y9p zjbhZ|60^8gi17KEDtjS!qgNmwMJTi&M~!t(3G0)9GF?eyd)|;)CC-KpJDiN!-Omv; zqPJf}N9JNIie#J)e*G^&KtBe=kLH)cxmYrgzroew7Z&ZSeoOy@m2lkOf`W@9HA&}HsYW!{`aLXT+qoMf$)i8_ zb;5x?jjg`ENbXnhZQJ%W__6K8ege(;Gc~02lSVY>uLkPV{c(fJdRO)(_mw zR0(dc3LEtCgzR1#=`L3vf?+jV*_r9$p5>G&*-Djc_N$n|fJ_m0Qd)q+rw@j3ao7mI zs_jX)?yS-VBj%6c7Dm)9RE8uHX}AF>Tof!?yUB%}(9IQjeDS*~D@7M=7j1?X>e?)R zum$ag0YYJqk&$Ejksv zlx#}|FVMLa-GDB3TXZ$dSzu_Yyme9KuNe9;0(2Wjn(z7bAqaHe-|ZQBf>=I!GK>Q~ zw!4DN@DT4Y#bAHLzBRTmo2f#aA@^xoI3yjQDr4}sQu4jbY7WnIupCBOAA|}mt;2H~ z93nGc!9FPgPe_+Axb!?}SXGw3kIU!TEiAFeQ7(5iH2Ty~n4HXg`*Jw@dngF?Y#NM* zeQZ)JVKdXGM3p;mFDqyW{feO{;0`ykB@qpyCUbk=j;vsnyO42xniJoFJF9c`cy*E} z5rW~PZf@4itMdKPuz0Cv!mZ(oN!(9D#Hdvt9wof}@^&!%%NuO8LMOy)#TpVvdCPP){mr=K-0`7K9cqU*EHMgIjmR}-cI_o&ul=M?Al$TV1w zO21{0a)U>i!s9Y+dcf3amM1@cscwoH@LiCnah^GLdiA4lAHqzBd`tLXYUQVuE5GOU za}xKDC5xr`xrqChmDf2?KGQ0b-&Yp=Ev=nhAop|PYCG4%q?nG9fBA#m=9zP>@`&e{ z!`ziuHhHe1JhP5MULP-PxhJ3Jq|IfvEzjkIjV$iTOEz+tptHF=^I#*PtlcPM6m=uT z-%3&q373!z_r>t>*@Crg;%H+N&o<1y+?a4O@#l|C4b%kjz^JcQLDd?rE}4-&bjScdN9$ThlPVlZcgfe@K#n)b7UJccG%?d*qxiVsA6uZyc*&{uji%J)XJHKn? zJntzT=e>)4FPpJTcnr~!vEXY<@q(`>q4PJEXYcaN8RZ;}pXz?!K|-b6m5j2R$Rpz4 zuY~5X1Vo$tSgJ|+>DG$Es}$b2FspE2p`^AbtO6A5Fy_dcJ-S`m+k<(BgVE9(MqXH- z=$CANNdl5865N?T7OtC2QfRuRBu8P~4G#RD4cu&%{~zN+Y{l1b<+^bE@`#6uqjcSo z&h@%S&WUmQHXsJ9hu6M%6(6W{%Uvm)}DS> za($PbeErFx^RjfVhQ5pIEm8+>=t&Bul&Dd`!0|r7(`lw>TLp(|%C)b{LfltFwcnvi zto@KadCEQEpILpsh$+*NV!Aku72ER^Zpc~f%A)aWe^oFXK&_U;T z?!mp7jK_%B$nv=^*T5-$T_}z1huCtdyD}AI!3{pLI;hN`_)X<~%g6kBL|l%!C{+ne zLtbE(c0f@f+=_c==Z=W;Qk_{OmhN{bJbl<@e+-mTT zt#NqQpcBkPHTULd{LaZcmloOO2GPNUv_eLsX?Y$JkWMAYkpq+y5*;$M4ai-BqWgU4 zC&I7i^JbsE2=JhW@p%espSOZ&-<2q2Osj48L1Osh!TIUpL?L;+zBSgkC;G=@+WcqN zj0}uW#h~$uCc7K+x(>0m=o{5^mT0bg-^@$cYxw&9x5;4HiJNqiY~_!khp8D)E{wm}u8 zL3XLCqxW1O>+cX!r~E z!`_+^6J{MT2NSAm_Ud7`$LhEmSh9Vht~1xs#hI;QvS}c5HPgKwI-Nh)@|YZI*V=bk zQ|O%zFHuNKutB21aBGOLn*q{_9-ykHW{8a^ZJIbk4o z6SoV2*I#z$oU&MRvZ2MmYjc)^u0s^yjI-bTbFP4^$gpNCtRZ$Q6C;Osb?c(c!}!t- zorDjW1Ts*zcBLC;X-`-~uO#k!B?w&V!O?73Wh65GYcxO2cUKCO-#yJ{JeyDrIO)n5 z8SuHq!+PM3BNp%kly_kk&Q97RPz60c-r)26T?;LOq%?MKsMSl*s7V%t#G0Y6pNc-C z(K{m04Bq`ZXcf`FS~6(r;7q5kiO7mPtX{BPha@Ocbt)^E?8nR z-F`X-h=6ImC$R_BTt-2sgAKxUWHAy@b)UwA=Lu081Zj#_t6xrS&5g+UbLsnC2^};a| zD#ZM%ua#1TmHe?7v1tS=1o*@>v*wv%dLs!w#K^2VI7F&@i zYYBujgQ|on6f|Cik#j#{{_$UbADER@iK^?EfU>MLHdt*Lqa_<2cuo_-Nhp^r-M{4#=1E2}_qxurUU0y!7Ids&_NR zk38!aA*t%eu*|kqjJk~h<7|;m&Cna`V36EMfnhd30s2{cir>6(9|^`uIOZcjx46*A z+Fu_`V(bVijmy~GK|=WIQGA-EFMv`iAJY*4J*vLZ=fopp9ja3cP4iqFfOwgeUl9zVwQZ2)P5C@K77?O*u4R*IXLWq4Z4dbw?G*wiw zu|!_@q^fL1w(aEPED_3fWR@-MD6{l%`MOre5 zyQ8yr5|KJ%6x*nkfm7IvNw)v z9S!WZcW39(F=8%<3HKm_PCD#_P64t9APuG*UYCV4?BvQ~cA=^~lC1natVOTb^s;pZ zBmZaZHAT9jSmm{9PaT$H7v3HZI-$Tj_b9p3|17pkv;PB5KS!35pa za#VDeK@Ne!5tO8y_CqwO5XxBp%rk>`hedUw>3C|bq$xmH<+K4g?+0L!Mwo%20uE@5 zh3tA%+gmIlr%!mPFs|H44sDB*5rTIO?dhvD1j`0ISVNO#!^d5alfCu(vOiT3Wt02{ zjY09N6omj8xMlrIR?)Ktw}Ph>xZ6UCkplI$tSP>z9#CbO&qY znkC6j!dfvK(q3JwzRQa2O2RYrBsg830(BQFprE}%ZOOkfE>|3?sV}raP29^&d7n@LGSj?~K+~X9C>t6JWkDHG zIy8=20bh!D`n1f|zf1g8gV$}tK7#mjXMg$&xKj0pSOHRbK6)+1%4vl@sR;zYj8My^ zZol+U5oOQ`lXRY=X`dYD9J^LaL?HRnY^>;T1 z@9WJLkAMQ6<*rUa6L4ufhDO9WG%uw^CRu+ftCwuU+4D^iv4eXHZWk`mHBp8{VN1ek z#wQs)u^QHCjd=DQ@GUx!_|>8UFhc|fP#FW^Z#qFYgKF9hN|9+lj5KvwqDLSMuY>kh zwsN4dVhOkQ?2bTaACqNQDerC^P~Qs zFa)qk`Tb^Yj`A63NPG!U+Xql=m=r0P=X5Zlhoim=n|%}I?Iz;$ayb%-Bv~Yn3WEhy zEb;L0r-+#U1CN zPCYy+sg#kLL12q4JEe#^Bf2r7oOfwerXBearhY&)n4RSd^!;X=?$Ra9 z%}(~+bb^apn91I0`9UBlA?v_;NpRl@jSfv->Tq*u)lvU6W`rAf%|_LfSjsYE34*LB z>D9#?nMa8A9NQyYSW|5VqgE(^`t0)V6p&M^cz(@^*?1Y!o>^+ zhm)t8Tr%QFR~7??7wzrg{HcgsUbOv*E_|}xr^SWl;>GY4iC*GT3(fQ`LXRhhRE{8_ z&u@1J;^tmHX6JI+5_VkRxx4Q($=BSr_r5DD}j33hl&g(ejMcrUpsf8rRfh4y;8hS4s(VsB{Zf z8YA+AN_FXOwkE$5Gq;g`$qG@Y#tQ`Y`rr8~PL+DB6@tLJ5~d=BqDwBGI#pORl(bs7 zylJfV&U%h7;xG#aTe+A?9p!R}jqMtO3_f~?E@RxCE`TaJ7=4QX+m!g@a{atJ?9S~9 zS2pcB+cMd*9CfOFq4eFyp%XhDtl_8~i?EmF$J{%uzq&o!F{Y8~evHwg=z)`?fA!49vR=~mn06ouJ)bM`ClccblQ$ju7$IZJ-E6m+R2Qt`>qNcoW6 zwPgR<`gB>4;~DKe z2aemEal^Cx?X&MgS~>Sy5_j?b*#Y{zf8f-%)j&1kooRRBwD3T}z>Y#8e=8Nr*Pz=m zwi{Paj6+Aak~|>bUVqLGrIIQ?Ca5~kzQHEoRytVZKiY0^e$%QzTan*7UY6=3x zp>Mr=s*7!9TKP?p$+~fBcb|K!@ioai`Yif?#Z-9xPY?2P_F&&iI7;rC@IV)nP&#NV ztMQux)0NqiS@%Ali7=?VGu6q)k_+zJofVpn@L}8h8oP3O#LTKZsNIwT9!a!6CKFLY zbyT#XyQ?*E$>3I1{lUd1r`56H@QxTxm7UfkKE~r_-gH9Kg=a7$mT=b*dj?3&)A8w7 z+Fc-i85UM(wKuNBUq)fn-|r$vy^7P`L}>5Pu5+#KB)T~lf+&I}p=L;^i1icEmfKCR zWpTu%;|<<3I3o#MGgXlDOpa@fZeyuy>2ZVMC-1s#EKnDA_TD`rQ-{Tp8GDrWB668J z+`#a0B-L-pG>x#1pX&13JF|T4{gdS>J6v}v+&8@+GzLwB781pj9i8KIU4~d3Go%g5 zqJz->%GLJCJ=CZ1)|ipt>eiC}(k>5s!Pe)kZ*Ouj1)Ju3ODK|KFPP?Q@rF!ATvxJQJQay{DB_=z>`d*9<{(L>18wtHTvSc3PIWvp zjqXN@IdqGdHACwKie>6dLTKfizV(J%7hk-$!5KUx=$P`+1OVmX5V-tf-$K=~XlI?L z?hNNu0PcZ~>}$p?QlkSBSX~9CB2Noz-VRlLU$r{-{Y@FY-ZjhL%&evgLjCCnx49{w z3~sW{ceQ3-{yO#A?_tCXCu0Ve0v7vWpj=fsU`~&#Q>tmABX5gW0ZikEyH5~ZXUhQI z3eEz=E_&jK+||3d{pY6o3WI;5?YEw(OTE6|srL~vS`>9>mw74EEC7Aw7pWW}8++9^x zcJZvZ_}#yCb2O1^l=yls)+dYXV6_=)A8TbRd{<|0(h=%hW7FHp99KG;PgsafPK__; z2uvpw(Mk=2!(lskNgB%%Yyn=9Z;Z~ZRL{mO9NK<2`Eru??U^fl{mSCCuhuK3zYno2 zbz8k{a|Z~(BBnBIsby+9B8!=6RxNv=_V-mz`JV37YN3iK-gJuJ_MxIj?S7fk z^uZHucJj!c<~b)s9=RDE{tYtq8nsm!^AFFYe5Bf<_dqNDSKHN_-kGCO?IRoR$vv-s z@I~9vTbE$^w4V8{{V<=-)xR_PUixptMW^Z>Z92>49@~BG?|(QR`@87AZ&s|__dc9l zoD079o*Q-ue5Idr)@6|WMVHFTS9cXxHf~I21hU`KB>TCa;ZX&X@0+gi4Bcpn2?vfV z5)ON`w9wX&rBLu(YHcw=8ICuy$d~?(0@QE5Aq#rv&0yY`g)#lMRZ}5I}OoxW;I*KPO?%8*u$fBO#UHo0HSSG>^bP+Bhj_!Vm5*; zV+oj#N9#aT!5anKZbbt6f#yo@BLf02Veby(x|FsSe2*0dAXPhi&z`Ee6F@zy9%eT^C#L(>7wMaiujioP(Rv|?ncuvN(;gp@{ zjI~ZTIB@VcW7T0*Giac91pxc!S{nG|TVC@i008UzTQ28Q>{DiW-KX$n4=xHV_0Hx1 zvm#49xgFqNV5!^gfT6y2C1+OubsuwLvo53YO~T=G;{)S( zYN-~|3TxVGb#?v@GdrzpucqJp!{h$E5#$!2=qVRt-qzkbn_gh9BKKPt2huz;d_B?3 z`bwXxdkq)Y!&~JMacp#?jzz*`Dz7sMgy|*B!x#yv0`#Un!8Y+1qc?Q%Q>unO;IpJO zd${WAReZ`QBe*HGEyYp{V&N$OQ9!Q001d0p5lBO{$Dx9pfIY;BqS;~5tth1#XIio4 z^qUjFU<|!bMT+e30w3KMW7l|A6I<*_W{m_1-cR)61bl-SouQyUiWla^=EE4$!z>QN zOIu@>A(od_Wlm!301yEo84lZo|6$%B$td&(Mnt;}Gbe%eYgoP-Cp)}xljKgg+Y1|zWLR?D*7&gJiQL`QT+Qw{!|N*N|r!4e(dRud#Hdi zCd+xT6s06-$wW(=wtLis%^uCVE9>#GU=i8^4`Lim)I>6!F7J`472;VVy5bCG+xww@ zraxhu``R!HMHmb6k0UNpqM$%NuRy*Mefw;xQX^_XRg?FPeQLm@$lv!z<4B)24Kpv+f1W6Q$fyj5DE0B^G zz$b6O1t_UxP)j4LIV2}rGnci@U#32=(pmOA45Ng->6xeMA7D6xP*Y@~6MADPAqgo+ z$9{Z>5BNKZRcB&WkBn{@s+TmDEXkHK;Xo1tU}FckmfKDZ@G7*@k5U!gp%Lq6c%SK* z(&)oP7!5zBEPfw3#HP!(}G9#K#mz>-|tQ^C+h4GBWSAI4v=>TQ1v>zPLN7W_I zzkNGi_OxE+TI(G+veKl})@J+YoR}~x)N?ucdURR#2glER6<=nOEE&PPH|G2Y@FfMh z5PQ3OM2L$-O@e40)6(*_$O8Uw36QEgYop~C_Yi3nHHc31rsHjM;WT59axrISaqF~3lE{C;2c#jtD9&cjOT(PEG?x4 zMVE)~9vkWDWXz)Jd1@r=NMP&er6dh8cD<1zg{aypB8u%$%!pgIPC1&k^2W|gp~E02 zky8-g{x!_91m0k5|4pvl9UKSEU#C%0l<=J=pRLtCnUCE*N&|zEcQ@_#`Bj|lupKiB zYz!Yf-SqFm`T%Y4un4(l^ND9Y%VDWe)WpjzJeM9hc8M@S?X8GR>qfHUOvy=$$tf}4 zy7yd56_n2Un`e88#C1{157(tE{q3;cb%Zga4db0Y&r~xH4kX!v_YrLZNlZYT#HT*W zL+h-eT#B`JiqRM(tgv;=vc47>EE+!B`RQkHm-9BCkp~|6qK&$}`9m3>2h5MZ9sl*@ zLHKcC@ZOg14?$f6Wa)uGe$D_$XhGUVUSRk#0tEKMyZaeR~r~5vJTxE>A6?X$Gr`5APy**;D2UQ9CRs<1{IVUJ1 zaR074*!&_nWBc-fy7&t5Fj{9wi&Ixdi}r9?ufv_=`}?jpp)aKirryaNB+mA1$3H9Y zEi|ktfCu|bD{<}qMkF44NY0B4ev)!mlauSwE%!B{ViyQ{acPj9Zlv*~<7H&3nrE4+0Nc-F19x=P3^K?V;Q!GxVe|NOCo{7#njkaN4vj1j4;bv)wm{s zUn{p6EwklSdgFJUDowBJ#EV;_m$1j_!2p?5LA{iSzKnz;?eCH_-!~g6dz=y2Rqoyd zswAeX%2ijc?#Z851+!ZynWR?g3Kz&kNil~hbN=ZLTWCT4U=iOU+2G!a7rs6dKW)-} zUr|(hAN})>6ayE}y%5A69>$U5tuLS~p8_1lHZSQ4PIPPTl#73%picg)p8>~ZGJev zbA135>~i*rny%h>(5OYRRVP#C{R5JfEB-x}_HDBqvm`FXY9F58K0MejUXV4l{VOD2 zYvb?fZht^-(`x2r86(O`-bg{fJqj`>MK>G{xQaWGKLHmr1OcABCl#$L3v|8irZS44 zrgy0hyFN+8+~L~8z839KX7(KEiE zc~3O-6%rC@Ys9D1e@c#65X^s>OQk>j=pgY{a$1e3W$OM5lbb`iRb)Pin~Ajy0 zqc;54d+4y5!yosoGZiTMt%9h`=JeK4YD1f5Tq|7# zA%)papkIp{%2*#RRm7Y@8bv;y3RLV|Jea@I2Ikes5&X8qCT??Jd_EW8`b!HT-{aA_ ziO3F7rCLNc&b(=JtcMt9lXWr$!mGwDS$HXLz4GO9Lk-R?aWhTKG-FC3gto?n!l@?p zvY_u-C0MeY>yuI2sBH0_OPlQ*-Pa)P;U36v`{*e!x={wN7T3L&7&_1*TVy+nc%q49 zaIyJ{!s1&_i6tjq_m|xXE5d44qIAl*T1CMYIn-NXE z9qj}BRX5m^((dJfkC`NQ5aj8fHj-mEgg#46`QIb<1#7)~;J44I_N50%Av!6)kD)gD zZ;fvUoyv-u{wWBkL$uV8_lj&GvA=J$0Ckoi-lYa93SKRZ>TL>Yj|U^uvFX^~z^6s$ z^5Zen^KsAbsGnI6`{OLtCze|k#S2=o_bI|^UH6K!V2ytn!8PtdmdcJb*aOM8{kN9*diqSfZh!=`r>(f?Pq;*22;L1JF^RKyOO`#V9 z&=XinCMbtHGah2Eph!UQpO1s@o>7R*|7?cs1F;cTKmFLPNIF#rkM?$qW*Z}s_OmBK zpGjR?|LUf*j~rNjx_@o)qq&8s2}O8t{15Sm5)U9ei3E zPc^cx1aoK6okb@%vJPTiSjg8Q1ou@^B>lM_9WNA2r`7BwL7ozriie)3x_UG2okHs2 zv3E@L=Np7;TKdV2;`Xaaz?wR^)k*Lmf?@B0dNHFB#)k96EAQ$gy(z+qQbJqQg68)v z203NVJ33Uzc@|G-k09LnYW#leVZZiAa8krh)6}y@=FfAi@>%=)^?`{tx&2jzzt&$^ z7e1 zp1gRbdGlw#WL6RKoGO^k9n(u2ZlO%)@}@K6vpYu~Y+tHlMG;Neyl3gAo)@Xl`N>kr&Ksj6W1ue{z zC-T;}UjLV@2)oR$%i9Q>K5GET<>fxIc5p`U;&w7xP#jRJxBjV^3+ZqE+W~EFf%O7$ zNr%^CsRVl|-}0Rt*j)HYa9GDZAAwc=(i_G7GEaGeFNV&ai9EdHe@B-gnIGL=;`qON z=0{e9vT%Y*2Nq^4l7S#R%(ifZtRz#Iu1+>gvV?rNoe}{LchJ-V;wDKDTKM&XtgJ+$ zcvC^BYYiq%IF#0aN40`w6UQ%5X6t}gWwq)G<%VLemt^2O0a)DE_+`exeLf&8SLUf_>*uc|KQf6iD zH);y=^4(MG9JgOve@%)qvb&y$%sp4;07O-$WKt$%M22NZ2Bjd8xXA6&h^l9!3 zvh=l2d+rMFtDmL&m1lbO4?W9xmcHBG|Lw?i!|&uI`Z-j5xzhE{>0asub;AwM^U~^b z`18Cw*=TW(UT@c1C_A&j5T&mluD&eq8s_HuZ$#8F#Lu^XSd+b=Ygxd&E!ZM0*)mkS zrrVf}1@YozJ2q>JrAQi^wW|1>N=DN|pf3O%+Y-`1CR|Z1Pl1heiyu$XJ>n3GOk|^g z-FqkM(HOo)Aqfm<@`iD!K&UM|Q)I^jT{)bhn#-B+%G(a7hmFHW_A2|n{u+oafw}1` zuU^kx&le)ti^D}j9C*a-(g6c1_f&2*K`VL&(EKb@XcN3oxhTWsC%X8RXLm?h1gc}o zE@<)|hOkVdv2>~K`qiVMeEVplixE;@?K|9ZjErT0L$;8&$GS;A)rgRsxzg!c^r>E4 zlBN1RRC}mN(a|e@Bv?f!*@{6QdM;my)GgVh+VECdEzf!H5#bTh&7QY&c-f%V$G~2Y zXwdcA%N!Bz13vzgL1i7Ecx?07m*qG$tv}xK+cc|PT65M2bbMzcYFFFVDqr-hWT84C zRnHi4sRJD43)@k^1!M=Fgz#`C3h4KnJfWK|Zvg77vIlx7lBbhadcCpwDtDuwW{gqB z8SF2*XZ<2<6E#kSsRAi9hQ@=2G=Z9+3yr36G?7a6YsWRD4u4##!kqUDv&5s7 zR7&HgTcnx#6D4e&9SRg#ywht~L2`!I-+gOuIV=Q#oIEldY$(jszVBx~ z#V02`W@+bQ^y?E0lSR|J=Y}L&D_xvn!31<9m5Qk0sQ2f_{bJzMFu60@aHU22m839* z_9lz5^L|Z^CZOD!FH35RGKG#;W26!b>R( zbgWAT)B2cT5g@koDHELs-|@is*%fg`JTg)kjB!m|)r|*lmP+GFpPPpQ80$0-6n;9- z_;C9EZQQKMc#XeIq@M4;cu&MSfY9bqGn~0?KhxyOrA7>YD z)0rn;s2Vd&*9=UrgR)EdQTcu|?BK2{XSR8mQNciYMDNS;BOMY0gn&J17G`er?5Pfo z0TN->6<*(IV6!oMvvVC{DTpL3b#hLOk?MKU12E6Bq5^*0g3U!WXt|5Yug+7tcApyC zvVhdoLxXT$fuUjaQt8ugilliBP`e0_z2|Ms*R!$rkJ=Vc>@6@X`j2^ zEg{dSro~#6SGKtFR%<%E{QW%hLS1QY3Bkk?Q&$aFE!#5ati!z>f&E-!f>YA5RPC8( zEvqB4-zlaev96r2C4@uTW>(fswIeF~BmYD{+M`oLTCO1YA@EAQyInU~ANzQy`mWxfqdR!2it zWmR#}s;!oagpCcau2yd?*4Q!GkeWWvODzf4o_QhotCGB*eQRx9v=*tePKuou4O>Fe zMn&)=dJvm%fAL7D5!Op*rGH)uk=9q(*|aEU{XAivblxQ~&ZZ9tI38TC=CSgk^F`3c z=W*+|0UI=;k%(6x>uqcQARc3F%_c&$q}oPS z>I8n0t<4Szg=TYsjwSm{qK~)j7}p6g$2Rn9Y@?fVolNPhN)YJjs%;0v&Njb83$3=p z^1Cn0RAbjHX%H(rHm=P@Qt3gLff)Nca2R537aZXO?6MSjQZ&3GD%Z1epd@(#g;>bs zPNKD(H;wO3XTvQheW z;~H6}q`4Zapfm|6yTIGUBRh6-cAl}=*}DPTW7kNo)!T=?Nc!EpZPuwGq#*b=59GV- zqPFhDAKt+{i~N+}1Fd2%A#Pj-i<(^qBJLsDuD7z6V~9XZzyI!d*^m+k{lgSd$uFJt z0&)%Um>Y-(&PGmi=2!44nOl>9BfsI6Gy=L`B^B;^P&xOjxr(?ilU2iQ=_*8|F5f}G zlPZo}o^UpmaTkhT6~|+02D7AmMe>y4OumQKuji#ap9WAskFu21i=7wgKlXIEyga@X zIbO@#@H$?}Gs?i($H32EGfecVxOqLVfG?U@y@?L5!Ewa%24%Uqwp=*S>$1ij68IaW z0OFkh`$vRHB@-y=^SLBN1TJ{pv9TK&C6oPR23*Bf zDNS6R!Gi{;)cyZ^tlG0-h!AQHT2P?~diO@=kk>vgCIJ%$Y&-RB-#D5@7XL{6bc<$49omeft1NDA#T(e zb<2E?pLs>=NjKp!QQ$J&at?@h=CgG^z%q}ep=5~gF=3rqcTy%Tti7_P6J_TSpNW(< zg~Zx5C^GKfZt42coqrVcc%__ z=uKkB=mH$Un>0qx>QB?CCvVv;?Pb3W=B6uOTWG8-E0n*;7-(8$5?122JPK23{6Fin z!1oLjk4;pbi+$w`MU+zR==7%FjIIGOCU2vjYFYMk)>2QMtk_E#%{0*C<1=}KP4pG)~ykmDDjT7SD2rjwhAd6EYO=EQ`S%=Fu~%BkmkjqdWA;m>k> zMkPjNYbbR%KPJF;^x2n*j4cN)LO4st83l6b2r3r~<0bRcg{yj%MC8L|!S!Cn;+ojL zIk#q@H`9w$UcJxBmoagKQ!w+F4tMj_xp!PBSod@}dRhtVkE9$%28ZE=^D=Zc^wyJ3IqVg^>6s0U9{tT)FijxWQT(ErP8MLC8&irzorF{c>I zYX2xrl;a7-0Uq{f5_{|qSYS7H)flsDkO_DQw=ck6mxDxo8KQyeYv&Ub% zuS5RFztIa{E&vcU2EaZ?{6xZ#H%e#OVXqoGoN+r63ppKuFuZ09-s126uy- z45-q~9vwI7#ve+T8U$+iHkB!u^bRW*QGdraG2UNPB*D-~&kYw7DEWmXfM`UL7V~33 z5CTQ0n@cN@0kmnHg-`e%XPo177e0U`4duXQX{0dS4KeF6F-v5^SdR(9(G&!a- z5lfq|8$1XFGzZk^i+#zn)y&H9%a0E_Bog;xOr10|10|&xJk=2Q*~EwAR54#xQ3b@x&RR{thf*-Sc1* zlLD>(Z?Jm@lk!x$Jq13(QLy!tcp~gdlYYWpdV2v~BvYQcWA{Q|lxMRl9GFqO+>26g zFNG5v;iBM*C$EWiy(}>!7d;l%i#a^u0rz~5wX-hInVmyc2vP|BN@cDeA_I92iK^_G{C)%MuaMH`!9xZsqeh7@ zYkytt)rdw^ucuPPATBqBY$PEOu}DP*(gxxMfvJl4LlnqOZ-CV0LM{HPEByy@3xa5^ zUj}6x>UVS&h3uXNTpx<^n+&Bmh0e{gjVRw@D*?R~#RE|g5OcG+sU6MbD80+*?KwAv z^bVBGO(^FIR1ExjRI5YpLLKsvhk8Nx$KTiF%rh|EsBKC^xfcx~$6cUyp#NnTv=Y9O zixxBs28|VC(cXK|R_!G-1fxU6f`n$E2i?6FUFemp-rxJsm+yip`g3d_(9AUa4&7YI zHdU#_6c!C%eJJPkVP;m#(dA2sK+HNH;d=KG^z>eG?_=1)sGb3b?vD2`Vh+d{$DpcV zG0*pY5IX_GwceK|Rxy`7upAW$Tr4CHEO8~~`v4|D#Vn>V1x6hJldEJngx~~}^HJ?O z40IxSVGlDY-eW+USOX5&T%91{48as@`yjYj#Lx3FdhhwT_K3oWT@aEdF0s@}l3h{Y z$3Ex2ajy@~vCn5$jAIj@o1M=XGpv&p79uWaQ3oZ zcR`U1(vsMX5ZHd|kx&5*c40qi?SJcW154dzSnYqzPi58C^wSIt6i6U~!6JdtB7zDg zz=8t6fG9l=8m``+ng02G2t_+Sz+2p+pTVU}dMQ}aWN`4m!2>#A0=^Ui+U`{Gii$r7 zvfYgUp$#z*0DiN`zz+o4?iBzR*|sbWA;1bksXWa{EK%FGk;UNF5d|VZIEVvbpb;^y z?Vt>7+!<7X14L5!a053;0EzH|q{2liB3?2mX*2CJBZ9QHtAc|A;&)7tuTTyeC@xH$*=-A+9N-Q6^Pt_&9y z(a!K`=W(=cnAjKD$*njkYsc*C+GT>%p1>my{KSfhJjHA?b#dn4sQ z1D241S9>Rw%&eA7Ng~bSxh^4)yn}H2NLSkT83p`P;@THEyV4!R+PBqZtqZ1*f^;!6 z^9vF+t9N-QeY>ELTKsf5Rg_j1woPfstvf_SNRbFN$ zIOZkBT|x|XmHwCYBC9doX0|qR@M~sjAe$kdorpw*VdEk5hW+^&X(hUtFjpOpy_s^^ zAi>8QC6ULonT5K*s>AlHXDsl<7nd1_rTz&{GbVUFbFc;Ka^_?@l@w>RK{GvOY_{1| z9ajfbmVXHe{^Dlr6uUc*Z!qAG&UePgz$y>CZt&pWFLHsJ8IPEpEjI+=h9%(ZOvD!Y z(hZSYZ2I}&M_cg8H(x$1hN(z?XJXJWlW-*)gLKd(7Vgi#F+|O=Mc(#!k*0k z$SkX_HX~ymiVnG(M#5J-I+!RnK=Nt0U=KZ4by{KosrMTvA>V%Ex7#r=`lqhRo?7Z) zL*%0Sa{D1D5O7k{cy8;zVD;F{as97z_-Ci#6rE2kM8Oui2z`G|K@Lo3)x8vgY~^-g zi>%ErSYZ0Zwnr1kg0$Y2v@4;R4m4V=t#f&mVi~%PnfIU#jSkoB=gXf9#nRK=Ut^=o zD%;Ps%!?BtdeLJ$pX`e~nc1?{T%m5*+~`B65o{adyeIUEJ=Q#Ji5@4W=5-KP{Vgwvj#3FL4?qTz%=IPlcm7kjng3h^GMbFaTV^F7! z=M_K(#%$h_k+vjb&&Tvg&*e=@t7x|{X>c2au6eA)xEdt^C6&<%6vf&{@e;&~Ov{ts z5+t^ZOoo*lsxf1smtYO0-zdp@sfUWy|29kN@laNsleMx16{eS)Dt%wgy&z>S1TC;# z5`&l(I`>Oa*kw_;u=D@jXF?8z6C1`V5&L#xiQ{l=AMf-0s#Up#XVb1wu1b5^VW?R(7O85PSnE|vu!%;mRHHh& zoVGA+oT_$9oSwynoV}l%T_z=8)A9MJ~~XnRU3Y% zHG$jprq=4Lt@m0yVf|&tFq3p8e!tX`1WBuU){|bcoD-qf7MiQ{W}U^8KDQX+_w;Fe zc>J_CnU$~49To9j{F&CCb@fNz`m>(?0>IZ>|MV6k(uAs^s+=1PYO&$2e)x9ZngMbV zE$QODwb~ZPK>x9E#)>B0DUO1_FcX15O%aQ5dJ^hBOca|`KCx-xCjqL5YQrxvFTitW zH3E&AAq(_Rn`c^+nz*Pw=N2b)+I~V$DcA~fll^f;Nt-#}t3W5M8kBxLD1_M=Ti)}w zkPxxek++zsB#LI3Lvmu9qpeAiF*?T-pu5=C> z2T4)v4ltVQ)o)Y*7AccqQi+mVw!kYZ#B<8C4L6wxqU~|zYyd(6854Wm9E2kM1+Y{| z0qVSEP1+lp#OX#l@nqIUGM)i$lR1E*CEaqrHtA?FQ6yO`(bA!n$D3fUfyqO>_qsyh(}XR|x3Q?3BfOkslSM%`3j|UGXW4IH zyP%PYwk7xD^@?(2Cf8UO*%<@6MNIZX?Lu6kmgHU4mKtR6CPA1owEczSwu<(9yB&W} zQ|JUd$aU#xzy~ez{6E3UPmXdpp(EE3xywo2k7~kaKk;*uasmk;(oqBxM8q&5Kaj3S zBqyh&{O^ND98V$%B$2ud$?>6N2gy~>(WGjvl0qiwWI35!r;t{EEf(5=9&U2xQ_)kYoq7PD@ld11H*@xjLwKPJZN; z6_A6=>kehkxHW~{ogFk947t$-G=v5RW+aDcqRoZ0uA6G^|JvI|=QH3Ux?M^a9W|@4uf9DlXLRJcib47^T(q`#FtX9f1N8KU?ZvtF z>za|z6y4>Q38U^!^7Mb_uF2)Jp17L%6t3%X*>lr1tlV4{{e$HUFK%S+yES*lWeZ~+ zO27IhWD+_AXz_wauG^~Qos^R6*`Hr-;gA{FshfGPy7BCc$+loL7l+8x=KY3so-~H5 zY+l0MbwdTtCeC;Wc?fXKwID#r{wRpV7LYAHK!Q5xX)jT3QhQ}_hD=6-~Clzs} zHGpovs{Z~2(|&z31B=K5&C<(%NV{OzZ){eZg{*+fps3G_6PoO|9E#Tg9N5Tyd$WPa ze#bf5H?je~)N8YDk+S*X;tFL4LdyWz-Rw6z+1nf-mj1W%C7}SwM@XO*2H77grIb<% zcl!${sQ>jnRBSvAkk)RF&?<5Emr%oS`zr`wsQtC4p%ED5ceB`>V6=ZwO5tT0CeLKp zYss>IRuK6uehh8?<0-2Lig7Cf(=YpEnlcv1Dw0t|;nTPYL0*1zdg1dr zHV7&xrJN{LZh0$EYoShtwnDY|qVeK}Z8le35cbgfUlHxxi+AoHC)|;e<^Xctnexu<+;!M4yeG=`oO)L!Qj>UJ@Y(DfUT%I^=Slij0!*C5`;D1#HCq68h1n?{)Ofb&m zB!riks<^enNyKz+sm$bfW)+W_XIUwwRK3D_{b3}p4!*!4ak9@Z)b`)Wj=?{joYJnzMK{1Wn#?HK*6zJy1xw?Uj#P|9 z*$E0l3JU|n2Y^`-cguo8eEeGMeWLZP8Q z4GF}p8`40uX}MKhF9ku)R%Yd9Zw1kdP>PMiq=e2j$uOal0kTzjd`@gmI0Gy&Jc;V=%leyaBZ()vYuHw~hqT$#1% zAd1>33Sn8UMpRe`Meu+aC9K426t!?H2m|~*fTF6pwFHEh%|WXY5%&U;a!)tUF4O5S zW+Tu*g-=09fF-SVNTZQP%f=qN;*850xJ!sc1~3rC8hkI1cIUGF(h>+5QMx2Z&uL+= z)aFvm5Q+*BvS`pmhb{&r2%rYOVgU~}0x{utO1G>eVTcGW92lz#qXbIz&9UhX85q*#wX7Ny+9|%Q_>1S+su*?$}<{VjI z3B<}W5Ys;qHgUr+I+_T=EgTy$vZkB_Vr4@$?$}9`Ee_d*@l15`D;(`!>PvrtahOul zttWLxcPg{H{j7FJ2w7y17MZn~Y%)nFpFDEOp+F9-Hy0_`KncY;(OM{L;Zn0%u56Xt z0Hxf^G``L+a%X!8A%vo*F3YJg$(TLkZ(gT}=CRh+da=B#E&Iyfyipob+&GGPRz5a4 zUKMkhN~`bEHA{;bhwV7#)!I_7M0d!qvS_|GMLAt7p4JSA?qx9FQ9j2N!01X=V4j+G zM*u@9f&S3sbkQ9X?PGRQA>)c>gihMByu}zjVJi%+6c*`apaSG9rWj^CGY8w(xVv7m zLXXFA^>{ZZ^rf->R$;b05f#GU)xf%aW0^^&+nMOyCrZq(LslijT|B7AH$|{BdzHv` z{NHY8UQaj$&zhccq-R}CKW0+TIb@xMYCV*z8ZX&rn>}_oV1_}9==$`sp%<^_R-CUJ z3PY#VC2QiDDDj_9%Hz4@5S-KxY?LV>AH-eDJ0t`zKR|IH{nMA9pK}5B`w2+XI!p8W z84!u=_e(o&sfc_7oZGLN25J^rD$fMlOzf0&+xcL6oxs|M;AE=V)@(kAM>-p2+k-%> z!`qTs5NCU`-Px|_G{_CU(xW+|nC(*_p+HQ%lb@=*RS0@z?$owQ*`hpmsaR!s9;}LX zHFJIwwbF{rzLIaNo6mUjJEQRq6${|9W9GBW*=m_f;FXtl*2O_H@^b);wQD!?5Xtf# z%RTO6!h@xcXq6)Zy_cB zUO*&Kj!gIScseD_dl;ps1HuGTcnNlh!r;=)_+DU7` zxN2;sVG62vdROWTnwA`<;-rmrYU_s*4w}yMaAw9j0dKohv$t!Yr&L$p?5oq+RVA!@ z@49WMl`vWDz|%oJT`~VfJ9gtyUe=+}#|X`|W%uj-HAQQEUbpF_yDM1Flt7rJkrhVh zr;lEK?!xWJFaz1c@LKDk(kF)UXXj?`T7SLZII4Z6vp#%%WIca@XN`sIV?O(7VKxVt zY^SVstTMxT4l;GB8_BRaWqxJ9knKfVWxG1g9`1(uAd>Q3Bri<7Qh9{W zbR1Bgehg*#dd6m1k}JU`C$LU$>tDsF4N^(Yg5MgCx&3|L%Di?5@3u-f=cF-xv$%Rx znk_tmKp`nH-BtpUDuGln+b4+>WVB5^x%!$rn?+zuz9Es{+B)${pEJ@*O}jx^zjdPJ z+dEJ6AjDJDZz3_`mx;mB4DDB6cK3MB%E3D2yRw^goM-SEx=Sy4valK(vU6oLr!Q!X43I1L5%SLKdmgFI#)AKz3Wg*{H8!bG+dGH1cTYpL zY6r|%jWt_up7#b@vZlS^gX9W+9RAoMtVPb>FFw~Q8xcLZ1>}MzZ!=ojFBrj+wxnq; zYPM_gGBiJya=X0#J+XQuKeQOpc_(swXseU1I_ezru>HXH?=MYHR0B z{PmLnyM8)8$En|KC>wLXjSnPmx22`{fifFx^HR0%vVEiMu(eSvx4WOB?d%6_*k|l} z9%$)g+Z?diLg7abIpo``-4`L)94TaLi5z zUSICV4%)xmmWU*WA13dgzZ|Ipm%S4|fA!Fbcvuw9hjR}bd)skmUHCRwuiGCts!OiA z;<{^YOnv*yc-Yz3%l$WC5Gw!NIQ}>j-L;9iJerSf=j%EjUAh~(w#@Al5z)l#(@|?4 z9EJaCr$^eV_5VX=|DV1bNrqnYh&u-l=K&RFynyrLqXnrgZ;;2FM@YrB65=2*x+}_1 zO+bllO|bY94zi^mTe$I;V1z4@do%JABm%nfdpBcR?7}hFcYP# z)ykZcpJcs@MJNj4kAVUCqX-m5bw(KVLfbhnjiRMw#337s7dC*|8A#rR4zi=h$bnqQ zY1tX6Q373o+i(l_aaYnK2~9}2?1nSdFp)(NJKP0v6|-1|DQN>VkD(argB;``*ZIY8 zl%Hv%1rbFlROp&?FYM%R4vsE~k*JMtgciD@9Xg;rI%Y2Df^KQFJ0v_%EO{B#gp+Bu zH$qPV&yGP4^tJ8TvY?}%$Fy!g!|9+`=9=1IK$x=+!YlO+Y%C%sm?}5y$D=wLtYTHC zp7MKQ4AljLFwpO)H!kjQH;MApD+I+jG|B>_`w1e9!Dx&<+zTH}q^QV_Ro@7R?8MH< zfu@)p1QE|XY!qgg9yz(|$v=?`gIIkjx>%PL&=U}ju*i!RA|~>oIfzCSVnsRxRiJ2FnAPd$njI18K%#ksT}lPV79F#;k;Sp+UdOO#?5WWmH0jC$9wQXAn$ zvT z`cz{odNMY7`)~;b&Ue>uW;>k{w zXVjae0BWKa5-HQUQ4kc=-1;Vvh-0|lUI8j<=@it;Do*BABnqoF(b$ixKhaa$h^SpO zQ`-|Gb+A8zx?bU#1mOrvU721S<=aOtr+`R^9&Ueuk5 zz<_Gin1-r}Qe_KA8pb3@!o;uFQWRd*Zb7q;(+V{|2!jYqHKma*9Szi;Bt;S=t7bHc zk|qUzK9AxF$*aXw_{-;{MylF} zRMdL28{306CPd$;tr$DHVdv8uztxW1;F4|#wy8b8KV}E#X9wY>0oW>;_0GgXQ%Fx8 zjTwC(j|UUZq_1{dug2B(2p2pFM`6)V;YJ2zL|QuJX8d`pJ6Mb~G(Y2}I?a*zg=wPr z9r^%%S0~;TruL+rfS;h3f9Eqz8dLUz((^uEl8ug126m})gy~r9ami^VIjaj-tt{*# z2eMa}nkI;1nvPoXBoBM4M4^HLP_M2OiU#s^G!UqKG=n(i#UW^S2h*p44H76I3zAd_ zx&mMbzJLWT_~7MLg|}xUrUnQb3Y}pi&7Naa-#P|5Q-E|RX1#x?Qknq;_75f>PyBL2II+h8{W?VSuTGsRCv}HLP_{^I@03Y3^y(cw}mQA<()HG%){&YVW26 z4VCTy4j=qxIC0tg94g&Yap*oXg0_6n-;`i@x}7)eqFU%4yN~(lcrBgQ+POPWcn&|)GN9wR*Ht*TwjeINl1EKD|#C`K?4JHOkzsX-_{6w ztsT?S?~JuKE9t|rj`6X9Q4zu7*c6deFg`l4T&UHUt?`&UANAkRD$I+Bwcudl*4N{Z zmbfo0U>Qqd?_v!sF|~0ThY?If zM4623!c+1J9(tN}e(NVpkP0ew$}8y4SxgPp#$7K&NbW zNsq8ukFq7Nv!hCLgq_l|+d}!O$M{BynsIhpF+4|-#o%cUu-{G*ALgy7^(((~m_xeB z&m7e){^TU7q?v(0>OZR6{G&T}34aFn?xJ{$)4D61B188*6Fv|TGD*;19Lbj)7tWK* zCH>7&^7IcUDbT-X7x~WzikF`c;s@zJJd5DFtRHmw^i0aTy$$>-zJTv%jGVB|UEyES z{fK{hJN0CMyzghA%B6(ZGg8@75|>m`^OpI(tfqctnM_`W?EGi`Yd{krOp$$$1e#JT z^x|`Sn0X~ryb>^He-h9SlWIE%zc#IihY0McdhRDO1z@@=3ubxE@-XoPG z%o(F}fJt1QN{ABd(kWpssDvw_(!?`8hNUpIuoC4m?&H*_zdhamZSw;^Cw_5n)~~E$Evs3_XY|v@0Gn7~4#m`y zLK?|5P(~vqJjf?bw7yfCF5nxDJ*xp|+7h?>>le=;sO_K)H?w3PqV6?z_#nP1iV%Q? z#U}LxU42LXGG1jjb6t6JP2+6LMEqcM>Q;vRy}*NbK{LLk$p1IWI_`zqAAR*@IjN+C z_L{Cyi5))Z#V>PhuxdavKn%I?5(p_L z55A~L`u}?Pqy+Z4-&-riy3g`WF6cz=XkxfxTj34YURE|G3Z3m3B=#hiN~^Uyt-JNL za_f&Z830j0uD@G{vkkPup~!7_Ypu^S0LlIeZM_lGTY@TG8G;KbgiuNrL@7#Drm~eK znAqZrD~@;)N+6kXB#~5#!s`iyTwu)#thdzCDo-Zm$|#EhjGG(aMktwR#)z_dyUP`T zaFu}&mNn9TEViHWe;RaBbT~FTqc^%^AVM(|2{t)lNpw`qm>+Y79qF7{kA5~gamfNP z|G9N>Y8KG>**GN6;!%?0Dvkj1%BAt`gq(s)NE|Q`8i_dq+Q=kn+Bj;>Xk1@?D;r}y?)4>Wa}~IJI^Og zzU<&PUfopKxbWT8`&agWxqJ6-z1;TOm99ijvbKvk#NBGYvb)DkY;_%;S$VPyS2 z&zNgFoaTDMNQ9a6*{e0o;mJ5X9_>tM27~nc8^oeee@NvLA4ZJYT0nFsRvxBcZ4#({ z7_WjpoWwFcXrUvuhJj*8S*uez&D|t8=nN51Yot1vBQKUlnX+jn(>It0?uM6Z5pEG) z&$}F=xQ|$9U$l-=sx>iIWt5oj%|IT)YcU$y4UZsejlf~!6pda;?TG^`{6M`Rzg??S z=bFV*V~~QbqXle1IS$`W7=<;;3Pi|VaE1}#8@1iFD&f*|)iJpeyerVn3Y%%`WDRo= zfPTbO?0Slio=NDW(zS@xW|??pc8cx^8)I6r3*DOHxnkgTrK%RJBhjnyvy^Be58N5g z?$zzo&+ecFQWs+~unoLS1mFPG>XfnX$egNXZx0z~=CYE5b_ z##pDpHxb7JC-969x^mI>^2G}(r{ki8T+umvfbmaf$d{am4#Iwy2y`0sm8CS8wWHl5s9F#?q@I2^J9r0QKi96vjLl6h zZK3F@G=oYz!O-DbXn+oIkB_ReLDdZ<(PYZHV=kohoUYp_We+xk&e*Nc^&rmf$*>CF zKTfNp^Dq+lMtZ$nX-C1A$r9j?@_ppp*R8@-vv`eaZp8QJNE>-o_E>Fg6M4xJ^0z|v z?C9H_pi<}E6~AV^8gmoGDu2(vfet@@*;i%N_sxrBY#cmjOW97D)IJ*j<|H?yf;S^H;bN30M?#Qvv04 z*<3{>*1Mk)DthD8d%@6uj1e`sAXDSdvdbFVelWZwk6KZcwxTUS?Z|EFj z3h!~Lom5@CSXa!ZTm1z=>3?4lMSru(Ux; zc+^B!-s}#Gm@1;?AP;-OB*w|8B`Uuic{VYQZlc0pO7S+HfAo zBgHO+-5h8ilBorwl~2LJ@fCSV8&h2mRNe)Ci6e{78*w$6pwUT z&AFWU!6f~)X@-rOFuh~!X8J!YLb{}8S+1fVjx&X*1zk+B#!hrjz(DrG940$k5z*%l zE+}&08J8JwB{IAYIqR7l<|77ZqspXzbiva_=nw3?Kl1&9eNf<2Y*0_=Atx)V9$34PSC2qJAGlE zK2XTT3pUi9xB#X+#XQrV-f%ne$59P0 zZsTAE_9<&@^gk4bOtk1cv;b;Vm3tFP;gLa~*Eo$7^-U1sUxoPy3ziIc00jU{ znZ2VJB%?`R9{VEdw%A@PNXqe@Vm++bFzNByk{~KSs`Jh_L2M-Mn6MO}C z!KF#yLRW~F#JTA+r1*a^m3-O^>n#>RyM5bP;t$qC%XFA9KmrvT`V$faMlYx9#bWlL?4ifmJ6I0Eo?HV6!mqsWhOVy;f;DgcU|0uNDAI0r_wL;xxEJ z&XUMyT4toMTN3xc?2mxZ<_-^%t=~uTXuqxR-kB3iZ= zD+>*1l&A(n3)Xuc!OR+NbmaPDb^m37Y1FvlK6%uwY`U^HSYHpbrvfDP!g`dv3U7RY zpu)yz{@!-F_G**$SM*jwQ>DmS)VJkp6_6$^*0OFV$N-a^IN?0bZ#wfr`<^Y@RhmWJ04Aal1=gQ;IkgQJ-T`B;RAeQtsOtd&vZu_ALr$1 zPz^>KZDfoVnR&)Vu?aFmMHmxq2QM3JO07NKwzXI|lo3)e`n{>C$&P`t?w(0T#Ug(G z)5-)(pHhm=AE22utFW&sl?zv8RFJ@ak%bp)O!@J=^j?cnnwmN8U!A5p=8Q=c&Rh4U zgiawkVE_s+bgYEbi8$vnVhl>Dkg{!{h3bKR@l#V=wOi5-_NpW#75`PYrNT>q6VX-1 z=0$2bh?|>Mr^(2w+Z`KSfxafUxT3ckxAC6?k6_Sp_`2a9l~NmF3RO4n#L~Ml z#STR)3FCx|Mw7AN=^-6f_XEKB7{k{G$kBqqPn>XM{e=O^Yxj*-^k$JtndY>~KJ}6b z+0ArV(x9jn4M-@PmZ*7{a$DqafQ(NE9Th@LH{c@4OimtxJlWgUMOs1NYl1)&y2(ct z0a(W+WfA5(EyT6vt^2k zeVfXHC@*1`5L_Z}5hXwhBFcu{*1ss95HA;|dFrhI%CPIVUD4d+KN<_REXf2R6jacI z4q`;0cGT)Cc8LppUz$F>*6CsI-sTMojPxHox0rWfo{`s@p)2S=tf{o$n$}^QWLF%^ zTu56tbz_~28q~XJzZBj*R++7z9)GKe_qA@?SWEW2O$LtLj6QC-`OXgY_owrZfJ@qT&jO2-)YJP$JuR{V@MFPgunh=7Er}lpCx@MvAt%hoVPJ9j$b_5j=u&rS zpAgdBlSigVg#?|UUgxg$<+4cIj)?G;M7gm$?3sooke230 zIyBoVW+$iP>r<5uLlL0o!?aAU70c6Bm9aS4!X5g+0FYO$H=c3bvnoL^c!7AEOIPsw zw}G1oq*t}8yX>%83I>Q(WE6z zO4rQ1-6p+!0OV0=jt<@~P7JK%$pZ&bvyb4%G41FJcN2^#TUQM~JR^g%U|360ciFa8 zN%ZmCxRbCDX>x;@^Yvw%r^)*gq&~8@-^Fm|Xc@mQsveWth1= zNaNEF9^Jq2E%(@Lq} zX1;uL!9D8@9rAy{LM~Np8x;IVi?LcW^-yn77k-r4Dk`}O)GdQ)@C|Hpp>8liVWxFn zH8X@3$awe-AUsG^P=Z3*$m?M3+`FL*gUGC?(pekqbzc4K8-uL`g<8|r=kzIS6GKfq zR0KpG#4M@0qRJ^$l|fzbD^919;#s5QCp1r0VN zSk0Id=&}WkEZ*Y< zqQDb`RqU0fi26&^WDnK6h|AmQHxV0E)ST<8nkY%-d`u$;2ZqM*0D;E!5`o^>_FVKy z{Fk0T`b9wmve(Oes{3wxOOfcjI3|AEs>}T%l$YA_eI6MQx96^2gvtltIC?)Ef>+V_ zp~g{5G+=ycky?(9o`2&X)Epidk(P-W^^*9~&rTFkkiHF0;OSjNuTGZ!=+j6lqn5>$~q)SG8}i%K2yJMmezeRS+i7 zYL~ev7}I7lF9<$25tf_PuR?)+i8cOqBQtwPx^nE*Mb8$BXxx0+;VB=H2m9xKp-VAA z+7qmv&%>_vH#_J4l?Xe&D0zJxIcmk{`RST+YWw4>DzkigN0M5(klE-%u=Be^K153G z+J$9kh?kI1r1&iE&6M^eYPk(V5du^Nr&;#?m@vm*IG}J`p;u+-6W-?j8 z;!$jd8>K_hhnA z)Ara)6#et`}13kYC9- zWxv6(3}sS5ofc^n6}gcrq*G3zR=HliR%EAr`9hBa(R;w}p;H{O_>tk%06@GEy^;Cmx?)!W*Ah=8fS z01}f-%`4sdc%cb1s;5dOiJRveCpjyxyNsvI-D<~C9fdvQ@#&Ktg|UhMijU!-g+C&O zufxz&&@hBiFyHuo4F;yb^4bL>0CASS*7DcO?a1w5JI+OjA|fofo8jwt_9Io?*;4ky zsMiL;cZY1vC-c1CbBAvkwbAKpeQKu>6Vquu6)1t7n!dXX_11K1B@*^`g`K+xZjCQE zLC2;O!+OzN*er&%k!#^Dl1$XeW~tj`$}MGCwhK4zh;)s!PJW%IB83cJMDr z{x0zsho-wM)l|Ro$Zz0=g;i-9KO`x-DJIgF--D(piA!2CayEZHdzjC;caN=Gxg*Vc3)u^5|c7?8@huBrJswP}n9c%Kp?GxR)zkU))QW+Ar|AX_Ck~L*( zcVF7eKBkI0ozWRvWU30wIVvH~ObvM}xAV5z${&ecEgOEL(HI(U6Wg|3JI=0WH?-T> z9qhJtce@v8NE7S@yK2|%o;|sz_xt7nIshtj0ciCzmjXuvVx=ppfsU#P`m44y+PsQ_ zs~=r+6I0g!Hq=_Z)W_;=87-jtEsUiJB2mP3I+wnROVuh*UDBGiwWqhM?*qp{n29HG zp0A3d5QhjMoME!5R{FEmv&0jWLkN@+u*G?bc*Y1aNWhap#I>cS!3QV5f7QX`+S3V@ zOa}}agJC3?jNySIi~gn)LFZ(a>yH%hG?sq6eE*;;@Joq{^2O2J2!6$6!WlbZomI!7wiOvxCKaKK7(Nlc4z z(=$PIKG0J&{tLv(&0wJbsWS(06N~b!|B`zX9bZEN=Pi&Xt^94oBwKQVge+P3`-nxq zcgMmXA2R?J61Vn=6$of^|vDE5yG`)MxQkbDK1Hf_L3JR zblRsQ%Zx!#0P*B`UEHg^)zm~`7>Z{_t?8KW0eZwj4SwetG~ zm&x*nWVtV=^H&cOIv>_C>`Qh(B8MMKtLt;ip!0)PuC2^D@m!?rU(X=l*u;hUx;)_G z`Jt!*dHDMIAns<){qqq>*uO40A1fQG`2bkkuqv||RX+0hNvjqCjvZ!O;V=|$0iUMr zk%RE1`3LiapaL*dMXGt)?XiPFBS6dkyH=Q;^Yi>g*Um5C(a8DT-5_KS%s(=70P{)) zqmguxb&HV{QLAE5;1)pW!59JVO@|u@`UrvQFa*6iQwK579g^zN@Jb;hSxE@!qiBvx z>EF{qim8f-p?w0C>k>c!xbQ*iBWsyRVFXA|=A=p{b1IGq<}_@1gu#wJL*$^IF{RJJ z8A&949%cl&xS0g-&Bat=fxa}3-M%75-zAT}WtLz<@Twp(*P98*uaW{pHX*TzLTctP zPC45N>l;FE$(2uFZUa>YxmP;#Bou{J_ia^Ou~O)ETy^%C~b~;rigi7-}<7LknSs>td@DHxPEC~Hx|z}Z;_>m z7fq>DTC`9?Dmyxw53xlbN=cCU*jq>`^gBo>nNL6&LBjMqOG&u3T1N9_%`#tQa=)*F z`Bq;zd?3=4J{bju%ug<-fDnnQFz9b%2sM8~jE4DJ)d;p23rIA=%)hFUSN{%5GXJY5 z$v_N81Y@!VB8Az7{}CHex>cGXi6I5Su7tQ6yImbeg~kvPxQL!4>aH@xLQ;S@uA~|4 zeRHCudYPof8WocC0L}`^0duCjfK>KxHf}Oh$;TYve?b*%y~IEJHn(7%LL-WSt2vVO5d%V&;^AHyyEhtn5 zLM;tQpV}iqYod`_HHQXmk`JRXvn`|vNmC0|i}qaFKPctU(f2ZWj%|^vKcowB1zA`L zGc~(&Wps-`q&tfet2T)qq^C2u+V$e#7(^VQBm8FQCoM3*e+nK+NP>ZcLCh#5+SxF6 z?0mvph9r()+_8oqBVLCXMu`|b0+Ca|6JsOYh%8Yx#$M2UQC+*!G2u+6O(|l5$%(eW zq{vEJOunGtjVZ+lNjAfj>NfR)8l!BaBMJhiN&9ldV`^@cF&%r%7~2mq3x|RvG0Uc0 z19L8>O|712bR=dPFyAr68VgRdE{v8}R||6zuv9Q_6eV&6xDKlZ3tRi6gOzKn|Lkgt zE>@kut{!95B}C6s+$d-O#a%iD4T`aJ=B8LznJ8`=)okG?O$-~wm{Dw^Sl9+3+?rc) z8|jKoYn)B7sZNNkQ-LduQJxSx6(0JZV14ZClMe1Ev4!HOa$B4dl<<&Uj!;9yUBF#+ zU^}dGF4nNM$BVgpi?Ts_VBhZk*q@_FdI=sHMM|f``fN`qj#VjR=_9?nlp)f`%G+d| zK>Bk}@kGcH$UV&p$DHQ)z+Z|PjBn65{5&T)o2i^hJU@6-svIZ-UAP4$@CeG#TBXO} zMV%$wNeb`Cn(j&~pT}a=$!ujIuP?eft0ymagiNlfJ6lPvS5#+^RCvQ0Zsu*m0?%{+ zHM7FpTJFk_jHMH|1sR&`k_GELc&;h9w>$#R2QT2bZzj)60ih`*%!8_>uZCP)v%41I zu`N9!ezD{wo{hsIOY;_GxdI=(s=>uYMl~)KukyNbnd{Ap^fo`%T^0zxI(+6SH6+bCi6(r}=~o-@4&M%8Z2kh{h&T!F zRHf7R21e<&4)K1~N(a-BiNpOArjry2$)=LxPY#2>CrM`!10Pc1KdLBZAcuK-(#c1< znA%h6+%F6$QW;~HZi5{GXy&B=*D+H8N%S%HR%yrZ)~LM>z^r1&0XW-96K?~EQU;(6 z0S0fAfCX@Dr^L&esi*c90Z$+p(w!y{X=eZmV0M`8?QS-msX*(_#WRjK14FX&4jbPx zyP!D-y8zCrmv%9tVOIe@E{73fQwX~olQ$rm`7*>YP2c50(%$K5B9ZOgq9%EJ1eqzj zm8u$p9B-dPq4o}-Ec4nL?Jl4)!sY9{0)m#HW6&F8Q4-nAfhn0Ogk{VL-91=K8+(UD zHo?JmU$TclnJ!CXm*6Uknj70=dP{6qc>5S=5CW1t<=yuJt-iRv5gMVbRc=Pa@`^oA za2l6~sOd$8V|z)2P;PBcN*vh9CREm!&vIcG)2kdu-LC>trnf~-C(r+MJA1)_pP)=xbN4q3WFKxzG{SSog)bRMxd;y*cgEOlL>kw^G!- zb1Ualbo33q{bgg1OkI5wdT3be4C#J27tu&jnp3u8XabsQ8J~Aw6StVPI|0o-3e&gZ z+W55tElj~O=S!d!_M6cf?F&L1nEt@@7pDJcey3y_svL`Ngy zBBQ%V=Izf}Pf~VE@Bn)9vWE*qMOK_x_XuwY%mhCA-oodY=&SD=FGoMSe?^ff{hd!y zu!%A-iRU!^VC*AuQNw^TL}2h!BS?r0Zho8)Euk=UqIx;p{pd!BFuZHrs6;;|xp6@o zFuebnB4$PuKoBV-iZhyoLBxPq5XXt|}olrrf<$CYV&*^uY z3LznzGE3yWo5?O4q6mgWj+!W$o1!7eNAEkoZC36Oh4k(iMMC?Y;(Dl}grpyG(WBgD zwaG zOF1r!I@FjUF19v1wRJPDE^?JMP@bg8u&prS9fEdZ2WKbIWv0~FUCzYU`#}7y(Zrog zpf(D3A;I2=yF_`VB_Zq?tH|EyXr#L)bxfipF(z>d!{C^}_*w-uNfq#h zeHo7>+6OXXJaHO$n#xnOl$6hm=iq|1|0QZFYG3_JyVR-XI(Vr=*48OsGiumX!E_Vw zy1Y$f%XfKeW>c@?X4=^S;`5}~_4K69asQh*ee9!~u8GP(J>V!{CQzkz_ zUGO<;5?^%imGP#?cLf+W-~7rCv20m4-TV73I-K}WuqS?IlKwX8)i_mrm5a)0K)vz1 zU!}gGlhn_c`r}XQBn@yX6v(fldZ$yN!HK`*X-r`$dV1m$rLG;za~zfM5BmC zGnxRl;sXX-OW20fWY837hD0-&X2G@++v%X$2JBe=S9%i?L)FTEJ9S+Th2D^zi5A#L z;dMhlon5m{7G><_u+WW%LK;A5*borRuURy2_4kZP=y+MuSPB@ z)lnUiDl}#wFrGhaWp&D`Ov9|?oK8SfsZ?5;%4q#pG?P~$ovE};vuc_%gDyl?rg@!o zQQ-77J3^~gS#;g3)h)K79HdQCwU;r1tCkH(#?*u8%DD_xcOFLQkX)9Pm*|lxLMNc7OlLYOgbI_+T?&@a{9Q#0(2IUM=o)KeZ6OCT-4Pcs9a=J& zB;)B*WzE5Z)DKIaE4#kLSyUb_BCi+grshIR^wrqEDp&IcYoM#s_bHTX*nS+mq}0mv zgJPn}d(+QaiyUGK*ODQ@mh=l;TcU5$rAq%YZ?@%J$E6ZN{U2~$byL@~T3f^7Eg692 za0lr6ipGZ<%J76M3v;*;49vq4Zmd=lEi9~4F6X9t+{|z!xjAzSgblByARO}Cq{|`* zIg$*+$4^=`g_AyRC7kQrwkxSM7cP<-iEu}FXoaWSs0(vU;}yBBP%E$2P=?>EorHhr zg*o7Ms8LjV9)T=^4G|%(0a{!#xFbT{!QohURCnf1)mP-sGP?W1U1a8SnA}y-eyu`8 zz+^1KId>Pegc?(TA>zS3nCZL$?x`gX)^jhBJZ&TQF2RtoDjAo=V8{e=ABt4QeTn-q zkp?YEF>92Y_;BNbFBzkznW7?39H$XR^}=E7#?&OyLRwln zwuXgyay(|kOidw~t~V^r#d;P|FV<*kQ*Y6RtPz%|q-ADgn_7aW4OC%`Dm=X}qPlvdmfM+G3k7Q`QZrfG#BfKQ+RQ~=w8$~EIP_%h;Qp3` zfmXIUJ3_dlq55b`!!y}LVpCywqj7wOm%(O^ftQE5;Z|FyPq!j08#Eh!1H>dM!msWn=*(Ts*WZQG;KI`}@BXeSvV*ijf^MV96*h8+}mV>sFE9i0_%8g9}H z|2Ff45mn!0Bc`*hPNh9~TX;rlL_kN|DSA03cV%&DxKOTj;e z1^)&bJ1jwnolCGILPB#JX;L?3S8^<>vmIgxg=9C(MJQn> zvB z5Ci&azGp^}!ORfB1XDMBaeN@AiZ6~Y9M|zNezsvE=;-g9M}Ol2N{7ZLIp40v06%Sm zsZqa=3H{U0jeIz2N*hW8{maxgZkkz8Y7>5Gl(bLd&$*;R8l}8Xz7$e~cnUw@1N?;S zA@$X!GQS83BFdqkh2xMAnPTPyAS&p8$RnyK1%HLq;a5AG<41t#Pl!k+keR^{qwW!k z@HkLbL=t}67Eq!+OD6!FlL_F9-kMkNJBz{ZGeDq6`7BdpCi8>0By;#<6QYXfqyM%U zWvY<`T5`yU^-oLi8e#(YQIw~IKDBZ8DB5k{5^aUBfIngmr&<+KLjUerBN!?K0xH3= z2}G<&$bgyF#HmbBD!2ulhz-h-%@#v!Rxj8A4)C~|(uK3P^B^_qM<5tOl0BSbeGF+h zXgM0_W0c24_~Jn_CN#{<71G2|ncGk0mJoKJWdI>K4`CprBUHCB-UMg~#2t7<+0{!7 zQ+F`T2UmIagbP|0z8J;{r>M*tWqonBq}uS8tWezpq|g0I$Z%xU@Ko%GwklHYTEKxA5AvDYZBH)6qh&2)d z*&w0ap<6BKA*Tp%5wJJKAL!?Y>JcC()kJoEorMV)A>k3C<*z~>W85L0#@7I5hv05J|t5)FxS1^}^`L-(>ogYEhV%aKxBG-pG?c zQOHvu&o~ssi{(%-;BxeW;0nN%C`kbdMUoE^d5FBq6+vDT^a#CS_%a?xi(^DNEn5Vx z0u(J0UmfY<3S4tDNIepWcQ&EUQHQ|w(+f%Ouc5e#aSX{&$EXVR?CtLzRYlb|eSicP zPv4pm`@J^dE~99qaw1Ydz7kb6M(C+4yGEdAm?$;q`3!_R+D0Z?)sGs>8)G#!9yn@Z zIPe^BH=j5O>AYGKPP#gDgfut>GElufr~M=1d&dm^1+saKCP?cj8F}cESThCPV@E9< z$`X(YwU3rT0wlve8D)%eWDB@5jsCGY-PA#b0I@I(|P$i`5LA~B7YTQ@z0j-4@ zM`Oo1()%3kJ$2AjMOIv{65fE(LVTXrb_eRIrt#>Xt-6 zQ#$HnbbtX<>ZPleoCCE+=3xt{&4W9Hr;|w{$z%}HZ^9{X-DK5(%1Jdq-Ph4FxvOHd zSp6j2Hni~6C}YZj(8plSKsJ@3l5(bto-Fpv_Sm$XQj8f1UV-aqMreo0sB(ar$Ql4< zyGnUM&56i7%(l{urxwtm2;S_#(l}uAj#w019@<7?NzEPR;DnYaV4YB_1XlxDE8)14 zTW$kdGmnMs=Y9s_@aLSY`|1*8+v$MpuxLSfc@B|1c*7Fek1qqXHJ*sb!C*5Fn1{qQ zhjxftL%0*6eXnMiK!>gdI8ZlfEd&HU68|03$(=l1RP=+cN*F_OwhFLJBxwGd3svpSffOqAe$l6{GQAM zc%`QxK$jAP(1{0;`m-g_8QZC~*XY zFbDYql*Z6pr`=}2KR8+v7PGrOmrz#iJLzRJF261o@!6Ev5H}-Jn!nDhkvm)xz`ecN#!pXR7XJ6!GvWtfrF$%L295?cJ(< zgeuFK1`^?`g%wsp2drvQJ$?dmg@2Bk^ij~xg0^DEcp#}Ty2)qhZgDmN^axY2wP+i5 zyiFB7Spr=u^pa@BPShIPnLe*gquSC|GJn%2Q-%KaVF(P6R}5q)TUF_z219g(p}lH= zVY(Hohx93f67D7>TE$3X>@>06YQ@eGSFUcO2Yo2tg8gCCC^w850qfE@fL-L7elS6Q zc?Tw4-4V*6GE-KOERxtuHR~e-^lG3$egka)GTQQk%yqf@K zjS`&HYbK9SGiMZWYPn>t3y^JcdBFT}V4f-U5@*D&-xQol-U)!S$Ty>SP1tpt;9LsJ z50;jH_1b`CW>XTrCQ&j#@dz!e>erSCcRm&77O*uD!J3L}V1c#AEXZIRxTM}v6T(6* zg+YQ#VKL)}AfnjV!_pgWB{YsOojQ{kSJQ6gydUj)kdFEkcD58O9QCl(b?gA$;jG8# z4P*<}gv*9Xr7x1jI@`dZaJzy2LfXinrHs_Fb8N00TP`zn>w%MP#VoeF=dD!P(Oc0h zy)D|sXg9mB@VL|N@U9NIC!XC!#-h$~8>);68>@VTzg}kYkcj)M+JS11fXU41XqB`f z43D(55?Npd5n=X9JO<_sL=-L_t@e0S8=V+g?Fjx9oMyPJ?ev!ooMohLFQ|^)UB3Ng*5VueiJ_mMw2^ zMn13<=!$nT-ot0XKde@|;8*e+^Hai+ZFwXzx4!N4a!d-*zH%E3gb^jcwtm^(^ zl~#8D6W>K^WFn2P$guhJFes0dv3fQ<6*c)2Uk+d*{Y6b*_~BVLYjNP}zKNivCb@T%i_ z)*#8<0^#6TN=}3kJqQ(0h2vG4CnXf_zzI0R2_W4D;v~cCt%HN%8kwT6)C?P+E90JU z3_L&vlQ|q@5jwf7!qFZLN)c%X11eDNV5r>2NVS!w z+>2&C@qyHR=wet(t_NBY6-Qc#;aM>-$udY#!3h!?-#kYAb#cIo2R?SGRqn8l1|`uL zUxI#$wemFON-FAFXk73MY{jJXT{5w8z)nogje}oGHaNy9T0AC7u9y}bui9izMyz88X0YhjF-4hMTGTO%wf#D(}1a`|S%3Df$G*UN> zHZf}0o5-baxkT!G*--HMo9Z7(FaZ#S1Ck3-{~9VPG0qTIf&Js&z(phO%G5%qtKm&d z8bZaG#+VMj<$h;~E4Wa>+hj#EvV;oj=54V5eT#$SCfTTtnt$eP;@nb+%QWw7sLC0Y zTG+c9!H{*b&(hZkv3vN}kmEnV#QWY^s$_BoNQkS}Nc>H1*|Ssw)p;MQt=KCsS`Z?# zTQi*hy^)nuyuHgD78iHkftI*&emmJ?z_LaeLSGHdjMus+6-6tG7p*Ly92&k5s~tgI zT^6dF*xy74AISd18m8{%DnJ~hG~|$GuA5X}N4pQfZSWCdJ<$dU)#RhZdy#s>$DGUJ zI5qB$R(dE~Ev!`9QG&88%2nD~LR_}6E8iBUl-5imPRnQj9Wc~qJQ5}$qch;cIGu&W z5;{kG1o4r!AsIu%gWPgEw`1Hdjq9jq>KT(l+&@akR7?$@$b`Dh@ySfjzvH3sQ8#!% zn$5O!FO%+P(1Y-3SrC^)aF%?FXVL}Mf#>s|6?qvW>BIbSURBocO5ja|)|t{6LyUJS z&GmR~-9hg;eK4kU4vXu3_zs`45I6W-U0b6S_@bQfT6`6JSGn!=3(&W$=(||9m6KmX zKkb_G1AD_S@H=uR{T(^zd!cfdKQVSxwXMnky?+>Rc!+g#Y#G=#RHL#)JoU_=+uk{)|X+17WVMzDpN;rczpx7pZQU>j}9vptb~K-BluJe1M~)f(y4%YHCRQv0mI$o=6E zsXB&*JTuDVWB<%-1SCPTkKztqx5 zzZ!B9C+V&dH$>esxrI!TUJmYWhy0k#NYJ5fZO|hNH{qQ+)+?gndzntCKwvQg1sV*w{nm!}EaCR2LXzmY zB~899@IZV*7X^`sBTVdh`WwPD16)L}Eu#u-lCY7yIfGmxZ_%5AcBw?W+o9fBncmOJ zlCtT$BK4z!(Fd!x<6{M@kM($B4e1jTmNI=7^b!p@)`Ydp+M<3Hd`KP&!8Iv)hjn2+ zk$zi0Cl@!+p$dV1WHz)yf48u++~|KJ?3xF3!g4_jVGkIn0CVHI92fz{HnE8+wE4iK zok`3AhH+CWRF%zO(V1JJsR^eR#LZK4xCL=baw|nJZY`YMCiM%jlk+~vZYx}>bHF9x zs+Tlwr#1Uf*t1jM?aZBH2gxZy2+x)duL2!+CcD6{fcG_ZUw3>v^k6smmFRCL{M}u? z#69FhK+6_#E9RaEYDEZcKuD4vlGX#G^cvJ2RJdJwB|Hc#)AxaJ4~EV|3O`R@2YL?; zg}Q_n5CtACI1S-t$o(6ww&(jfz*b|;XJR8m-p1VQ`WK7I{A60DmKM5^q z<}^PM+Ymky8w!*!( zbjKSk=*FUJi2Oxn4>rt7UW?Kj}#{@8+q|j?<4F*%7J%rkUz!|tL(` zh=P|T#Sdbs1k^lAf>&#LQVd@6*7;_NV`6(mQhOXc+vZY|vwBF;Bei|!2q}eWB&S4N zKSsLu^ZJvJ;iEaK&xclN$v>uy$f@l2;p2AblLPNdIJCb?d`fQ0Vh(ven-lrfX=>&N z_pwBn46$-T5I2MYn<0`AcBKBmf(41DhYgrc%Y0+rT4zVVf4V-v&`t*iq`6 z)&yG96AvuVn%DX?&|Xq&l_l<%IRy&E{c^4K+(m0snuPD>P})iwiMy5-V$V5DMqrwv zHBN@=P`L7IoMU7nk9ye|1nI5{)2N+6#~eYyCVt!GZ361uwC^RG+ew>Nq1 z5bx2V{|miiTCBo`0gdUH_2kQ7LxTB{8LKnD6L2V_!XV~G2>LLN?h2mh#j24S6%eyy zAy!5U8^=nliKWnq6)}ndY`r#yKrBrkeWeW`npg#s(1bY=5%Hr76EL;HfMMs=kYvV+Fr#MyK3Q$obdXKmt{7jp&1Seg;o&-|TWxkFqMiXr!z)oxXxj=-k4 zS~i+ap9TT?x6$^GQ#b_YI}PWNHOzM+yV9(JP-bzX=E#z`7DC82f#`7puo;xFpkWk!(2RM zzXyH~&W~Hyt`K7@UEn=7SJ3KZ(PT`a#h20zbYtKc4aGsG-Vb>0B40nNQbaw&t;TsX|8G1(FDE*Q-~`oVK9h)jO~HoLrNo){to3RW_PB2X_; z#r80tGaGU65X%m_1qUh5U`+RjJHfiLgzP($kWfVUmQcUe-cQ*FQW$dpH9*S007ohL z0I)O;SF6QgsNRuf4-OiD-HzAhQD zys=yxURbI-3}l@{4hypThz@THJHZR`!d%r~eXIY#TMhd;vf>s9%*kZn&MSME7@&}o z3rQx4G}3{M{Q-G$^O!Prx$-5iyZTBg<#dRX3ddQ;%a)9m3d=RA2}Plv7Oe!uDRzlU!sx5g zvX!I4_i!bv{~MYg)N*ZBF78Z3nl2By%S%aX=Bz1_0u>m4$cP$dZ=)8}gc?&TYD4X) zW!df2je1ab>O<3#iZi4hQsXxM+N5#yWu)w>(^_8Bp{0*4@CoskdfRUOzI^itB!Q;$ zEXWG7fEZB2usf|mGtdEa#{S$0uJjGe@Q?!2vLiKo6DcH+PAZuq zm12shq=E*T>3xmHdP76{vxMeZd3R@x{Et&up4$Q23+J)N2^T!@?p4={L8Hd+Fk=UC zgwD{1o#!cWG}A;UJq*xKAA>!|7}LzMz`;&2Kjn6UmPM!wp;QgY@|c1G6@KZ9IF)+o zmxL&f%BTS~rasi01}^m|LfJu?%2#<%k?PbHCS4i|>R2k)n5HzV`N^3UwXV~k698JR zgB7+o5=ipZQ&B=qaW(2ti)Ltt*65G^7)mfPM5nZ9fu)qwOzU+zIB9s$Y3XfK82OYv zIpQwEe(Ly~H>2!e{?gRBc^;m}oy+`a=FNPyxs%}=_ym03PUQM%q4wFg9K`Xnv$@=& z)L%cjOXdZ7GPlJ~v~G8z))f0@1sHiBFPJ;W9WrrSB~EmzL{F|ivs;`@zG!VduRQZN z%F3&UPcvGCyHH^4t|p#b#R+yXGF)2<`*>x4`Um$hew%QF zmVxD!>#@G&l!ZcY0ISEZabV;CN2!%cI4-bdmh0k%_N9)!6I9%nj@|lA0?vwo#Xk*P zJN=V|Z`O>z6H2?!sj7hmZy)a~UNuj4!}s%)xPk}hRNU!rL&;a*ysx+cuAe*O(4q#I z0ijjyBDOD9j`zeDD^kwY&t@B%-Ag_RsiXY&Gc(Yh?1!9%Gsj>f4f2molV#}~a zPtR}Sfb#%j)a1%a!8I0OA$n=Im5kNGIsq6g^f}t^V@fekG&4ReIQB`GBl8hc=<5NW z^3h0V*@Rs6K2>P1fq&flr?dxfhP@I^u;%nxx+uxLCX}%ntP$(Ts)3(@A2CzRIp!AQ zU?wRErirMh*-qAy^&NTQWf>(e25F{%N8neen?)1%DS8>kxSiu>n!CRy;jkBtV6CYK zva&2+&v01Z(#O^px+PX_MmZ{j{q+hSROXK-6eKb`vA?0$$hl(QC`zU|yj9^9FWZ3g z(PG}!p+OwxX&FnJf=x3oi*}5MS(Kwzn%R0eTb-N5`3C$nS2Xio719tj?7ksu;!euI z$)P?`xjf?TE8f9*1)EdRl30#02bU?fE!Q=RbzS$hb2#6M6FaHvEYi7vCMvWa=c!>N z;c!Jlq4;8$$GL+u3O4f_Dx^;(6|O>>2p615Rc0nwxtU`Xj;3+GlvijUqsyB7SfLu_ zN2@sOd6%cFbA^aeFH<^OFMlon9R>1pVcuEHmzLo^nVDE_)N=*VsnA}9_VM$>1=@wN zlxH{NEao^4PvDX?%kSlQ+Lu=pUtXz5unjmL$t$!EJ*sOS>#D7s*nLi)K&?%i?*-s;}jj+-&{g8-6|5+}Dk+SkAkHM^54sVm)8mU7?RZhiiOOJ-LrwjsAB zby;*1`g1$dEPnf_q*;x$e21c{ym3XM?32^G&2GT!=*(;=12l{t!#}P_wzs zvNU?{7WB*%AU@Z<=3e7w_H%eF$$~JjW`CFqQFXV}2An%}X%e6?*S$0w4BS4#l@J+G zH88~eXf1FpQi7|Y^=knaRsiC^Cszs*=Sn>B;%>p_P|^72(@EO>l`jx3CQob;s-_v1 za%4r=N!UrKeeeq>go(sS*awrD4~L({Ozt&LwMY)8DQPVEYR;xdAqZQ;f7@Esc-Tx~PY?t>L3`kEK|^E0D6m04 zKWjU^y1llZ(RhYEr4pg5Kl-a5wudAO4*d8WGjcm?^m(Ilb*m{}|FWP@n{}*BSfxQ& z$u<#g8*&I1e=oe2w7e;6spvx6=jVAvdMC^Nj^2L*XB;5pRk~tWc>cOBeDBJOD+88j z=oHGPyC6do5;Vvb29!)0Mz$`jlaeIMk&?sHG88!?onbMI>~7elA0?-CD0>~ASq|T< zhCgIAysQCtf?Ojwkn8m?a`Sj2H*%-Ez}=bLCrqa`c}NGzBZ{FcI4h%id~ne zEJjA7PRFRCQz3a($+_gFD)GB&0$k~_Xz9r3V#bEJdz&|P6v+2=lKeEzN}T*r=NJdF z4lr&&HZTAH$QFRoeKa$*(%fZ2o=Qnt*Ixa%d%jn_osQm)KX5i`|* zVwfMIVCuRhUF0IcOYVy@E#)VVqE(Q5kfL>gB@y~C#P;xl5y<3jdc{1r!mgGYGhai`DdMaz@3yF3=Yi zjulWjmb!o&^~Gxaa4Tv>75To}{R{`9;vb;O{Rf~f`L_=D(e9z|wY6^d^AzfFd*RPI zz*TjJXk+emR1FUSQlA?Rl>4>WMcV5M9Vx4H0x0v0s$1Blrb+y(1hi0TuqChrZO!s& z32P!EN0gP2DOJXHg&VShjR&n_Qysz%L|Hn}rrbh0bXqG-BWIB4tg`RN`xa|8psMuX zYE#A2>T1ZShYM0R$t4XCIE9PVmN=;K$UMU~N1laBU4B*_zS)4Oia zA$2*o2eZ;g?ofk2m7T!6{N?4N3ll;YCkj8GI6$k)czGgzpz=+ip5ojDaXAD-qGwTQS0zm$|`VW9ELb{`ixXk9;ZUO?ohbp z-OJm)Gq(F8@1TK0G6;kXGD*c{asasAy9BsP82}Gi^2Fuqzh-;KP2O{S_(C4*(>(kE zK%(vO2zo;d?xqHYHu|>ESPVWxVzV%x7iWg;vrjQZb{Q)BdYdY83uxM*hvrb8if$QQ zUT99{qky3fb8F1Yi^e3c0)~DW^Q_iw9%c_ri3`KDvCn?e53E7~Go%W;42M?{z_6i_ zDvWUi4BHu3RRRFal&Th~$js`DM1zslN3k}`=AC&s4M3OIOs`1|jQlXr88umfk#2z5 zH%E{j2#lb`0|6Tx+e!WuET?%OP*gIH1d3bDQ=pz3JR=A+r4PJl7YuTLj12LhTN2xR&nT{-YlJtS=WG`jpeH~9?^es;|rW)Tst^UcMK zN3r0^*JBkm8CB6j+F2q*OR}h&7%PUDsc}1XR_g89A?ITNmLfUzxa#`&pAK3`u)YcR zRU+m65`V&jlgd7#W8!R9V!oO-Sf$k5RQQ!j=<01sEvqh=*J%fk5x=KjMo9K%h_#Rj zN1AY(_;Z+PIpjz@^Pgn=e9>h6fXze_*$~k!Mu=np@~rv(x9h^TrB z5L2&-tvAHgdlU%uRW2dX&&Cw`PbKwM<`FY!lv)~BnJccT*#IwnN$;VIk9~-}p`CAZ z@PjVc$9nqvhS<#oO>S-m&T?BHbPo^EfIiWjiaSQ4=aZvibS(m*6 zbN9_?+hR*b?6<7kjHu1y_BZYiwsgq6=tPgumq}hSlqp9LCi8&lY*a_eFmx;BNT<<$;&%Nc;etMReoCk6v(VFB4DwiB2g6||iJH-%!8DRkRW5wPt@EXH>g zM0f+EFLLjiLR8e=qm;dOsd)ceQ(ADVO5Y~{#SgcKI~oM30-~}U9<4Q26gaE*X(&I; zGcUtx&8|#qaWui9jY;Cx#im}}Zy$otrtN_?H6hyl(KKEL-C>;mzA%yQQ$U)sfR?3} zJdI3W#I}rbD?6_8S!lJsq=xM>jn8MJ;?~{|%}@P+juF0f-P(5k+3!25x+ST;4fk3! zZGU@5OETZLOPJ#b>dAs3BNcR((u)-ANGH+YR3^dYUM9GeC5FEejknrx9T!48Vy!f$ zWAsSysznMVHdZ#q?;G1uy^bw0oClp+ilZxNy#x}SuNkz9^lKRd60>^H(5g%(mFl`w z1=&igJcDVaAjdY8NRs+n#w?5V|J+=Umz=WTV6e| zGG!j|h;ORrzFjs28$@r5PQpoIJ4<}$iaDV`kVgN~O^AXU02ax4JsZEudm3NwgP!nZ zz#4#+$q%zeNI6%Mpu~@~cER}7oXk6%>iVLf>5+MrbDkc0o=g9N{0LKT3JY(Ii#}_R zhK?~|Xn4jI=k zE$7YkRf}6hACYbL05t6Wmct)I+qU(XZ+5SG^uM3p{Y?O6k}Ue}j*67uh#tfS18i4b z0a_H**@Sg<8wggY8cm8CIa_y$!Vj*Eek*62Qj;A4yat`m`hbac&+_BwaXBdv(Kp}} zA{0Stx)P&oCgb~3<5&POL|irrP@I3CY&)jp=thUy_t~puUNgQTSfoX@zG!I}lUY8XPDG8dl zT-injF+-q3m*oNW#mKb@UU}tXPQAKcEV?xu-rhkZnWA^7S@) zOh5%-09)o`jy39#P+P{MWc0qx-D0|LVm3-LxSSS(c6QWEk;#CDG|Pu?I*^%-m2(`U zkPj+hTCo?~=O5gKlFlbF5`K$?p)|?V}DrmT?xCA;Hj%cwIN_r z7TGW&?c;d5dbIv3O-qlDmJi+eyanFUAbMXKHHacjm^3&euzt_1wUPE^I4bbf>S?%N zSVO(G=L{Gr>G=79@=r9@_bxCX7gXvAi$}7LfR)v4#4e@`vMlKVbKJ2434Dl`rxnCH z+=+C;I7%IZ=yiDnld_H}HZb!+W3^B&PfF^1K8;4y4^yFXo)+h}309#DctdJY+(r?5$m&JtFu;l48dW>w&80|*os<2c?wBy#yB3YiJ50!zt5k3tIizJUMbmR_`!xSuKjrq^#0~HgCR&RS>bi!2= zFe(0ZgTCg0oM2XP9tEP z@$v~l;5eG-IYn%Bl>ImO{GD(1vWZBSC*MC;Oj8)17|R^8%(drRn@AZwbDoqY1Hx%M z<1M>zGQY?6mL>q$I_m`>TRZAWvQJW@dZfe`>Vb~6D+>k z*UTF0-IP`XOR6Z;e60W2lLv+=d1c_C(oafN#*-wpnT$==uFIds7-$HRth=g+sef=D zk`tPs7}Ad9{C8sLLo5494?oq-J6iw>onxq3o&L~{MhT@x*KpBW=vHmRW*7DK7C zOvVEvFiIffs`Gg*by{N2e2Q|IPxW;f?oZW8jc0U(FrU0Oq^P0fdUhVL+N$y+%Qdk6 zsV7G>>0C@Ubn2ldcM7H-jfOkZ!@-R9rASu~Nq}3v%4Gy)j4*_4jt)a7V?YnJ>n#jF zW5Jj}=@6_;%GRTt7Qr}QzA%{dLs#3e{Tn2#60{{3;jEfctrrrg1+9YX6;I=MB*i*U z`$`>DNh zSA4l*QeWz&jv4(VsAmb--b%TX$^Wri%pPlE9Jgm^(Uz#qIC#M)QU`dl`EYQ!^dXeLk zyU7TTq>qJI16oX4cv{A!y+EiOtyU-L;#^YyoYkg5oBVMC#vin#?I7DNF_4t0x-eOo z$FhK?x389Muj^;y7&->c2qY5=60M>&7-CVIG$zubnBju9<8vxx8{9VTzM9ckgyZ!WD%;9q3)J>5x#1#{C-V5eMlAI4>)nZeOKQ665Fb$N;$u_pQ`sDg0g{jPn zkBLtJlaAt76nsI!_L^Q?;!p%t$`2oc1xL*cN%IB_o>+yfUbC~BI|R}uj!l+eHqm4D zO!!rSJxyuIr8H+6$fV)DB}B~pXi4pVz}tCKmfRdyod|Hd%kr&&sb%k)Rrw7a#fCxe zyGwd4E+BfWbu%a!hlGE%0asm|{YZP(<)9^9n1!YkpgTNX!GFe=|EgFL%Y5d6}Iv=n1~K2I!Q-wB3;n6o5Gq0 zgA|yiF6Jte?v5ejnQ%gm5P(*dgEiX>x*Y7F4cF$29*$B^oNByo$P0;B`k zo?vQ~ilQPjSQi&ZYuBB26>R#$~7gYqsIqE0Rtkh;jQNvUi{kVvs-D-Ry1` z`;@?}V096G8S%a!G=tfe#YzJ26=dl&AaMvx>ZU$)~8WL%MCB4z5JUzXH(ga^r z!Ebq&d1UoLJg_pso3gekwOW9q2`al?`4Z748<&&yHY1oCfmdYuT#z$)xlNgrSFaeQ zdwbBzo*{@&^H45B#|VvdVfkj-xhFT1NTRzC+(n1s#~4PUQ_*>xaGhB!2+ zfVyLROJ+C+W-l-I=5?5;l!NCxV?}8XWq4aQ&j|c(H(u;P#T@atSQg$mI=F=8i%_A! z`(M)WE5A$m>yM@sjN6l{gDA%^02WeHf7umzbRUHl_wozEZ6b2p& z2X=$C?jRGK0(U2#&&hxF- zZ4ln20bLoyV((+I6-A|{%O|XWREEpJ%qOdhIatTuN31famoClrOQH*S7322v%cK^4 z{Tb!>Bm0eecI2>kuZ=Ze4uE}r-*ny@-Xbns@8-uY>}&~A$Ir+2xoLFz;}ys1PnmFG z-eJ1G%zYr`HCOyv@u+eFIHuBN70`zIZdu_=hyT@F9AG9MFpHoTz?MF*D~VSja8Xp& z@d4mfpMCS*r;CV}jfuw!83-;1V<+bB| z@swvZ1&XX{Sj}=1X)2U^5~~$^yxvuXn}u1KjSKFn+0#foLbKfN#$zUK<|d{v16LT$ zH-_!UW8CRA*wO6lWW7_^ppSV)VY}EZFk*oH@<#QuS5CwrLvc}in3FVaH+|E_giOka z88T(YM0}Sm%a|F{rp)#g9F@~BdiAsw^5wNYx^=ZRGG!Pw(l$ulO-Ps4P;e9}k}atB z_Y`ScIz>}wNDyyWpFx1%5kYr+P!&`LsX+sr_8*+t_~182ok}=lc(?&jpqOX%VRZIH z8gUiRV{DTbhs5pecD;oo@0I{p9}%AJxuO5MA~|?;tbc^~)JUG-1J=#q zws*bpw8F;-Z?U(=lSaQCDVs#*v-JiKlz&XP92f?*7#B4Dvcmz7t&5NkW%j9+2)}Pa zg@RwKE+rnT=Ir}j{_zY((cc~(69aGd zjX9e+X!NQ8%ceTCoW5V*1qOWeCHUANwOAQxjv`gD|2a!}^f@oXrwrOt{zW-Io_`!Q zT5m&2QWLEBy*)q20~LmiFCkERhyzAmEv5gyA3fT(Gw!L8(yb_yE$ix9(h1ZKW)Qf) za&OCJaJF9AsyuMwmPZZ4A+A{u|?wlv$t ze11d6m&@Zdr+8({A51~=^z7=vrI&wWobWjNQ5J9QvcEYT=s)k>FhJ#f)m_DF*zX+b zAZh}sOeIhHy4#OVMLqvtR%l>I`O)BHP-BWI@N*a;3*71(%6dghZzA50{T{r{=iqRu z@B99lHCpdlR8cMkI01M~G2{z3dVvz?n_KS#&DE4+`P zE@)Y__T{9NN~fcXenn@B-dX}XS@b~^+{nmO zeGdfc4%}1hPe{UywgK!yc<$S_yWdP`o>=*>SE+RN6UmjR{aC@YdU(rkX{XTh3RbJ} zqPQ4ppXoUK^mna>Y7#(eL(S8m7Zy!un3Zy=Xv*Y+ta`4Lj7#+THXK>7@%f93HzF&R&kCp zltZgZ2H#3OuF!LvH^uh_RhshI)bR0d#zX;;#o2raWNOt3ID%FsB|{uGpTPkB$LO8f zeg+3e9BWT-XLo`IFllVH{$wX1e3C?_*r~Yk;I%MseEl{P*k4a=2{~Kk+^JI;hV0QB z2>0m`au;24BK?C*#K?H)pPW{W^tjfC`;`bt*sLbGbJ3Tzu#oBDXJw+R$^HzF9*Ib3 z+}`JZ*>&;7uK}zM5=(`j;>#+3TCZf*#JBwJr~K>0W=A4d>H?ABYNbqj`;P4QO=-{{g`Ar&|Q_T2sa3?@opF$N;7df9nJ9grXnZ{^;u z39EG=F&M;iR)aWd-{R!^D%Cq$@*I87IVNJ`*OOnP216b+OTZT=*lp*dRVqUTeKQYs z258YMs2u-&2n;-t)~CWMB+~JBTc8Eks+^$sQUqgrevPvf{%cJ8WPD9Sz7~z&1rwVw z=8^IfRn0q^GVEx_`yM4F$0ZJb50t(xuN8bz3YLp}Tg^x>(#rS@J0j4AurS6=B-6`o z7|Z{@*gi>6`}5(gD+C|9=ikI30*T4l3_rJl#3nS0`6<;KMHFxUoAg^Xh|j%t*Y8=& zB*zJ^ng|8FO9elv&MlUA@>QBc;`Gzsa5XIK=@0bgnw7yNM5a1cQ+Wme>n8^)}+a9?s5oJcn@o!j)wn7vT2it-#K9 zQY{rdF)uinS$4ZIkahdvu!BHaA0lC`b!$B@D!r2sn2p2FtC($M_-@Ud>#P^TmaUmx z=5m#d*KA-``R;L}B?@VO;m1$L6M|aMfiSc^$G>(IGkE*xt}7IYPd}(GprD`_-fibT zSp3RK?2oHu5S}U~6Q@C@12B#g8cK0^RZQLENPx7ej6#c&bG#S&TY$yS{cC%hN2+{4|;q*2gG5=>LNZvcNpv3&NABwrA?tm%Eqtm>49xndy z0D}%tGy$l0RV26#guxDBahZv_+Nv|AfY%s=mDU|)haZXw6mC|jd18~n{C;?5NY`~x zjFOU$l@YU%iy(@eG$VCU8puHs@9`C`Z3ls+UUM5ejY8j zT*gMJN30xgIk`0Q?EG23XFp{X7d|l+QmZMZ5j*nu~9)LJ#uTEe(<^Z8yqdGb7b&VZ3q=| z&&_DFyP-6*am8=ZN9QABk0GO%nxEg<-nt(4_*H#;VE&K7lfo1<0+zOhR?ME&%Uy7| zwe@%BJ%06oLHPw?#KYV9MGMMU|4Vqm54qAJHvX|nglqum!&3UDQIKq@@HJ+-M6J13Ii?`%qO|6coj8GWR&7;6qxOPX7RrR?Eq0uzrx|KcQ( zlQx@~RA5nCCp9diQn%y!Q|aofQ8V-QFL+04FaFYP6|42yXRiNt`~b+m=k@5)bQSMJSEWR|9x-!0Vo=jNwJbCdSfyb zs{dz&$kgseCs&G%gqQ~T^+XauW>Ci!4(=FOzlFIwxmW&b%_U2l2c9$#SbP~_xhC-r z95nD^J)Zm*cY3wrhN1$l$3*9rS)@>z0fY|q`NipazpKL`VJf$g!oy7n2jltphevI; zsxD#J@YF5%0YM5igRT1l`cEPjp+Umi4B`=z!_kq36gX>8FC%^x$AmwAoxSqw4!zT_ zUXP_Q+KWk}OJ?L9r&W6uH?%o$PN#02P1%MKh+k5@@NokSWgF21E5G2;759LH{3;== zKx}H-f<%uUEtaMX_i0$ijYcVFfX>BGgIAM>L z)Ufb6U2eg%&s1W>`LTZY?tiIk3Qnk%dLW8|>9JxsBk4?4Qi;hPV+HT7dWqPD1uZMi z)fk9uK3!b1YZ`7AXWq09X-@YGXusXzrdcU4K=w62JorwCyR0A{`jT?qBXZOD_7pLw ze2%#NXZts3F8uW2>%VMk2OT=0$VUJRb+rZl=}~EAGukgCQVg6TaAhSaF@+)**L1Hh zpTtS31|bd<{6iFWqynM?uNya{1^k{QzWUt{Bt|NdtTJVVFVlIS3x=JZm6%!sPil*z zgPe6mwbGVFzREmp0EULM$-KRn#_1=Tru;h4@7o9K3Fi~=SN$hF*G}oT3RaWE9RMu! zr2D;pK06AO6oL!^hTE`2rMy$223I==DQa#)H_yg!cA^=Eib6e{h&@8$hG-Vn7;dg7 zQ+n~lWp1DgNhCeLnnRCV%`%@Ww(q9rWflq55G)oF`_{K0;3b@-@P*eqMfmJO1{8>X z&{}Rdk*lCL8x7CQAAk+%N-Hjf6!sb|bg_8Egxu4mv9^eQWJzT#e>n zY6y>>;S3S*2{+u-E$}5}hOCve82*f+GmJQt&1Pm{DN^Za&8_yizo&-_LyEUQ7JC|K z>{(>AVT-m(p`+`X=0fa1l^?sPjh^_k89+`RMIgHIJ8vC2i_Z_ZdZgnMdz51Okn`(= zU~3@1sbMKOOr9$9TI@@#VumJW*8^;Tb0X6^P7G}RpmOLBi$}`_xDBl>pc+eq4cgO_ z%a>#0!spM0cW>JvJ7PSP3?H>l|BJxlx+4`<+JOL9e@Yy!A1N|D9q2a3wmb!wh1?M9 zCqMi=2ds_)gvf8=g&oz*0~d+?HxqaMGXbCp?fCWVC$l6j_@G%T0T5_@3p4Yj z%$jT&vq-`|3y0TXdtr}-?B~qddy~-lEoLh=6Na~Yb9k;iy9PC)VA)UZTq58RPIpweo$!KWyP;jxjbpJJ5lPFxDg4YTpI^z0PNfX5a;$gE)26* z5+pu=+5hX=_|3?!UqfL8?x_NAF`ITP_Vy&EQZli>^;biH3~GEr(NJRYx3SXxW=Jv# zpJJnVo+FnHzm{#>!dV?!g+6?Xs^t-2m*szW(=m?xoRP$1`B@;*R zDasCPQAc5-W4pvsQ=L*s>eeYLKDS2Tz2HDfSfga9_Ux!^fOxU6CfL~9xxrJi1+1Km z-i4Ya8Z)YS;bPM$F8pSjnXik+;6TP$QGn`(A-a44g8_~j(oL2CeW zmR&ACUkId$o1)$278N<~vtwRXx8zGpW<6^|+o|QHPF%p3UGN+M3U4_x)BBX~ri&YL z&|SFjd>A({xAY-RA8gAm7v5eFD$6}xysXBz5w(~aI+yl=74#5~RrFMfPnB)86sT4L zG^<|b>#Zf%PC=Mhe*5Dt-QCn#ZE<2@vd3?hW@9sH;I`bcJF%PcRPzPRbx9@u??^QfClEgo8+VCV3i_l#$ zK|H$~5#jy^BLtmzApQPLfX3&>sS!JYlXb0gy**eJcL`7=+n>=}VsAV+_9}v6?h)CK zH{-#Zrsb=Kk_{iJxwtVeAVvq;ce(G|#ZC*b% zy>mJ8k+ny<1?pYIH-Poy!}GqI-3urtaJL zCGzS{iesxCBXaM{&vf1Z?v!Sn-$>>rw;ayXUQyw3A%Co#K~gYwbLoX9u#+=a${vCo zC0dyeyU5C-YywbRkM}F-_4b=SaQ>bfFNmZ=^AU~}0y50Ge*tm{Cd$ci7G9Mk=t*eK zzX_XH`5+Ggp-RtQ|Aui=R}B8NzB@9p_ow&=g#6&BGJ>=EhNHV7 zms+T{*Us?(Gp|<#s;qMUFsl;K4VMnL^M>>p5r=uX*_rh|X=jcZOd&Z-Bd7lwCa2r{ zbe0v>YkDYokXbjPUxcQXe@7kNz!RT#MJ!mvuI`?MBzJD+3ZMp3rPMGdgS#zfKZMGK z8aac^CA9VZqGJ)#fp0$}kU0^Tgsv8I`F2n@yBrNh60?biG$}&x4V#bcLG6Q_$fY}b z24`T`5HM??!?CE~{#2Mmaq?*!&n|0{{yE)aNF=+&(rW5x`0|`dk3xMuMTx|ps{-}& zz%}Za#1>z2ax$cr(#l`R-7HSLDsqKf+z0ejYD&eE-wip-1ffE{x+?9+=N)xD1=TQ`kV!r>Z?6|rMN|04|->9%eHqkpuTn3FT z1Chkn8!eZyOhkI|zO+oNkW8B}61rRjks50|rx!M+cXuf98ke1{UvgUL_8JEVxc&;O zn=8TLhm<$o<)Jg5(n&K*_CBhobtM;Tcu}0~YBVpN*ON+P=RJCP6NUVQ(#H}q5Zk|S z`FK%B@+S;}3gh*Z3+M9^&jzPkU$WCf1Bj+wfBkx|TZv(i3{8VXcpi?z`hE^hSfbR=Bi0p028KJJj$Y8vB zO>5%>*@C)t2T1akW=rc91+{A;%0tEHuHe*rS6uyy+BR0W#v&D0&Qs;Dz#d}IYX%32 zy%XU=)UWqpa;rRC_&H=r;??DPAF)9|xi4RFAqk5{6_NU{Y)}>3 zeX_7bjiRzt*IgAp6;{8HMs?+LGa{6pY|$<+L}Z$fa&uA2ujw`mD(4B;IttkuS9p8$ zPtOb`PEZbzMKg%F3)_(mO)#KA0XyGvVDDDSO*R>nw=D~H&yJft1V+rPwJTMFbH1Nk z%k3gnFA;L4=GJ)HjG}h!4R)I07O(&O%*|L8?;wP;+k)(%oQ(=wMU(qI#f9P`9V4qTNMPzfZ3TB)MO~}tJrVsh_8}^y!zAvJ)88faG z5O;Z5fhX>j#@f5pHzF4O&f8;{vD$r7j_c6LhG|eiD@;v}3SJi-g+nUd#qES@)$qal@@nL&rr`Tuyy0 z>Wo}|XH?u@GOve{f^;;I`cQUF&UfuUnER<@(XsA$$!=PI*fE0n?TjGmdJ}Ppr%UB4 zOHld=y^C7EPoq;BLf!9k2+7~MywAE>o9?-uo>y1^4=-|5z7gVz#vLI!n@@dwUb)dZsb>`1K%R{Q29s7>>JBFrjjR5&tNp~vtXFX8^L*sE#>sB zb|pY#o*|%Vfh4>{(?SWhX0k=2&*s9CR)e!IZEXMi8`;v%%;U;sY}fAOrEv(q`gn2{ z&>i>F=SR>-Mf{=rVgCE~bWVCpGbJkMPITLC?fL$yzDUmbmE6xa;Z&fhnV60C=*aB% z%SmfS@FRb2DPwzoTFS1k`K#2N+V5t@!4Jq~&G&x0_p9hBxA)4!(&TwwZ+}Qr0zsGi z79S19t%~1JA9CSN;gm2^NLC$v9a!J%**TW&{nt8-1Ob`)tlg|y)DM(tCBa3 zE!PMo9F&RIpnGO=TVooIU2^U6@6RaB9+g}~)N(vO_6!~H5Z2tX=Zy}!h3Q4Cn|m9uF#|zu-5uT(toxpSK6)nfdP?jvcW|Zr0a9 z^(m1mekt^4kUOCDk#YTp+&ri_g%`$p-@!KDQkeTarGQ+!^qBv+Fq_Q+dYNz)pc&xU zn`piXgT{XrE!bhUS6aStsp;HyC;On=HCN^alcSdqn-qA%qc&5u%_EeezAD>(joJxKGdZry|fdKWflH1 z{)u&ZiZpmr-k;EPppLfB;ZGNRKPaZ>MMLrrd=Yq=rZo@O% zXu)DFpKo-2(GgpiDQcN&=wc%-UyDUJR-waqq7xo0$kxU!0J7|`2LO=i=?L@mh7)A3 zUb4d{I}!4`1%iR()#I6*$&GR$Hwi&*6O!Bw+?~k13LrBDkXb5aR{(&_1@670TM5wv zz{?~pOrt_4E8WQY-ujG0r{>@kRR(m@=N_sey=k?{sxRa zSzA}g|KxXo@uCn-kUcQ7Is_CO^w~;%vr6A*A!-H1E49rUiUaVE_V}&#&gMQ$UF##% zb?`Yk>d{=8>m8}hhUihp>M;PQuRys^8~sh&{arginG2{t${JmPpsX;3vcncE>QS^U zOVc8jN814_r0szf({2FJ4A3l_Y$BWDV79MGY(Iyw{TQjz)uC-1B^d91-=|h0d zrE;WaR%s&XQkK3LyQxm|cszY`jREycF97vQ&j59%=hneUz2}p4)%kb6%Z%^W6QKU* zp8(KBK)2KX$6TGiZHSwIAFLjTlbv^3;4fXRE3ujGz078D=?}P{9@yk0PHAq+l@4@f z+4XlMw}=iI+DV3?)~a~AP(0&}8$#Ww2Jyyuo73a=Xq3)=$)=B+ae&HBQu{gQ2TMU; zE9^%l{iGBy8i1+e9q zNoB90+YQ->CPNR*!7=lk`rMotD|W8k808D*PF^q%0)~DXvwUElsG9QBYsX{km{4UK zv0@odfSDF^ynZ#%_hxrKn7!-Y z`E)Q=Ba&`%DBb38x+CBq(A--<5aHt>CB{iPQ~*9XQVj#_91?a3X_&y5L&sM_4_^pj z4QoP#R}9~5OjmZ+T$RN~c3RE7&<<%#*P>x;>f7gvr9e(byafsDzDGj!NtnUQeR(I7 zm5^`1`j}FYDI1cERJpEQooTl#9d@UCu`^mopPj)=57E9q6B_Fvvtr?X$hiE7@%2&T z`w7{MBm-o?Z2)Wx2+a?@5_4K{b&yh_x(4wE>*F>!=Jbea?H ze4#GfSUFNEl2-Bg;Y^jzRwbKL=hB77v#tU4T|a1rmjzHD?*EX2u!DlM!-C4{XK^+~ z%PLT~<`Cke^bIJoY$6ns4=fEIRGL1j4B>XS&k#EJJ_@*90Rl9AY{Z-nVsnLZGcVTCq`}5-5{8 zPObZ{NI*@MQPY&5W+4sL9930|dXK!VHpmv~>Pf1!>eJ#%Wt7jt8`_?}w16TqefTR( zA3@(OsXMByjo+urU#l)vth!7nt_2Mg-Tk9$nuG<_eFM~{764XKTh>$C+BAf)sz*%x zLM&&9>rC*;SJy8n zk#SRMexVVLIGgz`uLm1yNgdihRec>go))^%n>)FF50gmomKBf7-V;}b7LHwHU|kMR zKMAiN74Sxz#D@EXxseQTBhwZR!Wt#}352teOxDeK2NnfD1RJ$>>CiRGFb>Lt5o?^A zVHG5kpIIwI)Cl6!2$G%1G!i#RcH%5&H!UEGQ237nz2|Bw!MGS!pBFC2h-u!Q5H;MQdRV`?CudT;h)( z{(tY^-hbgSUZ6x@|7irzG{2oWQ|n8D4m_%p-8h}Lz2nC?CMXO-1JIUWD$_miNtc!v z$yG`3z^~E1X}6y`iGZVmJ#!*<3E{1ejeaY8$+#%{BTI%^s(j?LU1-rY#-ofNq;h$a zy4$FS_RPDLM%k3|uY;Gc%7dm**__t#Upq54= ziVeDu`Wz+gC!1Z7bRx3&6aA;A)mUuE#YK4_qS6O=7|F1&Xo0)rOy2yrno0+I%+Ahp2LKE99E~$A!b23jw1XT2Pd$J3+Xo zqN38n0^`HT76vi7*`TaIqFqS|>Otxt0kuV6Wk_!+uNgqxJWeQ1Cvz{xk|~O%dZ`O1XvSu876?Tu=FOvEO;*;Qw`2g5y>$*~C|g&|Xrb zOtkJM%Sh&lF@e8~_Z(lSo6H+yYtS&0L&~`krmIUjPtT5oy9-eEx^yt;yM$A75AKOG ziZVJ#gS|uhg9Fs0Fcow``SIqKpqxCYwRqsiGYt&{OHE%L4Ty#h76!MDd3cGGnXbjv zqlIYR_Hr|n55x9BS=8u% zgX7$E??diZjZ4ms_9aW_@MteT6XPps=V3ZlrP7qp2Nn)%Xwowf+#hZdMX~7h^5FzgzzjmNlX5NodKTgJA}Hj9{T0IMhm~msXZi3#e^KR&d8;X4f%; zYtO*_bd29irQ9zBX$NZLCcabrFPB#~QmN;p1su@_J2!VHoh(9@SR0=Nx!B!x{@x;P zSuNWO#b<`0nFkyB4RLnl>HsZ*t>4QuZ}^Se)$sI^LYZc$FK#THh6TSGdH|%2rOU`o zBo1$D2O|aTWAFDR2yz?<%*>zx%Vm~~6@1r9b)a!L5or;aYDcU3oXDo?i?H5FxkK0bSSe1lt`O9tul0l&@&Dzy6kl1$8 zQ`qNGS?4v7MUcT5JlIkoq&r9W&@ch}sM*-h{@ICCAIcjExdI9u$jmmNla#j9 z1!!g~gv?LXq22Xe!r1)CPatY{&d>t(lDHHUax#+{+O6HhCACt_>zwA{ZzM`Js|o3# zDpn&W*Fu{{W4Im8-3R;%860)Y%MM9MB4V{xTEd+x2fXExy3=@!qNG5lgwE&BOb>a! ztIW_0zB~9zYQIu;z!8&_TNmQ4bng>S&Ulm=IK^v_WtT9Ns?!HnECvK*ZsuNmz7xLC zvo72ozyM8h>ov3h4>Xdh?ZlNy55D`g67!(m=@)7XW%5gaS`q=)^6qqgGwQMa_mq+l zDwxLcvm)u{@Xjt3O-ShQ+D+jg981;@Dle0+GrqQwD+6HGBCBXGVaFdT0Bm~-+!G@p zK=b?iecz`Ch29eWQ)~0?f#CuKpdK{5CB#s@!@)o}IEq$T1DJSX0SI0rHEoz!)&=B^ zS~a4Y7u(y|&U^Q)qgN9d7!PzEdrJjLWFu%&kQmq-N!OElzMy7^tRG%}cv+VfZZ=e! zwX3rr40mkB?muOs{J=S$FfnJV5HP5=BM?Xp$bJ$y=CClJ$)|c^T>~A17;+#lUQep> z+O9!1&eyEMUfq|A*zoYXHg!j%&ZhxuEDqMG^&$X$Pzu1LFS+x^a%>DERg)`kSP7)& zkA!Rj0Y$lu9r7Wm^66qC|33Y;yp?4p3Xdg?0IB?vMmwwvFpAe6RLsxTYLD8B2nScw z9wd9`htj7**Viu!_yUZLo>*L{pJrx>N3Y7bPW_|EWyP4LEMKqxl@d(9hE@hYtGIT- z6zS$~`nJ40o8UPDhpQM)I|bETA`2Rq;tZloudkYE4dIfUb-sup7%C6lbEYW_EAMcp z4~^cQ^@dUTelgex?@PIsqH;AL!2T|V6%nM7pl<((D!hhL(M*Zuuvw2yIISPGJ-20({o(TEw-S{hmC%=W-Q=tKFl z(Kk8pny`1`mghHIBue^d8j$Qw$sw)vs!**i6by*Yjcz_=v5C=bdQO$THgC-rB6kg1 zXai*3%q))3LxO)7n;%Q*s;@B<*2EN^X0y1NPLCa48F5Cv1#?SK$041@@AcYQ7h>QN$y?XiyWNcR#N&1(`e zlH~?wVm5QS99PWlGClEM5@+LbqWwY!sK5;%CsPb`#p6vJ)WZ-ZL(hq7wMqpHZNwWW zOcMWd)rQC@+fS>lR0SNDlCR>0V73*Af`u!OgxNi6N75SZ)u252US6kQ$?mbPtzK-? z=Pr5ApxA{NODp?r__RHobv=MJ+m_}pEnMd*N@XSv>VfmxbFk7(h>gqiUYS0&lizlg zEg^MV6R(sdVN}55Bvo?+Lm@JN7m9l?gDgaBv+Kae{fO&@q$G4Vg|*CMFf4N7W|B9z zDVNIhMH#C+*{TR_2(9R#Y@4OK+}_LO**g70gT&<%J!D-Bfh!=G%(jl*Pe**t-SYi+ z0%}LvSS1``d*Yq>8bv8pX&pFuvM5l?KQqX0-(8CMBT+2cKkKs|0lp~>*T@KOPyFm45ZaM8CZl$5y){j% z1-fW}jw-wTNOfyTT(lJ+IB3vivF1A59U^%qj5{ENA<`Fnn!o0)6rjXTaD)QCapkK8 z0<2fVxysAGfFp=8hso1H8}aQiMBW0hG;V#X5fm92MhS-{0=UHs&t++eMuqA&)Hgr3| zy=t@^0(R?>)beX54Y|fLRH1RQJl!vuAp>oiXqpRc9a}&i+zaB{bvOheP_upV2 = LazyLock::new(|| { + // We already know the bundle is valid, so we can skip validation and just + // unwrap everywhere. + + let compressed = include_bytes!("../../assets/cmaps.brotli"); + let mut decompressed = Vec::new(); + let mut reader = compressed.as_slice(); + + brotli::BrotliDecompress(&mut reader, &mut decompressed) + .ok() + .unwrap(); + + let mut reader = Reader::new(&decompressed); + let huff_size = reader.read_u32().unwrap() as usize; + let huff_data = reader.read_bytes(huff_size).unwrap(); + let (delta_table, count_table) = huffman::decode_tables(huff_data).unwrap(); + + let mut entries = Vec::new(); + + while !reader.at_end() { + let start = reader.position(); + + // Skip file magic and version. + reader.read_bytes(6).unwrap(); + let file_len = reader.read_u32().unwrap() as usize; + + reader.read_bytes(file_len - 10).unwrap(); + entries.push(start..start + file_len); + } + + Bundle { + data: decompressed, + delta_table, + count_table, + entries, + } +}); + +pub(super) struct Bundle { + data: Vec, + pub(super) delta_table: HuffmanTable, + pub(super) count_table: HuffmanTable, + entries: Vec>, +} + +/// Load the data for a cmap file, by name. +pub fn load_embedded(name: CMapName<'_>) -> Option<&'static [u8]> { + // Get the index of the font of the cmap in the bundle. They are sorted + // alphabetically. + let idx = match name { + CMapName::N83pvRksjH => 0, + CMapName::N90msRksjH => 1, + CMapName::N90msRksjV => 2, + CMapName::N90mspRksjH => 3, + CMapName::N90mspRksjV => 4, + CMapName::N90pvRksjH => 5, + CMapName::AddRksjH => 6, + CMapName::AddRksjV => 7, + CMapName::AdobeCns1Ucs2 => 8, + CMapName::AdobeGb1Ucs2 => 9, + CMapName::AdobeJapan1Ucs2 => 10, + CMapName::AdobeKorea1Ucs2 => 11, + CMapName::B5pcH => 12, + CMapName::B5pcV => 13, + CMapName::CnsEucH => 14, + CMapName::CnsEucV => 15, + CMapName::ETenB5H => 16, + CMapName::ETenB5V => 17, + CMapName::ETenmsB5H => 18, + CMapName::ETenmsB5V => 19, + CMapName::EucH => 20, + CMapName::EucV => 21, + CMapName::ExtRksjH => 22, + CMapName::ExtRksjV => 23, + CMapName::GbEucH => 24, + CMapName::GbEucV => 25, + CMapName::GbkEucH => 26, + CMapName::GbkEucV => 27, + CMapName::Gbk2kH => 28, + CMapName::Gbk2kV => 29, + CMapName::GbkpEucH => 30, + CMapName::GbkpEucV => 31, + CMapName::GbpcEucH => 32, + CMapName::GbpcEucV => 33, + CMapName::H => 34, + CMapName::HKscsB5H => 35, + CMapName::HKscsB5V => 36, + CMapName::IdentityH => 37, + CMapName::IdentityV => 38, + CMapName::KscEucH => 39, + CMapName::KscEucV => 40, + CMapName::KscmsUhcH => 41, + CMapName::KscmsUhcHwH => 42, + CMapName::KscmsUhcHwV => 43, + CMapName::KscmsUhcV => 44, + CMapName::KscpcEucH => 45, + CMapName::UniCnsUcs2H => 46, + CMapName::UniCnsUcs2V => 47, + CMapName::UniCnsUtf16H => 48, + CMapName::UniCnsUtf16V => 49, + CMapName::UniGbUcs2H => 50, + CMapName::UniGbUcs2V => 51, + CMapName::UniGbUtf16H => 52, + CMapName::UniGbUtf16V => 53, + CMapName::UniJisUcs2H => 54, + CMapName::UniJisUcs2HwH => 55, + CMapName::UniJisUcs2HwV => 56, + CMapName::UniJisUcs2V => 57, + CMapName::UniJisUtf16H => 58, + CMapName::UniJisUtf16V => 59, + CMapName::UniKsUcs2H => 60, + CMapName::UniKsUcs2V => 61, + CMapName::UniKsUtf16H => 62, + CMapName::UniKsUtf16V => 63, + CMapName::V => 64, + CMapName::Custom(_) => return None, + }; + + let range = BUNDLE.entries.get(idx)?; + + Some(&BUNDLE.data[range.clone()]) +} diff --git a/vendor/hayro-cmap/src/bcmap/huffman.rs b/vendor/hayro-cmap/src/bcmap/huffman.rs new file mode 100644 index 000000000..b6e8e9ab5 --- /dev/null +++ b/vendor/hayro-cmap/src/bcmap/huffman.rs @@ -0,0 +1,180 @@ +use alloc::vec; +use alloc::vec::Vec; +use core::num::NonZeroU32; + +use super::reader::Reader; + +#[derive(Debug, Clone, Copy)] +pub(super) enum HuffmanNode { + /// Intermediate node with optional children. + Intermediate { + /// Index of the zero-child. + zero: Option, + /// Index of the one-child. + one: Option, + }, + /// Leaf node holding a decoded symbol value. + Leaf(u32), +} + +/// A Huffman table built from canonical codes. +#[derive(Debug)] +pub(super) struct HuffmanTable { + nodes: Vec, +} + +impl HuffmanTable { + /// Decode a single symbol from the bit reader. + pub(super) fn decode(&self, reader: &mut Reader<'_>) -> Option { + let mut idx = 0_u32; + + loop { + match self.nodes[idx as usize] { + HuffmanNode::Intermediate { zero, one } => { + let bit = reader.read_bit()?; + let child = if bit == 0 { zero } else { one }; + idx = child?.get(); + } + HuffmanNode::Leaf(symbol) => return Some(symbol), + } + } + } + + fn insert_code( + nodes: &mut Vec, + node_index: u32, + code: u32, + length: u8, + symbol: u32, + ) { + if length == 0 { + nodes[node_index as usize] = HuffmanNode::Leaf(symbol); + + return; + } + + let bit = (code >> (length - 1)) & 1; + let remaining = code & ((1 << (length - 1)) - 1); + + let child_index = match &nodes[node_index as usize] { + HuffmanNode::Intermediate { zero, one } => { + let existing = if bit == 0 { *zero } else { *one }; + match existing { + Some(idx) => idx, + None => { + let new_idx = NonZeroU32::new(nodes.len() as u32).unwrap(); + nodes.push(HuffmanNode::Intermediate { + zero: None, + one: None, + }); + + match &mut nodes[node_index as usize] { + HuffmanNode::Intermediate { zero, one } => { + if bit == 0 { + *zero = Some(new_idx); + } else { + *one = Some(new_idx); + } + } + _ => unreachable!(), + } + new_idx + } + } + } + _ => return, + }; + + Self::insert_code(nodes, child_index.get(), remaining, length - 1, symbol); + } + + fn build(code_lengths: &[u8], symbols: &[u32]) -> Self { + debug_assert_eq!(code_lengths.len(), symbols.len()); + + if symbols.is_empty() { + return Self { + nodes: vec![HuffmanNode::Intermediate { + zero: None, + one: None, + }], + }; + } + + let max_length = *code_lengths.iter().max().unwrap_or(&0) as usize; + + let mut codes = vec![0_u32; symbols.len()]; + let mut code = 0_u32; + + for length in 1..=max_length { + for (i, &cl) in code_lengths.iter().enumerate() { + if cl as usize == length { + codes[i] = code; + code += 1; + } + } + code <<= 1; + } + + let mut nodes = vec![HuffmanNode::Intermediate { + zero: None, + one: None, + }]; + + for (i, &symbol) in symbols.iter().enumerate() { + if code_lengths[i] == 0 { + continue; + } + + Self::insert_code(&mut nodes, 0, codes[i], code_lengths[i], symbol); + } + + Self { nodes } + } +} + +pub(super) fn decode_tables(data: &[u8]) -> Option<(HuffmanTable, HuffmanTable)> { + let mut reader = Reader::new(data); + let delta = decode_single_table(&mut reader, |r| r.read_u32())?; + let count = decode_single_table(&mut reader, |r| r.read_u8().map(u32::from))?; + + Some((delta, count)) +} + +fn decode_single_table( + reader: &mut Reader<'_>, + read_sym: fn(&mut Reader<'_>) -> Option, +) -> Option { + let n_symbols = reader.read_u16()? as usize; + + if n_symbols == 0 { + let _max_len = reader.read_u8()?; + + return Some(HuffmanTable { + nodes: vec![HuffmanNode::Intermediate { + zero: None, + one: None, + }], + }); + } + + let max_code_length = reader.read_u8()? as usize; + + let mut counts = Vec::with_capacity(max_code_length); + for _ in 0..max_code_length { + counts.push(reader.read_u16()?); + } + + let mut code_lengths = Vec::with_capacity(n_symbols); + let mut symbols = Vec::with_capacity(n_symbols); + + for (length_idx, &count) in counts.iter().enumerate() { + let code_len = (length_idx + 1) as u8; + + for _ in 0..count { + symbols.push(read_sym(reader)?); + code_lengths.push(code_len); + } + } + + Some(HuffmanTable::build(&code_lengths, &symbols)) +} diff --git a/vendor/hayro-cmap/src/bcmap/mod.rs b/vendor/hayro-cmap/src/bcmap/mod.rs new file mode 100644 index 000000000..997bc00d8 --- /dev/null +++ b/vendor/hayro-cmap/src/bcmap/mod.rs @@ -0,0 +1,416 @@ +//! Inspired by pdf.js, we use a custom binary format to optionally embed +//! the required cmaps in a space-efficient manner. However, we don't use +//! their format description but have a custom one. For more information, +//! see . + +pub(crate) mod embedded; +pub(crate) mod huffman; +pub(crate) mod reader; + +use alloc::boxed::Box; +use alloc::vec::Vec; + +pub use embedded::load_embedded; + +use crate::bcmap::embedded::BUNDLE; +use crate::{ + BfRange, CMap, CMapName, CharacterCollection, CidFamily, CidRange, CodespaceRange, Metadata, + PartitionedRanges, Range, WritingMode, parse, +}; +use huffman::HuffmanTable; +use reader::Reader; + +const BCMAP_MAGIC: &[u8] = b"bcmap"; +const BCMAP_VERSION: u8 = 0x01; +const BCMAP_FILE_HEADER_SIZE: usize = 10; +const SEG_HEADER_SIZE: usize = 5; + +const SEGMENT_RANGE_1B: u8 = 0x01; +const SEGMENT_SINGLE_1B: u8 = 0x02; +const SEGMENT_RANGE_2B: u8 = 0x03; +const SEGMENT_SINGLE_2B: u8 = 0x04; +const SEGMENT_RANGE_3B: u8 = 0x05; +const SEGMENT_SINGLE_3B: u8 = 0x06; +const SEGMENT_RANGE_4B: u8 = 0x07; +const SEGMENT_SINGLE_4B: u8 = 0x08; +const SEGMENT_USECMAP: u8 = 0x09; +const SEGMENT_NOTDEF: u8 = 0x0A; +const SEGMENT_WMODE: u8 = 0x0B; +const SEGMENT_CODESPACE: u8 = 0x0C; +const SEGMENT_NAME: u8 = 0x0D; +const SEGMENT_CID_SYSTEM_INFO: u8 = 0x0E; +const SEGMENT_BF_RANGE_VARIABLE: u8 = 0x0F; +const SEGMENT_BF_SINGLE_VARIABLE: u8 = 0x10; +const SEGMENT_BF_SINGLE_1U: u8 = 0x11; +const SEGMENT_BF_SINGLE_2U: u8 = 0x12; +const SEGMENT_BF_SINGLE_3U: u8 = 0x13; +const SEGMENT_BF_SINGLE_4U: u8 = 0x14; +const SEGMENT_BF_RANGE_1U: u8 = 0x15; +const SEGMENT_BF_RANGE_2U: u8 = 0x16; + +pub(crate) fn parse<'a>( + data: &[u8], + get_cmap: impl Fn(CMapName<'_>) -> Option<&'a [u8]> + Clone + 'a, + depth: u32, +) -> Option { + // While in theory we can assume that all binary cmaps are valid, it can + // of course happen that an invalid one has been passed from outside, so + // we still need to do proper validation. + + if data.get(..5)? != BCMAP_MAGIC || *data.get(5)? != BCMAP_VERSION { + return None; + } + + let mut reader = Reader::new(data); + reader.read_bytes(6)?; // Skip magic + version. + + let file_len = reader.read_u32()? as usize; + + let delta_table = &BUNDLE.delta_table; + let count_table = &BUNDLE.count_table; + + let mut cmap_name = None; + let mut character_collection = None; + let mut writing_mode = None; + let mut base: Option> = None; + let mut _codespace_ranges = Vec::new(); + let mut cid_ranges = PartitionedRanges::new(); + let mut notdef_ranges = PartitionedRanges::new(); + let mut bf_entries = Vec::new(); + + // Start parsing all segments of the file. + let mut reader = Reader::new(data.get(BCMAP_FILE_HEADER_SIZE..file_len)?); + + while !reader.at_end() { + let seg_type = reader.read_u8()?; + let seg_len = reader.read_u32()? as usize; + + let payload = reader.read_bytes(seg_len.checked_sub(SEG_HEADER_SIZE)?)?; + + match seg_type { + SEGMENT_NAME => { + cmap_name = Some(Vec::from(payload)); + } + SEGMENT_CID_SYSTEM_INFO => { + // Format: Each string is 0-terminated. + let mut r = Reader::new(payload); + let registry = Vec::from(r.eat_until(|b| b == 0)); + r.read_u8()?; + + let ordering = Vec::from(r.eat_until(|b| b == 0)); + r.read_u8()?; + let supplement = r.read_u16()? as i32; + + character_collection = Some(CharacterCollection { + family: CidFamily::from_registry_ordering(®istry, &ordering), + supplement, + }); + } + SEGMENT_USECMAP => { + let base_data = get_cmap(CMapName::from_bytes(payload))?; + + base = Some(Box::new(parse::parse_inner( + base_data, + get_cmap.clone(), + depth, + )?)); + } + SEGMENT_WMODE => { + writing_mode = match payload.first()? { + 0 => Some(WritingMode::Horizontal), + 1 => Some(WritingMode::Vertical), + _ => None, + }; + } + SEGMENT_CODESPACE => { + parse_codespace(payload, &mut _codespace_ranges)?; + } + SEGMENT_NOTDEF => { + parse_notdef(payload, &mut notdef_ranges)?; + } + SEGMENT_RANGE_1B | SEGMENT_RANGE_2B | SEGMENT_RANGE_3B | SEGMENT_RANGE_4B => { + let bw = seg_type.div_ceil(2) as usize; + parse_cid_segment(payload, &mut cid_ranges, bw, delta_table, Some(count_table))?; + } + SEGMENT_SINGLE_1B | SEGMENT_SINGLE_2B | SEGMENT_SINGLE_3B | SEGMENT_SINGLE_4B => { + let bw = seg_type.div_ceil(2) as usize; + parse_cid_segment(payload, &mut cid_ranges, bw, delta_table, None)?; + } + SEGMENT_BF_RANGE_VARIABLE => { + parse_bf_segment( + payload, + &mut bf_entries, + delta_table, + Some(count_table), + None, + )?; + } + SEGMENT_BF_SINGLE_VARIABLE => { + parse_bf_segment(payload, &mut bf_entries, delta_table, None, None)?; + } + SEGMENT_BF_RANGE_1U => { + parse_bf_segment( + payload, + &mut bf_entries, + delta_table, + Some(count_table), + Some(1), + )?; + } + SEGMENT_BF_RANGE_2U => { + parse_bf_segment( + payload, + &mut bf_entries, + delta_table, + Some(count_table), + Some(2), + )?; + } + SEGMENT_BF_SINGLE_1U => { + parse_bf_segment(payload, &mut bf_entries, delta_table, None, Some(1))?; + } + SEGMENT_BF_SINGLE_2U => { + parse_bf_segment(payload, &mut bf_entries, delta_table, None, Some(2))?; + } + SEGMENT_BF_SINGLE_3U => { + parse_bf_segment(payload, &mut bf_entries, delta_table, None, Some(3))?; + } + SEGMENT_BF_SINGLE_4U => { + parse_bf_segment(payload, &mut bf_entries, delta_table, None, Some(4))?; + } + _ => { + return None; + } + } + } + + cid_ranges.sort(); + notdef_ranges.sort(); + bf_entries.sort_by(|a, b| a.range.start.cmp(&b.range.start)); + + Some(CMap { + metadata: Metadata { + character_collection, + name: cmap_name, + writing_mode, + }, + _codespace_ranges, + cid_ranges, + notdef_ranges, + bf_entries, + base, + }) +} + +fn parse_codespace(payload: &[u8], ranges: &mut Vec) -> Option<()> { + let mut r = Reader::new(payload); + let n_ranges = r.read_u8()? as usize; + + for _ in 0..n_ranges { + let bw = r.read_u8()? as usize; + let low = r.read_n_bytes(bw)?; + let high = r.read_n_bytes(bw)?; + + ranges.push(CodespaceRange { + number_bytes: bw as u8, + low, + high, + }); + } + + Some(()) +} + +fn parse_notdef(payload: &[u8], ranges: &mut PartitionedRanges) -> Option<()> { + let mut r = Reader::new(payload); + let bw = r.read_u8()? as usize; + let n_entries = r.read_u16()? as usize; + + for _ in 0..n_entries { + let start = r.read_n_bytes(bw)?; + let end = r.read_n_bytes(bw)?; + let cid = r.read_u16()? as u32; + + ranges.push( + bw, + CidRange { + range: Range { start, end }, + cid_start: cid, + }, + ); + } + + Some(()) +} + +fn parse_cid_segment( + payload: &[u8], + ranges: &mut PartitionedRanges, + byte_width: usize, + delta_table: &HuffmanTable, + count_table: Option<&HuffmanTable>, +) -> Option<()> { + let mut r = Reader::new(payload); + let n_entries = r.read_u16()? as usize; + + if n_entries == 0 { + return Some(()); + } + + // There are two types of CID segments, all of which are stored in a columnar + // fastion in the file: + // For the first type, we have a simple mapping from a single code to a single + // CID. In this case, the data stream first contains all codes using delta-coding, + // encoded using huffman coding. This is followed by all CIDs, stored as + // u16 (either as a raw CID, or as `0` in case the CID is +1 the previous CID). + // + // For the second type, we have a range of consecutive codes, which map + // to a range of consecutive CIDs. In this case, we have an additional "count" + // column that stores how many codes are mapped consecutively. These are also + // encoded using huffman coding. + + // Read deltas for the start code points. + let delta_len = r.read_u32()? as usize; + let delta_data = r.read_bytes(delta_len)?; + + let mut delta_reader = Reader::new(delta_data); + let mut deltas = Vec::with_capacity(n_entries); + + for _ in 0..n_entries { + deltas.push(delta_table.decode(&mut delta_reader)?); + } + + // Read the counts for each consecutive code range. + let mut counts = Vec::new(); + if let Some(ct) = count_table { + let count_len = r.read_u32()? as usize; + let count_data = r.read_bytes(count_len)?; + + let mut count_reader = Reader::new(count_data); + counts.reserve(n_entries); + for _ in 0..n_entries { + counts.push(ct.decode(&mut count_reader)?); + } + } + + // Read the CIDs and reconstruct the character codes that map to them. + let is_range = count_table.is_some(); + let mut prev_end: Option = None; + let mut prev_cid: Option = None; + let mut prev_range_len: u32 = 0; + + for i in 0..n_entries { + let raw_cid = r.read_u16()?; + + // Note that start deltas and counts are encoded minus one, so we + // need to add one when reconstructing them. + + // Reconstruct start code. + let start = if let Some(pe) = prev_end { + pe + 1 + deltas[i] + } else { + deltas[i] + }; + + // Reconstruct end code. + let end = if is_range { + start + counts[i] + 1 + } else { + start + }; + + // CID 0 means it's consecutive to the last seen CID, plus 1. + let cid = if raw_cid == 0 { + if let Some(pc) = prev_cid { + pc + prev_range_len + 1 + } else { + 0 + } + } else { + raw_cid as u32 + }; + + ranges.push( + byte_width, + CidRange { + range: Range { start, end }, + cid_start: cid, + }, + ); + + prev_end = Some(end); + prev_cid = Some(cid); + prev_range_len = end - start; + } + + Some(()) +} + +fn parse_bf_segment( + payload: &[u8], + entries: &mut Vec, + delta_table: &HuffmanTable, + count_table: Option<&HuffmanTable>, + fixed_units: Option, +) -> Option<()> { + let mut r = Reader::new(payload); + let n_entries = r.read_u16()? as usize; + + let delta_len = r.read_u32()? as usize; + let delta_data = r.read_bytes(delta_len)?; + + let mut delta_reader = Reader::new(delta_data); + let mut deltas = Vec::with_capacity(n_entries); + for _ in 0..n_entries { + deltas.push(delta_table.decode(&mut delta_reader)?); + } + + // Read counts for range segments. + let mut counts = Vec::new(); + if let Some(ct) = count_table { + let count_len = r.read_u32()? as usize; + let count_data = r.read_bytes(count_len)?; + + let mut count_reader = Reader::new(count_data); + counts.reserve(n_entries); + for _ in 0..n_entries { + counts.push(ct.decode(&mut count_reader)?); + } + } + + let is_range = count_table.is_some(); + let mut prev_end: Option = None; + + for i in 0..n_entries { + // Fixed-width segments encode destinations without a length prefix. + let n_units = match fixed_units { + Some(nu) => nu, + None => r.read_u8()? as usize, + }; + let mut dst_base = Vec::with_capacity(n_units); + for _ in 0..n_units { + dst_base.push(r.read_u16()?); + } + + let start = if let Some(pe) = prev_end { + pe + 1 + deltas[i] + } else { + deltas[i] + }; + + let end = if is_range { + start + counts[i] + 1 + } else { + start + }; + + entries.push(BfRange { + range: Range { start, end }, + dst_base, + }); + + prev_end = Some(end); + } + + Some(()) +} diff --git a/vendor/hayro-cmap/src/bcmap/reader.rs b/vendor/hayro-cmap/src/bcmap/reader.rs new file mode 100644 index 000000000..25d663910 --- /dev/null +++ b/vendor/hayro-cmap/src/bcmap/reader.rs @@ -0,0 +1,123 @@ +pub(super) struct Reader<'a> { + data: &'a [u8], + byte_pos: usize, + bit_pos: u8, +} + +impl<'a> Reader<'a> { + #[inline] + pub(super) fn new(data: &'a [u8]) -> Self { + Self { + data, + byte_pos: 0, + bit_pos: 0, + } + } + + #[inline] + pub(super) fn read_bit(&mut self) -> Option { + let byte = *self.data.get(self.byte_pos)?; + let bit = (byte >> (7 - self.bit_pos)) & 1; + + self.bit_pos += 1; + + if self.bit_pos == 8 { + self.bit_pos = 0; + self.byte_pos += 1; + } + + Some(bit) + } + + #[inline] + pub(super) fn read_u16(&mut self) -> Option { + debug_assert_eq!(self.bit_pos, 0, "read_u16 called at non-byte boundary"); + + let bytes: [u8; 2] = self + .data + .get(self.byte_pos..self.byte_pos + 2)? + .try_into() + .ok()?; + self.byte_pos += 2; + + Some(u16::from_be_bytes(bytes)) + } + + #[inline] + pub(super) fn read_u32(&mut self) -> Option { + debug_assert_eq!(self.bit_pos, 0, "read_u32 called at non-byte boundary"); + + let bytes: [u8; 4] = self + .data + .get(self.byte_pos..self.byte_pos + 4)? + .try_into() + .ok()?; + self.byte_pos += 4; + + Some(u32::from_be_bytes(bytes)) + } + + #[inline] + pub(super) fn read_u8(&mut self) -> Option { + debug_assert_eq!(self.bit_pos, 0, "read_u8 called at non-byte boundary"); + + let val = *self.data.get(self.byte_pos)?; + self.byte_pos += 1; + + Some(val) + } + + #[inline] + pub(super) fn read_n_bytes(&mut self, n: usize) -> Option { + debug_assert_eq!( + self.bit_pos, 0, + "read_n_bytes_be called at non-byte boundary" + ); + + match n { + 1 => Some(self.read_u8()? as u32), + 2 => Some(self.read_u16()? as u32), + 3 => { + let bytes = self.read_bytes(3)?; + Some((bytes[0] as u32) << 16 | (bytes[1] as u32) << 8 | bytes[2] as u32) + } + 4 => self.read_u32(), + _ => None, + } + } + + #[inline] + pub(super) fn read_bytes(&mut self, n: usize) -> Option<&'a [u8]> { + debug_assert_eq!(self.bit_pos, 0, "read_bytes called at non-byte boundary"); + + let bytes = self.data.get(self.byte_pos..self.byte_pos + n)?; + self.byte_pos += n; + + Some(bytes) + } + + #[inline] + pub(super) fn position(&self) -> usize { + self.byte_pos + } + + #[inline] + pub(super) fn at_end(&self) -> bool { + self.byte_pos >= self.data.len() + } + + #[inline] + pub(super) fn eat_until(&mut self, f: impl Fn(u8) -> bool) -> &'a [u8] { + debug_assert_eq!(self.bit_pos, 0, "eat_until called at non-byte boundary"); + + let start = self.byte_pos; + while let Some(&b) = self.data.get(self.byte_pos) { + if f(b) { + break; + } + self.byte_pos += 1; + } + + &self.data[start..self.byte_pos] + } +} diff --git a/vendor/hayro-cmap/src/lib.rs b/vendor/hayro-cmap/src/lib.rs new file mode 100644 index 000000000..aed021eee --- /dev/null +++ b/vendor/hayro-cmap/src/lib.rs @@ -0,0 +1,1620 @@ +/*! +A parser for cmap files, as they are found in PDFs. + +This crate provides a parser for cmap files and allows you to +- Map character codes from text-showing operators to CID identifiers. +- Map CIDs to Unicode characters or strings. + +## Safety +This crate forbids unsafe code via a crate-level attribute. +*/ + +#![cfg_attr(not(feature = "std"), no_std)] +#![forbid(unsafe_code)] +#![deny(missing_docs)] + +extern crate alloc; + +#[cfg(feature = "embed-cmaps")] +mod bcmap; +mod parse; + +#[cfg(feature = "embed-cmaps")] +pub use bcmap::load_embedded; + +/// Look up an embedded binary cmap by name. +/// +/// Returns `None` when the `embed-cmaps` feature is not enabled. +#[cfg(not(feature = "embed-cmaps"))] +pub fn load_embedded(_name: CMapName<'_>) -> Option<&'static [u8]> { + None +} + +use alloc::boxed::Box; +use alloc::string::String; +use alloc::vec::Vec; + +/// A CID (Character Identifier). +pub type Cid = u32; + +/// The name of the cmap. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum CMapName<'a> { + // ── Adobe-Japan1 (Japanese) ── Shift-JIS (RKSJ) encodings ── + /// `83pv-RKSJ-H` — Adobe-Japan1, Mac Shift-JIS (`KanjiTalk` 6), horizontal. + N83pvRksjH, + /// `90ms-RKSJ-H` — Adobe-Japan1, Windows Shift-JIS (code page 932), horizontal. + N90msRksjH, + /// `90ms-RKSJ-V` — Adobe-Japan1, Windows Shift-JIS (code page 932), vertical. + N90msRksjV, + /// `90msp-RKSJ-H` — Adobe-Japan1, Windows Shift-JIS with proportional Roman, horizontal. + N90mspRksjH, + /// `90msp-RKSJ-V` — Adobe-Japan1, Windows Shift-JIS with proportional Roman, vertical. + N90mspRksjV, + /// `90pv-RKSJ-H` — Adobe-Japan1, Mac Shift-JIS (`KanjiTalk` 7), horizontal. + N90pvRksjH, + /// `Add-RKSJ-H` — Adobe-Japan1, Fujitsu FMR Shift-JIS, horizontal. + AddRksjH, + /// `Add-RKSJ-V` — Adobe-Japan1, Fujitsu FMR Shift-JIS, vertical. + AddRksjV, + + // ── Adobe UCS2 (CID → Unicode) ── + // Note that these are not part of the predefined CMaps, but we use them + // so we can more easily map CIDs back to Unicode. + /// `Adobe-CNS1-UCS2` — Adobe-CNS1, CID-to-Unicode mapping. + AdobeCns1Ucs2, + /// `Adobe-GB1-UCS2` — Adobe-GB1, CID-to-Unicode mapping. + AdobeGb1Ucs2, + /// `Adobe-Japan1-UCS2` — Adobe-Japan1, CID-to-Unicode mapping. + AdobeJapan1Ucs2, + /// `Adobe-Korea1-UCS2` — Adobe-Korea1, CID-to-Unicode mapping. + AdobeKorea1Ucs2, + + // ── Adobe-CNS1 (Traditional Chinese) ── Big Five encodings ── + /// `B5pc-H` — Adobe-CNS1, Mac Big Five (`ETen` extensions), horizontal. + B5pcH, + /// `B5pc-V` — Adobe-CNS1, Mac Big Five (`ETen` extensions), vertical. + B5pcV, + /// `CNS-EUC-H` — Adobe-CNS1, CNS 11643 EUC encoding, horizontal. + CnsEucH, + /// `CNS-EUC-V` — Adobe-CNS1, CNS 11643 EUC encoding, vertical. + CnsEucV, + /// `ETen-B5-H` — Adobe-CNS1, `ETen` Big Five extensions, horizontal. + ETenB5H, + /// `ETen-B5-V` — Adobe-CNS1, `ETen` Big Five extensions, vertical. + ETenB5V, + /// `ETenms-B5-H` — Adobe-CNS1, `ETen` Big Five with Microsoft symbol extensions, horizontal. + ETenmsB5H, + /// `ETenms-B5-V` — Adobe-CNS1, `ETen` Big Five with Microsoft symbol extensions, vertical. + ETenmsB5V, + + // ── Adobe-Japan1 (Japanese) ── EUC and extended Shift-JIS encodings ── + /// `EUC-H` — Adobe-Japan1, JIS X 0208 EUC-JP encoding, horizontal. + EucH, + /// `EUC-V` — Adobe-Japan1, JIS X 0208 EUC-JP encoding, vertical. + EucV, + /// `Ext-RKSJ-H` — Adobe-Japan1, Shift-JIS with NEC/IBM extensions, horizontal. + ExtRksjH, + /// `Ext-RKSJ-V` — Adobe-Japan1, Shift-JIS with NEC/IBM extensions, vertical. + ExtRksjV, + + // ── Adobe-GB1 (Simplified Chinese) ── + /// `GB-EUC-H` — Adobe-GB1, GB 2312-80 EUC encoding, horizontal. + GbEucH, + /// `GB-EUC-V` — Adobe-GB1, GB 2312-80 EUC encoding, vertical. + GbEucV, + /// `GBK-EUC-H` — Adobe-GB1, GBK encoding (Microsoft code page 936), horizontal. + GbkEucH, + /// `GBK-EUC-V` — Adobe-GB1, GBK encoding (Microsoft code page 936), vertical. + GbkEucV, + /// `GBK2K-H` — Adobe-GB1, GB 18030-2000 encoding, horizontal. + Gbk2kH, + /// `GBK2K-V` — Adobe-GB1, GB 18030-2000 encoding, vertical. + Gbk2kV, + /// `GBKp-EUC-H` — Adobe-GB1, GBK with proportional Roman, horizontal. + GbkpEucH, + /// `GBKp-EUC-V` — Adobe-GB1, GBK with proportional Roman, vertical. + GbkpEucV, + /// `GBpc-EUC-H` — Adobe-GB1, Mac GB 2312 (simplified) EUC encoding, horizontal. + GbpcEucH, + /// `GBpc-EUC-V` — Adobe-GB1, Mac GB 2312 (simplified) EUC encoding, vertical. + GbpcEucV, + + // ── Adobe-Japan1 (Japanese) ── JIS encoding ── + /// `H` — Adobe-Japan1, JIS X 0208 row-cell encoding, horizontal. + H, + + // ── Adobe-CNS1 (Traditional Chinese) ── Hong Kong SCS ── + /// `HKscs-B5-H` — Adobe-CNS1, Hong Kong SCS (Big Five with HKSCS extensions), horizontal. + HKscsB5H, + /// `HKscs-B5-V` — Adobe-CNS1, Hong Kong SCS (Big Five with HKSCS extensions), vertical. + HKscsB5V, + + // ── Adobe-Identity ── + /// `Identity-H` — Adobe-Identity, two-byte identity mapping, horizontal. + /// Character codes map directly to CIDs (i.e. CID = character code). + IdentityH, + /// `Identity-V` — Adobe-Identity, two-byte identity mapping, vertical. + /// Character codes map directly to CIDs (i.e. CID = character code). + IdentityV, + + // ── Adobe-Korea1 (Korean) ── + /// `KSC-EUC-H` — Adobe-Korea1, KS X 1001:1992 EUC-KR encoding, horizontal. + KscEucH, + /// `KSC-EUC-V` — Adobe-Korea1, KS X 1001:1992 EUC-KR encoding, vertical. + KscEucV, + /// `KSCms-UHC-H` — Adobe-Korea1, Microsoft UHC (Unified Hangul Code, code page 949), horizontal. + KscmsUhcH, + /// `KSCms-UHC-HW-H` — Adobe-Korea1, Microsoft UHC with half-width Roman, horizontal. + KscmsUhcHwH, + /// `KSCms-UHC-HW-V` — Adobe-Korea1, Microsoft UHC with half-width Roman, vertical. + KscmsUhcHwV, + /// `KSCms-UHC-V` — Adobe-Korea1, Microsoft UHC (Unified Hangul Code, code page 949), vertical. + KscmsUhcV, + /// `KSCpc-EUC-H` — Adobe-Korea1, Mac KS X 1001:1992 EUC-KR encoding, horizontal. + KscpcEucH, + + // ── Adobe-CNS1 (Traditional Chinese) ── Unicode encodings ── + /// `UniCNS-UCS2-H` — Adobe-CNS1, Unicode UCS-2 encoding, horizontal. + UniCnsUcs2H, + /// `UniCNS-UCS2-V` — Adobe-CNS1, Unicode UCS-2 encoding, vertical. + UniCnsUcs2V, + /// `UniCNS-UTF16-H` — Adobe-CNS1, Unicode UTF-16 encoding, horizontal. + UniCnsUtf16H, + /// `UniCNS-UTF16-V` — Adobe-CNS1, Unicode UTF-16 encoding, vertical. + UniCnsUtf16V, + + // ── Adobe-GB1 (Simplified Chinese) ── Unicode encodings ── + /// `UniGB-UCS2-H` — Adobe-GB1, Unicode UCS-2 encoding, horizontal. + UniGbUcs2H, + /// `UniGB-UCS2-V` — Adobe-GB1, Unicode UCS-2 encoding, vertical. + UniGbUcs2V, + /// `UniGB-UTF16-H` — Adobe-GB1, Unicode UTF-16 encoding, horizontal. + UniGbUtf16H, + /// `UniGB-UTF16-V` — Adobe-GB1, Unicode UTF-16 encoding, vertical. + UniGbUtf16V, + + // ── Adobe-Japan1 (Japanese) ── Unicode encodings ── + /// `UniJIS-UCS2-H` — Adobe-Japan1, Unicode UCS-2 encoding, horizontal. + UniJisUcs2H, + /// `UniJIS-UCS2-HW-H` — Adobe-Japan1, Unicode UCS-2 with half-width Roman, horizontal. + UniJisUcs2HwH, + /// `UniJIS-UCS2-HW-V` — Adobe-Japan1, Unicode UCS-2 with half-width Roman, vertical. + UniJisUcs2HwV, + /// `UniJIS-UCS2-V` — Adobe-Japan1, Unicode UCS-2 encoding, vertical. + UniJisUcs2V, + /// `UniJIS-UTF16-H` — Adobe-Japan1, Unicode UTF-16 encoding, horizontal. + UniJisUtf16H, + /// `UniJIS-UTF16-V` — Adobe-Japan1, Unicode UTF-16 encoding, vertical. + UniJisUtf16V, + + // ── Adobe-Korea1 (Korean) ── Unicode encodings ── + /// `UniKS-UCS2-H` — Adobe-Korea1, Unicode UCS-2 encoding, horizontal. + UniKsUcs2H, + /// `UniKS-UCS2-V` — Adobe-Korea1, Unicode UCS-2 encoding, vertical. + UniKsUcs2V, + /// `UniKS-UTF16-H` — Adobe-Korea1, Unicode UTF-16 encoding, horizontal. + UniKsUtf16H, + /// `UniKS-UTF16-V` — Adobe-Korea1, Unicode UTF-16 encoding, vertical. + UniKsUtf16V, + + // ── Adobe-Japan1 (Japanese) ── JIS encoding ── + /// `V` — Adobe-Japan1, JIS X 0208 row-cell encoding, vertical. + V, + + /// A custom (non-predefined) `CMap` name. + Custom(&'a [u8]), +} + +impl<'a> CMapName<'a> { + /// Create a `CMapType` from raw bytes. + pub fn from_bytes(name: &'a [u8]) -> Self { + match name { + b"83pv-RKSJ-H" => Self::N83pvRksjH, + b"90ms-RKSJ-H" => Self::N90msRksjH, + b"90ms-RKSJ-V" => Self::N90msRksjV, + b"90msp-RKSJ-H" => Self::N90mspRksjH, + b"90msp-RKSJ-V" => Self::N90mspRksjV, + b"90pv-RKSJ-H" => Self::N90pvRksjH, + b"Add-RKSJ-H" => Self::AddRksjH, + b"Add-RKSJ-V" => Self::AddRksjV, + b"Adobe-CNS1-UCS2" => Self::AdobeCns1Ucs2, + b"Adobe-GB1-UCS2" => Self::AdobeGb1Ucs2, + b"Adobe-Japan1-UCS2" => Self::AdobeJapan1Ucs2, + b"Adobe-Korea1-UCS2" => Self::AdobeKorea1Ucs2, + b"B5pc-H" => Self::B5pcH, + b"B5pc-V" => Self::B5pcV, + b"CNS-EUC-H" => Self::CnsEucH, + b"CNS-EUC-V" => Self::CnsEucV, + b"ETen-B5-H" => Self::ETenB5H, + b"ETen-B5-V" => Self::ETenB5V, + b"ETenms-B5-H" => Self::ETenmsB5H, + b"ETenms-B5-V" => Self::ETenmsB5V, + b"EUC-H" => Self::EucH, + b"EUC-V" => Self::EucV, + b"Ext-RKSJ-H" => Self::ExtRksjH, + b"Ext-RKSJ-V" => Self::ExtRksjV, + b"GB-EUC-H" => Self::GbEucH, + b"GB-EUC-V" => Self::GbEucV, + b"GBK-EUC-H" => Self::GbkEucH, + b"GBK-EUC-V" => Self::GbkEucV, + b"GBK2K-H" => Self::Gbk2kH, + b"GBK2K-V" => Self::Gbk2kV, + b"GBKp-EUC-H" => Self::GbkpEucH, + b"GBKp-EUC-V" => Self::GbkpEucV, + b"GBpc-EUC-H" => Self::GbpcEucH, + b"GBpc-EUC-V" => Self::GbpcEucV, + b"H" => Self::H, + b"HKscs-B5-H" => Self::HKscsB5H, + b"HKscs-B5-V" => Self::HKscsB5V, + b"Identity-H" => Self::IdentityH, + b"Identity-V" => Self::IdentityV, + b"KSC-EUC-H" => Self::KscEucH, + b"KSC-EUC-V" => Self::KscEucV, + b"KSCms-UHC-H" => Self::KscmsUhcH, + b"KSCms-UHC-HW-H" => Self::KscmsUhcHwH, + b"KSCms-UHC-HW-V" => Self::KscmsUhcHwV, + b"KSCms-UHC-V" => Self::KscmsUhcV, + b"KSCpc-EUC-H" => Self::KscpcEucH, + b"UniCNS-UCS2-H" => Self::UniCnsUcs2H, + b"UniCNS-UCS2-V" => Self::UniCnsUcs2V, + b"UniCNS-UTF16-H" => Self::UniCnsUtf16H, + b"UniCNS-UTF16-V" => Self::UniCnsUtf16V, + b"UniGB-UCS2-H" => Self::UniGbUcs2H, + b"UniGB-UCS2-V" => Self::UniGbUcs2V, + b"UniGB-UTF16-H" => Self::UniGbUtf16H, + b"UniGB-UTF16-V" => Self::UniGbUtf16V, + b"UniJIS-UCS2-H" => Self::UniJisUcs2H, + b"UniJIS-UCS2-HW-H" => Self::UniJisUcs2HwH, + b"UniJIS-UCS2-HW-V" => Self::UniJisUcs2HwV, + b"UniJIS-UCS2-V" => Self::UniJisUcs2V, + b"UniJIS-UTF16-H" => Self::UniJisUtf16H, + b"UniJIS-UTF16-V" => Self::UniJisUtf16V, + b"UniKS-UCS2-H" => Self::UniKsUcs2H, + b"UniKS-UCS2-V" => Self::UniKsUcs2V, + b"UniKS-UTF16-H" => Self::UniKsUtf16H, + b"UniKS-UTF16-V" => Self::UniKsUtf16V, + b"V" => Self::V, + _ => Self::Custom(name), + } + } + + /// Convert the `CMapType` back to its raw byte representation. + pub fn to_bytes(&self) -> &[u8] { + match self { + Self::N83pvRksjH => b"83pv-RKSJ-H", + Self::N90msRksjH => b"90ms-RKSJ-H", + Self::N90msRksjV => b"90ms-RKSJ-V", + Self::N90mspRksjH => b"90msp-RKSJ-H", + Self::N90mspRksjV => b"90msp-RKSJ-V", + Self::N90pvRksjH => b"90pv-RKSJ-H", + Self::AddRksjH => b"Add-RKSJ-H", + Self::AddRksjV => b"Add-RKSJ-V", + Self::AdobeCns1Ucs2 => b"Adobe-CNS1-UCS2", + Self::AdobeGb1Ucs2 => b"Adobe-GB1-UCS2", + Self::AdobeJapan1Ucs2 => b"Adobe-Japan1-UCS2", + Self::AdobeKorea1Ucs2 => b"Adobe-Korea1-UCS2", + Self::B5pcH => b"B5pc-H", + Self::B5pcV => b"B5pc-V", + Self::CnsEucH => b"CNS-EUC-H", + Self::CnsEucV => b"CNS-EUC-V", + Self::ETenB5H => b"ETen-B5-H", + Self::ETenB5V => b"ETen-B5-V", + Self::ETenmsB5H => b"ETenms-B5-H", + Self::ETenmsB5V => b"ETenms-B5-V", + Self::EucH => b"EUC-H", + Self::EucV => b"EUC-V", + Self::ExtRksjH => b"Ext-RKSJ-H", + Self::ExtRksjV => b"Ext-RKSJ-V", + Self::GbEucH => b"GB-EUC-H", + Self::GbEucV => b"GB-EUC-V", + Self::GbkEucH => b"GBK-EUC-H", + Self::GbkEucV => b"GBK-EUC-V", + Self::Gbk2kH => b"GBK2K-H", + Self::Gbk2kV => b"GBK2K-V", + Self::GbkpEucH => b"GBKp-EUC-H", + Self::GbkpEucV => b"GBKp-EUC-V", + Self::GbpcEucH => b"GBpc-EUC-H", + Self::GbpcEucV => b"GBpc-EUC-V", + Self::H => b"H", + Self::HKscsB5H => b"HKscs-B5-H", + Self::HKscsB5V => b"HKscs-B5-V", + Self::IdentityH => b"Identity-H", + Self::IdentityV => b"Identity-V", + Self::KscEucH => b"KSC-EUC-H", + Self::KscEucV => b"KSC-EUC-V", + Self::KscmsUhcH => b"KSCms-UHC-H", + Self::KscmsUhcHwH => b"KSCms-UHC-HW-H", + Self::KscmsUhcHwV => b"KSCms-UHC-HW-V", + Self::KscmsUhcV => b"KSCms-UHC-V", + Self::KscpcEucH => b"KSCpc-EUC-H", + Self::UniCnsUcs2H => b"UniCNS-UCS2-H", + Self::UniCnsUcs2V => b"UniCNS-UCS2-V", + Self::UniCnsUtf16H => b"UniCNS-UTF16-H", + Self::UniCnsUtf16V => b"UniCNS-UTF16-V", + Self::UniGbUcs2H => b"UniGB-UCS2-H", + Self::UniGbUcs2V => b"UniGB-UCS2-V", + Self::UniGbUtf16H => b"UniGB-UTF16-H", + Self::UniGbUtf16V => b"UniGB-UTF16-V", + Self::UniJisUcs2H => b"UniJIS-UCS2-H", + Self::UniJisUcs2HwH => b"UniJIS-UCS2-HW-H", + Self::UniJisUcs2HwV => b"UniJIS-UCS2-HW-V", + Self::UniJisUcs2V => b"UniJIS-UCS2-V", + Self::UniJisUtf16H => b"UniJIS-UTF16-H", + Self::UniJisUtf16V => b"UniJIS-UTF16-V", + Self::UniKsUcs2H => b"UniKS-UCS2-H", + Self::UniKsUcs2V => b"UniKS-UCS2-V", + Self::UniKsUtf16H => b"UniKS-UTF16-H", + Self::UniKsUtf16V => b"UniKS-UTF16-V", + Self::V => b"V", + Self::Custom(name) => name, + } + } +} + +/// Let's limit the number of nested `usecmap` references to 16. +const MAX_NESTING_DEPTH: u32 = 16; + +/// A parsed cmap. +#[derive(Debug, Clone)] +pub struct CMap { + metadata: Metadata, + // Note that we don't actually use this, because Acrobat _seems_ to ignore + // it, too. + _codespace_ranges: Vec, + cid_ranges: PartitionedRanges, + notdef_ranges: PartitionedRanges, + bf_entries: Vec, + base: Option>, +} + +impl CMap { + /// Parse a cmap from raw bytes. + /// + /// The `get_cmap` callback is used to recursively resolve cmaps that + /// are referenced via `usecmap`. + pub fn parse<'a>( + data: &[u8], + get_cmap: impl Fn(CMapName<'_>) -> Option<&'a [u8]> + Clone + 'a, + ) -> Option { + parse::parse_inner(data, get_cmap, 0) + } + + /// Create an Identity-H cmap. + pub fn identity_h() -> Self { + Self::identity(WritingMode::Horizontal, b"Identity-H") + } + + /// Create an Identity-V cmap. + pub fn identity_v() -> Self { + Self::identity(WritingMode::Vertical, b"Identity-V") + } + + fn identity(writing_mode: WritingMode, name: &[u8]) -> Self { + Self { + metadata: Metadata { + character_collection: Some(CharacterCollection { + family: CidFamily::AdobeIdentity, + supplement: 0, + }), + name: Some(Vec::from(name)), + writing_mode: Some(writing_mode), + }, + _codespace_ranges: Vec::new(), + cid_ranges: { + let mut r = PartitionedRanges::new(); + r.push( + 2, + CidRange { + range: Range { + start: 0, + end: 0xFFFF, + }, + cid_start: 0, + }, + ); + r + }, + notdef_ranges: PartitionedRanges::new(), + bf_entries: Vec::new(), + base: None, + } + } + + /// Return the metadata of this cmap. + pub fn metadata(&self) -> &Metadata { + &self.metadata + } + + /// Look up the CID code of a character code. + /// + /// Returns `None` if the code does not match any range for the given byte length. + pub fn lookup_cid_code(&self, code: u32, byte_len: u8) -> Option { + // Note that, in theory, we are supposed to first check the code space range + // whether the entry exists in the first place. However, from my experiments + // Acrobat mostly seems to ignore this, so we do that as well. + + let cid_ranges = self.cid_ranges.get(byte_len)?; + let notdef_ranges = self.notdef_ranges.get(byte_len)?; + + if let Some(entry) = find_in_ranges(cid_ranges, code) { + let offset = code.checked_sub(entry.range.start)?; + return entry.cid_start.checked_add(offset); + } else if let Some(entry) = find_in_ranges(notdef_ranges, code) { + // For `.notdef` ranges, all codes map to the same `.notdef` CID, so + // no adding of the offset here. + return Some(entry.cid_start); + } + + // See pdfjs_bug920426, it uses bf chars for encoding the characters + // of a text-showing operator, so try that as well. We don't want to + // recurse though, only check the bf chars of this level. + if let Some(BfString::Char(lookup)) = self.lookup_bf_string_inner(code, false) { + return Some(lookup as u32); + } + + // If we still haven't found anything, check the base cmap. + self.base + .as_ref() + .and_then(|b| b.lookup_cid_code(code, byte_len)) + } + + /// Look up a bf string in the cmap. This is usually + /// used for mapping character codes to Unicode codepoints in a + /// `ToUnicode` cmap. + /// + /// Returns `None` if no mapping is available. + pub fn lookup_bf_string(&self, code: u32) -> Option { + self.lookup_bf_string_inner(code, true) + } + + fn lookup_bf_string_inner(&self, code: u32, recurse: bool) -> Option { + if let Some(entry) = find_in_ranges(&self.bf_entries, code) { + let offset = u16::try_from(code - entry.range.start).ok()?; + + fn decode_utf16(units: &[u16]) -> Option { + let mut iter = core::char::decode_utf16(units.iter().copied()); + let first = iter.next()?.ok()?; + + if iter.next().is_none() { + Some(BfString::Char(first)) + } else { + let s = String::from_utf16(units).ok()?; + Some(BfString::String(s)) + } + } + + return if offset == 0 { + Some(decode_utf16(&entry.dst_base)?) + } else { + let mut units = entry.dst_base.clone(); + *units.last_mut()? = units.last()?.checked_add(offset)?; + Some(decode_utf16(&units)?) + }; + } + + if recurse { + self.base.as_ref()?.lookup_bf_string(code) + } else { + None + } + } +} + +trait HasRange { + fn range(&self) -> &Range; +} + +fn find_in_ranges(entries: &[T], code: u32) -> Option<&T> { + let idx = entries + .binary_search_by(|entry| { + let r = entry.range(); + if code < r.start { + core::cmp::Ordering::Greater + } else if code > r.end { + core::cmp::Ordering::Less + } else { + core::cmp::Ordering::Equal + } + }) + .ok()?; + + Some(&entries[idx]) +} + +/// A range with a start and end code. +#[derive(Debug, Clone)] +pub(crate) struct Range { + pub(crate) start: u32, + pub(crate) end: u32, +} + +/// A range of character codes mapped to CIDs. +#[derive(Debug, Clone)] +pub struct CidRange { + pub(crate) range: Range, + pub(crate) cid_start: Cid, +} + +impl HasRange for CidRange { + fn range(&self) -> &Range { + &self.range + } +} + +#[derive(Debug, Clone)] +pub(crate) struct PartitionedRanges { + buckets: [Vec; 4], +} + +impl PartitionedRanges { + pub(crate) fn new() -> Self { + Self { + buckets: [Vec::new(), Vec::new(), Vec::new(), Vec::new()], + } + } + + // PdfCraft patch: needed by the truncated-stream guard in parse.rs. + pub(crate) fn is_empty(&self) -> bool { + self.buckets.iter().all(|b| b.is_empty()) + } + + pub(crate) fn push(&mut self, byte_len: usize, range: CidRange) { + if let Some(bucket) = byte_len + .checked_sub(1) + .and_then(|i| self.buckets.get_mut(i)) + { + bucket.push(range); + } + } + + pub(crate) fn get(&self, byte_len: u8) -> Option<&[CidRange]> { + let idx = (byte_len as usize).checked_sub(1)?; + self.buckets.get(idx).map(|v| v.as_slice()) + } + + pub(crate) fn sort(&mut self) { + for bucket in &mut self.buckets { + bucket.sort_by(|a, b| a.range.start.cmp(&b.range.start)); + } + } +} + +#[derive(Debug, Clone)] +pub(crate) struct BfRange { + pub(crate) range: Range, + pub(crate) dst_base: Vec, +} + +impl HasRange for BfRange { + fn range(&self) -> &Range { + &self.range + } +} + +/// A codespace range defining valid character code byte sequences. +#[derive(Debug, Clone)] +#[allow(dead_code)] +pub(crate) struct CodespaceRange { + pub(crate) number_bytes: u8, + pub(crate) low: u32, + pub(crate) high: u32, +} + +/// A Unicode value decoded from a cmap. +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum BfString { + /// A single character. + Char(char), + /// A UTF-8 string. + String(String), +} + +/// Metadata extracted from a cmap file. +#[derive(Debug, Clone)] +pub struct Metadata { + /// The referenced character collection. + pub character_collection: Option, + /// The cmap name. + pub name: Option>, + /// The writing mode. + pub writing_mode: Option, +} + +/// The registry+ordering family of a CID character collection. +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum CidFamily { + /// Adobe-Japan1 + AdobeJapan1, + /// Adobe-GB1 + AdobeGB1, + /// Adobe-CNS1 + AdobeCNS1, + /// Adobe-Korea1 + AdobeKorea1, + /// Adobe-Identity + AdobeIdentity, + /// A non-predefined registry/ordering pair. + Custom { + /// The registry name. + registry: Vec, + /// The ordering name. + ordering: Vec, + }, +} + +impl CidFamily { + /// Create a `CidFamily` from raw registry and ordering byte strings. + pub fn from_registry_ordering(registry: &[u8], ordering: &[u8]) -> Self { + if registry == b"Adobe" { + match ordering { + b"Japan1" => Self::AdobeJapan1, + b"GB1" => Self::AdobeGB1, + b"CNS1" => Self::AdobeCNS1, + b"Korea1" => Self::AdobeKorea1, + b"Identity" => Self::AdobeIdentity, + _ => Self::Custom { + registry: registry.to_vec(), + ordering: ordering.to_vec(), + }, + } + } else { + Self::Custom { + registry: registry.to_vec(), + ordering: ordering.to_vec(), + } + } + } + + /// Return the predefined UTF-16 `CMap` that maps Unicode code points to CIDs + /// for this family, if one exists. + pub fn unicode_cmap(&self) -> Option> { + match self { + Self::AdobeJapan1 => Some(CMapName::UniJisUtf16H), + Self::AdobeGB1 => Some(CMapName::UniGbUtf16H), + Self::AdobeCNS1 => Some(CMapName::UniCnsUtf16H), + Self::AdobeKorea1 => Some(CMapName::UniKsUtf16H), + Self::AdobeIdentity | Self::Custom { .. } => None, + } + } + + /// Return the `UCS2` cmap corresponding to the given character collection. + pub fn ucs2_cmap(&self) -> Option> { + match self { + Self::AdobeJapan1 => Some(CMapName::AdobeJapan1Ucs2), + Self::AdobeGB1 => Some(CMapName::AdobeGb1Ucs2), + Self::AdobeCNS1 => Some(CMapName::AdobeCns1Ucs2), + Self::AdobeKorea1 => Some(CMapName::AdobeKorea1Ucs2), + Self::AdobeIdentity | Self::Custom { .. } => None, + } + } +} + +/// A CID character collection identifying the character set and ordering. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct CharacterCollection { + /// The registry+ordering family. + pub family: CidFamily, + /// The supplement number. + pub supplement: i32, +} + +/// The writing mode of a cmap. +#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)] +pub enum WritingMode { + /// Horizontal writing mode. + #[default] + Horizontal, + /// Vertical writing mode. + Vertical, +} + +#[cfg(test)] +mod tests { + use super::*; + + // Note that those cmaps might not be completely valid according to the rules + // of cmap/Postscript, but since our parser is very lenient and doesn't run a real + // interpreter we can shorten them by a lot. + + const PREAMBLE: &[u8] = br#"/CIDSystemInfo 3 dict dup begin + /Registry (Adobe) def + /Ordering (Japan1) def + /Supplement 0 def +end def +/CMapName /Test def +/WMode 0 def +2 begincodespacerange +<00> +<0000> +endcodespacerange +"#; + + fn parse_with_preamble(body: &[u8]) -> CMap { + let mut data = Vec::new(); + data.extend_from_slice(PREAMBLE); + data.extend_from_slice(body); + CMap::parse(&data, |_| None).unwrap() + } + + #[test] + fn metadata_parsing() { + let data = br#" +/CIDInit /ProcSet findresource begin +12 dict begin +begincmap +/CIDSystemInfo 3 dict dup begin + /Registry (Adobe) def + /Ordering (Japan1) def + /Supplement 6 def +end def +/CMapName /Adobe-Japan1-H def +/CMapType 1 def +/WMode 0 def +endcmap"#; + + let cmap = CMap::parse(data, |_| None).unwrap(); + let cc = cmap.metadata().character_collection.as_ref().unwrap(); + assert_eq!(cc.family, CidFamily::AdobeJapan1); + assert_eq!(cc.supplement, 6); + assert_eq!( + cmap.metadata().name.as_deref(), + Some(b"Adobe-Japan1-H".as_slice()) + ); + assert_eq!(cmap.metadata().writing_mode, Some(WritingMode::Horizontal)); + } + + #[test] + fn vertical_writing_mode() { + let data = br#" +/CIDSystemInfo 3 dict dup begin + /Registry (Adobe) def + /Ordering (Japan1) def + /Supplement 6 def +end def +/CMapName /Adobe-Japan1-V def +/WMode 1 def +"#; + + let cmap = CMap::parse(data, |_| None).unwrap(); + assert_eq!(cmap.metadata().writing_mode, Some(WritingMode::Vertical)); + assert_eq!( + cmap.metadata().name.as_deref(), + Some(b"Adobe-Japan1-V".as_slice()) + ); + } + + #[test] + fn cid_range_lookup() { + let cmap = parse_with_preamble( + br#" +3 begincidrange +<0000> <00FF> 0 +<0100> <01FF> 256 +<8140> <817E> 633 +endcidrange +"#, + ); + + assert_eq!(cmap.lookup_cid_code(0x0000, 2), Some(0)); + assert_eq!(cmap.lookup_cid_code(0x0042, 2), Some(0x42)); + assert_eq!(cmap.lookup_cid_code(0x00FF, 2), Some(0xFF)); + + assert_eq!(cmap.lookup_cid_code(0x0100, 2), Some(256)); + assert_eq!(cmap.lookup_cid_code(0x01FF, 2), Some(511)); + + assert_eq!(cmap.lookup_cid_code(0x8140, 2), Some(633)); + assert_eq!(cmap.lookup_cid_code(0x817E, 2), Some(633 + 62)); + } + + #[test] + fn cid_char_lookup() { + let cmap = parse_with_preamble( + br#" +3 begincidchar +<03> 1 +<04> 2 +<20> 50 +endcidchar +"#, + ); + + assert_eq!(cmap.lookup_cid_code(0x03, 1), Some(1)); + assert_eq!(cmap.lookup_cid_code(0x04, 1), Some(2)); + assert_eq!(cmap.lookup_cid_code(0x20, 1), Some(50)); + } + + #[test] + fn lookup_miss() { + let cmap = parse_with_preamble( + br#" +1 begincidrange +<0100> <01FF> 0 +endcidrange +"#, + ); + + assert_eq!(cmap.lookup_cid_code(0x00FF, 2), None); + assert_eq!(cmap.lookup_cid_code(0x0200, 2), None); + assert_eq!(cmap.lookup_cid_code(0xFFFF, 2), None); + } + + #[test] + fn multiple_sections() { + let cmap = parse_with_preamble( + br#" +2 begincidrange +<0000> <00FF> 0 +<0100> <01FF> 256 +endcidrange +1 begincidchar +<0200> 600 +endcidchar +1 begincidrange +<8140> <817E> 633 +endcidrange +"#, + ); + + assert_eq!(cmap.lookup_cid_code(0x0000, 2), Some(0)); + assert_eq!(cmap.lookup_cid_code(0x0100, 2), Some(256)); + assert_eq!(cmap.lookup_cid_code(0x0200, 2), Some(600)); + assert_eq!(cmap.lookup_cid_code(0x8140, 2), Some(633)); + } + + #[test] + fn single_byte_codes() { + let cmap = parse_with_preamble( + br#" +2 begincidrange +<00> <7F> 0 +<80> 200 +endcidrange +"#, + ); + + assert_eq!(cmap.lookup_cid_code(0x00, 1), Some(0)); + assert_eq!(cmap.lookup_cid_code(0x41, 1), Some(0x41)); + assert_eq!(cmap.lookup_cid_code(0x80, 1), Some(200)); + assert_eq!(cmap.lookup_cid_code(0xFF, 1), Some(200 + 127)); + } + + #[test] + fn dict_style_cidsysteminfo() { + let data = br#" +/CIDInit /ProcSet findresource begin +10 dict begin +begincmap +/CIDSystemInfo +<< /Registry (Adobe) +/Ordering (UCS) +/Supplement 0 +>> def +/CMapName /Adobe-Identity-UCS def +/CMapType 2 def +1 begincodespacerange +<0000> +endcodespacerange +2 beginbfchar +<001F> +<002A> +endbfchar +endcmap +CMapName currentdict /CMap defineresource pop +end +end +"#; + let cmap = CMap::parse(data, |_| None).unwrap(); + let cc = cmap.metadata().character_collection.as_ref().unwrap(); + assert_eq!( + cc.family, + CidFamily::Custom { + registry: b"Adobe".to_vec(), + ordering: b"UCS".to_vec(), + } + ); + assert_eq!(cc.supplement, 0); + assert_eq!( + cmap.metadata().name.as_deref(), + Some(b"Adobe-Identity-UCS".as_slice()) + ); + assert_eq!( + cmap.lookup_bf_string(0x001F), + Some(BfString::Char('\u{F049}')) + ); + assert_eq!( + cmap.lookup_bf_string(0x002A), + Some(BfString::Char('\u{F055}')) + ); + } + + #[test] + fn usecmap_chaining() { + let base_data = br#" +/CIDSystemInfo 3 dict dup begin + /Registry (Adobe) def + /Ordering (Japan1) def + /Supplement 0 def +end def +/CMapName /Base def +/WMode 0 def +1 begincodespacerange +<0000> +endcodespacerange +1 begincidrange +<0000> <00FF> 0 +endcidrange +"#; + + let child_data = br#" +/Base usecmap +/CIDSystemInfo 3 dict dup begin + /Registry (Adobe) def + /Ordering (Japan1) def + /Supplement 0 def +end def +/CMapName /Child def +/WMode 0 def +1 begincodespacerange +<0000> +endcodespacerange +1 begincidrange +<0100> <01FF> 256 +endcidrange +"#; + + let cmap = CMap::parse(child_data, |name| { + if name.to_bytes() == b"Base" { + Some(base_data.as_slice()) + } else { + None + } + }) + .unwrap(); + + assert_eq!(cmap.lookup_cid_code(0x0100, 2), Some(256)); + assert_eq!(cmap.lookup_cid_code(0x01FF, 2), Some(511)); + assert_eq!(cmap.lookup_cid_code(0x0000, 2), Some(0)); + assert_eq!(cmap.lookup_cid_code(0x00FF, 2), Some(0xFF)); + + assert_eq!(cmap.lookup_cid_code(0x0200, 2), None); + } + + #[test] + fn usecmap_partial_override() { + let base_data = br#" +/CIDSystemInfo 3 dict dup begin + /Registry (Adobe) def + /Ordering (Japan1) def + /Supplement 0 def +end def +/CMapName /Base def +/WMode 0 def +1 begincodespacerange +<0000> +endcodespacerange +1 begincidrange +<0000> <00FF> 0 +endcidrange +"#; + + let child_data = br#" +/Base usecmap +/CIDSystemInfo 3 dict dup begin + /Registry (Adobe) def + /Ordering (Japan1) def + /Supplement 0 def +end def +/CMapName /Child def +/WMode 0 def +1 begincodespacerange +<0000> +endcodespacerange +1 begincidrange +<0040> <007F> 500 +endcidrange +"#; + + let cmap = CMap::parse(child_data, |name| { + if name.to_bytes() == b"Base" { + Some(base_data.as_slice()) + } else { + None + } + }) + .unwrap(); + + assert_eq!(cmap.lookup_cid_code(0x0000, 2), Some(0)); + assert_eq!(cmap.lookup_cid_code(0x003F, 2), Some(0x3F)); + assert_eq!(cmap.lookup_cid_code(0x0040, 2), Some(500)); + assert_eq!(cmap.lookup_cid_code(0x007F, 2), Some(563)); + assert_eq!(cmap.lookup_cid_code(0x0080, 2), Some(0x80)); + assert_eq!(cmap.lookup_cid_code(0x00FF, 2), Some(0xFF)); + } + + #[test] + fn notdef_char_lookup() { + let cmap = parse_with_preamble( + br#" +2 beginnotdefchar +<03> 10 +<20> 20 +endnotdefchar +"#, + ); + + assert_eq!(cmap.lookup_cid_code(0x03, 1), Some(10)); + assert_eq!(cmap.lookup_cid_code(0x20, 1), Some(20)); + assert_eq!(cmap.lookup_cid_code(0x04, 1), None); + } + + #[test] + fn notdef_range_lookup() { + let cmap = parse_with_preamble( + br#" +1 beginnotdefrange +<0000> <001F> 100 +endnotdefrange +"#, + ); + + assert_eq!(cmap.lookup_cid_code(0x0000, 2), Some(100)); + assert_eq!(cmap.lookup_cid_code(0x0001, 2), Some(100)); + assert_eq!(cmap.lookup_cid_code(0x001F, 2), Some(100)); + assert_eq!(cmap.lookup_cid_code(0x0020, 2), None); + } + + #[test] + fn bfchar_lookup() { + let cmap = parse_with_preamble( + br#" +2 beginbfchar +<0041> <0048> +<0042> <0065> +endbfchar +"#, + ); + + assert_eq!(cmap.lookup_bf_string(0x0041), Some(BfString::Char('H'))); + assert_eq!(cmap.lookup_bf_string(0x0042), Some(BfString::Char('e'))); + assert_eq!(cmap.lookup_bf_string(0x0043), None); + } + + #[test] + fn bfchar_ligature() { + let cmap = parse_with_preamble( + br#" +1 beginbfchar +<005F> <00660066> +endbfchar +"#, + ); + + assert_eq!( + cmap.lookup_bf_string(0x005F), + Some(BfString::String(String::from("ff"))) + ); + } + + // PdfCraft patch: malformed pieces of a CMap are skipped, not fatal - the reading + // path used to discard the whole CMap and fall back to the font encoding for every + // code, so search and search-redaction missed text the edit layer could see (#3177). + + #[test] + fn bad_bfchar_entry_is_skipped() { + let cmap = parse_with_preamble( + br#" +2 beginbfchar +<41> +<42> <0059> +endbfchar +"#, + ); + assert_eq!(cmap.lookup_bf_string(0x41), None); + assert_eq!( + cmap.lookup_bf_string(0x42), + Some(BfString::Char('Y')) + ); + } + + #[test] + fn unresolvable_usecmap_is_skipped() { + let cmap = parse_with_preamble( + br#" +/Bogus usecmap +1 beginbfchar +<41> <0058> +endbfchar +"#, + ); + assert_eq!( + cmap.lookup_bf_string(0x41), + Some(BfString::Char('X')) + ); + } + + #[test] + fn bfchar_surrogate_pair() { + let cmap = parse_with_preamble( + br#" +1 beginbfchar +<3A51> +endbfchar +"#, + ); + + assert_eq!( + cmap.lookup_bf_string(0x3A51), + Some(BfString::Char('\u{2003E}')) + ); + } + + #[test] + fn bfrange_incrementing() { + let cmap = parse_with_preamble( + br#" +1 beginbfrange +<0000> <0004> <0041> +endbfrange +"#, + ); + + assert_eq!(cmap.lookup_bf_string(0x0000), Some(BfString::Char('A'))); + assert_eq!(cmap.lookup_bf_string(0x0001), Some(BfString::Char('B'))); + assert_eq!(cmap.lookup_bf_string(0x0004), Some(BfString::Char('E'))); + assert_eq!(cmap.lookup_bf_string(0x0005), None); + } + + #[test] + fn bfrange_array() { + let cmap = parse_with_preamble( + br#" +1 beginbfrange +<005F> <0061> [<00660066> <00660069> <0066006C>] +endbfrange +"#, + ); + + // ff, fi, fl ligatures + assert_eq!( + cmap.lookup_bf_string(0x005F), + Some(BfString::String(String::from("ff"))) + ); + assert_eq!( + cmap.lookup_bf_string(0x0060), + Some(BfString::String(String::from("fi"))) + ); + assert_eq!( + cmap.lookup_bf_string(0x0061), + Some(BfString::String(String::from("fl"))) + ); + } + + #[test] + fn unicode_lookup_miss() { + let cmap = parse_with_preamble( + br#" +1 beginbfchar +<0041> <0048> +endbfchar +"#, + ); + + assert_eq!(cmap.lookup_bf_string(0x0000), None); + assert_eq!(cmap.lookup_bf_string(0x0042), None); + } + + #[test] + fn identity_h() { + let cmap = CMap::identity_h(); + assert_eq!( + cmap.metadata().name.as_deref(), + Some(b"Identity-H".as_slice()) + ); + assert_eq!(cmap.metadata().writing_mode, Some(WritingMode::Horizontal)); + + assert_eq!(cmap.lookup_cid_code(0x0041, 2), Some(0x0041)); + assert_eq!(cmap.lookup_cid_code(0x1234, 2), Some(0x1234)); + assert_eq!(cmap.lookup_cid_code(0xFFFF, 2), Some(0xFFFF)); + + assert_eq!(cmap.lookup_cid_code(0x0041, 1), None); + assert_eq!(cmap.lookup_cid_code(0x0041, 3), None); + } + + #[test] + fn identity_v() { + let cmap = CMap::identity_v(); + assert_eq!( + cmap.metadata().name.as_deref(), + Some(b"Identity-V".as_slice()) + ); + assert_eq!(cmap.metadata().writing_mode, Some(WritingMode::Vertical)); + + assert_eq!(cmap.lookup_cid_code(0x0041, 2), Some(0x0041)); + assert_eq!(cmap.lookup_cid_code(0xFFFF, 2), Some(0xFFFF)); + } + + #[test] + fn minimal_cmap_no_name_no_wmode() { + // Extracted from corpus PDF 0500013. + let data = br#" +/CIDInit /ProcSet findresource begin +12 dict begin +begincmap +/CMapType 2 def +1 begincodespacerange +<00> +endcodespacerange +1 beginbfrange +<01><01><0020> +endbfrange +endcmap +"#; + let cmap = CMap::parse(data, |_| None).unwrap(); + assert_eq!(cmap.metadata().name, None); + assert_eq!(cmap.metadata().character_collection, None); + assert_eq!(cmap.metadata().writing_mode, None); + assert_eq!(cmap.lookup_bf_string(0x01), Some(BfString::Char(' '))); + } + + #[test] + fn registry_as_name() { + // Extracted from corpus PDF 0875241, uses names instead of strings for + // strings for metadata. + let data = br#" +/CIDSystemInfo +<< /Registry /ABCDEF+SimSun +/Ordering (pdfbeaninc) +/Supplement 0 +>> def +/CMapName /ABCDEF+SimSun def +/CMapType 2 def +1 begincodespacerange +<0000> +endcodespacerange +1 beginbfrange +<0000> <0000> <6881> +endbfrange +"#; + let cmap = CMap::parse(data, |_| None).unwrap(); + let cc = cmap.metadata().character_collection.as_ref().unwrap(); + assert_eq!( + cc.family, + CidFamily::Custom { + registry: b"ABCDEF+SimSun".to_vec(), + ordering: b"pdfbeaninc".to_vec(), + } + ); + assert_eq!(cc.supplement, 0); + assert_eq!( + cmap.lookup_bf_string(0x0000), + Some(BfString::Char('\u{6881}')) + ); + } + + #[test] + fn single_byte_bfrange_destination() { + // See for example corpus test case 0625248. + let cmap = parse_with_preamble( + br#" +1 beginbfrange +<00> <7f> <00> +endbfrange +"#, + ); + + assert_eq!( + cmap.lookup_bf_string(0x00), + Some(BfString::Char('\u{0000}')) + ); + assert_eq!(cmap.lookup_bf_string(0x41), Some(BfString::Char('A'))); + assert_eq!( + cmap.lookup_bf_string(0x7f), + Some(BfString::Char('\u{007F}')) + ); + } +} + +#[cfg(all(test, feature = "embed-cmaps"))] +mod bcmap_tests { + use super::*; + + fn get_embedded_cmap(name: CMapName<'_>) -> Option<&'static [u8]> { + load_embedded(name) + } + + #[test] + fn embedded_h_cmap() { + let data = load_embedded(CMapName::H).expect("embedded H cmap not found"); + let cmap = CMap::parse(data, get_embedded_cmap).expect("failed to parse H cmap"); + assert_eq!(cmap.metadata().writing_mode, None); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeJapan1, + supplement: 1, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x2121, 2), Some(633)); + assert_eq!(cmap.lookup_cid_code(0x2221, 2), Some(727)); + } + + #[test] + fn embedded_90ms_rksj_h() { + let data = + load_embedded(CMapName::N90msRksjH).expect("embedded 90ms-RKSJ-H cmap not found"); + let cmap = CMap::parse(data, get_embedded_cmap).expect("failed to parse 90ms-RKSJ-H cmap"); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeJapan1, + supplement: 2, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x20, 1), Some(231)); + assert_eq!(cmap.lookup_cid_code(0x8140, 2), Some(633)); + } + + #[test] + fn embedded_v_cmap() { + let data = load_embedded(CMapName::V).expect("embedded V cmap not found"); + let cmap = CMap::parse(data, get_embedded_cmap).expect("failed to parse V cmap"); + assert_eq!(cmap.metadata().writing_mode, Some(WritingMode::Vertical)); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeJapan1, + supplement: 1, + }) + ); + // Inherited from base "H" via usecmap. + assert_eq!(cmap.lookup_cid_code(0x2121, 2), Some(633)); + } + + #[test] + fn embedded_gbk_euc_h() { + let data = load_embedded(CMapName::GbkEucH).expect("embedded GBK-EUC-H not found"); + let cmap = CMap::parse(data, get_embedded_cmap).expect("failed to parse GBK-EUC-H"); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeGB1, + supplement: 2, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x20, 1), Some(7716)); + assert_eq!(cmap.lookup_cid_code(0x21, 1), Some(814)); + assert_eq!(cmap.lookup_cid_code(0x7E, 1), Some(814 + 0x7E - 0x21)); + assert_eq!(cmap.lookup_cid_code(0xFE80, 2), Some(22094)); + assert_eq!(cmap.lookup_cid_code(0xFEA0, 2), Some(22094 + 0xA0 - 0x80)); + } + + #[test] + fn embedded_ksc_euc_h() { + let data = load_embedded(CMapName::KscEucH).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeKorea1, + supplement: 0, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x20, 1), Some(8094)); + assert_eq!(cmap.lookup_cid_code(0x41, 1), Some(8094 + 0x41 - 0x20)); + assert_eq!(cmap.lookup_cid_code(0xA1A1, 2), Some(101)); + assert_eq!(cmap.lookup_cid_code(0xA1FE, 2), Some(101 + 0xFE - 0xA1)); + assert_eq!(cmap.lookup_cid_code(0xFDA1, 2), Some(7962)); + } + + #[test] + fn embedded_b5pc_h() { + let data = load_embedded(CMapName::B5pcH).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeCNS1, + supplement: 0, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x20, 1), Some(1)); + assert_eq!(cmap.lookup_cid_code(0x41, 1), Some(1 + 0x41 - 0x20)); + assert_eq!(cmap.lookup_cid_code(0x80, 1), Some(61)); + assert_eq!(cmap.lookup_cid_code(0xD140, 2), Some(7251)); + assert_eq!(cmap.lookup_cid_code(0xD17E, 2), Some(7251 + 0x7E - 0x40)); + assert_eq!(cmap.lookup_cid_code(0xF9D5, 2), Some(13642 + 3)); + } + + #[test] + fn embedded_unijis_utf16_h() { + let data = load_embedded(CMapName::UniJisUtf16H).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeJapan1, + supplement: 7, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x20, 2), Some(1)); + assert_eq!(cmap.lookup_cid_code(0x41, 2), Some(1 + 0x41 - 0x20)); + assert_eq!(cmap.lookup_cid_code(0x5C, 2), Some(97)); + assert_eq!(cmap.lookup_cid_code(0x5BCC, 2), Some(3531)); + assert_eq!(cmap.lookup_cid_code(0xD884DF50, 4), Some(19130)); + } + + #[test] + fn embedded_identity_h() { + let data = load_embedded(CMapName::IdentityH).expect("embedded Identity-H not found"); + let cmap = CMap::parse(data, get_embedded_cmap).expect("failed to parse Identity-H cmap"); + assert_eq!(cmap.metadata().writing_mode, None); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeIdentity, + supplement: 0, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x0000, 2), Some(0)); + assert_eq!(cmap.lookup_cid_code(0x0041, 2), Some(0x41)); + assert_eq!(cmap.lookup_cid_code(0xFFFF, 2), Some(0xFFFF)); + } + + #[test] + fn embedded_identity_v() { + let data = load_embedded(CMapName::IdentityV).expect("embedded Identity-V not found"); + let cmap = CMap::parse(data, get_embedded_cmap).expect("failed to parse Identity-V cmap"); + assert_eq!(cmap.metadata().writing_mode, Some(WritingMode::Vertical)); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeIdentity, + supplement: 0, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x0000, 2), Some(0)); + assert_eq!(cmap.lookup_cid_code(0x0041, 2), Some(0x41)); + assert_eq!(cmap.lookup_cid_code(0xFFFF, 2), Some(0xFFFF)); + } + + // Note: Some tests AI-generated, haven't double-checked them. + + #[test] + fn embedded_adobe_gb1_ucs2() { + let data = load_embedded(CMapName::AdobeGb1Ucs2).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::Custom { + registry: b"Adobe".to_vec(), + ordering: b"Adobe_GB1_UCS2".to_vec(), + }, + supplement: 5, + }) + ); + + assert_eq!( + cmap.lookup_bf_string(0x0063), + Some(BfString::Char('\u{00B7}')) + ); + assert_eq!(cmap.lookup_bf_string(0x0001), Some(BfString::Char(' '))); + assert_eq!(cmap.lookup_bf_string(0x0022), Some(BfString::Char('A'))); + assert_eq!(cmap.lookup_bf_string(0x005F), Some(BfString::Char('~'))); + assert_eq!( + cmap.lookup_bf_string(0x560F), + Some(BfString::String(String::from("\u{90CE}\u{FE00}"))) + ); + } + + #[test] + fn embedded_adobe_japan1_ucs2() { + let data = load_embedded(CMapName::AdobeJapan1Ucs2).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::Custom { + registry: b"Adobe".to_vec(), + ordering: b"Adobe_Japan1_UCS2".to_vec(), + }, + supplement: 7, + }) + ); + + assert_eq!( + cmap.lookup_bf_string(0x003D), + Some(BfString::Char('\u{00A5}')) + ); + assert_eq!(cmap.lookup_bf_string(0x0001), Some(BfString::Char(' '))); + assert_eq!(cmap.lookup_bf_string(0x003C), Some(BfString::Char('['))); + assert_eq!( + cmap.lookup_bf_string(0x0476), + Some(BfString::String(String::from("\u{82A6}\u{E0100}"))) + ); + assert_eq!( + cmap.lookup_bf_string(0x265B), + Some(BfString::String(String::from("10/11"))) + ); + assert_eq!( + cmap.lookup_bf_string(0x2E6B), + Some(BfString::String(String::from("オングストローム"))) + ); + } + + #[test] + fn embedded_adobe_korea1_ucs2() { + let data = load_embedded(CMapName::AdobeKorea1Ucs2).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::Custom { + registry: b"Adobe".to_vec(), + ordering: b"Adobe_Korea1_UCS2".to_vec(), + }, + supplement: 2, + }) + ); + + assert_eq!( + cmap.lookup_bf_string(0x0060), + Some(BfString::Char('\u{20A9}')) + ); + assert_eq!( + cmap.lookup_bf_string(0x2072), + Some(BfString::String(String::from("[10]"))) + ); + assert_eq!( + cmap.lookup_bf_string(0x2073), + Some(BfString::String(String::from("[11]"))) + ); + } + + #[test] + fn embedded_adobe_cns1_ucs2() { + let data = load_embedded(CMapName::AdobeCns1Ucs2).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::Custom { + registry: b"Adobe".to_vec(), + ordering: b"Adobe_CNS1_UCS2".to_vec(), + }, + supplement: 7, + }) + ); + + assert_eq!( + cmap.lookup_bf_string(0x0060), + Some(BfString::Char('\u{00A9}')) + ); + assert_eq!(cmap.lookup_bf_string(0x0001), Some(BfString::Char(' '))); + assert_eq!(cmap.lookup_bf_string(0x005F), Some(BfString::Char('~'))); + assert_eq!( + cmap.lookup_bf_string(0x36B0), + Some(BfString::Char('\u{200CC}')) + ); + } + + #[test] + fn embedded_eten_b5_h() { + let data = load_embedded(CMapName::ETenB5H).unwrap(); + let cmap = CMap::parse(data, get_embedded_cmap).unwrap(); + assert_eq!( + cmap.metadata().character_collection, + Some(CharacterCollection { + family: CidFamily::AdobeCNS1, + supplement: 0, + }) + ); + assert_eq!(cmap.lookup_cid_code(0x20, 1), Some(13648)); + assert_eq!(cmap.lookup_cid_code(0x7E, 1), Some(13648 + 0x7E - 0x20)); + assert_eq!(cmap.lookup_cid_code(0xA140, 2), Some(99)); + assert_eq!(cmap.lookup_cid_code(0xA158, 2), Some(99 + 0x58 - 0x40)); + assert_eq!(cmap.lookup_cid_code(0xD040, 2), Some(7094)); + assert_eq!(cmap.lookup_cid_code(0xF9FE, 2), Some(14056 + 0xFE - 0xD6)); + } +} diff --git a/vendor/hayro-cmap/src/parse.rs b/vendor/hayro-cmap/src/parse.rs new file mode 100644 index 000000000..96eda4d50 --- /dev/null +++ b/vendor/hayro-cmap/src/parse.rs @@ -0,0 +1,409 @@ +use alloc::boxed::Box; +use alloc::vec::Vec; + +use hayro_postscript::{Object, Scanner}; + +#[cfg(feature = "embed-cmaps")] +use crate::bcmap; +use crate::{ + BfRange, CMap, CMapName, CharacterCollection, CidFamily, CidRange, CodespaceRange, + MAX_NESTING_DEPTH, Metadata, PartitionedRanges, Range, WritingMode, +}; + +struct Context { + buf: Vec, + get_cmap: F, +} + +pub(crate) fn parse_inner<'a>( + data: &[u8], + get_cmap: impl Fn(CMapName<'_>) -> Option<&'a [u8]> + Clone + 'a, + depth: u32, +) -> Option { + // Prevent stack overflow for malicious cmap files or circular references. + if depth >= MAX_NESTING_DEPTH { + return None; + } + + // Check if it's in our custom embedded cmap format. + #[cfg(feature = "embed-cmaps")] + if data.starts_with(b"bcmap") { + return bcmap::parse(data, get_cmap, depth); + } + + let mut scanner = Scanner::new(data); + let mut ctx = Context { + buf: Vec::new(), + get_cmap, + }; + let mut _codespace_ranges = Vec::new(); + let mut ranges = PartitionedRanges::new(); + let mut notdef_ranges = PartitionedRanges::new(); + let mut bf_entries = Vec::new(); + let mut base = None; + + let mut registry = None; + let mut ordering = None; + let mut supplement = None; + let mut cmap_name = None; + let mut writing_mode = None; + let mut last_name: Option> = None; + + // PdfCraft patch: track whether the scan completed. A CMap with an unparseable + // tail keeps the entries read so far instead of being discarded wholesale. + let mut completed = false; + while !scanner.at_end() { + let Ok(obj) = scanner.parse_object() else { break }; + + let Object::Name(name) = &obj else { continue }; + + if name.is_literal() { + match name.as_str() { + // Strictly speaking, Registry and Ordering should be strings, + // but some PDF generators emit names instead, so be lenient + // and try both. + Some("Registry") => { + registry = parse_string_or_name(&mut scanner); + } + Some("Ordering") => { + ordering = parse_string_or_name(&mut scanner); + } + Some("Supplement") => { + supplement = scanner.parse_number().ok().map(|n| n.as_i32()); + } + Some("CMapName") => { + cmap_name = scanner.parse_name().ok().and_then(|n| n.decode().ok()); + } + Some("WMode") => { + writing_mode = parse_writing_mode(&mut scanner); + } + _ => { + last_name = name.decode().ok(); + } + } + } else { + match name.as_str() { + Some("begincodespacerange") => { + parse_codespace_range(&mut scanner, &mut _codespace_ranges, &mut ctx)?; + } + Some("begincidrange") => { + parse_range(&mut scanner, &mut ranges, &mut ctx, "endcidrange")?; + } + Some("begincidchar") => { + parse_char(&mut scanner, &mut ranges, &mut ctx, "endcidchar")?; + } + Some("beginnotdefrange") => { + parse_range(&mut scanner, &mut notdef_ranges, &mut ctx, "endnotdefrange")?; + } + Some("beginnotdefchar") => { + parse_char(&mut scanner, &mut notdef_ranges, &mut ctx, "endnotdefchar")?; + } + Some("beginbfchar") => { + parse_bf_char(&mut scanner, &mut bf_entries, &mut ctx)?; + } + Some("beginbfrange") => { + parse_bf_range(&mut scanner, &mut bf_entries, &mut ctx)?; + } + Some("usecmap") => { + // PdfCraft patch: an unresolvable or unparseable `usecmap` + // reference drops only the base, not the whole CMap. + if let Some(name) = last_name.as_deref() { + if let Some(nested_data) = (ctx.get_cmap)(CMapName::from_bytes(name)) { + base = parse_inner(nested_data, ctx.get_cmap.clone(), depth + 1).map(Box::new); + } + } + } + _ => {} + } + } + } + + completed = true; + + // Since we will use binary search for finding the correct entry, sort now. + ranges.sort(); + notdef_ranges.sort(); + bf_entries.sort_by(|a, b| a.range.start.cmp(&b.range.start)); + + // See PDFJS-3323, which has an invalid CIDSystemInfo entry. + // We ignore it if it's invalid. + let character_collection = if let (Some(registry), Some(ordering), Some(supplement)) = + (registry, ordering, supplement) + { + Some(CharacterCollection { + family: CidFamily::from_registry_ordering(®istry, &ordering), + supplement, + }) + } else { + None + }; + + let metadata = Metadata { + character_collection, + name: cmap_name, + writing_mode, + }; + + // PdfCraft patch: a truncated CMap that yielded nothing still reports failure, + // so callers fall back to the font's own encoding exactly as before. + if !completed + && ranges.is_empty() + && notdef_ranges.is_empty() + && bf_entries.is_empty() + && base.is_none() + && metadata.character_collection.is_none() + { + return None; + } + + Some(CMap { + metadata, + _codespace_ranges, + cid_ranges: ranges, + notdef_ranges, + bf_entries, + base, + }) +} + +fn parse_writing_mode(scanner: &mut Scanner<'_>) -> Option { + match scanner.parse_number().ok()?.as_i32() { + 0 => Some(WritingMode::Horizontal), + 1 => Some(WritingMode::Vertical), + _ => None, + } +} + +fn parse_string_or_name(scanner: &mut Scanner<'_>) -> Option> { + match scanner.parse_object().ok()? { + Object::String(s) => s.decode().ok(), + Object::Name(n) => n.decode().ok(), + _ => None, + } +} + +fn parse_codespace_range( + scanner: &mut Scanner<'_>, + ranges: &mut Vec, + ctx: &mut Context, +) -> Option<()> { + loop { + let obj = scanner.parse_object().ok()?; + + if name_matches(&obj, "endcodespacerange") { + return Some(()); + } + + // PdfCraft patch: a malformed codespace entry is skipped, not fatal. + let Some(low) = extract_u32_code(&obj, &mut ctx.buf) else { continue }; + let Ok(n_bytes) = u8::try_from(ctx.buf.len()) else { continue }; + let Some(high) = read_u32_code(scanner, &mut ctx.buf) else { continue }; + + if ctx.buf.len() != usize::from(n_bytes) { + continue; + } + + ranges.push(CodespaceRange { + number_bytes: n_bytes, + low, + high, + }); + } +} + +fn parse_range( + scanner: &mut Scanner<'_>, + ranges: &mut PartitionedRanges, + ctx: &mut Context, + end_marker: &str, +) -> Option<()> { + loop { + let obj = scanner.parse_object().ok()?; + + if name_matches(&obj, end_marker) { + return Some(()); + } + + // PdfCraft patch: a malformed range entry is skipped, not fatal. + let Some(start) = extract_u32_code(&obj, &mut ctx.buf) else { continue }; + let byte_len = ctx.buf.len(); + let Some(end) = read_u32_code(scanner, &mut ctx.buf) else { continue }; + let Some(cid_start) = scanner.parse_number().ok().and_then(|n| u32::try_from(n.as_i32()).ok()) else { continue }; + + ranges.push( + byte_len, + CidRange { + range: Range { start, end }, + cid_start, + }, + ); + } +} + +fn parse_char( + scanner: &mut Scanner<'_>, + ranges: &mut PartitionedRanges, + ctx: &mut Context, + end_marker: &str, +) -> Option<()> { + loop { + let obj = scanner.parse_object().ok()?; + + if name_matches(&obj, end_marker) { + return Some(()); + } + + // PdfCraft patch: a malformed char entry is skipped, not fatal. + let Some(code) = extract_u32_code(&obj, &mut ctx.buf) else { continue }; + let byte_len = ctx.buf.len(); + let Some(cid_start) = scanner.parse_number().ok().and_then(|n| u32::try_from(n.as_i32()).ok()) else { continue }; + + ranges.push( + byte_len, + CidRange { + range: Range { + start: code, + end: code, + }, + cid_start, + }, + ); + } +} + +fn parse_bf_char( + scanner: &mut Scanner<'_>, + entries: &mut Vec, + ctx: &mut Context, +) -> Option<()> { + loop { + let obj = scanner.parse_object().ok()?; + + if name_matches(&obj, "endbfchar") { + return Some(()); + } + + // PdfCraft patch: a malformed bfchar entry (e.g. a non-hex destination) + // is skipped, not fatal - the entries after it still apply. + let Some(code) = extract_u32_code(&obj, &mut ctx.buf) else { continue }; + let Ok(dst) = scanner.parse_string() else { continue }; + if dst.decode_into(&mut ctx.buf).is_err() { continue } + let Some(dst_base) = decode_be(&ctx.buf) else { continue }; + + entries.push(BfRange { + range: Range { + start: code, + end: code, + }, + dst_base, + }); + } +} + +fn parse_bf_range( + scanner: &mut Scanner<'_>, + entries: &mut Vec, + ctx: &mut Context, +) -> Option<()> { + loop { + let obj = scanner.parse_object().ok()?; + + if name_matches(&obj, "endbfrange") { + return Some(()); + } + + // PdfCraft patch: a malformed bfrange entry is skipped, not fatal. + let Some(start) = extract_u32_code(&obj, &mut ctx.buf) else { continue }; + let Some(end) = read_u32_code(scanner, &mut ctx.buf) else { continue }; + + let Ok(next) = scanner.parse_object() else { continue }; + + match &next { + Object::String(s) => { + if s.decode_into(&mut ctx.buf).is_err() { continue } + let Some(dst_base) = decode_be(&ctx.buf) else { continue }; + + entries.push(BfRange { + range: Range { start, end }, + dst_base, + }); + } + Object::Array(array) => { + let mut array_scanner = array.objects(); + + for code in start..=end { + // A bad array member drops only that member. + let Ok(s) = array_scanner.parse_string() else { continue }; + if s.decode_into(&mut ctx.buf).is_err() { continue } + let Some(dst_base) = decode_be(&ctx.buf) else { continue }; + + entries.push(BfRange { + range: Range { + start: code, + end: code, + }, + dst_base, + }); + } + } + _ => continue, + } + } +} + +/// Convert the buffer into native-endian u16, so that we can use `String::from_utf16`. +fn decode_be(bytes: &[u8]) -> Option> { + if bytes.is_empty() { + return None; + } + + let mut out = Vec::with_capacity(bytes.len().div_ceil(2)); + let mut i = 0; + + // My understanding is that the bf strings should always be UTF16-BE encoded, + // but it seems like some PDFs only have a single byte there? I guess we just + // pad them? Wasn't able to find anything specific in the specifications. + if !bytes.len().is_multiple_of(2) { + out.push(u16::from(bytes[0])); + i = 1; + } + + while i < bytes.len() { + out.push(u16::from_be_bytes([bytes[i], bytes[i + 1]])); + i += 2; + } + + Some(out) +} + +#[inline] +fn read_u32_code(scanner: &mut Scanner<'_>, buf: &mut Vec) -> Option { + let s = scanner.parse_string().ok()?; + s.decode_into(buf).ok()?; + bytes_to_u32(buf) +} + +#[inline] +fn extract_u32_code(obj: &Object<'_>, buf: &mut Vec) -> Option { + let Object::String(s) = obj else { return None }; + s.decode_into(buf).ok()?; + bytes_to_u32(buf) +} + +#[inline] +fn bytes_to_u32(bytes: &[u8]) -> Option { + if bytes.is_empty() || bytes.len() > 4 { + return None; + } + + let mut val = 0_u32; + for &b in bytes { + val = (val << 8) | b as u32; + } + + Some(val) +} + +fn name_matches(obj: &Object<'_>, expected: &str) -> bool { + matches!(obj, Object::Name(name) if !name.is_literal() && name.as_str() == Some(expected)) +} From e70450420410fc68dd03a196b7718db8e02217b1 Mon Sep 17 00:00:00 2001 From: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> Date: Sun, 18 Oct 2026 13:21:57 +0530 Subject: [PATCH 03/34] hayro-cmap patch: fix guard, entry alignment and array drain - The truncated-stream guard was dead: completed was set on the break path too. Detect truncation from the scanner position so a stream that yields nothing still reports failure and callers fall back to the font encoding. - A skipped bfchar/bfrange entry consumed a different number of objects than a parsed one, shifting every later pair onto the wrong code. Skip the entry's remaining objects as well. - A bfrange array that runs out or breaks cannot be resynced to a member boundary; stop the loop instead of spinning over the rest of the range. Signed-off-by: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> --- vendor/hayro-cmap/src/lib.rs | 39 ++++++++++++++++++++++++++++++ vendor/hayro-cmap/src/parse.rs | 43 +++++++++++++++++++++++++--------- 2 files changed, 71 insertions(+), 11 deletions(-) diff --git a/vendor/hayro-cmap/src/lib.rs b/vendor/hayro-cmap/src/lib.rs index aed021eee..c45cc0ae5 100644 --- a/vendor/hayro-cmap/src/lib.rs +++ b/vendor/hayro-cmap/src/lib.rs @@ -1117,6 +1117,45 @@ endbfchar ); } + #[test] + fn truncated_stream_with_no_entries_reports_failure() { + // The tolerant parser keeps collected entries, but a stream that + // breaks before any entry must still report failure so callers + // fall back to the font's own encoding. + assert!(CMap::parse(b"<00", |_| None).is_none()); + } + + #[test] + fn bad_bfchar_code_keeps_later_entries_aligned() { + // The malformed 5-byte code still has a destination object; both + // must be skipped or every later pair shifts by one object. + let cmap = parse_with_preamble( + br#" +2 beginbfchar +<0102030405> <0041> +<42> <0059> +endbfchar +"#, + ); + assert_eq!(cmap.lookup_bf_string(0x41), None); + assert_eq!(cmap.lookup_bf_string(0x42), Some(BfString::Char('Y'))); + } + + #[test] + fn bfrange_array_stops_at_exhaustion() { + // One member for 256 codes: the array cannot be resynced, so the + // loop must stop rather than spin over the remaining codes. + let cmap = parse_with_preamble( + br#" +1 beginbfrange +<0000> <00FF> [<0041>] +endbfrange +"#, + ); + assert_eq!(cmap.lookup_bf_string(0x00), Some(BfString::Char('A'))); + assert_eq!(cmap.lookup_bf_string(0x01), None); + } + #[test] fn unresolvable_usecmap_is_skipped() { let cmap = parse_with_preamble( diff --git a/vendor/hayro-cmap/src/parse.rs b/vendor/hayro-cmap/src/parse.rs index 96eda4d50..9450bbbe2 100644 --- a/vendor/hayro-cmap/src/parse.rs +++ b/vendor/hayro-cmap/src/parse.rs @@ -49,11 +49,15 @@ pub(crate) fn parse_inner<'a>( let mut writing_mode = None; let mut last_name: Option> = None; - // PdfCraft patch: track whether the scan completed. A CMap with an unparseable - // tail keeps the entries read so far instead of being discarded wholesale. - let mut completed = false; + // PdfCraft patch: track whether the scan was cut short. A CMap with an + // unparseable tail keeps the entries read so far instead of being + // discarded wholesale; one that yielded nothing still reports failure. + let mut truncated = false; while !scanner.at_end() { - let Ok(obj) = scanner.parse_object() else { break }; + let Ok(obj) = scanner.parse_object() else { + truncated = true; + break; + }; let Object::Name(name) = &obj else { continue }; @@ -118,7 +122,6 @@ pub(crate) fn parse_inner<'a>( } } - completed = true; // Since we will use binary search for finding the correct entry, sort now. ranges.sort(); @@ -146,7 +149,7 @@ pub(crate) fn parse_inner<'a>( // PdfCraft patch: a truncated CMap that yielded nothing still reports failure, // so callers fall back to the font's own encoding exactly as before. - if !completed + if truncated && ranges.is_empty() && notdef_ranges.is_empty() && bf_entries.is_empty() @@ -285,7 +288,12 @@ fn parse_bf_char( // PdfCraft patch: a malformed bfchar entry (e.g. a non-hex destination) // is skipped, not fatal - the entries after it still apply. - let Some(code) = extract_u32_code(&obj, &mut ctx.buf) else { continue }; + let Some(code) = extract_u32_code(&obj, &mut ctx.buf) else { + // The malformed code is still followed by its destination + // object; skip it so later entries stay aligned. + let _ = scanner.parse_object(); + continue; + }; let Ok(dst) = scanner.parse_string() else { continue }; if dst.decode_into(&mut ctx.buf).is_err() { continue } let Some(dst_base) = decode_be(&ctx.buf) else { continue }; @@ -313,8 +321,18 @@ fn parse_bf_range( } // PdfCraft patch: a malformed bfrange entry is skipped, not fatal. - let Some(start) = extract_u32_code(&obj, &mut ctx.buf) else { continue }; - let Some(end) = read_u32_code(scanner, &mut ctx.buf) else { continue }; + let Some(start) = extract_u32_code(&obj, &mut ctx.buf) else { + // Skip the range's end and destination objects too, so later + // entries stay aligned. + let _ = scanner.parse_object(); + let _ = scanner.parse_object(); + continue; + }; + let Some(end) = read_u32_code(scanner, &mut ctx.buf) else { + // Skip the destination object as well to stay aligned. + let _ = scanner.parse_object(); + continue; + }; let Ok(next) = scanner.parse_object() else { continue }; @@ -332,8 +350,11 @@ fn parse_bf_range( let mut array_scanner = array.objects(); for code in start..=end { - // A bad array member drops only that member. - let Ok(s) = array_scanner.parse_string() else { continue }; + // An exhausted or broken array cannot be resynced to a + // member boundary; stop instead of spinning over the + // remaining codes. A decodable-but-bad member still + // drops only itself. + let Ok(s) = array_scanner.parse_string() else { break }; if s.decode_into(&mut ctx.buf).is_err() { continue } let Some(dst_base) = decode_be(&ctx.buf) else { continue }; From e08950c8e67362b0f55f5101502b985e5f870fbc Mon Sep 17 00:00:00 2001 From: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> Date: Sun, 18 Oct 2026 13:34:49 +0530 Subject: [PATCH 04/34] hayro-cmap patch: truncated metadata-only streams report failure Character-collection metadata is not a mapping: a truncated stream that parsed only /Registry //Ordering //Supplement produced an empty CMap, which callers accept and which then blocks the encoding fallback paths. Drop the metadata exemption so a truncated stream with no mappings returns no CMap, exactly as a full parse failure did before. Signed-off-by: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> --- vendor/hayro-cmap/src/lib.rs | 12 ++++++++++++ vendor/hayro-cmap/src/parse.rs | 1 - 2 files changed, 12 insertions(+), 1 deletion(-) diff --git a/vendor/hayro-cmap/src/lib.rs b/vendor/hayro-cmap/src/lib.rs index c45cc0ae5..4eb88f505 100644 --- a/vendor/hayro-cmap/src/lib.rs +++ b/vendor/hayro-cmap/src/lib.rs @@ -1125,6 +1125,18 @@ endbfchar assert!(CMap::parse(b"<00", |_| None).is_none()); } + #[test] + fn truncated_stream_with_only_metadata_reports_failure() { + // Character-collection metadata alone is not a usable CMap: an empty + // metadata-only map would block the caller's fallback paths, so a + // truncated stream with no mappings must report failure here too. + assert!(CMap::parse( + b"/Registry (Adobe) /Ordering (Japan1) /Supplement 0 <00", + |_| None + ) + .is_none()); + } + #[test] fn bad_bfchar_code_keeps_later_entries_aligned() { // The malformed 5-byte code still has a destination object; both diff --git a/vendor/hayro-cmap/src/parse.rs b/vendor/hayro-cmap/src/parse.rs index 9450bbbe2..2a89fdbac 100644 --- a/vendor/hayro-cmap/src/parse.rs +++ b/vendor/hayro-cmap/src/parse.rs @@ -154,7 +154,6 @@ pub(crate) fn parse_inner<'a>( && notdef_ranges.is_empty() && bf_entries.is_empty() && base.is_none() - && metadata.character_collection.is_none() { return None; } From fb1f72e9ed805692cbe19d31cedb932769ffad79 Mon Sep 17 00:00:00 2001 From: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> Date: Sun, 18 Oct 2026 15:29:20 +0530 Subject: [PATCH 05/34] Attribute the vendored cmaps.brotli as a repository asset With hayro-cmap vendored, its Adobe-authored CMap data no longer arrives through a crates.io dependency: record it as an [[asset]] (same SHA-256, licence_file at vendor/hayro-cmap/assets/LICENSE.txt) instead of the [[bundled]] entry, and add the vendored path to the closed first-party list in xtask so cargo xtask assets passes. Signed-off-by: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> --- ATTRIBUTION.md | 6 +++--- ATTRIBUTION.toml | 9 ++++----- xtask/src/assets.rs | 5 ++++- 3 files changed, 11 insertions(+), 9 deletions(-) diff --git a/ATTRIBUTION.md b/ATTRIBUTION.md index 903d13284..709017835 100644 --- a/ATTRIBUTION.md +++ b/ATTRIBUTION.md @@ -4,7 +4,7 @@ Every asset PdfCraft includes, bundles or uses to build its published material, with its author, source and licence. The policy is in [AGENTS.md](AGENTS.md) §1. The machine-readable list, with SHA-256 hashes, is [ATTRIBUTION.toml](ATTRIBUTION.toml). Licence texts are kept beside the assets and summarised in [NOTICE](NOTICE). -## In this repository (272) +## In this repository (273) | Asset | Title | Author | Licence | Source | Used for | |---|---|---|---|---|---| @@ -224,6 +224,7 @@ Every asset PdfCraft includes, bundles or uses to build its published material, | `vendor/hayro-interpret/assets/FoxitSymbol.pfb` | FoxitSymbol (standard-14 substitute font) | Foxit Software Inc. / PDFium Authors | BSD-3-Clause | https://pdfium.googlesource.com/pdfium (via hayro-interpret 0.7.0 assets/) | Renders non-embedded standard-14 fonts (vendored hayro-interpret) | | `vendor/hayro-interpret/assets/CGATS001Compat-v2-micro.icc` | CGATS001Compat v2 micro ICC profile | Clinton Ingram (saucecontrol/Compact-ICC-Profiles) | CC0-1.0 | https://github.com/saucecontrol/Compact-ICC-Profiles | Default CMYK colour conversion (vendored hayro-interpret) | | `crates/fonts/src/std14_metrics.rs` | Standard-14 font advance widths and encodings (as Rust tables) | Adobe (Core14 AFM metrics, PDF specification encodings); tables generated by the PdfCraft contributors | MIT OR Apache-2.0 | ISO 32000-2 Annex D and the Adobe Core14 AFM metrics. Generated by joining the published WinAnsi, Symbol and ZapfDingbats encodings (code to glyph name) with the per-glyph advance widths, then spot-checked against the specification by hand in the tests of crates/fonts/src/std14.rs. Data only: the code that reads these tables lives in std14.rs, so that ordinary edits do not churn this file's hash. | Advance widths for the fourteen standard fonts, which a PDF may name without embedding. Non-visual data, not a typeface: the item AGENTS.md 1.1 permits, held first-party so generated appearance streams and the renderer agree. | +| `vendor/hayro-cmap/assets/cmaps.brotli` | Adobe CMap resources (predefined CJK CMaps) | Adobe | BSD-3-Clause | https://github.com/adobe-type-tools/cmap-resources (via hayro-cmap 0.1.0 assets/) | Character-code mapping tables required to read CJK PDFs. Non-visual data, allowed by AGENTS.md §1.1 | | `docs/brand/artcraft-logo-white.png` | ArtCraft brand: artcraft-logo-white | ArtCraft Team | LicenseRef-Storyteller-Trademark | ArtCraft Team brand assets (provided by the project owner, 2026-10-01; identical copies in storytold/photocraft) | ArtCraft branding next to the community links (About dialog, Help menu, home screen) and in docs. ArtCraft Team trademark, not open source; may be used only unmodified within PdfCraft; forks and modified versions must remove it. | | `docs/brand/artcraft-logo-white.svg` | ArtCraft brand: artcraft-logo-white | ArtCraft Team | LicenseRef-Storyteller-Trademark | ArtCraft Team brand assets (provided by the project owner, 2026-10-01; identical copies in storytold/photocraft) | ArtCraft branding next to the community links (About dialog, Help menu, home screen) and in docs. ArtCraft Team trademark, not open source; may be used only unmodified within PdfCraft; forks and modified versions must remove it. | | `docs/brand/artcraft-logo.png` | ArtCraft brand: artcraft-logo | ArtCraft Team | LicenseRef-Storyteller-Trademark | ArtCraft Team brand assets (provided by the project owner, 2026-10-01; identical copies in storytold/photocraft) | ArtCraft branding next to the community links (About dialog, Help menu, home screen) and in docs. ArtCraft Team trademark, not open source; may be used only unmodified within PdfCraft; forks and modified versions must remove it. | @@ -281,7 +282,7 @@ Every asset PdfCraft includes, bundles or uses to build its published material, | `crates/ui-egui/src/i18n/sl.tsv` | Slovene interface translation catalog | PdfCraft contributors | MIT OR Apache-2.0 | Contributor-original: clean-room translations of PdfCraft's English UI labels, no proprietary localisation resources | Slovene UI strings, compiled in by crates/ui-egui/src/i18n/mod.rs | | `vendor/vello_common/assets/probe.png` | vello_common probe reference image (rendered from probe.rgba) | Linebender vello contributors | MIT OR Apache-2.0 | vello_common 0.0.8 (crates.io), vendored in vendor/vello_common; regenerated by its vello_sparse_tests regenerate_probe_reference binary | Reference picture of the vendored vello_common probe module, for people only; the build reads probe.rgba, not this file | -## Compiled in through dependencies (6) +## Compiled in through dependencies (5) | Crate | File | Title | Author | Licence | Source | Used for | |---|---|---|---|---|---|---| @@ -289,7 +290,6 @@ Every asset PdfCraft includes, bundles or uses to build its published material, | `epaint_default_fonts` 0.36.2 | `fonts/Hack-Regular.ttf` | Hack Regular | Source Foundry Authors (derived from Bitstream Vera / DejaVu) | MIT AND Bitstream-Vera | https://github.com/source-foundry/Hack | egui fallback fonts (eframe feature default_fonts) | | `epaint_default_fonts` 0.36.2 | `fonts/NotoEmoji-Regular.ttf` | Noto Emoji (monochrome) | Google Inc. | OFL-1.1 | https://github.com/googlefonts/noto-emoji | egui fallback fonts (eframe feature default_fonts) | | `epaint_default_fonts` 0.36.2 | `fonts/emoji-icon-font.ttf` | emoji-icon-font | John Slegers | MIT | https://github.com/jslegers/emoji-icon-font | egui fallback fonts (eframe feature default_fonts) | -| `hayro-cmap` 0.1.0 | `assets/cmaps.brotli` | Adobe CMap resources (predefined CJK CMaps) (non-visual data, AGENTS.md §1.1) | Adobe | BSD-3-Clause | https://github.com/adobe-type-tools/cmap-resources (via hayro-cmap) | Character-code mapping tables required to read CJK PDFs. Non-visual data, allowed by AGENTS.md §1.1 | | `hayro-interpret` 0.7.0 | `src/font/generated/metrics.rs` | Standard-14 font metrics and encodings (as Rust tables) (non-visual data, AGENTS.md §1.1) | Adobe (Core 14 AFM metrics, PDF specification encodings); tables by The Hayro Authors | MIT OR Apache-2.0 | ISO 32000-2 Annex D; Adobe Core14 AFM files (via vendored hayro-interpret) | Glyph widths and encodings for non-embedded standard fonts. Non-visual data, allowed by AGENTS.md §1.1 | ## Downloaded at build time (22) diff --git a/ATTRIBUTION.toml b/ATTRIBUTION.toml index 34f17b513..c276ee643 100644 --- a/ATTRIBUTION.toml +++ b/ATTRIBUTION.toml @@ -2436,14 +2436,13 @@ kind = "font" usage = "egui fallback fonts (eframe feature default_fonts)" sha256 = "f426bec371f484646d002717e2c555a95cd5145e635d83a48bbef83f3abf18ca" -[[bundled]] -crate = "hayro-cmap" -version = "0.1.0" -path = "assets/cmaps.brotli" +[[asset]] +path = "vendor/hayro-cmap/assets/cmaps.brotli" title = "Adobe CMap resources (predefined CJK CMaps)" author = "Adobe" -source = "https://github.com/adobe-type-tools/cmap-resources (via hayro-cmap)" +source = "https://github.com/adobe-type-tools/cmap-resources (via hayro-cmap 0.1.0 assets/)" licence = "BSD-3-Clause" +licence_file = "vendor/hayro-cmap/assets/LICENSE.txt" kind = "data" adobe_data = true usage = "Character-code mapping tables required to read CJK PDFs. Non-visual data, allowed by AGENTS.md §1.1" diff --git a/xtask/src/assets.rs b/xtask/src/assets.rs index 1352a64c8..21dc6dea7 100644 --- a/xtask/src/assets.rs +++ b/xtask/src/assets.rs @@ -47,7 +47,10 @@ pub const ADOBE_DATA_ALLOWED: &[(&str, &str)] = &[("hayro-cmap", "assets/cmaps.b /// technical data we hold ourselves rather than through a dependency. Each still needs an /// `[[asset]]` entry with `adobe_data = true`, so its SHA-256 is checked and any change to the /// table has to be re-attested deliberately. -pub const ADOBE_DATA_FILES: &[&str] = &["crates/fonts/src/std14_metrics.rs"]; +pub const ADOBE_DATA_FILES: &[&str] = &[ + "crates/fonts/src/std14_metrics.rs", + "vendor/hayro-cmap/assets/cmaps.brotli", +]; /// File extensions that count as assets wherever they appear in the repository. const ASSET_EXTENSIONS: &[&str] = &[ From dd64f1797ee2cefa59cb2c8e093fb0a02f9b3cae Mon Sep 17 00:00:00 2001 From: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> Date: Sun, 18 Oct 2026 16:43:48 +0530 Subject: [PATCH 06/34] hayro-cmap patch: an unresolved usecmap with no mappings reports failure --- vendor/hayro-cmap/src/parse.rs | 31 ++++++++++++++++++++++++------- 1 file changed, 24 insertions(+), 7 deletions(-) diff --git a/vendor/hayro-cmap/src/parse.rs b/vendor/hayro-cmap/src/parse.rs index 2a89fdbac..1137ab078 100644 --- a/vendor/hayro-cmap/src/parse.rs +++ b/vendor/hayro-cmap/src/parse.rs @@ -53,6 +53,12 @@ pub(crate) fn parse_inner<'a>( // unparseable tail keeps the entries read so far instead of being // discarded wholesale; one that yielded nothing still reports failure. let mut truncated = false; + + // PdfCraft patch: track a `usecmap` reference that could not be resolved + // or parsed. When the stream yields no mappings of its own, the caller + // must see failure rather than an empty map, so its encoding fallback + // paths run exactly as they did before the tolerant-usecmap change. + let mut usecmap_failed = false; while !scanner.at_end() { let Ok(obj) = scanner.parse_object() else { truncated = true; @@ -110,11 +116,20 @@ pub(crate) fn parse_inner<'a>( } Some("usecmap") => { // PdfCraft patch: an unresolvable or unparseable `usecmap` - // reference drops only the base, not the whole CMap. - if let Some(name) = last_name.as_deref() { - if let Some(nested_data) = (ctx.get_cmap)(CMapName::from_bytes(name)) { - base = parse_inner(nested_data, ctx.get_cmap.clone(), depth + 1).map(Box::new); + // reference drops only the base, not the whole CMap, but + // is remembered so a stream that then yields no mappings + // at all still reports failure. + let resolved = last_name.as_deref().and_then(|name| { + (ctx.get_cmap)(CMapName::from_bytes(name)) + }); + match resolved { + Some(nested_data) => { + match parse_inner(nested_data, ctx.get_cmap.clone(), depth + 1) { + Some(nested) => base = Some(Box::new(nested)), + None => usecmap_failed = true, + } } + None => usecmap_failed = true, } } _ => {} @@ -147,9 +162,11 @@ pub(crate) fn parse_inner<'a>( writing_mode, }; - // PdfCraft patch: a truncated CMap that yielded nothing still reports failure, - // so callers fall back to the font's own encoding exactly as before. - if truncated + // PdfCraft patch: a CMap that was cut short or whose `usecmap` base could + // not be resolved, and that yielded no mappings of its own, still reports + // failure, so callers fall back to the font's own encoding exactly as + // before. + if (truncated || usecmap_failed) && ranges.is_empty() && notdef_ranges.is_empty() && bf_entries.is_empty() From f6521cf64eed4492eea908c3c2323ace56aba35b Mon Sep 17 00:00:00 2001 From: Charan Rathore <180254320+charan-rathore@users.noreply.github.com> Date: Sun, 18 Oct 2026 16:44:19 +0530 Subject: [PATCH 07/34] hayro-cmap patch: tests for the unresolved-usecmap failure guard --- vendor/hayro-cmap/src/lib.rs | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/vendor/hayro-cmap/src/lib.rs b/vendor/hayro-cmap/src/lib.rs index 4eb88f505..4786e75e5 100644 --- a/vendor/hayro-cmap/src/lib.rs +++ b/vendor/hayro-cmap/src/lib.rs @@ -1137,6 +1137,27 @@ endbfchar .is_none()); } + #[test] + fn unresolved_usecmap_with_no_mappings_reports_failure() { + // A stream whose only content is a `usecmap` reference that cannot be + // resolved yields no mappings at all; returning an empty map here + // would block the caller's encoding fallback paths, so it must + // report failure exactly as a full parse failure did before. + assert!(CMap::parse(b"/Bogus usecmap", |_| None).is_none()); + } + + #[test] + fn unresolved_usecmap_keeps_local_mappings() { + // The same unresolved reference must not discard mappings the stream + // carries itself: only the missing base is dropped. + let cmap = CMap::parse( + b"/Bogus usecmap 1 beginbfchar <41> <0041> endbfchar", + |_| None, + ) + .expect("local mappings survive an unresolved usecmap"); + assert!(cmap.lookup_bf_string(0x41).is_some()); + } + #[test] fn bad_bfchar_code_keeps_later_entries_aligned() { // The malformed 5-byte code still has a destination object; both From 227b4bfcf093bf171390e345a0d352843ab2e2ed Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Thu, 15 Oct 2026 06:24:22 +0000 Subject: [PATCH 08/34] Issue #2365: a rebuilt file keeps the /Info and /ID of its cross-reference stream When the cross-reference chain is unusable (here: a second, truncated PDF appended after the first, so no startxref is in reach), the index is rebuilt from object headers and the trailer taken from the last `trailer` dictionary with a usable /Root. A file written with cross-reference streams has no `trailer` keyword, so the rebuild fell through to the catalog alone and /Info (and /ID, /Encrypt) were lost: Document Properties showed nothing, and a full save after editing one Info field wrote only that field and /ModDate. reconstruct() now also considers every /Type /XRef stream dictionary with a usable /Root, taking whichever trailer comes last in the file. Regression test: document::tests::reconstruction_takes_the_trailer_from_a_cross_reference_stream. Verified on the issue's fixture (openpreserve PDF-HUL-138/6.2017-0960.pdf): Creator, Producer, CreationDate and ModDate are read and kept by a full save. Co-Authored-By: Claude Fable 5.1 Claude-Session: https://claude.ai/code/session_011K1peyrh5MdZ6ttHkVq8Gh --- crates/cos/src/document.rs | 68 ++++++++++++++++++++++++++++++++++---- 1 file changed, 62 insertions(+), 6 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index e94355590..51358f525 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -1867,6 +1867,9 @@ impl Document { let mut entries: BTreeMap = BTreeMap::new(); let mut i = 0; let mut objstms = Vec::new(); + // Offsets of objects that may be cross-reference streams: their dictionaries are the + // trailers of files that have no `trailer` keyword (§7.5.8). + let mut xref_streams = Vec::new(); while i < data.len() { // Object headers start at a line start (or after whitespace). let at_start = i == 0 || is_whitespace(data[i - 1]) || data[i - 1] == b'>'; @@ -1882,6 +1885,9 @@ impl Document { entries.insert(num, XrefEntry::InFile { offset: i as u64, generation }); objstms.push(num); } + if find(&data, b"/XRef", lx.pos, lx.pos + 1024).is_some() { + xref_streams.push(i); + } i = lx.pos; continue; } @@ -1908,20 +1914,35 @@ impl Document { let object_count = entries.len(); self.entries = Arc::new(XrefIndex::Materialized(Arc::new(entries))); self.cache.lock().map_err(|_| CosError::Poisoned)?.clear(); - // Trailer: the last `trailer` dictionary with a usable /Root, else the catalog itself. - let mut trailer = None; + // Trailer: the last `trailer` dictionary or cross-reference stream dictionary with a + // usable /Root (last in the file, whichever kind), else the catalog itself. A file whose + // only trailers are cross-reference streams (and whose `startxref` is gone or points + // nowhere) keeps its /Info, /ID and /Encrypt this way. + let usable = |d: &Dict| d.reference(b"Root").is_some_and(|r| matches!(self.try_get(r.num).as_deref(), Ok(Object::Dict(_)))); + let mut trailer: Option<(usize, Dict)> = None; let mut from = 0; while let Some(p) = find(&data, b"trailer", from, data.len()) { from = p + 7; let mut lx = Lexer::new(&data, from); if let Ok(Object::Dict(d)) = lx.object() - && d.reference(b"Root").is_some_and(|r| matches!(self.try_get(r.num).as_deref(), Ok(Object::Dict(_)))) + && usable(&d) { - trailer = Some(d); + trailer = Some((p, d)); + } + } + for at in xref_streams { + if trailer.as_ref().is_some_and(|(p, _)| *p > at) { + continue; + } + if let Ok((_, Object::Stream(s))) = parse_indirect_shared(&data, at, &|_| None) + && s.dict.name(b"Type") == Some(b"XRef") + && usable(&s.dict) + { + trailer = Some((at, s.dict)); } } let trailer = match trailer { - Some(t) => t, + Some((_, t)) => t, None => { let catalog = self .entries @@ -1935,7 +1956,9 @@ impl Document { } }; let mut trailer = trailer; - for k in [&b"Prev"[..], b"XRefStm"] { + // What a cross-reference stream's dictionary carries for itself goes, as the chain + // readers drop it: a save writes its own. + for k in [&b"Prev"[..], b"XRefStm", b"Type", b"W", b"Index", b"Filter", b"DecodeParms", b"Length"] { trailer.remove(k); } self.trailer = trailer; @@ -2696,6 +2719,39 @@ mod tests { } } + /// A file whose only trailers are cross-reference stream dictionaries, with its `startxref` + /// out of reach (here a second file appended after it, cut off before its own + /// cross-reference data, as in #2365), is rebuilt from its object headers. The rebuild used + /// to look only for the `trailer` keyword, so /Info and /ID were lost, and a full save after + /// editing one Info field dropped every other. + #[test] + fn reconstruction_takes_the_trailer_from_a_cross_reference_stream() { + let table = build( + &["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>", "<< /Title (Kept) /Creator (Me) >>"], + "/Root 1 0 R /Info 3 0 R", + ); + let packed = crate::write_full(&Document::open(Arc::new(table)).unwrap(), &crate::SaveOptions::default()).unwrap(); + let text = String::from_utf8_lossy(&packed); + assert!(text.contains("/XRef") && !text.contains("trailer"), "a cross-reference stream is the only trailer"); + let mut bytes = packed.clone(); + let appended = format!("%PDF-1.3\n90 0 obj\n<< /Length 5000 >>\nstream\n{}\nendstream\nendobj\n", "x".repeat(5000)); + bytes.extend_from_slice(appended.as_bytes()); + let doc = Document::open(Arc::new(bytes)).unwrap(); + assert!(doc.revisions().is_empty(), "rebuilt: {:?}", doc.repair_log()); + let info = doc.trailer().reference(b"Info").expect("the cross-reference stream's /Info"); + let field = |d: &Document, key: &[u8]| d.get(info).as_dict().unwrap().get(key).unwrap().as_string().unwrap().bytes.clone(); + assert_eq!(field(&doc, b"Title"), b"Kept"); + assert!(doc.trailer().get(b"ID").is_some(), "and its /ID"); + let mut edited = doc.clone(); + edited.update_dict(info, |d| d.set(b"Subject".to_vec(), Object::String(crate::PdfString::literal("new")))).unwrap(); + let saved = crate::write_full(&edited, &crate::SaveOptions::default()).unwrap(); + let back = Document::open(Arc::new(saved)).unwrap(); + assert!(back.repair_log().is_empty(), "{:?}", back.repair_log()); + let info = back.trailer().reference(b"Info").expect("the full save keeps /Info"); + let field = |key: &[u8]| back.get(info).as_dict().unwrap().get(key).unwrap().as_string().unwrap().bytes.clone(); + assert_eq!((field(b"Title"), field(b"Creator"), field(b"Subject")), (b"Kept".to_vec(), b"Me".to_vec(), b"new".to_vec())); + } + #[test] fn damaged_xref_is_reconstructed() { let mut bytes = build(&["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>"], "/Root 1 0 R"); From 1c126150320bdabadce93949b094600cc59bb50b Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Fri, 16 Oct 2026 15:16:16 +0000 Subject: [PATCH 09/34] lopdf: a rebuild takes the trailer from a cross-reference stream (PdfCraft patch 6) Document Properties are read through the inspector: the lazy cos reader refuses a file that needed repair, and lopdf's rebuild only looked for the trailer keyword, so a file whose only trailer is a cross-reference stream failed to load there and every field showed empty. lopdf's rebuild now takes a /Type /XRef stream dictionary as the trailer, the latest in the file winning as a later trailer keyword would. --- crates/render/src/inspect.rs | 33 +++++++++++++++++++++++ vendor/README.md | 2 +- vendor/lopdf/src/reader.rs | 51 +++++++++++++++++++++++++++++++----- 3 files changed, 78 insertions(+), 8 deletions(-) diff --git a/crates/render/src/inspect.rs b/crates/render/src/inspect.rs index 3c3988b7b..2b9684210 100644 --- a/crates/render/src/inspect.rs +++ b/crates/render/src/inspect.rs @@ -2392,6 +2392,39 @@ trailer << /Root 1 0 R >> assert!(reads < 20 * (PAGES + PAGES * PER_PAGE), "{reads} object reads"); } + /// #2365: the issue's fixture is two PDFs concatenated, the second cut off before its + /// cross-reference data, and the first's only trailer is a cross-reference stream. The + /// lazy reader refuses a file that needed repair, and lopdf's rebuild found no `trailer` + /// keyword and failed, so Document Properties showed every field empty. The rebuild now + /// takes the cross-reference stream's dictionary as the trailer. + #[test] + fn rebuilt_file_keeps_the_info_of_its_cross_reference_stream() { + let header = "%PDF-1.5\n"; + let obj1 = "1 0 obj<>endobj\n"; + let obj2 = "2 0 obj<>endobj\n"; + let obj3 = "3 0 obj<>endobj\n"; + let obj4 = "4 0 obj<>endobj\n"; + let o1 = header.len(); + let o2 = o1 + obj1.len(); + let o3 = o2 + obj2.len(); + let o4 = o3 + obj3.len(); + let o5 = o4 + obj4.len(); + let rows: Vec = + [[0u8, 0, 0, 255], [1, 0, o1 as u8, 0], [1, 0, o2 as u8, 0], [1, 0, o3 as u8, 0], [1, 0, o4 as u8, 0], [1, (o5 >> 8) as u8, o5 as u8, 0]] + .concat(); + let mut bytes = format!("{header}{obj1}{obj2}{obj3}{obj4}").into_bytes(); + bytes.extend_from_slice(format!("5 0 obj<>stream\n", rows.len()).as_bytes()); + bytes.extend_from_slice(&rows); + // A wrong startxref, and a second file after %%EOF: every reader rebuilds the table. + bytes.extend_from_slice(b"\nendstream\nendobj\nstartxref\n999999\n%%EOF\n%PDF-1.3\n"); + bytes.extend_from_slice(&[b'x'; 1024]); + + let info = inspect(Arc::new(bytes), None).unwrap(); + assert_eq!(info.pages.len(), 1); + assert_eq!(info.title.as_deref(), Some("Kept")); + assert_eq!(info.creator.as_deref(), Some("Me")); + } + #[test] fn lazy_structure_matches_the_compatibility_inspector() { // Each fixture as lopdf writes it, and with its objects in object streams. diff --git a/vendor/README.md b/vendor/README.md index 2d42b2628..57ddebe68 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -20,7 +20,7 @@ A PdfCraft patch in `hayro-syntax::Page::base_dimensions` expresses the existing | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height` | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it. | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed` -| lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace` | +| lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) direct arrays and dictionaries share a reader-local nesting cap of 128, counted while skipping and while parsing a dictionary (its lenient repair reads, rather than skips, an object found where a key belongs, so `<< << …` recursed through that read); deeper objects return `None` instead of overflowing the stack (including unused trailer values, trailer entries, content operands and marked-content properties); a chain of 65 direct stitching functions now stops at this cap, before the 64-function guard | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `raster::tests::deeply_nested_direct_objects_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_content_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_the_trailer_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_as_marked_content_properties_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_are_read_up_to_the_nesting_cap`, `raster::tests::nesting_at_the_cap_fits_a_512_kib_stack` (native only), `raster::tests::function_limits_depth_guard_refuses_a_chain_of_references` | diff --git a/vendor/lopdf/src/reader.rs b/vendor/lopdf/src/reader.rs index df471371d..be2ac354b 100644 --- a/vendor/lopdf/src/reader.rs +++ b/vendor/lopdf/src/reader.rs @@ -1629,6 +1629,12 @@ impl Reader<'_> { /// Newest backwards-scanned `trailer` whose `/Root` references a scanned object. fn find_latest_trailer(&self, xref: &Xref) -> Option<(usize, Dictionary)> { + let usable = |dict: &Dictionary| { + dict.get(b"Root") + .and_then(Object::as_reference) + .is_ok_and(|root| xref.entries.contains_key(&root.0)) + }; + let mut latest = None; let mut bound = self.buffer.len(); for _ in 0..MAX_TRAILER_CANDIDATES { let Some(pos) = Reader::search_substring(&self.buffer[..bound], b"trailer", 0) else { @@ -1643,15 +1649,46 @@ impl Reader<'_> { let Ok((_, dict)) = parser::dictionary(&after_keyword[dict_start..]) else { continue; }; - if dict - .get(b"Root") - .and_then(Object::as_reference) - .is_ok_and(|root| xref.entries.contains_key(&root.0)) - { - return Some((pos, dict)); + if usable(&dict) { + latest = Some((pos, dict)); + break; } } - None + // PdfCraft patch: (#2365) a file whose only trailer is a cross-reference stream has + // no `trailer` keyword, so the scan above found nothing and the rebuild failed; its + // /Info and /ID were then lost. Such a stream's dictionary is the trailer, and the + // latest one in the file wins, as a later `trailer` keyword would. + const HEADER_WINDOW: usize = 32; + const DICT_WINDOW: usize = 4096; + for entry in xref.entries.values() { + let XrefEntry::Normal { offset, .. } = entry else { + continue; + }; + let offset = *offset as usize; + if latest.as_ref().is_some_and(|(pos, _)| *pos >= offset) { + continue; + } + let Some(head) = self.buffer.get(offset..) else { + continue; + }; + let Some(dict_start) = head[..cmp::min(head.len(), HEADER_WINDOW)] + .windows(2) + .position(|window| window == b"<<") + else { + continue; + }; + let dict_bytes = &head[dict_start..cmp::min(head.len(), dict_start + DICT_WINDOW)]; + if !dict_bytes.windows(5).any(|window| window == b"/XRef") { + continue; + } + let Ok((_, dict)) = parser::dictionary(&head[dict_start..]) else { + continue; + }; + if dict.has_type(b"XRef") && usable(&dict) { + latest = Some((offset, dict)); + } + } + latest } /// Some generators write `startxref` (or trailer `Prev`) values that are From d1b0293a5d9cce58126e60a6f673027dd4a80d4b Mon Sep 17 00:00:00 2001 From: Ozt-y Date: Sun, 18 Oct 2026 19:05:46 +0530 Subject: [PATCH 10/34] fix(content): accept inline image EI without preceding whitespace (#3108) --- crates/content/src/lib.rs | 8 ++++---- crates/content/src/tests.rs | 25 +++++++++++++++++++++++++ 2 files changed, 29 insertions(+), 4 deletions(-) diff --git a/crates/content/src/lib.rs b/crates/content/src/lib.rs index c7c8bca9e..143387c59 100644 --- a/crates/content/src/lib.rs +++ b/crates/content/src/lib.rs @@ -293,23 +293,23 @@ fn unfiltered_image_end(data: &[u8], begin: usize, dict: &Dict) -> Option<(usize let row = width.checked_mul(components)?.checked_mul(bits)?.div_ceil(8); let end_data = begin.checked_add(row.checked_mul(height)?)?; let after = data.get(end_data..)?; - if !after.first().is_some_and(|b| is_ws(*b)) { + let separator = after.iter().position(|b| !is_ws(*b))?; + if separator == 0 && data.get(end_data.checked_sub(1)?).is_some_and(|b| is_ws(*b)) { return None; } - let separator = after.iter().position(|b| !is_ws(*b))?; let ei = end_data.checked_add(separator)?; let end = ei.checked_add(2)?; if data.get(ei..end) == Some(b"EI") && data.get(end).is_none_or(|b| is_ws(*b) || is_delim(*b)) { Some((end_data, end)) } else { None } } /// Where inline image data starting at `begin` ends: (end of the data, position after `EI`). The -/// data ends at whitespace + `EI` + (whitespace, a delimiter or the end). +/// data ends at whitespace or `>` + `EI` + (whitespace, a delimiter or the end). fn end_of_image(data: &[u8], begin: usize) -> Option<(usize, usize)> { let mut i = begin; while i + 2 <= data.len() { if data[i] == b'E' && data.get(i + 1) == Some(&b'I') - && (i == begin || is_ws(data[i - 1])) + && (i == begin || is_ws(data[i - 1]) || data[i - 1] == b'>') && data.get(i + 2).is_none_or(|b| is_ws(*b) || is_delim(*b)) { let end_data = if i > begin && is_ws(data[i - 1]) { i - 1 } else { i }; diff --git a/crates/content/src/tests.rs b/crates/content/src/tests.rs index 1119bdaf9..2ce806fba 100644 --- a/crates/content/src/tests.rs +++ b/crates/content/src/tests.rs @@ -388,3 +388,28 @@ fn piece_lookup_preserves_exact_boundary_and_empty_piece_mapping() { assert_eq!(empty.piece_of(0), 0); assert_eq!(empty.piece_of(usize::MAX), 0); } + +#[test] +fn inline_image_terminator_without_preceding_whitespace() { + let cases: &[(&[u8], &[u8])] = &[ + (b"BI /W 4 /H 1 /BPC 8 /CS /G /F /AHx ID 2060A0E0>EI Q BT /F1 18 Tf (After) Tj ET", b"2060A0E0>"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G /Filter /ASCIIHexDecode ID 2060A0E0>EI Q BT /F1 18 Tf (After) Tj ET", b"2060A0E0>"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G ID ab()EI Q BT /F1 18 Tf (After) Tj ET", b"ab()"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G ID ab[]EI Q BT /F1 18 Tf (After) Tj ET", b"ab[]"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G ID \x20\x60\xa0\xe0EI Q BT /F1 18 Tf (After) Tj ET", b"\x20\x60\xa0\xe0"), + ]; + + for (src, expected_data) in cases { + let p = parse(src); + assert_eq!(p.skipped, 0, "stream should have 0 skipped bytes: {}", String::from_utf8_lossy(src)); + let names: Vec<&[u8]> = p.ops.iter().map(|o| o.op.as_slice()).collect(); + assert_eq!( + names, + [&b"BI"[..], b"Q", b"BT", b"Tf", b"Tj", b"ET"], + "operators after inline image must be preserved: {}", + String::from_utf8_lossy(src) + ); + let (_, img) = p.ops[0].inline.as_ref().unwrap(); + assert_eq!(img, expected_data, "inline image data must match: {}", String::from_utf8_lossy(src)); + } +} From 494567771767b4d8c39c25fd359ce19bf5220ee4 Mon Sep 17 00:00:00 2001 From: Ozt-y Date: Sun, 18 Oct 2026 19:32:09 +0530 Subject: [PATCH 11/34] fix(content): restrict '>EI' terminator detection to ASCIIHex filter --- crates/content/src/lib.rs | 25 ++++++++++++++++++++----- crates/content/src/tests.rs | 11 +++++++++++ 2 files changed, 31 insertions(+), 5 deletions(-) diff --git a/crates/content/src/lib.rs b/crates/content/src/lib.rs index 143387c59..be6e7c36c 100644 --- a/crates/content/src/lib.rs +++ b/crates/content/src/lib.rs @@ -219,19 +219,34 @@ fn inline_image(data: &[u8], pos: usize) -> Result<(Dict, Vec, usize), usize return Ok((dict, Vec::new(), data.len())); } // Without an `EI` after it, the image runs to the end of the stream. - return Err(end_of_image(data, at).map_or(data.len(), |(_, end)| end)); + return Err(end_of_image(data, at, false).map_or(data.len(), |(_, end)| end)); }; dict.set(k, v); } // One whitespace byte separates ID from the data. let begin = (lx.pos + 1).min(data.len()); - let end = unfiltered_image_end(data, begin, &dict).or_else(|| end_of_image(data, begin)); + let end = unfiltered_image_end(data, begin, &dict).or_else(|| end_of_image(data, begin, is_ascii_hex_filter(&dict))); Ok(match end { Some((end_data, end)) => (dict, data[begin..end_data].to_vec(), end), None => (dict, data[begin..].to_vec(), data.len()), }) } +fn is_ascii_hex_filter(dict: &Dict) -> bool { + for key in [b"Filter".as_slice(), b"F"] { + if let Some(filter) = dict.get(key) { + match filter { + Object::Name(name) => return matches!(name.as_slice(), b"ASCIIHexDecode" | b"AHx"), + Object::Array(filters) => { + return matches!(filters.first().and_then(Object::as_name), Some(b"ASCIIHexDecode" | b"AHx")); + } + _ => return false, + } + } + } + false +} + /// An unfiltered image's samples determine its length, including byte padding per row. /// Binary samples may themselves contain whitespace + `EI` + whitespace, so use the known /// length when the actual terminator follows it. Unknown or damaged layouts keep the heuristic. @@ -303,13 +318,13 @@ fn unfiltered_image_end(data: &[u8], begin: usize, dict: &Dict) -> Option<(usize } /// Where inline image data starting at `begin` ends: (end of the data, position after `EI`). The -/// data ends at whitespace or `>` + `EI` + (whitespace, a delimiter or the end). -fn end_of_image(data: &[u8], begin: usize) -> Option<(usize, usize)> { +/// data ends at whitespace (or `>` for ASCIIHex) + `EI` + (whitespace, a delimiter or the end). +fn end_of_image(data: &[u8], begin: usize, allow_gt_ei: bool) -> Option<(usize, usize)> { let mut i = begin; while i + 2 <= data.len() { if data[i] == b'E' && data.get(i + 1) == Some(&b'I') - && (i == begin || is_ws(data[i - 1]) || data[i - 1] == b'>') + && (i == begin || is_ws(data[i - 1]) || (allow_gt_ei && data[i - 1] == b'>')) && data.get(i + 2).is_none_or(|b| is_ws(*b) || is_delim(*b)) { let end_data = if i > begin && is_ws(data[i - 1]) { i - 1 } else { i }; diff --git a/crates/content/src/tests.rs b/crates/content/src/tests.rs index 2ce806fba..e450d5a5f 100644 --- a/crates/content/src/tests.rs +++ b/crates/content/src/tests.rs @@ -413,3 +413,14 @@ fn inline_image_terminator_without_preceding_whitespace() { assert_eq!(img, expected_data, "inline image data must match: {}", String::from_utf8_lossy(src)); } } + +#[test] +fn ascii85_inline_image_does_not_terminate_at_gt_ei_inside_payload() { + let src = b"BI /W 4 /H 1 /BPC 8 /CS /G /F /A85 ID !!!!>EI !!~>\nEI Q BT /F1 18 Tf (After) Tj ET"; + let p = parse(src); + assert_eq!(p.skipped, 0); + let names: Vec<&[u8]> = p.ops.iter().map(|o| o.op.as_slice()).collect(); + assert_eq!(names, [&b"BI"[..], b"Q", b"BT", b"Tf", b"Tj", b"ET"]); + let (_, img) = p.ops[0].inline.as_ref().unwrap(); + assert_eq!(img, b"!!!!>EI !!~>"); +} From e336b48a35766f7977c7b7ff9713fab86871404e Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Sun, 18 Oct 2026 14:53:39 +0000 Subject: [PATCH 12/34] cos: a rebuild prefers a trailer whose /Root is the catalog (#2365 review) A later cross-reference stream (or trailer) whose /Root points at another dictionary, such as /Info, replaced an earlier good trailer, and a full save then wrote that dictionary as /Root. A candidate whose /Root passes the root_is_catalog rule (/Type /Catalog or /Pages) now beats one whose /Root is only some dictionary, wherever it is in the file. The looser rule still applies when nothing better is there, which FDF files (whose /Root has neither key) and the repair path for a non-catalog /Root rely on. Test: reconstruction_skips_a_trailer_whose_root_is_not_the_catalog --- crates/cos/src/document.rs | 45 +++++++++++++++++++++++++++++--------- 1 file changed, 35 insertions(+), 10 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index 51358f525..062c29d40 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -1915,30 +1915,39 @@ impl Document { self.entries = Arc::new(XrefIndex::Materialized(Arc::new(entries))); self.cache.lock().map_err(|_| CosError::Poisoned)?.clear(); // Trailer: the last `trailer` dictionary or cross-reference stream dictionary with a - // usable /Root (last in the file, whichever kind), else the catalog itself. A file whose - // only trailers are cross-reference streams (and whose `startxref` is gone or points - // nowhere) keeps its /Info, /ID and /Encrypt this way. - let usable = |d: &Dict| d.reference(b"Root").is_some_and(|r| matches!(self.try_get(r.num).as_deref(), Ok(Object::Dict(_)))); - let mut trailer: Option<(usize, Dict)> = None; + // usable /Root (last in the file, whichever kind), else the catalog itself. One whose + // /Root is a catalog (as `root_is_catalog` judges it) beats one whose /Root is another + // dictionary, such as /Info, wherever it is; an FDF's /Root (no /Type, no /Pages) still + // counts when nothing better is there. + // A file whose only trailers are cross-reference streams (and whose `startxref` is gone + // or points nowhere) keeps its /Info, /ID and /Encrypt this way. + let rank = |at: usize, d: &Dict| { + let root = self.try_get(d.reference(b"Root")?.num).ok()?; + let Object::Dict(c) = &*root else { return None }; + Some((c.name(b"Type") == Some(b"Catalog") || c.contains(b"Pages"), at)) + }; + let mut trailer: Option<((bool, usize), Dict)> = None; let mut from = 0; while let Some(p) = find(&data, b"trailer", from, data.len()) { from = p + 7; let mut lx = Lexer::new(&data, from); if let Ok(Object::Dict(d)) = lx.object() - && usable(&d) + && let Some(r) = rank(p, &d) + && trailer.as_ref().is_none_or(|(best, _)| r > *best) { - trailer = Some((p, d)); + trailer = Some((r, d)); } } for at in xref_streams { - if trailer.as_ref().is_some_and(|(p, _)| *p > at) { + if trailer.as_ref().is_some_and(|((catalog, p), _)| *catalog && *p > at) { continue; } if let Ok((_, Object::Stream(s))) = parse_indirect_shared(&data, at, &|_| None) && s.dict.name(b"Type") == Some(b"XRef") - && usable(&s.dict) + && let Some(r) = rank(at, &s.dict) + && trailer.as_ref().is_none_or(|(best, _)| r > *best) { - trailer = Some((at, s.dict)); + trailer = Some((r, s.dict)); } } let trailer = match trailer { @@ -2752,6 +2761,22 @@ mod tests { assert_eq!((field(b"Title"), field(b"Creator"), field(b"Subject")), (b"Kept".to_vec(), b"Me".to_vec(), b"new".to_vec())); } + /// A rebuild prefers a trailer whose /Root is the catalog (as `root_is_catalog` judges it): a + /// later cross-reference stream whose /Root points at the Info dictionary does not replace an + /// earlier good trailer, or a full save would write the Info dictionary as /Root. + #[test] + fn reconstruction_skips_a_trailer_whose_root_is_not_the_catalog() { + let mut bytes = + build(&["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>", "<< /Title (Kept) >>"], "/Root 1 0 R /Info 3 0 R"); + bytes.extend_from_slice(b"4 0 obj\n<< /Type /XRef /Size 5 /W [1 1 1] /Root 3 0 R /Length 3 >>\nstream\nabc\nendstream\nendobj\n"); + let appended = format!("%PDF-1.3\n90 0 obj\n<< /Length 5000 >>\nstream\n{}\nendstream\nendobj\n", "x".repeat(5000)); + bytes.extend_from_slice(appended.as_bytes()); + let doc = Document::open(Arc::new(bytes)).unwrap(); + assert!(doc.repair_log().iter().any(|l| l.contains("rebuilt")), "{:?}", doc.repair_log()); + assert_eq!(doc.trailer().reference(b"Root").map(|r| r.num), Some(1), "the catalog, not the Info dictionary"); + assert_eq!(doc.trailer().reference(b"Info").map(|r| r.num), Some(3)); + } + #[test] fn damaged_xref_is_reconstructed() { let mut bytes = build(&["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>"], "/Root 1 0 R"); From d75d440ad995c10b6dc87510cfd5696cb9dff200 Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Sun, 18 Oct 2026 15:23:38 +0000 Subject: [PATCH 13/34] lopdf: a rebuild prefers a trailer whose /Root is the catalog (#2365 review) The mirror of the cos change for render::inspect: a later cross-reference stream whose /Root points at /Info replaced the earlier trailer, and Document Properties lost the catalog. A candidate whose /Root is /Type /Catalog or has /Pages now wins; another dictionary is still taken when nothing better is there (FDF). Test: inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog --- crates/render/src/inspect.rs | 26 ++++++++++++++++++++++++ vendor/README.md | 2 +- vendor/lopdf/src/reader.rs | 38 +++++++++++++++++++++++++++++------- 3 files changed, 58 insertions(+), 8 deletions(-) diff --git a/crates/render/src/inspect.rs b/crates/render/src/inspect.rs index 2b9684210..ad236e7cc 100644 --- a/crates/render/src/inspect.rs +++ b/crates/render/src/inspect.rs @@ -2425,6 +2425,32 @@ trailer << /Root 1 0 R >> assert_eq!(info.creator.as_deref(), Some("Me")); } + /// #2365 review: the rebuild prefers a trailer whose /Root is the catalog. A later + /// cross-reference stream whose /Root points at the Info dictionary used to replace the + /// earlier `trailer`, so Document Properties lost the catalog it should have read. + #[test] + fn rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog() { + let mut bytes = b"%PDF-1.5\n".to_vec(); + for body in [ + "<>", + "<>", + "<>", + "<>", + ] { + let n = bytes.iter().filter(|b| **b == b'\n').count(); + bytes.extend_from_slice(format!("{n} 0 obj{body}endobj\n").as_bytes()); + } + bytes.extend_from_slice(b"trailer\n<>\n"); + bytes.extend_from_slice(b"5 0 obj<>stream\n\x01\x00\x09\x00\nendstream\nendobj\n"); + // A wrong startxref, and a second file after %%EOF: every reader rebuilds the table. + bytes.extend_from_slice(b"startxref\n999999\n%%EOF\n%PDF-1.3\n"); + bytes.extend_from_slice(&[b'x'; 1024]); + + let info = inspect(Arc::new(bytes), None).unwrap(); + assert_eq!(info.pages.len(), 1); + assert_eq!(info.title.as_deref(), Some("Kept")); + } + #[test] fn lazy_structure_matches_the_compatibility_inspector() { // Each fixture as lopdf writes it, and with its objects in object streams. diff --git a/vendor/README.md b/vendor/README.md index 57ddebe68..923baa4c5 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -20,7 +20,7 @@ A PdfCraft patch in `hayro-syntax::Page::base_dimensions` expresses the existing | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height` | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it. | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed` -| lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream` | +| lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) direct arrays and dictionaries share a reader-local nesting cap of 128, counted while skipping and while parsing a dictionary (its lenient repair reads, rather than skips, an object found where a key belongs, so `<< << …` recursed through that read); deeper objects return `None` instead of overflowing the stack (including unused trailer values, trailer entries, content operands and marked-content properties); a chain of 65 direct stitching functions now stops at this cap, before the 64-function guard | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `raster::tests::deeply_nested_direct_objects_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_content_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_the_trailer_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_as_marked_content_properties_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_are_read_up_to_the_nesting_cap`, `raster::tests::nesting_at_the_cap_fits_a_512_kib_stack` (native only), `raster::tests::function_limits_depth_guard_refuses_a_chain_of_references` | diff --git a/vendor/lopdf/src/reader.rs b/vendor/lopdf/src/reader.rs index be2ac354b..d9e58c0d9 100644 --- a/vendor/lopdf/src/reader.rs +++ b/vendor/lopdf/src/reader.rs @@ -1627,14 +1627,30 @@ impl Reader<'_> { Some((number, generation)) } - /// Newest backwards-scanned `trailer` whose `/Root` references a scanned object. + /// Newest backwards-scanned `trailer` whose `/Root` references a scanned object, one whose + /// `/Root` is the catalog first. fn find_latest_trailer(&self, xref: &Xref) -> Option<(usize, Dictionary)> { let usable = |dict: &Dictionary| { dict.get(b"Root") .and_then(Object::as_reference) .is_ok_and(|root| xref.entries.contains_key(&root.0)) }; - let mut latest = None; + // PdfCraft patch: (#2365) a candidate whose /Root is the catalog (`/Type /Catalog` or + // `/Pages`) beats one whose /Root is another dictionary, such as /Info, wherever it is; + // the other kind is still taken when nothing better is there (an FDF's /Root). + let is_catalog = |dict: &Dictionary| { + let Ok(root) = dict.get(b"Root").and_then(Object::as_reference) else { + return false; + }; + let Some(XrefEntry::Normal { offset, .. }) = xref.entries.get(&root.0) else { + return false; + }; + parser::indirect_object(self.buffer, *offset as usize, None, self, &mut HashSet::new(), None) + .ok() + .and_then(|(_, object)| object.as_dict().map(|d| d.has_type(b"Catalog") || d.has(b"Pages")).ok()) + .unwrap_or(false) + }; + let mut latest: Option<(bool, usize, Dictionary)> = None; let mut bound = self.buffer.len(); for _ in 0..MAX_TRAILER_CANDIDATES { let Some(pos) = Reader::search_substring(&self.buffer[..bound], b"trailer", 0) else { @@ -1650,8 +1666,13 @@ impl Reader<'_> { continue; }; if usable(&dict) { - latest = Some((pos, dict)); - break; + let catalog = is_catalog(&dict); + if latest.is_none() || catalog { + latest = Some((catalog, pos, dict)); + } + if catalog { + break; + } } } // PdfCraft patch: (#2365) a file whose only trailer is a cross-reference stream has @@ -1665,7 +1686,7 @@ impl Reader<'_> { continue; }; let offset = *offset as usize; - if latest.as_ref().is_some_and(|(pos, _)| *pos >= offset) { + if latest.as_ref().is_some_and(|(catalog, pos, _)| *catalog && *pos >= offset) { continue; } let Some(head) = self.buffer.get(offset..) else { @@ -1685,10 +1706,13 @@ impl Reader<'_> { continue; }; if dict.has_type(b"XRef") && usable(&dict) { - latest = Some((offset, dict)); + let rank = (is_catalog(&dict), offset); + if latest.as_ref().is_none_or(|(catalog, pos, _)| rank > (*catalog, *pos)) { + latest = Some((rank.0, offset, dict)); + } } } - latest + latest.map(|(_, pos, dict)| (pos, dict)) } /// Some generators write `startxref` (or trailer `Prev`) values that are From a2bc0d3570b46998e6792985b599fb11fa7d6908 Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Sun, 18 Oct 2026 16:05:33 +0000 Subject: [PATCH 14/34] lopdf: a rebuilt trailer's /Root must match its generation (#2365 review) usable() and is_catalog() checked the object number only, so a later stream with /Root 1 1 R outranked the trailer naming 1 0 R; get_offset then found no such object and Document Properties lost the catalog. cos resolves by number alone, so its rank needs no change. Test: inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog --- crates/render/src/inspect.rs | 8 +++++++- vendor/README.md | 2 +- vendor/lopdf/src/reader.rs | 12 ++++++++++-- 3 files changed, 18 insertions(+), 4 deletions(-) diff --git a/crates/render/src/inspect.rs b/crates/render/src/inspect.rs index ad236e7cc..14fbb4cf8 100644 --- a/crates/render/src/inspect.rs +++ b/crates/render/src/inspect.rs @@ -2446,9 +2446,15 @@ trailer << /Root 1 0 R >> bytes.extend_from_slice(b"startxref\n999999\n%%EOF\n%PDF-1.3\n"); bytes.extend_from_slice(&[b'x'; 1024]); - let info = inspect(Arc::new(bytes), None).unwrap(); + let info = inspect(Arc::new(bytes.clone()), None).unwrap(); assert_eq!(info.pages.len(), 1); assert_eq!(info.title.as_deref(), Some("Kept")); + // A later stream whose /Root names the catalog's number with another generation does + // not count either: loading `1 1 R` finds no such object. + let at = bytes.windows(10).position(|w| w == b"/Root 4 0 ").unwrap(); + bytes[at..at + 10].copy_from_slice(b"/Root 1 1 "); + let info = inspect(Arc::new(bytes), None).unwrap(); + assert_eq!((info.pages.len(), info.title.as_deref()), (1, Some("Kept"))); } #[test] diff --git a/vendor/README.md b/vendor/README.md index 923baa4c5..e0f1bc154 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -20,7 +20,7 @@ A PdfCraft patch in `hayro-syntax::Page::base_dimensions` expresses the existing | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height` | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it. | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed` -| lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` | +| lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary, and an in-file /Root must match its generation (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) direct arrays and dictionaries share a reader-local nesting cap of 128, counted while skipping and while parsing a dictionary (its lenient repair reads, rather than skips, an object found where a key belongs, so `<< << …` recursed through that read); deeper objects return `None` instead of overflowing the stack (including unused trailer values, trailer entries, content operands and marked-content properties); a chain of 65 direct stitching functions now stops at this cap, before the 64-function guard | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `raster::tests::deeply_nested_direct_objects_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_content_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_the_trailer_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_as_marked_content_properties_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_are_read_up_to_the_nesting_cap`, `raster::tests::nesting_at_the_cap_fits_a_512_kib_stack` (native only), `raster::tests::function_limits_depth_guard_refuses_a_chain_of_references` | diff --git a/vendor/lopdf/src/reader.rs b/vendor/lopdf/src/reader.rs index d9e58c0d9..d7e33b7dc 100644 --- a/vendor/lopdf/src/reader.rs +++ b/vendor/lopdf/src/reader.rs @@ -1630,10 +1630,15 @@ impl Reader<'_> { /// Newest backwards-scanned `trailer` whose `/Root` references a scanned object, one whose /// `/Root` is the catalog first. fn find_latest_trailer(&self, xref: &Xref) -> Option<(usize, Dictionary)> { + // PdfCraft patch: (#2365) an in-file /Root must also match its generation, as + // `get_offset` requires when the root is loaded. let usable = |dict: &Dictionary| { dict.get(b"Root") .and_then(Object::as_reference) - .is_ok_and(|root| xref.entries.contains_key(&root.0)) + .is_ok_and(|root| match xref.entries.get(&root.0) { + Some(XrefEntry::Normal { generation, .. }) => *generation == root.1, + entry => entry.is_some(), + }) }; // PdfCraft patch: (#2365) a candidate whose /Root is the catalog (`/Type /Catalog` or // `/Pages`) beats one whose /Root is another dictionary, such as /Info, wherever it is; @@ -1642,9 +1647,12 @@ impl Reader<'_> { let Ok(root) = dict.get(b"Root").and_then(Object::as_reference) else { return false; }; - let Some(XrefEntry::Normal { offset, .. }) = xref.entries.get(&root.0) else { + let Some(XrefEntry::Normal { offset, generation }) = xref.entries.get(&root.0) else { return false; }; + if *generation != root.1 { + return false; + } parser::indirect_object(self.buffer, *offset as usize, None, self, &mut HashSet::new(), None) .ok() .and_then(|(_, object)| object.as_dict().map(|d| d.has_type(b"Catalog") || d.has(b"Pages")).ok()) From 04697ad4924a040eae8e59c6a7d1d309b051aca7 Mon Sep 17 00:00:00 2001 From: Luigi Pinna Date: Thu, 15 Oct 2026 12:09:29 +0200 Subject: [PATCH 15/34] fix(content): an inline image ends at its own EI when later text holds a non-ASCII byte The end-of-data look-ahead rejected a valid EI candidate when any byte above 0x7F lay between it and the next EI. Those bytes are the content stream that follows the image, and content streams carry such bytes inside strings, so no candidate qualified and the image data ran to the end of the stream. The operator check that follows now decides alone, bounded as before by MAX_EI_CANDIDATES. Fixes #2456 --- crates/render/src/text.rs | 12 ++++++++ vendor/README.md | 2 +- vendor/hayro-syntax/src/content/mod.rs | 42 ++++++++++++++++++++------ 3 files changed, 46 insertions(+), 10 deletions(-) diff --git a/crates/render/src/text.rs b/crates/render/src/text.rs index 6784f14eb..1e88bbcc8 100644 --- a/crates/render/src/text.rs +++ b/crates/render/src/text.rs @@ -1779,6 +1779,18 @@ mod tests { extract_page(pdf, page, &InterpreterSettings::default()).unwrap() } + #[test] + fn text_after_an_inline_image_survives_a_non_ascii_byte_before_a_later_ei() { + // Everyday content both ways: a string with a byte above 0x7F, and a later word holding + // a capital "EI". The end-of-data look-ahead used to read the string's bytes as image + // data, so the image swallowed the rest of the page. + let pdf = marked_pages(&[ + "BI /W 1 /H 1 /BPC 8 /CS /DeviceGray ID \0 EI BT /F1 20 Tf 1 0 0 1 20 200 Tm (Gr\u{fc}\u{df}e) Tj 1 0 0 1 20 150 Tm (EINGANG) Tj ET", + ]); + + assert!(marked_text(&pdf, 0).plain_text().contains("EINGANG"), "the text after the inline image was not extracted"); + } + #[test] fn actual_text_replaces_the_glyphs_it_marks_in_every_writing_direction() { // Chrome marks each shaped Thai cluster this way: "น้ำ" (U+0E19 U+0E49 U+0E33) here stands diff --git a/vendor/README.md b/vendor/README.md index 7b2c37f02..0cfb53e9e 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -15,7 +15,7 @@ A PdfCraft patch in `hayro-syntax::Page::base_dimensions` expresses the existing | hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour, planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk, interpolated, axis-aligned image is snapped to whole device pixels: shrunk once to its rounded device rectangle and drawn 1:1 at the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat); (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group with its background painted inside it as the initial backdrop, so a Multiply highlight blends with a dark reading-colours background instead of staying an opaque block (identity on white pages); (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349) | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_samples_the_same_columns`, `raster::tests::non_integer_minification_samples_colour_and_mask_from_the_same_columns`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::interpolated_minified_image_edges_snap_to_whole_pixels` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path` | | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap, page bitmap, halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer) | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | -| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306) | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`) +| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306); (22) the inline-image end-of-data look-ahead no longer rejects a candidate `EI` because a byte above 0x7F lies between it and the next one: when the candidate is the end, those bytes are the content stream that follows the image, and its strings carry such bytes (WinAnsi `(Grüße) Tj`, the two-byte codes of any Type 0 font), so the operator check that follows decides alone, bounded as in (4) (a scanned page whose OCR text held an accented word and a later capital "EI" — EINGANG, BEI, KLEIN — drew and extracted nothing after the image: 40 of 275 pages across five real documents) | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (22) `text::tests::text_after_an_inline_image_survives_a_non_ascii_byte_before_a_later_ei`, `content::tests::an_inline_image_ends_at_its_own_ei_when_later_text_holds_a_non_ascii_byte` (hayro-syntax) | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it. | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed` | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it` | diff --git a/vendor/hayro-syntax/src/content/mod.rs b/vendor/hayro-syntax/src/content/mod.rs index 8e47324d0..5e26b4793 100644 --- a/vendor/hayro-syntax/src/content/mod.rs +++ b/vendor/hayro-syntax/src/content/mod.rs @@ -299,16 +299,21 @@ impl<'a> UntypedIter<'a> { find_reader.read_bytes(next_pos)?; if is_ei { - let analyze_data = &tail[..find_reader.offset()]; + // PdfCraft patch: a byte above 0x7F between the two "EI"s + // used to reject the candidate ("binary data in-between, so + // we for sure have not reached the end"). When the candidate + // is the real end, those bytes are the content stream that + // follows the image, and a content stream carries bytes above + // 0x7F inside its strings: WinAnsi text such as `(Grüße) Tj`, + // or the two-byte codes of any Type 0 font. A page whose text + // held one, with a capital "EI" in a later word (EINGANG, + // BEI, KLEIN), had no valid candidate left, so the image data + // ran to the end of the stream and nothing after the image + // was drawn or extracted. The operator check below is the + // discriminator, and `MAX_EI_CANDIDATES` already bounds how + // often it runs. - // If there is any binary data in-between, we for sure - // have not reached the end. - if analyze_data.iter().any(|c| !c.is_ascii()) { - self.reader.read_bytes(2)?; - continue 'outer; - } - - // Otherwise, the only possibility that we reached an + // The only possibility that we reached an // "EI", even though the previous one was valid, is // that it's part of a string in the content // stream that follows the inline image. Therefore, @@ -872,6 +877,25 @@ mod tests { assert_eq!(operators(data), [b"TJ".to_vec(), b"BI".to_vec(), b"q".to_vec(), b"Q".to_vec()]); } + #[test] + fn an_inline_image_ends_at_its_own_ei_when_later_text_holds_a_non_ascii_byte() { + // `(Grüße)` in WinAnsi, then a word holding a capital "EI": both are everyday content, + // and the look-ahead used to read the non-ASCII bytes of the string as image data. + let data = b"BI /W 1 /H 1 /BPC 8 /CS /DeviceGray ID \x00 EI BT /F1 12 Tf (Gr\xFC\xDFe) Tj (EINGANG) Tj ET"; + + assert_eq!( + operators(data), + [ + b"BI".to_vec(), + b"BT".to_vec(), + b"Tf".to_vec(), + b"Tj".to_vec(), + b"Tj".to_vec(), + b"ET".to_vec() + ] + ); + } + #[test] fn tokenizer_interface_preserves_every_byte_boundary() { let data = b"% comment\n1 -2.5 /Name /A#20B (a\\(b\\) \\101) <4869> [1 2] << /Key (value) >> [(a) 1] TJ BI /W 1 /H 1 /BPC 8 /CS /DeviceGray ID \x00 EI q Q"; From 15f92032436fea0c1cd804e440e7ad00b07e7392 Mon Sep 17 00:00:00 2001 From: oracola91 <118899664+oracola91@users.noreply.github.com> Date: Tue, 20 Oct 2026 10:30:20 +0300 Subject: [PATCH 16/34] Page-tree walks in annot, forms and organize skip repeated children annot::page_refs, the forms page_refs and organize's walk queued every /Kids entry before the seen check, and organize cloned the inherited attributes, Resources included, for each queued entry. A page tree whose /Kids lists the same child many times, or whose nodes share one /Kids array, held all of those entries at once before they were dropped as already seen. Each walk is now a small iterator with one cursor per open /Kids array, the way pdfcraft_model::pages has walked since #1926. organize carries inherited attributes as the nodes that set them and copies them only for the pages it returns. Page order, the pages found, the inherited values and the million-node cap are unchanged. --- crates/annot/src/lib.rs | 61 ++++++++++++++++++------- crates/annot/src/tests.rs | 17 +++++++ crates/forms/src/lib.rs | 61 ++++++++++++++++++------- crates/forms/src/tests.rs | 16 +++++++ crates/organize/src/lib.rs | 88 ++++++++++++++++++++++++++---------- crates/organize/src/tests.rs | 23 ++++++++++ 6 files changed, 209 insertions(+), 57 deletions(-) diff --git a/crates/annot/src/lib.rs b/crates/annot/src/lib.rs index 587cd4466..0edcb0871 100644 --- a/crates/annot/src/lib.rs +++ b/crates/annot/src/lib.rs @@ -651,25 +651,54 @@ impl ReviewState { pub fn page_refs(doc: &Document) -> Result, AnnotError> { let root = doc.root().ok_or(AnnotError::NoPageTree)?; let pages = doc.get(root).as_dict().and_then(|d| d.reference(b"Pages")).ok_or(AnnotError::NoPageTree)?; - let mut out = Vec::new(); - let mut seen = std::collections::HashSet::new(); - let mut stack = vec![pages]; - while let Some(node) = stack.pop() { - if !seen.insert(node) || seen.len() > 1_000_000 { - continue; - } - let obj = doc.get(node); - let Some(d) = obj.as_dict() else { continue }; - match d.get(b"Kids").map(|k| doc.resolve(k)) { - Some(kids) if d.name(b"Type") != Some(b"Page") => { - if let Some(a) = kids.as_array() { - stack.extend(a.iter().rev().filter_map(|k| k.as_ref())); - } + Ok(PageWalk::new(doc, pages).collect()) +} + +/// Depth-first walk over the leaf pages under `top`. It keeps one cursor per open `/Kids` array, +/// never one entry per child, so a child listed many times or shared between nodes is skipped by +/// the seen check without being queued first (`pdfcraft_model::pages` walks the same way). +/// Cycles end at the seen check and at most a million nodes are visited. +struct PageWalk<'a> { + doc: &'a Document, + seen: std::collections::HashSet, + pending: Vec<(std::sync::Arc, usize)>, +} + +impl<'a> PageWalk<'a> { + fn new(doc: &'a Document, top: ObjRef) -> Self { + let top = std::sync::Arc::new(Object::Array(vec![Object::Ref(top)])); + Self { doc, seen: Default::default(), pending: vec![(top, 0)] } + } +} + +impl Iterator for PageWalk<'_> { + type Item = ObjRef; + + fn next(&mut self) -> Option { + while let Some((kids, next)) = self.pending.last_mut() { + let Some(kid) = kids.as_array().and_then(|a| a.get(*next)) else { + self.pending.pop(); + continue; + }; + let node = kid.as_ref(); + *next += 1; + let Some(node) = node else { continue }; + if self.seen.len() >= 1_000_000 { + self.pending.clear(); + return None; + } + if !self.seen.insert(node) { + continue; + } + let obj = self.doc.get(node); + let Some(d) = obj.as_dict() else { continue }; + match d.get(b"Kids").map(|k| self.doc.resolve(k)) { + Some(kids) if d.name(b"Type") != Some(b"Page") => self.pending.push((kids, 0)), + _ => return Some(node), } - _ => out.push(node), } + None } - Ok(out) } fn page_ref(doc: &Document, page: usize) -> Result { diff --git a/crates/annot/src/tests.rs b/crates/annot/src/tests.rs index 4a4dfb106..7d5399276 100644 --- a/crates/annot/src/tests.rs +++ b/crates/annot/src/tests.rs @@ -1382,3 +1382,20 @@ fn free_text_properties_show_legacy_or_absent_border_width() { .unwrap(); assert_eq!(props(&doc, 0, index).unwrap().width, Some(4.0)); } + +#[test] +fn page_walk_skips_repeated_and_cyclic_kids_without_queueing_them() { + let mut doc = fixture(); + let root = ObjRef { num: 2, generation: 0 }; + let (first, second) = (ObjRef { num: 3, generation: 0 }, ObjRef { num: 4, generation: 0 }); + // A cycle back to the root, then both pages 128 times over. + let mut kids = vec![Object::Ref(root)]; + kids.extend([Object::Ref(first), Object::Ref(second)].into_iter().cycle().take(256)); + doc.update_dict(root, |d| d.set(b"Kids".to_vec(), Object::Array(kids))).unwrap(); + let mut walk = PageWalk::new(&doc, root); + assert_eq!(walk.next(), Some(first)); + assert!(walk.pending.len() <= 2, "repeated kids retained {} pending entries", walk.pending.len()); + assert_eq!(walk.next(), Some(second)); + assert_eq!(walk.next(), None); + assert_eq!(page_refs(&doc).unwrap(), [first, second]); +} diff --git a/crates/forms/src/lib.rs b/crates/forms/src/lib.rs index 1dfa76726..629489898 100644 --- a/crates/forms/src/lib.rs +++ b/crates/forms/src/lib.rs @@ -452,25 +452,54 @@ fn nums(doc: &Document, o: Option<&Object>) -> Option> { pub(crate) fn page_refs(doc: &Document) -> Vec { let Some(root) = doc.root() else { return Vec::new() }; let Some(pages) = doc.get(root).as_dict().and_then(|d| d.reference(b"Pages")) else { return Vec::new() }; - let mut out = Vec::new(); - let mut seen = std::collections::HashSet::new(); - let mut stack = vec![pages]; - while let Some(node) = stack.pop() { - if !seen.insert(node) || seen.len() > 1_000_000 { - continue; - } - let obj = doc.get(node); - let Some(d) = obj.as_dict() else { continue }; - match d.get(b"Kids").map(|k| doc.resolve(k)) { - Some(kids) if d.name(b"Type") != Some(b"Page") => { - if let Some(a) = kids.as_array() { - stack.extend(a.iter().rev().filter_map(|k| k.as_ref())); - } + PageWalk::new(doc, pages).collect() +} + +/// Depth-first walk over the leaf pages under `top`. It keeps one cursor per open `/Kids` array, +/// never one entry per child, so a child listed many times or shared between nodes is skipped by +/// the seen check without being queued first (`pdfcraft_model::pages` walks the same way). +/// Cycles end at the seen check and at most a million nodes are visited. +struct PageWalk<'a> { + doc: &'a Document, + seen: std::collections::HashSet, + pending: Vec<(std::sync::Arc, usize)>, +} + +impl<'a> PageWalk<'a> { + fn new(doc: &'a Document, top: ObjRef) -> Self { + let top = std::sync::Arc::new(Object::Array(vec![Object::Ref(top)])); + Self { doc, seen: Default::default(), pending: vec![(top, 0)] } + } +} + +impl Iterator for PageWalk<'_> { + type Item = ObjRef; + + fn next(&mut self) -> Option { + while let Some((kids, next)) = self.pending.last_mut() { + let Some(kid) = kids.as_array().and_then(|a| a.get(*next)) else { + self.pending.pop(); + continue; + }; + let node = kid.as_ref(); + *next += 1; + let Some(node) = node else { continue }; + if self.seen.len() >= 1_000_000 { + self.pending.clear(); + return None; + } + if !self.seen.insert(node) { + continue; + } + let obj = self.doc.get(node); + let Some(d) = obj.as_dict() else { continue }; + match d.get(b"Kids").map(|k| self.doc.resolve(k)) { + Some(kids) if d.name(b"Type") != Some(b"Page") => self.pending.push((kids, 0)), + _ => return Some(node), } - _ => out.push(node), } + None } - out } /// The AcroForm dictionary. diff --git a/crates/forms/src/tests.rs b/crates/forms/src/tests.rs index fef5bcfbd..968bfb201 100644 --- a/crates/forms/src/tests.rs +++ b/crates/forms/src/tests.rs @@ -2277,3 +2277,19 @@ fn document_scripts_read_each_name_tree_node_once() { ]); assert_eq!(document_scripts_named(&doc), [("init".to_string(), "var ready = true;".to_string())]); } + +#[test] +fn page_walk_skips_repeated_and_cyclic_kids_without_queueing_them() { + let mut doc = fixture(); + let root = ObjRef { num: 2, generation: 0 }; + let page = ObjRef { num: 3, generation: 0 }; + // A cycle back to the root, then the page 128 times over. + let mut kids = vec![Object::Ref(root)]; + kids.extend(std::iter::repeat_n(Object::Ref(page), 128)); + doc.update_dict(root, |d| d.set(b"Kids".to_vec(), Object::Array(kids))).unwrap(); + let mut walk = PageWalk::new(&doc, root); + assert_eq!(walk.next(), Some(page)); + assert!(walk.pending.len() <= 2, "repeated kids retained {} pending entries", walk.pending.len()); + assert_eq!(walk.next(), None); + assert_eq!(page_refs(&doc), [page]); +} diff --git a/crates/organize/src/lib.rs b/crates/organize/src/lib.rs index 99c228c60..fd1e4ecff 100644 --- a/crates/organize/src/lib.rs +++ b/crates/organize/src/lib.rs @@ -72,36 +72,74 @@ fn pages_root(doc: &Document) -> Result { /// All leaf pages in order, with their effective inherited attributes. fn walk(doc: &Document) -> Result, OrganizeError> { - let root = pages_root(doc)?; - let mut out = Vec::new(); - let mut seen = std::collections::HashSet::new(); - // (node, inherited attributes from ancestors) - let mut stack: Vec<(ObjRef, Dict)> = vec![(root, Dict::new())]; - while let Some((node, inherited)) = stack.pop() { - if !seen.insert(node) || seen.len() > 1_000_000 { - continue; // cycles in broken trees - } - let obj = doc.get(node); - let Some(d) = obj.as_dict() else { continue }; - let mut attrs = inherited.clone(); - for k in INHERITABLE { - if let Some(v) = d.get(k) { - attrs.set(k.to_vec(), v.clone()); + Ok(PageWalk::new(doc, pages_root(doc)?).collect()) +} + +/// For each of `INHERITABLE`, the nearest node that sets it, shared rather than copied. +type Owners = [Option>; 4]; + +/// Depth-first walk over the leaf pages under `top`. It keeps one cursor per open `/Kids` array, +/// never one entry per child, and carries inherited attributes as the nodes that set them, so a +/// child listed many times or shared between nodes is skipped by the seen check before anything +/// is queued or copied (`pdfcraft_model::pages` walks the same way). Cycles in broken trees end +/// at the seen check and at most a million nodes are visited. +struct PageWalk<'a> { + doc: &'a Document, + seen: std::collections::HashSet, + pending: Vec<(std::sync::Arc, usize, Owners)>, +} + +impl<'a> PageWalk<'a> { + fn new(doc: &'a Document, top: ObjRef) -> Self { + let top = std::sync::Arc::new(Object::Array(vec![Object::Ref(top)])); + Self { doc, seen: Default::default(), pending: vec![(top, 0, Default::default())] } + } +} + +impl Iterator for PageWalk<'_> { + type Item = (ObjRef, Dict); + + fn next(&mut self) -> Option<(ObjRef, Dict)> { + while let Some((kids, next, inherited)) = self.pending.last_mut() { + let Some(kid) = kids.as_array().and_then(|a| a.get(*next)) else { + self.pending.pop(); + continue; + }; + let node = kid.as_ref(); + *next += 1; + let Some(node) = node else { continue }; + if self.seen.len() >= 1_000_000 { + self.pending.clear(); + return None; } - } - let is_pages = d.name(b"Type") == Some(b"Pages") || (d.contains(b"Kids") && d.name(b"Type") != Some(b"Page")); - if is_pages { - let kids = d.get(b"Kids").map(|k| doc.resolve(k)).and_then(|k| k.as_array().cloned()).unwrap_or_default(); - for k in kids.iter().rev() { - if let Some(r) = k.as_ref() { - stack.push((r, attrs.clone())); + if !self.seen.insert(node) { + continue; + } + let obj = self.doc.get(node); + let Some(d) = obj.as_dict() else { continue }; + let mut owners = inherited.clone(); + for (k, owner) in INHERITABLE.iter().zip(&mut owners) { + if d.contains(k) { + *owner = Some(obj.clone()); } } - } else { - out.push((node, attrs)); + let is_pages = d.name(b"Type") == Some(b"Pages") || (d.contains(b"Kids") && d.name(b"Type") != Some(b"Page")); + if is_pages { + if let Some(kids) = d.get(b"Kids") { + self.pending.push((self.doc.resolve(kids), 0, owners)); + } + } else { + let mut attrs = Dict::new(); + for (k, owner) in INHERITABLE.iter().zip(&owners) { + if let Some(v) = owner.as_ref().and_then(|o| o.as_dict()).and_then(|o| o.get(k)) { + attrs.set(k.to_vec(), v.clone()); + } + } + return Some((node, attrs)); + } } + None } - Ok(out) } /// Leaf pages in document order. diff --git a/crates/organize/src/tests.rs b/crates/organize/src/tests.rs index 9bd3eb396..4f681367d 100644 --- a/crates/organize/src/tests.rs +++ b/crates/organize/src/tests.rs @@ -2821,3 +2821,26 @@ fn rotate_repeated_indices_accumulate_from_the_current_page_value() { } } } + +#[test] +fn page_walk_skips_repeated_and_cyclic_kids_without_copying_attributes() { + let mut doc = open(fixture()); + let (root, branch) = (ObjRef { num: 2, generation: 0 }, ObjRef { num: 3, generation: 0 }); + let (first, second) = (ObjRef { num: 4, generation: 0 }, ObjRef { num: 5, generation: 0 }); + // The branch node carries Resources for its pages: a cycle back to it, then both pages + // 128 times over. + let mut kids = vec![Object::Ref(branch)]; + kids.extend([Object::Ref(first), Object::Ref(second)].into_iter().cycle().take(256)); + doc.update_dict(branch, |d| d.set(b"Kids".to_vec(), Object::Array(kids))).unwrap(); + let mut walk = PageWalk::new(&doc, root); + let (page, attrs) = walk.next().unwrap(); + assert_eq!(page, first); + assert!(walk.pending.len() <= 3, "repeated kids retained {} pending entries", walk.pending.len()); + assert!(attrs.get(b"Resources").and_then(|r| r.as_dict()).is_some_and(|r| r.contains(b"Font")), "Resources from the branch"); + assert_eq!(attrs.get(b"Rotate").and_then(Object::as_int), Some(90), "Rotate from the root"); + assert_eq!(walk.next().map(|(p, _)| p), Some(second)); + let (_, last) = walk.next().unwrap(); + assert_eq!(last.get(b"MediaBox").and_then(Object::as_array).map(Vec::len), Some(4), "own MediaBox kept"); + assert!(walk.next().is_none()); + assert_eq!(labels(&doc), ["Page 1", "Page 2", "Page 3"]); +} From 586246d69aea65814bbdd83d43fb6dc61f5d992e Mon Sep 17 00:00:00 2001 From: oracola91 <118899664+oracola91@users.noreply.github.com> Date: Tue, 20 Oct 2026 11:23:29 +0300 Subject: [PATCH 17/34] organize: check the page's own MediaBox values in the walk test The root's MediaBox and the page's own both have four numbers, so the length check passed even if the page lost its own value. Compare the values instead. --- crates/organize/src/tests.rs | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/crates/organize/src/tests.rs b/crates/organize/src/tests.rs index 4f681367d..224863af1 100644 --- a/crates/organize/src/tests.rs +++ b/crates/organize/src/tests.rs @@ -2840,7 +2840,8 @@ fn page_walk_skips_repeated_and_cyclic_kids_without_copying_attributes() { assert_eq!(attrs.get(b"Rotate").and_then(Object::as_int), Some(90), "Rotate from the root"); assert_eq!(walk.next().map(|(p, _)| p), Some(second)); let (_, last) = walk.next().unwrap(); - assert_eq!(last.get(b"MediaBox").and_then(Object::as_array).map(Vec::len), Some(4), "own MediaBox kept"); + let media_box = last.get(b"MediaBox").and_then(Object::as_array).map(|m| m.iter().filter_map(Object::as_int).collect::>()); + assert_eq!(media_box, Some(vec![0, 0, 500, 500]), "own MediaBox kept over the root's"); assert!(walk.next().is_none()); assert_eq!(labels(&doc), ["Page 1", "Page 2", "Page 3"]); } From 766ff9d05a51e943396d42de1dcf8d5c12ce4fbd Mon Sep 17 00:00:00 2001 From: Hamza Date: Tue, 20 Oct 2026 14:47:52 +0500 Subject: [PATCH 18/34] Render: a doubled sign is one number, so a cm with it still has a matrix A number written with a doubled sign was read by the renderer's lexer as two numbers. `--280` set the sign, found no digit before the second `-`, and returned 0, leaving `-280` to be read next: a `cm` carrying it got seven operands, kept the last six, and the matrix `0 0 1 0 -280 10` hid everything drawn under it. The page rendered blank while text extraction still reported the text, because the glyphs were there and only the matrix was wrong. Measured on the report's own file, at 72 dpi, counting pixels darker than mid-grey: double_minus.pdf before: 0 dark px (blank) after: 581, leftmost at x=121 control.pdf (-280) 581, leftmost at x=121 -- unchanged so the two files now render identically, which is what the report asks for and what MuPDF does. One extra sign is absorbed and the value is negative if either sign was a minus, matching the COS lexer since #3299 (`--280` and `+-280` are both -280). `Number::skip` already consumed the whole token, so read and skip had disagreed about where the number ended. Fixes #3565. #3299 fixed the same input in the COS lexer only. --- vendor/hayro-syntax/src/object/number.rs | 44 ++++++++++++++++++------ 1 file changed, 33 insertions(+), 11 deletions(-) diff --git a/vendor/hayro-syntax/src/object/number.rs b/vendor/hayro-syntax/src/object/number.rs index bd37b19c3..7d3f05b4e 100644 --- a/vendor/hayro-syntax/src/object/number.rs +++ b/vendor/hayro-syntax/src/object/number.rs @@ -113,17 +113,24 @@ impl Readable<'_> for Number { #[inline(always)] fn read_inner(r: &mut Reader<'_>) -> Option { - let negative = match r.peek_byte()? { - b'-' => { - r.forward(); - true - } - b'+' => { - r.forward(); - false - } - _ => false, - }; + // PdfCraft patch: a doubled sign is one number, not two. `--280` left `negative` set with no + // digits read, so it returned 0 and the reader stopped on `-280`, which became a second + // number: a `cm` with it got seven operands, kept the last six, and the matrix + // `0 0 1 0 -280 10` hid everything drawn under it (#3565). `Number::skip` above already + // consumes the whole token, so read and skip disagreed about where the number ended. + // + // One extra sign is absorbed and the value is negative if either sign was a minus, which is + // what the COS lexer does since #3299 (`--280` and `+-280` are both -280). A third sign is + // left to the digit loop's trailing-minus arm, as before. + let mut negative = false; + let mut signs = 0; + while signs < 2 + && let Some(b @ (b'-' | b'+')) = r.peek_byte() + { + r.forward(); + negative |= b == b'-'; + signs += 1; + } // PdfCraft patch: high-precision float mantissa overflow (jsPDF double-precision coordinates). // Digits that don't fit in `mantissa` are dropped: past the dot they are only precision; @@ -348,6 +355,21 @@ mod tests { use crate::reader::Reader; use crate::reader::ReaderExt; + /// PdfCraft patch: a doubled sign is one number. `--280` returned 0 and left `-280` to be + /// read as a second number, so a `cm` carrying it got seven operands and a degenerate matrix + /// (#3565). The COS lexer reads these the same way since #3299. + #[test] + fn a_doubled_sign_is_one_number() { + for (src, want) in [("--280", -280), ("+-280", -280), ("-+280", -280), ("++280", 280), ("-280", -280), ("+280", 280)] { + let mut r = Reader::new(src.as_bytes()); + assert_eq!(r.read_without_context::().unwrap(), want, "{src}"); + assert!(r.at_end(), "{src}: the whole number was read, leaving nothing for a second one"); + } + // The fraction too, and a lone sign is still zero. + assert_eq!(Reader::new("--280.5".as_bytes()).read_without_context::().unwrap(), -280.5); + assert_eq!(Reader::new("-".as_bytes()).read_without_context::().unwrap(), 0); + } + #[test] fn int_1() { assert_eq!( From 0e8ebf00e6afd61b292f05b3bebe028e1b1a170e Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Thu, 15 Oct 2026 06:31:48 +0000 Subject: [PATCH 19/34] Issue #2366: a trailer whose /Prev is a reference still reaches the earlier sections The JHOVE PDF-HUL-157 files (tiff2pdf output) end their first trailer with `/Prev 297298 0 R`: the offset written as a reference. The chain reader took /Prev as an integer only, so the chain stopped after the first section without a word, and every object of the earlier sections (here pages 2-4 and their images) read as null. The document showed one page where Poppler and MuPDF show four, and a full save wrote the missing kids as null, so pages 2-4 were unreadable afterwards ("cycle in page tree", 1x1 renders). A reference-valued /Prev is now read as the offset it names, with a repair-log note; any other non-offset value is noted and ends the chain as before. Regression test: document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections. Verified on JA3-1998-PDF.pdf: four pages read, and after a Subject edit and a full save Poppler renders all four pages without a warning (before: "Kid object (page 2) is not an indirect reference (null)"). Co-Authored-By: Claude Fable 5.1 Claude-Session: https://claude.ai/code/session_011K1peyrh5MdZ6ttHkVq8Gh --- crates/cos/src/document.rs | 60 ++++++++++++++++++++++++++++++++++++-- 1 file changed, 58 insertions(+), 2 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index 639edf582..af9997716 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -1476,7 +1476,7 @@ impl Document { } } sections.push(XrefRevisionSection { primary, hybrid }); - next = sect_trailer.int(b"Prev").filter(|p| *p >= 0).map(|p| p as u64); + next = prev_offset(§_trailer, off, log); if trailer.is_none() { trailer = Some(sect_trailer); } @@ -1764,7 +1764,7 @@ impl Document { for (num, e) in sect { entries.entry(num).or_insert(e); } - next = sect_trailer.int(b"Prev").filter(|p| *p >= 0).map(|p| p as u64); + next = prev_offset(§_trailer, off, log); if trailer.is_none() { trailer = Some(sect_trailer); } @@ -1990,6 +1990,25 @@ impl Document { } } +/// The offset of the section before the one whose trailer this is (`/Prev`), if any. Some writers +/// (tiff2pdf; JHOVE's PDF-HUL-157 files) give it as a reference, `N 0 R`: other readers rebuild +/// such a file, and the sections before are at N, so it is read as that, with a note. Any other +/// value that is not an offset is noted and ends the chain. +fn prev_offset(trailer: &Dict, off: u64, log: &mut Vec) -> Option { + match trailer.get(b"Prev") { + Some(Object::Int(p)) if *p >= 0 => Some(*p as u64), + Some(Object::Ref(r)) => { + log.push(format!("the trailer at {off} gives /Prev as the reference {} 0 R; reading it as offset {}", r.num, r.num)); + Some(u64::from(r.num)) + } + Some(_) => { + log.push(format!("the trailer at {off} has a /Prev that is not an offset; the sections before it are unreachable")); + None + } + None => None, + } +} + pub(crate) fn find(hay: &[u8], needle: &[u8], from: usize, to: usize) -> Option { let to = to.min(hay.len()); if from >= to || needle.is_empty() || to - from < needle.len() { @@ -2834,6 +2853,43 @@ endobj } } + /// A trailer whose /Prev is written as the reference `N 0 R` (tiff2pdf did this; the files + /// are JHOVE's PDF-HUL-157 cases, #2366) used to end the cross-reference chain silently, so + /// every object of the earlier sections read as null: a four-page file showed one page, and + /// a full save wrote the three missing kids as null. The offset is read from the reference. + #[test] + fn a_prev_written_as_a_reference_still_reaches_the_earlier_sections() { + let base = build( + &["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [3 0 R] /Count 1 >>", "<< /Type /Page /Parent 2 0 R >>"], + "/Root 1 0 R", + ); + let prev = String::from_utf8_lossy(&base).rfind("xref\n0 4").unwrap(); + let update = |prev_entry: &str| { + let mut bytes = base.clone(); + let at = bytes.len(); + bytes.extend_from_slice(b"4 0 obj\n<< /Title (Update) >>\nendobj\n"); + let xref = bytes.len(); + bytes.extend_from_slice( + format!( + "xref\n4 1\n{at:010} 00000 n \ntrailer\n<< /Size 5 /Root 1 0 R /Info 4 0 R /Prev {prev_entry} >>\nstartxref\n{xref}\n%%EOF\n" + ) + .as_bytes(), + ); + Document::open(Arc::new(bytes)).unwrap() + }; + let good = update(&prev.to_string()); + let bad = update(&format!("{prev} 0 R")); + assert!(good.repair_log().is_empty(), "{:?}", good.repair_log()); + assert_eq!(bad.revisions().len(), 2, "{:?}", bad.repair_log()); + assert_eq!(bad.repair_log().len(), 1, "a note about the /Prev: {:?}", bad.repair_log()); + assert_eq!(bad.xref_entry(3), good.xref_entry(3), "the page of the first section is reached"); + assert_eq!(bad.get(ObjRef::new(3, 0)).as_dict().and_then(|d| d.name(b"Type")), Some(&b"Page"[..])); + assert_eq!( + bad.get(ObjRef::new(4, 0)).as_dict().and_then(|d| d.get(b"Title").and_then(|t| t.as_string().map(|s| s.to_text()))), + Some("Update".into()) + ); + } + #[test] fn damaged_xref_is_reconstructed() { let mut bytes = build(&["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>"], "/Root 1 0 R"); From c351febe85c03f7f6143697ca8f490de4ca03daa Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Fri, 16 Oct 2026 13:07:48 +0000 Subject: [PATCH 20/34] hayro-syntax: a trailer /Prev written as a reference reaches the earlier sections (PdfCraft patch 22) The cos reader learnt this in the previous commit, but the viewer and the save's reopen check count pages with hayro, which still stopped at the reference-valued /Prev: a four-page file showed one page, and the correctly written full save was refused as a page-count change. --- vendor/README.md | 2 +- vendor/hayro-syntax/src/xref.rs | 53 ++++++++++++++++++++++++++++++--- 2 files changed, 50 insertions(+), 5 deletions(-) diff --git a/vendor/README.md b/vendor/README.md index 824e84419..4a78d1263 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -17,7 +17,7 @@ In `hayro-interpret`, Type 3 glyph bounding boxes are transformed to device spac | hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour, planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk, interpolated, axis-aligned image is snapped to whole device pixels: shrunk once to its rounded device rectangle and drawn 1:1 at the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat); (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group with its background painted inside it as the initial backdrop, so a Multiply highlight blends with a dark reading-colours background instead of staying an opaque block (identity on white pages); (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349) | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_samples_the_same_columns`, `raster::tests::non_integer_minification_samples_colour_and_mask_from_the_same_columns`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::interpolated_minified_image_edges_snap_to_whole_pixels` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path` | | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap, page bitmap, halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer) | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | -| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306) | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`) +| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306) (22) `xref::prev_offset`: a trailer `/Prev` written as a reference `N 0 R` (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366) is read as offset N, as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`) (22) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections` | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it. | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed` | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it` | diff --git a/vendor/hayro-syntax/src/xref.rs b/vendor/hayro-syntax/src/xref.rs index 20e33a410..865fc19c2 100644 --- a/vendor/hayro-syntax/src/xref.rs +++ b/vendor/hayro-syntax/src/xref.rs @@ -758,6 +758,17 @@ fn populate_xref_impl<'a>(data: &'a [u8], pos: usize, xref_map: &mut XrefMap) -> populate_xref_impl_inner(data, pos, xref_map, &mut visited) } +/// The offset of the previous xref section (`Prev`). +// PdfCraft patch: (#2366) some writers (tiff2pdf, in JHOVE's PDF-HUL-157 files) give `Prev` as +// a reference `N 0 R`; its object number is the offset, as other readers take it. +fn prev_offset(trailer: &Dict<'_>) -> Option { + let prev = trailer + .get::(PREV) + .or_else(|| trailer.get_ref(PREV).map(|r| r.obj_number))?; + + usize::try_from(prev).ok() +} + /// Maximum number of allowed xref `Prev` pointers before we abort. const MAX_XREF_CHAIN_DEPTH: usize = 256; @@ -834,9 +845,9 @@ fn populate_from_xref_table<'a>( let mut max_obj = 0; - if let Some(prev) = trailer.get::(PREV) { + if let Some(prev) = prev_offset(&trailer) { // First insert the entries from any previous xref tables. - populate_xref_impl_inner(data, prev as usize, insert_map, visited)?; + populate_xref_impl_inner(data, prev, insert_map, visited)?; } // In hybrid files, entries in `XRefStm` should have higher priority, therefore we insert them @@ -880,9 +891,9 @@ fn populate_from_xref_stream<'a>( .read_with_context::>>(&ReaderContext::dummy())? .get(); - if let Some(prev) = stream.dict().get::(PREV) { + if let Some(prev) = prev_offset(stream.dict()) { // First insert the entries from any previous xref tables. - let _ = populate_xref_impl_inner(data, prev as usize, insert_map, visited)?; + let _ = populate_xref_impl_inner(data, prev, insert_map, visited)?; } let size = stream.dict().get::(SIZE)?; @@ -1357,6 +1368,40 @@ mod tests { let _result = populate_xref_impl(pdf.as_ref(), xref_pos, &mut xref_map); } + #[test] + fn prev_as_reference_reaches_the_earlier_section() { + let mut pdf = b"%PDF-1.0\n1 0 obj\n<< /Type /Catalog /Pages 2 0 R >>\nendobj\n".to_vec(); + let first_xref_pos = pdf.len(); + pdf.extend_from_slice( + format!( + "xref\n\ + 0 2\n\ + 0000000000 65535 f \n\ + 0000000009 00000 n \n\ + trailer\n<< /Size 2 /Root 1 0 R >>\n\ + startxref\n{first_xref_pos}\n%%EOF\n" + ) + .as_bytes(), + ); + let second_xref_pos = pdf.len(); + pdf.extend_from_slice( + format!( + "xref\n\ + 0 1\n\ + 0000000000 65535 f \n\ + trailer\n<< /Size 2 /Root 1 0 R /Prev {first_xref_pos} 0 R >>\n\ + startxref\n{second_xref_pos}\n%%EOF" + ) + .as_bytes(), + ); + + let mut xref_map = FxHashMap::default(); + let xref_pos = find_last_xref_pos(pdf.as_ref()).unwrap(); + assert_eq!(xref_pos, second_xref_pos); + populate_xref_impl(pdf.as_ref(), xref_pos, &mut xref_map).unwrap(); + assert!(xref_map.contains_key(&ObjectIdentifier::new(1, 0))); + } + #[test] fn find_last_xref_uses_last_startxref() { let pdf = b"%PDF-1.0\nstartxref\n5\n%%EOF\nstartxref\n42\n%%EOF"; From 3af350026ffbeaefcfd3cfdfdf79b84b4486affc Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Sun, 18 Oct 2026 11:25:29 +0000 Subject: [PATCH 21/34] Issue #2366: a /Prev reference to an object holding the offset reads that offset iText's indirectPrev.pdf writes /Prev 6 0 R, and object 6, listed in the same update, holds 470, the offset of the first section. Both readers took the object number as the offset, so cos found no section at 6, rebuilt the file from object headers and lost its revision list, and hayro's chain stopped there. When /Prev is a reference, the sections read so far (cos) or the section's own entries (hayro, which now reads them before following /Prev) are asked for that object first. If it holds a non-negative integer, that is the offset; otherwise the object number still is, as the tiff2pdf files need. Co-Authored-By: Claude Opus 5.5 (1M context) Claude-Session: https://claude.ai/code/session_01NH3itd85jPuM6wNjgcGfEH --- crates/cos/src/document.rs | 70 +++++++++++++++--- vendor/README.md | 2 +- vendor/hayro-syntax/src/xref.rs | 123 +++++++++++++++++++++++--------- 3 files changed, 152 insertions(+), 43 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index af9997716..7299056b2 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -1476,7 +1476,7 @@ impl Document { } } sections.push(XrefRevisionSection { primary, hybrid }); - next = prev_offset(§_trailer, off, log); + next = prev_offset(data, §_trailer, off, |num| sections.iter().find_map(|s| s.lookup(num, data)), log); if trailer.is_none() { trailer = Some(sect_trailer); } @@ -1764,7 +1764,7 @@ impl Document { for (num, e) in sect { entries.entry(num).or_insert(e); } - next = prev_offset(§_trailer, off, log); + next = prev_offset(data, §_trailer, off, |num| entries.get(&num).copied(), log); if trailer.is_none() { trailer = Some(sect_trailer); } @@ -1991,15 +1991,36 @@ impl Document { } /// The offset of the section before the one whose trailer this is (`/Prev`), if any. Some writers -/// (tiff2pdf; JHOVE's PDF-HUL-157 files) give it as a reference, `N 0 R`: other readers rebuild -/// such a file, and the sections before are at N, so it is read as that, with a note. Any other -/// value that is not an offset is noted and ends the chain. -fn prev_offset(trailer: &Dict, off: u64, log: &mut Vec) -> Option { +/// give it as a reference, `N 0 R`. When the sections read so far (`lookup`) place that object +/// and it holds an offset (iText's `indirectPrev.pdf`), that is the offset; otherwise, as the +/// tiff2pdf files in JHOVE's PDF-HUL-157 cases need, the sections before are at N. Either way the +/// reference is noted. Any other value that is not an offset is noted and ends the chain. +fn prev_offset(data: &[u8], trailer: &Dict, off: u64, lookup: impl Fn(u32) -> Option, log: &mut Vec) -> Option { match trailer.get(b"Prev") { Some(Object::Int(p)) if *p >= 0 => Some(*p as u64), Some(Object::Ref(r)) => { - log.push(format!("the trailer at {off} gives /Prev as the reference {} 0 R; reading it as offset {}", r.num, r.num)); - Some(u64::from(r.num)) + let held = match lookup(r.num) { + Some(XrefEntry::InFile { offset, generation }) if generation == r.generation => { + usize::try_from(offset).ok().and_then(|at| parse_indirect(data, at, &|_| None).ok()).and_then(|(at, value)| match value { + Object::Int(p) if at == *r && p >= 0 => Some(p as u64), + _ => None, + }) + } + _ => None, + }; + match held { + Some(p) => { + log.push(format!( + "the trailer at {off} gives /Prev as the reference {} {} R; reading the offset {p} it holds", + r.num, r.generation + )); + Some(p) + } + None => { + log.push(format!("the trailer at {off} gives /Prev as the reference {} 0 R; reading it as offset {}", r.num, r.num)); + Some(u64::from(r.num)) + } + } } Some(_) => { log.push(format!("the trailer at {off} has a /Prev that is not an offset; the sections before it are unreachable")); @@ -2890,6 +2911,39 @@ endobj ); } + /// iText's `indirectPrev.pdf` in miniature: the update's trailer gives `/Prev 4 0 R`, and + /// object 4, listed in that update, holds the offset of the first section. Read as offset 4 + /// (inside the header), the chain broke and the file was rebuilt without its revisions. + #[test] + fn a_prev_reference_to_an_object_holding_the_offset_reads_that_offset() { + let mut bytes = build( + &["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [3 0 R] /Count 1 >>", "<< /Type /Page /Parent 2 0 R >>"], + "/Root 1 0 R", + ); + let prev = String::from_utf8_lossy(&bytes).rfind("xref\n0 4").unwrap(); + let mut at = Vec::new(); + for (num, body) in + [(2, "<< /Type /Pages /Kids [3 0 R 5 0 R] /Count 2 >>".to_string()), (4, prev.to_string()), (5, "<< /Type /Page /Parent 2 0 R >>".into())] + { + at.push(bytes.len()); + bytes.extend_from_slice(format!("{num} 0 obj\n{body}\nendobj\n").as_bytes()); + } + let xref = bytes.len(); + bytes.extend_from_slice( + format!( + "xref\n2 1\n{:010} 00000 n \n4 2\n{:010} 00000 n \n{:010} 00000 n \ntrailer\n<< /Size 6 /Root 1 0 R /Prev 4 0 R >>\nstartxref\n{xref}\n%%EOF\n", + at[0], at[1], at[2] + ) + .as_bytes(), + ); + let doc = Document::open(Arc::new(bytes)).unwrap(); + assert_eq!(doc.revisions().len(), 2, "{:?}", doc.repair_log()); + assert_eq!(doc.repair_log().len(), 1, "a note about the /Prev, and no rebuild: {:?}", doc.repair_log()); + assert!(doc.repair_log()[0].contains(&format!("the offset {prev}")), "{:?}", doc.repair_log()); + assert_eq!(doc.get(ObjRef::new(3, 0)).as_dict().and_then(|d| d.name(b"Type")), Some(&b"Page"[..])); + assert_compact_matches_materialized(&doc); + } + #[test] fn damaged_xref_is_reconstructed() { let mut bytes = build(&["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>"], "/Root 1 0 R"); diff --git a/vendor/README.md b/vendor/README.md index 4a78d1263..87545e7e5 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -17,7 +17,7 @@ In `hayro-interpret`, Type 3 glyph bounding boxes are transformed to device spac | hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour, planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk, interpolated, axis-aligned image is snapped to whole device pixels: shrunk once to its rounded device rectangle and drawn 1:1 at the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat); (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group with its background painted inside it as the initial backdrop, so a Multiply highlight blends with a dark reading-colours background instead of staying an opaque block (identity on white pages); (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349) | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_samples_the_same_columns`, `raster::tests::non_integer_minification_samples_colour_and_mask_from_the_same_columns`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::interpolated_minified_image_edges_snap_to_whole_pixels` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path` | | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap, page bitmap, halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer) | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | -| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306) (22) `xref::prev_offset`: a trailer `/Prev` written as a reference `N 0 R` (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366) is read as offset N, as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`) (22) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections` +| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306) (22) `xref::prev_offset`: a trailer `/Prev` written as a reference `N G R` is read as the offset that object holds when the section's own entries place it (iText's `indirectPrev.pdf`; each section's entries are read before its `Prev` is followed), and otherwise as offset N (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366), as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`) (22) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `xref::tests::prev_as_reference_to_an_object_holding_the_offset_reads_that_offset`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections`, `document::tests::a_prev_reference_to_an_object_holding_the_offset_reads_that_offset` | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it. | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed` | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it` | diff --git a/vendor/hayro-syntax/src/xref.rs b/vendor/hayro-syntax/src/xref.rs index 865fc19c2..fe0337c73 100644 --- a/vendor/hayro-syntax/src/xref.rs +++ b/vendor/hayro-syntax/src/xref.rs @@ -759,14 +759,31 @@ fn populate_xref_impl<'a>(data: &'a [u8], pos: usize, xref_map: &mut XrefMap) -> } /// The offset of the previous xref section (`Prev`). -// PdfCraft patch: (#2366) some writers (tiff2pdf, in JHOVE's PDF-HUL-157 files) give `Prev` as -// a reference `N 0 R`; its object number is the offset, as other readers take it. -fn prev_offset(trailer: &Dict<'_>) -> Option { - let prev = trailer - .get::(PREV) - .or_else(|| trailer.get_ref(PREV).map(|r| r.obj_number))?; +// PdfCraft patch: (#2366) some writers give `Prev` as a reference `N G R`. When this section's +// own entries (`own`) place that object and it holds an offset (iText's `indirectPrev.pdf`), that +// is the offset; otherwise its object number is (tiff2pdf, in JHOVE's PDF-HUL-157 files), as +// `pdfcraft-cos` and other readers take it. +fn prev_offset(data: &[u8], trailer: &Dict<'_>, own: &XrefMap) -> Option { + if let Some(prev) = trailer.get::(PREV) { + return usize::try_from(prev).ok(); + } - usize::try_from(prev).ok() + let r = trailer.get_ref(PREV)?; + let id = ObjectIdentifier::new(r.obj_number, r.gen_number); + let held = match own.get(&id) { + Some(EntryType::Normal(at)) => { + let mut reader = Reader::new(data); + reader.jump(*at); + reader + .read_with_context::>(&ReaderContext::dummy()) + .filter(|o| o.id() == &id) + .map(|o| o.get()) + .filter(|p| *p >= 0) + } + _ => None, + }; + + usize::try_from(held.unwrap_or(r.obj_number)).ok() } /// Maximum number of allowed xref `Prev` pointers before we abort. @@ -844,17 +861,8 @@ fn populate_from_xref_table<'a>( reader.skip_white_spaces(); let mut max_obj = 0; - - if let Some(prev) = prev_offset(&trailer) { - // First insert the entries from any previous xref tables. - populate_xref_impl_inner(data, prev, insert_map, visited)?; - } - - // In hybrid files, entries in `XRefStm` should have higher priority, therefore we insert them - // after looking at `PREV`. - if let Some(xref_stm) = trailer.get::(XREF_STM) { - populate_xref_impl_inner(data, xref_stm as usize, insert_map, visited)?; - } + // PdfCraft patch: (#2366) this section's entries are read first, for `prev_offset`. + let mut own = XrefMap::default(); while let Some(header) = reader.read_without_context::() { reader.skip_white_spaces(); @@ -870,7 +878,7 @@ fn populate_from_xref_table<'a>( // Specification says we should ignore any object number > SIZE, but probably // not important? if entry.used { - insert_map.insert( + own.insert( ObjectIdentifier::new(obj_number as i32, entry.gen_number), EntryType::Normal(entry.offset), ); @@ -878,6 +886,19 @@ fn populate_from_xref_table<'a>( } } + if let Some(prev) = prev_offset(data, &trailer, &own) { + // First insert the entries from any previous xref tables. + populate_xref_impl_inner(data, prev, insert_map, visited)?; + } + + // In hybrid files, entries in `XRefStm` should have higher priority, therefore we insert them + // after looking at `PREV`. + if let Some(xref_stm) = trailer.get::(XREF_STM) { + populate_xref_impl_inner(data, xref_stm as usize, insert_map, visited)?; + } + + insert_map.extend(own); + Some(trailer.data()) } @@ -891,11 +912,6 @@ fn populate_from_xref_stream<'a>( .read_with_context::>>(&ReaderContext::dummy())? .get(); - if let Some(prev) = prev_offset(stream.dict()) { - // First insert the entries from any previous xref tables. - let _ = populate_xref_impl_inner(data, prev, insert_map, visited)?; - } - let size = stream.dict().get::(SIZE)?; let [f1_len, f2_len, f3_len] = stream.dict().get::<[u8; 3]>(W)?; @@ -912,6 +928,8 @@ fn populate_from_xref_stream<'a>( } let xref_data = stream.decoded().ok()?; + // PdfCraft patch: (#2366) this section's entries are read first, for `prev_offset`. + let mut own = XrefMap::default(); let mut xref_reader = Reader::new(xref_data.as_ref()); if let Some(arr) = stream.dict().get::>(INDEX) { @@ -925,21 +943,20 @@ fn populate_from_xref_stream<'a>( f1_len, f2_len, f3_len, - insert_map, + &mut own, )?; } } else { - xref_stream_subsection( - &mut xref_reader, - 0, - size, - f1_len, - f2_len, - f3_len, - insert_map, - )?; + xref_stream_subsection(&mut xref_reader, 0, size, f1_len, f2_len, f3_len, &mut own)?; } + if let Some(prev) = prev_offset(data, stream.dict(), &own) { + // First insert the entries from any previous xref tables. + let _ = populate_xref_impl_inner(data, prev, insert_map, visited)?; + } + + insert_map.extend(own); + Some(stream.dict().data()) } @@ -1402,6 +1419,44 @@ mod tests { assert!(xref_map.contains_key(&ObjectIdentifier::new(1, 0))); } + #[test] + fn prev_as_reference_to_an_object_holding_the_offset_reads_that_offset() { + let mut pdf = b"%PDF-1.0\n1 0 obj\n<< /Type /Catalog /Pages 2 0 R >>\nendobj\n".to_vec(); + let first_xref_pos = pdf.len(); + pdf.extend_from_slice( + format!( + "xref\n\ + 0 2\n\ + 0000000000 65535 f \n\ + 0000000009 00000 n \n\ + trailer\n<< /Size 2 /Root 1 0 R >>\n\ + startxref\n{first_xref_pos}\n%%EOF\n" + ) + .as_bytes(), + ); + // Object 3 holds the offset; read as offset 3, `Prev` would point inside the header. + let held_pos = pdf.len(); + pdf.extend_from_slice(format!("3 0 obj\n{first_xref_pos}\nendobj\n").as_bytes()); + let second_xref_pos = pdf.len(); + pdf.extend_from_slice( + format!( + "xref\n\ + 3 1\n\ + {held_pos:010} 00000 n \n\ + trailer\n<< /Size 4 /Root 1 0 R /Prev 3 0 R >>\n\ + startxref\n{second_xref_pos}\n%%EOF" + ) + .as_bytes(), + ); + + let mut xref_map = FxHashMap::default(); + let xref_pos = find_last_xref_pos(pdf.as_ref()).unwrap(); + assert_eq!(xref_pos, second_xref_pos); + populate_xref_impl(pdf.as_ref(), xref_pos, &mut xref_map).unwrap(); + assert!(xref_map.contains_key(&ObjectIdentifier::new(1, 0))); + assert!(xref_map.contains_key(&ObjectIdentifier::new(3, 0))); + } + #[test] fn find_last_xref_uses_last_startxref() { let pdf = b"%PDF-1.0\nstartxref\n5\n%%EOF\nstartxref\n42\n%%EOF"; From 386c92c034d03d3fbfa1c589c8583ec02249ad87 Mon Sep 17 00:00:00 2001 From: Hamza Date: Tue, 20 Oct 2026 17:12:11 +0500 Subject: [PATCH 22/34] Render: skipping a number covers the signs reading accepts `read_inner` now reads `-+280` and `++280` as well as `--280`, but `Number::skip` consumed one sign and then required a digit, a dot or a further `-`. So a skip failed where a read succeeded, and `ReaderExt::skip` puts the reader back where it started -- `Object::skip` would leave the number in place rather than step over it. `--280` was already covered, by the `b'-'` arm. One more sign is consumed, and the six sign forms are asserted skippable to the end of the token. Without this the test says `-+280: not skipped`. A third sign is still read and skipped differently (`---280` reads as 0 and skips whole), which was already so before this branch and is left alone. Found in review of #3630. --- vendor/hayro-syntax/src/object/number.rs | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/vendor/hayro-syntax/src/object/number.rs b/vendor/hayro-syntax/src/object/number.rs index 7d3f05b4e..9b89ed709 100644 --- a/vendor/hayro-syntax/src/object/number.rs +++ b/vendor/hayro-syntax/src/object/number.rs @@ -70,6 +70,13 @@ impl Number { impl Skippable for Number { fn skip(r: &mut Reader<'_>, _: bool) -> Option<()> { let has_sign = r.forward_if(|b| b == b'+' || b == b'-').is_some(); + // PdfCraft patch: one more sign, so that skipping a number covers what `read_inner` + // reads. `--280` was already covered by the `b'-'` arm below, but `-+280` and `++280` + // were not, and a skip that fails where a read succeeds puts the reader back where it + // started (#3630 review). + if has_sign { + r.forward_if(|b| b == b'+' || b == b'-'); + } // Some PDFs have weird trailing minuses, so try to accept those as well. match r.peek_byte()? { @@ -368,6 +375,15 @@ mod tests { // The fraction too, and a lone sign is still zero. assert_eq!(Reader::new("--280.5".as_bytes()).read_without_context::().unwrap(), -280.5); assert_eq!(Reader::new("-".as_bytes()).read_without_context::().unwrap(), 0); + + // Skipping covers what reading accepts: a skip that fails where a read succeeds puts the + // reader back where it started, so `Object::skip` would leave the number in place + // (#3630 review). + for src in ["--280", "+-280", "-+280", "++280", "-280", "+280", "--280.5"] { + let mut r = Reader::new(src.as_bytes()); + assert!(r.skip::(false).is_some(), "{src}: not skipped"); + assert!(r.at_end(), "{src}: only part of the number was skipped"); + } } #[test] From 801980ba352a0d1a9d2628a8948f2008a440ae96 Mon Sep 17 00:00:00 2001 From: wstock <1394687+wstock@users.noreply.github.com> Date: Tue, 20 Oct 2026 12:43:21 +0000 Subject: [PATCH 23/34] cos: the /Prev repair note names the reference's own generation (#2366) --- crates/cos/src/document.rs | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index 7299056b2..b1c0e1648 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -2017,7 +2017,10 @@ fn prev_offset(data: &[u8], trailer: &Dict, off: u64, lookup: impl Fn(u32) -> Op Some(p) } None => { - log.push(format!("the trailer at {off} gives /Prev as the reference {} 0 R; reading it as offset {}", r.num, r.num)); + log.push(format!( + "the trailer at {off} gives /Prev as the reference {} {} R; reading it as offset {}", + r.num, r.generation, r.num + )); Some(u64::from(r.num)) } } From 07f3c9749139bd82b60294a8a6164e3d89959b14 Mon Sep 17 00:00:00 2001 From: NetterSilver <133535268+Silvertree2010@users.noreply.github.com> Date: Tue, 20 Oct 2026 12:57:55 +0000 Subject: [PATCH 24/34] Keep filtered inline images whole when their data contains EI Fixes #3579 Co-Authored-By: Claude Haiku 5.5 Co-Authored-By: Claude Opus 5.5 --- crates/content/src/lib.rs | 38 ++++++++++++++++++++++++++++++++++++- crates/content/src/tests.rs | 26 +++++++++++++++++++++++++ 2 files changed, 63 insertions(+), 1 deletion(-) diff --git a/crates/content/src/lib.rs b/crates/content/src/lib.rs index c7c8bca9e..0d65519ec 100644 --- a/crates/content/src/lib.rs +++ b/crates/content/src/lib.rs @@ -225,7 +225,7 @@ fn inline_image(data: &[u8], pos: usize) -> Result<(Dict, Vec, usize), usize } // One whitespace byte separates ID from the data. let begin = (lx.pos + 1).min(data.len()); - let end = unfiltered_image_end(data, begin, &dict).or_else(|| end_of_image(data, begin)); + let end = unfiltered_image_end(data, begin, &dict).or_else(|| declared_image_end(data, begin, &dict)).or_else(|| image_data_end(data, begin)); Ok(match end { Some((end_data, end)) => (dict, data[begin..end_data].to_vec(), end), None => (dict, data[begin..].to_vec(), data.len()), @@ -302,6 +302,42 @@ fn unfiltered_image_end(data: &[u8], begin: usize, dict: &Dict) -> Option<(usize if data.get(ei..end) == Some(b"EI") && data.get(end).is_none_or(|b| is_ws(*b) || is_delim(*b)) { Some((end_data, end)) } else { None } } +/// The length declared by `/L` (or `/Length`, ISO 32000-2 §8.9.7), used when `EI` follows the +/// declared bytes after optional whitespace. A declared length without that `EI` is ignored. +fn declared_image_end(data: &[u8], begin: usize, dict: &Dict) -> Option<(usize, usize)> { + let length = dict.get(b"L").or_else(|| dict.get(b"Length"))?.as_int()?; + let end_data = begin.checked_add(usize::try_from(length).ok()?)?; + let after = data.get(end_data..)?; + let ei = end_data.checked_add(after.iter().position(|b| !is_ws(*b))?)?; + let end = ei.checked_add(2)?; + if data.get(ei..end) == Some(b"EI") && data.get(end).is_none_or(|b| is_ws(*b) || is_delim(*b)) { Some((end_data, end)) } else { None } +} + +/// Where inline image data of unknown length ends. A candidate `EI` followed by binary bytes before +/// the next `EI` lies inside the (compressed) data, unless another inline image starts first. +fn image_data_end(data: &[u8], begin: usize) -> Option<(usize, usize)> { + let mut candidate = end_of_image(data, begin)?; + loop { + let (end_data, end) = candidate; + let Some(next) = end_of_image(data, end) else { + return Some(candidate); + }; + let between = data.get(end..next.0).unwrap_or_default(); + if !between.iter().any(|b| !b.is_ascii()) || starts_inline_image(between) { + return Some((end_data, end)); + } + candidate = next; + } +} + +/// Whether `bytes` hold a whitespace-delimited `BI` operator. +fn starts_inline_image(bytes: &[u8]) -> bool { + bytes + .windows(2) + .enumerate() + .any(|(i, pair)| pair == b"BI" && (i == 0 || is_ws(bytes[i - 1])) && bytes.get(i + 2).is_none_or(|b| is_ws(*b) || is_delim(*b))) +} + /// Where inline image data starting at `begin` ends: (end of the data, position after `EI`). The /// data ends at whitespace + `EI` + (whitespace, a delimiter or the end). fn end_of_image(data: &[u8], begin: usize) -> Option<(usize, usize)> { diff --git a/crates/content/src/tests.rs b/crates/content/src/tests.rs index 1119bdaf9..599e6696e 100644 --- a/crates/content/src/tests.rs +++ b/crates/content/src/tests.rs @@ -151,6 +151,32 @@ fn filtered_unknown_and_damaged_inline_images_keep_the_existing_fallback() { } } +/// A zlib stream (one stored block) for 64 DeviceGray samples whose bytes hold " EI\n" at offset 41, +/// the way a stored deflate block copies incompressible samples verbatim (#3579). +const FLATE_WITH_EI: &str = "789c014000bfffed746159d50788f7ea604984cd08e00b4c8983207f56fe82c77ef8dfb8b9a3a586382045490a3120\ + 31206c2025630aaa4353e6813e233b96539c18bc3a2c5009e9d51b81"; + +#[test] +fn flate_inline_image_is_not_cut_at_an_ei_inside_its_compressed_bytes() { + let flate: Vec = (0..FLATE_WITH_EI.len()).step_by(2).map(|i| u8::from_str_radix(&FLATE_WITH_EI[i..i + 2], 16).unwrap()).collect(); + assert_eq!(flate.len(), 75); + assert_eq!(&flate[41..45], b" EI\n"); + for parameters in [b"/W 64 /H 1 /BPC 8 /CS /G /F /Fl".as_slice(), b"/W 64 /H 1 /BPC 8 /CS /G /F /Fl /L 75"] { + let mut src = b"q 200 0 0 200 100 300 cm BI ".to_vec(); + src.extend_from_slice(parameters); + src.extend_from_slice(b" ID "); + src.extend_from_slice(&flate); + src.extend_from_slice(b"\nEI Q BT /F1 24 Tf 72 200 Td (AFTER) Tj ET"); + let parsed = parse(&src); + let label = String::from_utf8_lossy(parameters); + assert_eq!(parsed.ops[2].inline.as_ref().unwrap().1.len(), 75, "{label}"); + let names: Vec<&[u8]> = parsed.ops.iter().map(|o| o.op.as_slice()).collect(); + assert_eq!(names, [&b"q"[..], b"cm", b"BI", b"Q", b"BT", b"Tf", b"Td", b"Tj", b"ET"], "{label}"); + assert_eq!(parsed.skipped, 0, "{label}"); + assert_eq!(parse(&serialize_ops(&parsed.ops)).ops[2].inline, parsed.ops[2].inline, "{label}"); + } +} + fn inline_pdf(content: &[u8], object_streams: bool) -> Vec { use pdfcraft_cos::{Document, SaveOptions, Stream, write_full}; From 9451e0652c79bc827c4b08092dbfebd17c1c3ba5 Mon Sep 17 00:00:00 2001 From: Ozt-y Date: Tue, 20 Oct 2026 18:57:43 +0530 Subject: [PATCH 25/34] fix(content): recognize ~>EI terminator for ASCII85 inline images and cover filter arrays --- crates/content/src/lib.rs | 33 +++++++++++++++++++-------------- crates/content/src/tests.rs | 23 ++++++++++++++++------- 2 files changed, 35 insertions(+), 21 deletions(-) diff --git a/crates/content/src/lib.rs b/crates/content/src/lib.rs index be6e7c36c..b65d157a0 100644 --- a/crates/content/src/lib.rs +++ b/crates/content/src/lib.rs @@ -219,32 +219,36 @@ fn inline_image(data: &[u8], pos: usize) -> Result<(Dict, Vec, usize), usize return Ok((dict, Vec::new(), data.len())); } // Without an `EI` after it, the image runs to the end of the stream. - return Err(end_of_image(data, at, false).map_or(data.len(), |(_, end)| end)); + return Err(end_of_image(data, at, b"").map_or(data.len(), |(_, end)| end)); }; dict.set(k, v); } // One whitespace byte separates ID from the data. let begin = (lx.pos + 1).min(data.len()); - let end = unfiltered_image_end(data, begin, &dict).or_else(|| end_of_image(data, begin, is_ascii_hex_filter(&dict))); + let end = unfiltered_image_end(data, begin, &dict).or_else(|| end_of_image(data, begin, filter_eod(&dict))); Ok(match end { Some((end_data, end)) => (dict, data[begin..end_data].to_vec(), end), None => (dict, data[begin..].to_vec(), data.len()), }) } -fn is_ascii_hex_filter(dict: &Dict) -> bool { +/// The end-of-data marker (`>` for ASCIIHex, `~>` for ASCII85) of the outer stream filter, if any. +fn filter_eod(dict: &Dict) -> &'static [u8] { for key in [b"Filter".as_slice(), b"F"] { if let Some(filter) = dict.get(key) { - match filter { - Object::Name(name) => return matches!(name.as_slice(), b"ASCIIHexDecode" | b"AHx"), - Object::Array(filters) => { - return matches!(filters.first().and_then(Object::as_name), Some(b"ASCIIHexDecode" | b"AHx")); - } - _ => return false, - } + let name = match filter { + Object::Name(name) => Some(name.as_slice()), + Object::Array(filters) => filters.first().and_then(Object::as_name), + _ => None, + }; + return match name { + Some(b"ASCIIHexDecode" | b"AHx") => b">", + Some(b"ASCII85Decode" | b"A85") => b"~>", + _ => b"", + }; } } - false + b"" } /// An unfiltered image's samples determine its length, including byte padding per row. @@ -318,13 +322,14 @@ fn unfiltered_image_end(data: &[u8], begin: usize, dict: &Dict) -> Option<(usize } /// Where inline image data starting at `begin` ends: (end of the data, position after `EI`). The -/// data ends at whitespace (or `>` for ASCIIHex) + `EI` + (whitespace, a delimiter or the end). -fn end_of_image(data: &[u8], begin: usize, allow_gt_ei: bool) -> Option<(usize, usize)> { +/// data ends at whitespace (or `eod`: `>` for ASCIIHex, `~>` for ASCII85) + `EI` + (whitespace, +/// a delimiter or the end). +fn end_of_image(data: &[u8], begin: usize, eod: &[u8]) -> Option<(usize, usize)> { let mut i = begin; while i + 2 <= data.len() { if data[i] == b'E' && data.get(i + 1) == Some(&b'I') - && (i == begin || is_ws(data[i - 1]) || (allow_gt_ei && data[i - 1] == b'>')) + && (i == begin || is_ws(data[i - 1]) || (!eod.is_empty() && data[begin..i].ends_with(eod))) && data.get(i + 2).is_none_or(|b| is_ws(*b) || is_delim(*b)) { let end_data = if i > begin && is_ws(data[i - 1]) { i - 1 } else { i }; diff --git a/crates/content/src/tests.rs b/crates/content/src/tests.rs index e450d5a5f..911cce53d 100644 --- a/crates/content/src/tests.rs +++ b/crates/content/src/tests.rs @@ -394,6 +394,11 @@ fn inline_image_terminator_without_preceding_whitespace() { let cases: &[(&[u8], &[u8])] = &[ (b"BI /W 4 /H 1 /BPC 8 /CS /G /F /AHx ID 2060A0E0>EI Q BT /F1 18 Tf (After) Tj ET", b"2060A0E0>"), (b"BI /W 4 /H 1 /BPC 8 /CS /G /Filter /ASCIIHexDecode ID 2060A0E0>EI Q BT /F1 18 Tf (After) Tj ET", b"2060A0E0>"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G /F [/AHx] ID 2060A0E0>EI Q BT /F1 18 Tf (After) Tj ET", b"2060A0E0>"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G /F [/AHx /RL] ID 2060A0E0>EI Q BT /F1 18 Tf (After) Tj ET", b"2060A0E0>"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G /F /A85 ID 5s,^g~>EI Q BT /F1 18 Tf (After) Tj ET", b"5s,^g~>"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G /Filter /ASCII85Decode ID 5s,^g~>EI Q BT /F1 18 Tf (After) Tj ET", b"5s,^g~>"), + (b"BI /W 4 /H 1 /BPC 8 /CS /G /F [/A85 /RL] ID 5s,^g~>EI Q BT /F1 18 Tf (After) Tj ET", b"5s,^g~>"), (b"BI /W 4 /H 1 /BPC 8 /CS /G ID ab()EI Q BT /F1 18 Tf (After) Tj ET", b"ab()"), (b"BI /W 4 /H 1 /BPC 8 /CS /G ID ab[]EI Q BT /F1 18 Tf (After) Tj ET", b"ab[]"), (b"BI /W 4 /H 1 /BPC 8 /CS /G ID \x20\x60\xa0\xe0EI Q BT /F1 18 Tf (After) Tj ET", b"\x20\x60\xa0\xe0"), @@ -416,11 +421,15 @@ fn inline_image_terminator_without_preceding_whitespace() { #[test] fn ascii85_inline_image_does_not_terminate_at_gt_ei_inside_payload() { - let src = b"BI /W 4 /H 1 /BPC 8 /CS /G /F /A85 ID !!!!>EI !!~>\nEI Q BT /F1 18 Tf (After) Tj ET"; - let p = parse(src); - assert_eq!(p.skipped, 0); - let names: Vec<&[u8]> = p.ops.iter().map(|o| o.op.as_slice()).collect(); - assert_eq!(names, [&b"BI"[..], b"Q", b"BT", b"Tf", b"Tj", b"ET"]); - let (_, img) = p.ops[0].inline.as_ref().unwrap(); - assert_eq!(img, b"!!!!>EI !!~>"); + for src in [ + b"BI /W 4 /H 1 /BPC 8 /CS /G /F /A85 ID !!!!>EI !!~>\nEI Q BT /F1 18 Tf (After) Tj ET".as_slice(), + b"BI /W 4 /H 1 /BPC 8 /CS /G /F [/A85 /AHx] ID !!!!>EI !!~>\nEI Q BT /F1 18 Tf (After) Tj ET".as_slice(), + ] { + let p = parse(src); + assert_eq!(p.skipped, 0); + let names: Vec<&[u8]> = p.ops.iter().map(|o| o.op.as_slice()).collect(); + assert_eq!(names, [&b"BI"[..], b"Q", b"BT", b"Tf", b"Tj", b"ET"]); + let (_, img) = p.ops[0].inline.as_ref().unwrap(); + assert_eq!(img, b"!!!!>EI !!~>"); + } } From b85af272c3223932addfe879935db25bff484719 Mon Sep 17 00:00:00 2001 From: NetterSilver <133535268+Silvertree2010@users.noreply.github.com> Date: Tue, 20 Oct 2026 14:11:33 +0000 Subject: [PATCH 26/34] Accept a delimiter after EI when ending an inline image Fixes #3578 Co-Authored-By: Claude Haiku 5.5 Co-Authored-By: Claude Sonnet 5.5 --- vendor/hayro-syntax/src/content/mod.rs | 46 ++++++++++++++++++++++++-- 1 file changed, 43 insertions(+), 3 deletions(-) diff --git a/vendor/hayro-syntax/src/content/mod.rs b/vendor/hayro-syntax/src/content/mod.rs index 8e47324d0..9d5faaab2 100644 --- a/vendor/hayro-syntax/src/content/mod.rs +++ b/vendor/hayro-syntax/src/content/mod.rs @@ -40,7 +40,7 @@ use crate::object::name::{Name, skip_name_like}; use crate::object::{Array, Null, Number, Object, Stream}; use crate::reader::Reader; use crate::reader::{Readable, ReaderContext, ReaderExt, Skippable}; -use crate::trivia::{Comment, is_white_space_character}; +use crate::trivia::{Comment, is_regular_character, is_white_space_character}; use crate::util::find_needle; use core::array; use core::fmt::{Debug, Formatter}; @@ -246,11 +246,12 @@ impl<'a> UntypedIter<'a> { self.reader.read_bytes(pos)?; if self.reader.peek_bytes(2) == Some(b"EI") { - // If the following character is not a whitespace character, then we are in a ASCII-85 stream. + // If the following character is a regular character, then the token + // is not `EI` (white space and delimiters both end it). if self .reader .peek_bytes(3) - .is_some_and(|b| !is_white_space_character(b[2])) + .is_some_and(|b| is_regular_character(b[2])) { self.reader.read_bytes(3)?; @@ -872,6 +873,45 @@ mod tests { assert_eq!(operators(data), [b"TJ".to_vec(), b"BI".to_vec(), b"q".to_vec(), b"Q".to_vec()]); } + #[test] + fn inline_image_ei_may_be_followed_by_a_delimiter() { + let mut image = b"BI /W 16 /H 4 /BPC 8 /CS /G ID ".to_vec(); + image.extend([0x80; 64]); + image.extend(b"\nEI"); + let tail = b" Q BT /F1 24 Tf 72 200 Td (AFTER) Tj ET"; + let ops = |suffix: &[u8]| { + let mut data = image.clone(); + data.extend(suffix); + data.extend(tail); + operators(&data) + }; + let control = ops(b""); + assert_eq!(control.len(), 7); + assert_eq!(ops(b"%c\n"), control); + assert_eq!(ops(b"[] 0 d"), [&control[..1], &[b"d".to_vec()], &control[1..]].concat()); + } + + #[test] + fn inline_image_ei_followed_by_a_regular_character_is_not_the_end() { + let data = b"BI /W 1 /H 1 /BPC 8 /CS /G ID \x80 EIQ \x81 EI Q BT (a) Tj ET"; + // `EIQ` is one token and stays part of the image data. + let expected: Vec> = + ["BI", "Q", "BT", "Tj", "ET"].iter().map(|s| s.as_bytes().to_vec()).collect(); + assert_eq!(operators(data), expected); + } + + #[test] + fn ascii85_inline_image_data_may_contain_ei_before_a_delimiter() { + for glue in [&b"EI["[..], b"EI/", b"EI%", b"EI(", b"EI<"] { + let mut data = b"BI /W 4 /H 4 /BPC 8 /CS /G /F /A85 ID 9jqo^".to_vec(); + data.extend(glue); + data.extend(b"87cURD]i,\"Ebo80~>EI Q BT /F1 12 Tf (AFTER) Tj ET"); + let expected: Vec> = + ["BI", "Q", "BT", "Tf", "Tj", "ET"].iter().map(|s| s.as_bytes().to_vec()).collect(); + assert_eq!(operators(&data), expected, "{}", String::from_utf8_lossy(glue)); + } + } + #[test] fn tokenizer_interface_preserves_every_byte_boundary() { let data = b"% comment\n1 -2.5 /Name /A#20B (a\\(b\\) \\101) <4869> [1 2] << /Key (value) >> [(a) 1] TJ BI /W 1 /H 1 /BPC 8 /CS /DeviceGray ID \x00 EI q Q"; From 4e94aa1cbb8f344be40304b2110ffe839e44ceff Mon Sep 17 00:00:00 2001 From: Marco Arnulfo Date: Wed, 14 Oct 2026 19:31:16 +0200 Subject: [PATCH 27/34] fix(cos): find objects a broken cross-reference table leaves out (fixes #1628) --- crates/cos/src/document.rs | 193 +++++++++++++++++++++++++++++++---- crates/organize/src/tests.rs | 27 +++++ 2 files changed, 199 insertions(+), 21 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index 639edf582..e9818454b 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -386,6 +386,10 @@ pub struct Document { revisions: Arc>, /// See [`Self::xref_oddities`]. xref_oddities: Arc>, + /// Where the file's object headers are, as (offset, generation) of the last header of each + /// number. Filled only when the cross-reference data proved inconsistent; see + /// `fill_xref_gaps`. + located: Arc>, repair_log: Arc>, cache: Arc>>>, objstms: Arc>>>, @@ -508,6 +512,7 @@ impl Document { trailer: Dict::new(), revisions: Arc::new(Vec::new()), xref_oddities: Arc::default(), + located: Arc::default(), repair_log: Arc::new(Vec::new()), cache: Arc::default(), objstms: Arc::default(), @@ -589,6 +594,7 @@ impl Document { trailer: Dict::new(), revisions: Arc::new(Vec::new()), xref_oddities: Arc::default(), + located: Arc::default(), repair_log: Arc::new(Vec::new()), cache: Arc::default(), objstms: Arc::default(), @@ -616,6 +622,7 @@ impl Document { doc.trailer = trailer; doc.revisions = Arc::new(revisions); doc.xref_oddities = Arc::new(oddities); + doc.fill_xref_gaps(&mut log); // Authenticate before checking the catalog: it may sit in an encrypted object // stream, which reads as garbage until the key is known. if let Some(enc) = doc.trailer.get(b"Encrypt").cloned() { @@ -1151,6 +1158,7 @@ impl Document { // Offsets relative to a shifted header, or simply wrong: try both fixes. at(off) .or_else(|| if self.header_offset == 0 { None } else { at(off.saturating_add(self.header_offset)) }) + .or_else(|| self.located.get(&num).and_then(|(o, _)| usize::try_from(*o).ok()).and_then(at)) .or_else(|| self.scan_for(num).map(|(id, o)| self.decrypted(id, o))) .ok_or(CosError::MissingObject(num)) } @@ -1906,32 +1914,48 @@ impl Document { Ok((s.dict.clone(), true)) } + /// Cross-reference data that leaves out objects the file contains: a subsection numbered one + /// too low (object 0 in use, the last object without an entry), or an in-use entry at offset 0 + /// that nothing else replaces. (A number no section lists is not evidence: it is a null + /// object, §7.3.10, and valid files have them below an oversized `/Size`.) Viewers rebuild + /// the table and show these objects; reading them as null would drop them from every copy. + /// When one of those shows, the file is scanned once for object headers, numbers without an + /// entry take the last header's location, and wrong offsets are looked up there (`load`). + /// Existing entries are never replaced: an object freed by an update stays freed. + fn fill_xref_gaps(&mut self, log: &mut Vec) { + let entries = &self.entries; + let shifted = matches!(entries.get(&0), Some(XrefEntry::InFile { .. })); + let set_aside = self.xref_oddities.iter().any(|(_, num)| !entries.contains_key(num)); + if !(shifted || set_aside) { + return; + } + let mut located = BTreeMap::new(); + for (num, offset, generation) in object_headers(&self.data) { + located.insert(num, (offset, generation)); + } + let mut filled = entries.as_ref().clone(); + let mut found = 0usize; + for (&num, &(offset, generation)) in &located { + if num != 0 && !filled.contains_key(&num) { + filled.insert(num, XrefEntry::InFile { offset, generation }); + found += 1; + } + } + if found > 0 { + log.push(format!("the cross-reference data leaves out {found} objects the file contains; located them by their object headers")); + self.entries = Arc::new(filled); + } + self.located = Arc::new(located); + } + /// Rebuild the cross-reference index by scanning for `N G obj` headers (§7.5.4 recovery). fn reconstruct(&mut self, log: &mut Vec) -> Result<(), CosError> { let data = self.data.clone(); let mut entries: BTreeMap = BTreeMap::new(); - let mut i = 0; let mut objstms = Vec::new(); - while i < data.len() { - // Object headers start at a line start (or after whitespace). - let at_start = i == 0 || is_whitespace(data[i - 1]) || data[i - 1] == b'>'; - if at_start && data[i].is_ascii_digit() { - let mut lx = Lexer::new(&data, i); - let n = lx.token(); - lx.skip_ws(); - let g = lx.token(); - if !n.is_empty() && !g.is_empty() && n.iter().all(u8::is_ascii_digit) && g.iter().all(u8::is_ascii_digit) && lx.eat_keyword(b"obj") { - let num = std::str::from_utf8(n).ok().and_then(|s| s.parse::().ok()); - let generation = std::str::from_utf8(g).ok().and_then(|s| s.parse::().ok()).unwrap_or(0); - if let Some(num) = num { - entries.insert(num, XrefEntry::InFile { offset: i as u64, generation }); - objstms.push(num); - } - i = lx.pos; - continue; - } - } - i += 1; + for (num, offset, generation) in object_headers(&data) { + entries.insert(num, XrefEntry::InFile { offset, generation }); + objstms.push(num); } if entries.is_empty() { return Err(CosError::Syntax { offset: 0, detail: "no objects found".into() }); @@ -1990,6 +2014,33 @@ impl Document { } } +/// Every `N G obj` header in `data`, in file order, as (number, offset, generation). +fn object_headers(data: &[u8]) -> Vec<(u32, u64, u16)> { + let mut headers = Vec::new(); + let mut i = 0; + while i < data.len() { + // Object headers start at a line start (or after whitespace). + let at_start = i == 0 || is_whitespace(data[i - 1]) || data[i - 1] == b'>'; + if at_start && data[i].is_ascii_digit() { + let mut lx = Lexer::new(data, i); + let n = lx.token(); + lx.skip_ws(); + let g = lx.token(); + if !n.is_empty() && !g.is_empty() && n.iter().all(u8::is_ascii_digit) && g.iter().all(u8::is_ascii_digit) && lx.eat_keyword(b"obj") { + let num = std::str::from_utf8(n).ok().and_then(|s| s.parse::().ok()); + let generation = std::str::from_utf8(g).ok().and_then(|s| s.parse::().ok()).unwrap_or(0); + if let Some(num) = num { + headers.push((num, i as u64, generation)); + } + i = lx.pos; + continue; + } + } + i += 1; + } + headers +} + pub(crate) fn find(hay: &[u8], needle: &[u8], from: usize, to: usize) -> Option { let to = to.min(hay.len()); if from >= to || needle.is_empty() || to - from < needle.len() { @@ -2297,6 +2348,106 @@ endobj assert!(!doc.repair_log().iter().any(|l| l.contains("numbered from 1"))); } + /// A one-page scan: the page draws image 5 with content stream 4. + fn scanned_page() -> Vec { + build( + &[ + "<< /Type /Catalog /Pages 2 0 R >>", + "<< /Type /Pages /Kids [3 0 R] /Count 1 >>", + "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 10 10] /Resources << /XObject << /Im0 5 0 R >> >> /Contents 4 0 R >>", + "<< /Length 21 >>\nstream\n10 0 0 10 0 0 cm /Im0 Do\nendstream", + "<< /Type /XObject /Subtype /Image /Width 1 /Height 1 /ColorSpace /DeviceGray /BitsPerComponent 8 /Length 1 >>\nstream\nA\nendstream", + ], + "/Root 1 0 R", + ) + } + + /// `bytes` with its cross-reference table replaced by `table` (the lines after `xref`). + fn with_table(bytes: &[u8], table: &[String], size: usize) -> Vec { + let text = String::from_utf8_lossy(bytes).into_owned(); + let at = text.rfind("\nxref\n").unwrap() + 1; + let mut out = bytes[..at].to_vec(); + out.extend_from_slice(format!("xref\n{}\ntrailer\n<< /Size {size} /Root 1 0 R >>\nstartxref\n{at}\n%%EOF\n", table.join("\n")).as_bytes()); + out + } + + /// The table's entry lines, object 0 first. + fn table_entries(bytes: &[u8]) -> Vec { + let text = String::from_utf8_lossy(bytes).into_owned(); + let at = text.rfind("\nxref\n").unwrap() + 1; + text[at..].lines().skip(2).take_while(|l| !l.starts_with("trailer")).map(str::to_string).collect() + } + + /// The classic off-by-one: the subsection says it starts at 0 but lists objects 1 to 5, so + /// the last object has no entry. Viewers rebuild the table and show the page; a reader that + /// took the missing entry for a null object dropped the image when pages were copied (#1628). + #[test] + fn an_off_by_one_xref_subsection_still_finds_the_last_object() { + let base = scanned_page(); + let entries = table_entries(&base); + let mut table = vec!["0 5".to_string()]; + table.extend(entries[1..6].iter().cloned()); + let doc = Document::open(Arc::new(with_table(&base, &table, 5))).unwrap(); + let image = doc.get(ObjRef::new(5, 0)); + assert_eq!(image.as_dict().and_then(|d| d.name(b"Subtype")), Some(&b"Image"[..]), "{image:?}"); + assert!(matches!(&*image, Object::Stream(_))); + assert!(matches!(&*doc.get(ObjRef::new(4, 0)), Object::Stream(s) if &s.raw[..] == b"10 0 0 10 0 0 cm /Im0 Do")); + assert!(doc.repair_log().iter().any(|l| l.contains("object headers")), "{:?}", doc.repair_log()); + } + + /// An "in use" entry at offset 0 says nothing about where the object is: the object is + /// found by its header, and the repair is recorded. + #[test] + fn an_object_listed_at_offset_zero_is_found_by_its_header() { + let base = scanned_page(); + for num in [4usize, 5] { + let mut table = vec!["0 6".to_string()]; + table.extend(table_entries(&base)); + table[num + 1] = "0000000000 00000 n ".into(); + let doc = Document::open(Arc::new(with_table(&base, &table, 6))).unwrap(); + assert!(matches!(&*doc.get(ObjRef::new(num as u32, 0)), Object::Stream(_)), "object {num}"); + assert!(doc.repair_log().iter().any(|l| l.contains("object headers")), "object {num}: {:?}", doc.repair_log()); + } + } + + /// A valid table that simply does not list every number below `/Size` is not repaired: an + /// unlisted number is a null object (§7.3.10), even when an old body with that number is + /// still in the file. + #[test] + fn a_consistent_table_with_unlisted_numbers_is_taken_as_it_is() { + let base = build(&["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>", "(three)"], "/Root 1 0 R"); + let at = String::from_utf8_lossy(&base).rfind("\nxref\n").unwrap() + 1; + let mut bytes = base[..at].to_vec(); + bytes.extend_from_slice(b"4 0 obj\n(stale)\nendobj\n"); + let mut table = vec!["0 4".to_string()]; + table.extend(table_entries(&base)); + let doc = Document::open(Arc::new(with_table(&[&bytes[..], &base[at..]].concat(), &table, 6))).unwrap(); + assert!(doc.repair_log().is_empty(), "{:?}", doc.repair_log()); + assert_eq!(*doc.get(ObjRef::new(4, 0)), Object::Null); + assert!(matches!(&*doc.get(ObjRef::new(3, 0)), Object::String(_))); + } + + /// Filling the gaps of a damaged table never brings back an object that an incremental + /// update freed: its body is still in the file, but the newer section says it is gone. + #[test] + fn an_object_freed_by_an_update_stays_freed_when_the_table_is_repaired() { + let base = scanned_page(); + // The original table lists the content stream at offset 0, so the gaps get filled. + let mut table = vec!["0 6".to_string()]; + table.extend(table_entries(&base)); + table[5] = "0000000000 00000 n ".into(); + let mut bytes = with_table(&base, &table, 6); + let prev = String::from_utf8_lossy(&bytes).rfind("\nxref\n").unwrap() + 1; + let section = bytes.len(); + bytes.extend_from_slice( + format!("xref\n5 1\n0000000000 00001 f \ntrailer\n<< /Size 6 /Root 1 0 R /Prev {prev} >>\nstartxref\n{section}\n%%EOF\n").as_bytes(), + ); + let doc = Document::open(Arc::new(bytes)).unwrap(); + assert!(matches!(&*doc.get(ObjRef::new(4, 0)), Object::Stream(_)), "the gap is filled"); + assert_eq!(*doc.get(ObjRef::new(5, 0)), Object::Null, "the freed image stays freed"); + assert_eq!(doc.xref_entry(5), Some(XrefEntry::Free { next_generation: 1 })); + } + #[test] fn opening_an_exhausted_classic_xref_repairs_and_allocates() { // The extra in-use row need not be loaded to exhaust the next object number. diff --git a/crates/organize/src/tests.rs b/crates/organize/src/tests.rs index 9bd3eb396..00b2abc76 100644 --- a/crates/organize/src/tests.rs +++ b/crates/organize/src/tests.rs @@ -774,6 +774,33 @@ fn resources_split_between_page_and_tree_travel_together() { } } +#[test] +fn a_scan_whose_xref_misses_the_image_combines_with_the_image() { + // A cheap image-to-PDF converter's table: "0 5" followed by the entries of objects 1 to 5, + // so the image (object 5) has none. Viewers rebuild the table and show the scan; the + // combined copy got a null image and came out blank (#1628). + let b: Vec = vec![ + "<< /Type /Catalog /Pages 2 0 R >>".into(), + "<< /Type /Pages /Kids [3 0 R] /Count 1 >>".into(), + "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 100 100] /Resources << /XObject << /Im0 5 0 R >> >> /Contents 4 0 R >>".into(), + "<< /Length 26 >>\nstream\n100 0 0 100 0 0 cm /Im0 Do\nendstream".into(), + "<< /Type /XObject /Subtype /Image /Width 1 /Height 1 /ColorSpace /DeviceGray /BitsPerComponent 8 /Length 1 >>\nstream\nA\nendstream".into(), + ]; + let good = build(&b, "/Root 1 0 R"); + let text = String::from_utf8_lossy(&good).into_owned(); + let at = text.rfind("\nxref\n").unwrap() + 1; + let entries: Vec<&str> = text[at..].lines().skip(3).take(5).collect(); + let mut bytes = good[..at].to_vec(); + bytes.extend_from_slice(format!("xref\n0 5\n{}\ntrailer\n<< /Size 5 /Root 1 0 R >>\nstartxref\n{at}\n%%EOF\n", entries.join("\n")).as_bytes()); + let src = open(bytes); + let out = full_roundtrip(&combine(&[("scan", &src)]).unwrap()); + let res = out.dict(page_dict(&out, 0).get(b"Resources").unwrap()).unwrap(); + let image = res.get(b"XObject").and_then(|x| out.dict(x)).and_then(|d| d.get(b"Im0").map(|v| out.resolve(v))).expect("the page lists its image"); + let Object::Stream(image) = &*image else { panic!("the image is a stream, not {image:?}") }; + assert_eq!(image.dict.name(b"Subtype"), Some(&b"Image"[..])); + assert_eq!(image.decoded().unwrap(), b"A"); +} + #[test] fn links_are_rewired_to_copied_pages_or_dropped() { let a = doc_a(); From 062270da50a411cfd811c752dc935e9be7162552 Mon Sep 17 00:00:00 2001 From: Marco Arnulfo Date: Thu, 15 Oct 2026 21:23:28 +0200 Subject: [PATCH 28/34] fix(cos): fill xref gaps on the compact cross-reference index (refs #1628) main now keeps valid tables as a compact index; the gap filling reads entries through it and materializes them only when it repairs. --- crates/cos/src/document.rs | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index e9818454b..e102993ae 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -268,6 +268,20 @@ impl XrefIndex { } } + /// Every entry, newest section first, as the materialized reader keeps them. + fn to_map(&self, file: &[u8]) -> BTreeMap { + match self { + Self::Materialized(entries) => entries.as_ref().clone(), + Self::Compact { revisions, .. } => { + let mut numbers = Vec::new(); + for revision in revisions { + revision.extend_numbers(&mut numbers); + } + numbers.into_iter().filter_map(|num| self.lookup(num, file).map(|entry| (num, entry))).collect() + } + } + } + fn max_num(&self) -> u32 { match self { Self::Compact { max_num, .. } => *max_num, @@ -1924,8 +1938,8 @@ impl Document { /// Existing entries are never replaced: an object freed by an update stays freed. fn fill_xref_gaps(&mut self, log: &mut Vec) { let entries = &self.entries; - let shifted = matches!(entries.get(&0), Some(XrefEntry::InFile { .. })); - let set_aside = self.xref_oddities.iter().any(|(_, num)| !entries.contains_key(num)); + let shifted = matches!(entries.lookup(0, &self.data), Some(XrefEntry::InFile { .. })); + let set_aside = self.xref_oddities.iter().any(|(_, num)| entries.lookup(*num, &self.data).is_none()); if !(shifted || set_aside) { return; } @@ -1933,7 +1947,7 @@ impl Document { for (num, offset, generation) in object_headers(&self.data) { located.insert(num, (offset, generation)); } - let mut filled = entries.as_ref().clone(); + let mut filled = entries.to_map(&self.data); let mut found = 0usize; for (&num, &(offset, generation)) in &located { if num != 0 && !filled.contains_key(&num) { @@ -1943,7 +1957,7 @@ impl Document { } if found > 0 { log.push(format!("the cross-reference data leaves out {found} objects the file contains; located them by their object headers")); - self.entries = Arc::new(filled); + self.entries = Arc::new(XrefIndex::Materialized(Arc::new(filled))); } self.located = Arc::new(located); } From 57ab6fcb2a8e08891cd6cb974adc110b46ee75a6 Mon Sep 17 00:00:00 2001 From: Marco Arnulfo Date: Thu, 15 Oct 2026 21:47:01 +0200 Subject: [PATCH 29/34] fix(cos): the renumbering check follows a shifted header (refs #1628) With junk before %PDF the table's offsets are relative to the header, so the "1 G obj" check is also tried there, as load does; otherwise object 1 stayed free and a full save dropped the image. --- crates/cos/src/document.rs | 93 +++++++++++++++++++++++++++++++++----- 1 file changed, 82 insertions(+), 11 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index e102993ae..5089cfe3c 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -1548,7 +1548,7 @@ impl Document { if !valid_xref_range(start, count) || count > 10_000_000 { return Err(CosError::Syntax { offset: lx.pos, detail: "invalid xref object range".into() }); } - let shifted = subsection_start(data, lx.pos, start, count); + let shifted = subsection_start(data, self.header_offset, lx.pos, start, count); renumbered |= shifted != start; let start = shifted; let Some(next_rows) = usize::try_from(count).ok().and_then(|count| row_count.checked_add(count)) else { return Ok(None) }; @@ -1824,7 +1824,7 @@ impl Document { if !valid_xref_range(start, count) || count > 10_000_000 { return Err(CosError::Syntax { offset: lx.pos, detail: "invalid xref object range".into() }); } - let shifted = subsection_start(data, lx.pos, start, count); + let shifted = subsection_start(data, self.header_offset, lx.pos, start, count); if shifted != start { log.push(format!("xref subsection at {off} is numbered from 1 but begins with object 0's entry; read from 0")); } @@ -2085,9 +2085,10 @@ fn generated_id(seed: &[u8; 32]) -> Vec { /// The object number a classic xref subsection really starts at. A common producer bug heads the /// table `1 n` and then writes object 0's free entry (`0000000000 65535 f`) as its first row, so /// every row would be read one number too high and object 1 as free. Such a subsection is read -/// from 0 when its second row, read that way, locates an object that is indeed numbered 1. `rows` -/// is the position just after the subsection header. -fn subsection_start(data: &[u8], rows: usize, start: i64, count: i64) -> i64 { +/// from 0 when its second row, read that way, locates an object that is indeed numbered 1, at +/// that offset or, as `load` reads offsets, relative to a header shifted by `header_offset`. +/// `rows` is the position just after the subsection header. +fn subsection_start(data: &[u8], header_offset: usize, rows: usize, start: i64, count: i64) -> i64 { if start != 1 || count < 2 { return start; } @@ -2102,12 +2103,15 @@ fn subsection_start(data: &[u8], rows: usize, start: i64, count: i64) -> i64 { }; let (Some(0), Some(65535), true) = row() else { return start }; let (Some(second), _, false) = row() else { return start }; - let Some(at) = usize::try_from(second).ok().filter(|at| *at < data.len()) else { return start }; - let mut obj = Lexer::new(data, at); - match (obj.object(), obj.object()) { - (Ok(Object::Int(1)), Ok(Object::Int(_))) if obj.eat_keyword(b"obj") => 0, - _ => start, - } + let object_one = |at: usize| { + if at >= data.len() { + return false; + } + let mut obj = Lexer::new(data, at); + matches!((obj.object(), obj.object()), (Ok(Object::Int(1)), Ok(Object::Int(_)))) && obj.eat_keyword(b"obj") + }; + let Ok(second) = usize::try_from(second) else { return start }; + if object_one(second) || (header_offset != 0 && object_one(second.saturating_add(header_offset))) { 0 } else { start } } /// Object numbers must fit the representation used by references and the xref map. @@ -2462,6 +2466,73 @@ endobj assert_eq!(doc.xref_entry(5), Some(XrefEntry::Free { next_generation: 1 })); } + /// The image the first page draws as `/Im1`, followed through the page tree. + fn first_page_image(doc: &Document) -> Arc { + let catalog = doc.get(doc.root().unwrap()); + let pages = doc.resolve(catalog.as_dict().unwrap().get(b"Pages").unwrap()); + let page = doc.resolve(&pages.as_dict().unwrap().get(b"Kids").unwrap().as_array().unwrap()[0]); + let resources = doc.resolve(page.as_dict().unwrap().get(b"Resources").unwrap()); + let xobjects = doc.resolve(resources.as_dict().unwrap().get(b"XObject").unwrap()); + doc.resolve(xobjects.as_dict().unwrap().get(b"Im1").unwrap()) + } + + /// The opposite off-by-one (#1628): the subsection says it starts at 1, but its first entry + /// is object 0's free-list head, so every entry lands one number too high and object 1, the + /// image, reads as free. The offset the table gives object 2 holds `1 0 obj`, so the + /// subsection is read from 0, and the image survives a full save. + #[test] + fn a_subsection_from_one_that_opens_with_the_free_list_head_is_read_from_zero() { + let base = build( + &[ + "<< /Type /XObject /Subtype /Image /Width 2 /Height 2 /ColorSpace /DeviceRGB /BitsPerComponent 8 /Length 12 >>\nstream\nABCDEFGHIJKL\nendstream", + "<< /Length 29 >>\nstream\nq 72 0 0 72 0 0 cm /Im1 Do Q\nendstream", + "<< /Type /Page /Parent 4 0 R /MediaBox [0 0 72 72] /Resources << /XObject << /Im1 1 0 R >> >> /Contents 2 0 R >>", + "<< /Type /Pages /Kids [3 0 R] /Count 1 >>", + "<< /Type /Catalog /Pages 4 0 R >>", + "<< /Subject (synthetic) >>", + ], + "/Root 5 0 R /Info 6 0 R", + ); + // /Size 7 is one short of the shifted numbers; /Size 8 covers them all. Junk before the + // header moves every object past the offsets the table gives; it is not a comment, which + // the lexer would skip on its way to `1 0 obj` anyway. + for (size, junk) in [(7, ""), (8, ""), (7, "Content-Type: application/pdf\n\n")] { + let text = String::from_utf8_lossy(&base).replace("\nxref\n0 7\n", "\nxref\n1 7\n").replace("/Size 7 ", &format!("/Size {size} ")); + let doc = Document::open(Arc::new(format!("{junk}{text}").into_bytes())).unwrap(); + let image = first_page_image(&doc); + assert_eq!(image.as_dict().and_then(|d| d.name(b"Subtype")), Some(&b"Image"[..]), "/Size {size} {junk:?}: {image:?}"); + assert_eq!(doc.xref_entry(0), Some(XrefEntry::Free { next_generation: 65535 }), "/Size {size} {junk:?}"); + assert!(doc.repair_log().iter().any(|l| l.contains("from 0")), "/Size {size} {junk:?}: {:?}", doc.repair_log()); + let saved = Document::open(Arc::new(crate::write_full(&doc, &crate::SaveOptions::default()).unwrap())).unwrap(); + let image = first_page_image(&saved); + assert!(matches!(&*image, Object::Stream(s) if &s.raw[..] == b"ABCDEFGHIJKL"), "/Size {size} {junk:?}: {image:?}"); + } + } + + /// Object 1 really freed: the subsection from 1 opens with a free entry, but the next entry's + /// offset holds object 2, not object 1. The table is taken as it is, and the old body of + /// object 1 still in the file stays freed. + #[test] + fn a_subsection_from_one_whose_next_offset_holds_object_two_is_taken_as_it_is() { + let base = build(&["(freed)", "<< /Type /Catalog /Pages 3 0 R >>", "<< /Type /Pages /Kids [] /Count 0 >>"], "/Root 2 0 R"); + let entries = table_entries(&base); + let mut table = vec!["1 3".to_string(), "0000000000 65535 f ".to_string()]; + table.extend(entries[2..].iter().cloned()); + let text = String::from_utf8_lossy(&base).into_owned(); + let at = text.rfind("\nxref\n").unwrap() + 1; + for junk in ["", "Content-Type: application/pdf\n\n"] { + let mut bytes = junk.as_bytes().to_vec(); + bytes.extend_from_slice(&base[..at]); + bytes.extend_from_slice(format!("xref\n{}\ntrailer\n<< /Size 4 /Root 2 0 R >>\nstartxref\n{at}\n%%EOF\n", table.join("\n")).as_bytes()); + let doc = Document::open(Arc::new(bytes)).unwrap(); + assert_eq!(doc.xref_entry(1), Some(XrefEntry::Free { next_generation: 65535 }), "{junk:?}"); + assert_eq!(*doc.get(ObjRef::new(1, 0)), Object::Null, "{junk:?}: the freed object stays freed"); + assert!(matches!(&*doc.get(ObjRef::new(2, 0)), Object::Dict(d) if d.name(b"Type") == Some(&b"Catalog"[..])), "{junk:?}"); + assert!(!doc.repair_log().iter().any(|l| l.contains("from 0")), "{junk:?}: {:?}", doc.repair_log()); + assert!(!junk.is_empty() || doc.repair_log().is_empty(), "{:?}", doc.repair_log()); + } + } + #[test] fn opening_an_exhausted_classic_xref_repairs_and_allocates() { // The extra in-use row need not be loaded to exhaust the next object number. From 8444a08c0002b37bcc59e6a8aee85861072d6b97 Mon Sep 17 00:00:00 2001 From: Marco Arnulfo Date: Wed, 21 Oct 2026 10:45:37 +0200 Subject: [PATCH 30/34] fix(cos): a repaired xref table fills only the numbers the damage leaves out (refs #1628) --- crates/cos/src/document.rs | 90 ++++++++++++++++++++++++++++++++++++-- 1 file changed, 86 insertions(+), 4 deletions(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index 5089cfe3c..81c4d5d32 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -1933,9 +1933,11 @@ impl Document { /// that nothing else replaces. (A number no section lists is not evidence: it is a null /// object, §7.3.10, and valid files have them below an oversized `/Size`.) Viewers rebuild /// the table and show these objects; reading them as null would drop them from every copy. - /// When one of those shows, the file is scanned once for object headers, numbers without an - /// entry take the last header's location, and wrong offsets are looked up there (`load`). - /// Existing entries are never replaced: an object freed by an update stays freed. + /// When one of those shows, the file is scanned once for object headers, and wrong offsets + /// are looked up there (`load`). Only the numbers the damage leaves out take the last + /// header's location: one whose entry was set aside, or the last object of a subsection + /// numbered one too low. Any other unlisted number stays null, and existing + /// entries are never replaced: an object freed by an update stays freed. fn fill_xref_gaps(&mut self, log: &mut Vec) { let entries = &self.entries; let shifted = matches!(entries.lookup(0, &self.data), Some(XrefEntry::InFile { .. })); @@ -1948,9 +1950,22 @@ impl Document { located.insert(num, (offset, generation)); } let mut filled = entries.to_map(&self.data); + // A subsection `0 N` whose rows are those of objects 1 to N leaves out object N, whatever + // newer sections say about its neighbours. + let past_shift: HashSet = if shifted { + self.revisions + .iter() + .filter(|r| !r.is_stream) + .filter_map(|r| usize::try_from(r.xref_offset).ok().map(|o| if r.header_relative { o.saturating_add(self.header_offset) } else { o })) + .filter_map(|at| shifted_subsection_end(&self.data, at)) + .collect() + } else { + HashSet::new() + }; let mut found = 0usize; for (&num, &(offset, generation)) in &located { - if num != 0 && !filled.contains_key(&num) { + let left_out = past_shift.contains(&num) || self.xref_oddities.iter().any(|(_, n)| *n == num); + if num != 0 && !filled.contains_key(&num) && left_out { filled.insert(num, XrefEntry::InFile { offset, generation }); found += 1; } @@ -2028,6 +2043,25 @@ impl Document { } } +/// For a classic table at `at` whose first subsection is `0 N` and opens with an in-use row +/// (the rows of objects 1 to N), the number N its rows really end at. +fn shifted_subsection_end(data: &[u8], at: usize) -> Option { + let mut lx = Lexer::new(data, at); + if !lx.eat_keyword(b"xref") { + return None; + } + let (Ok(Object::Int(0)), Ok(Object::Int(count))) = (lx.object(), lx.object()) else { return None }; + lx.skip_ws(); + let offset = std::str::from_utf8(lx.token()).ok().and_then(|s| s.parse::().ok()); + lx.skip_ws(); + lx.token(); + lx.skip_ws(); + match (offset, lx.token()) { + (Some(offset), b"n") if offset != 0 => u32::try_from(count).ok().filter(|n| *n > 0), + _ => None, + } +} + /// Every `N G obj` header in `data`, in file order, as (number, offset, generation). fn object_headers(data: &[u8]) -> Vec<(u32, u64, u16)> { let mut headers = Vec::new(); @@ -2466,6 +2500,54 @@ endobj assert_eq!(doc.xref_entry(5), Some(XrefEntry::Free { next_generation: 1 })); } + /// Repairing a damaged table fills only the numbers the damage leaves out: an old body of a + /// number no section lists, under an oversized `/Size`, stays a null object. + #[test] + fn a_repaired_table_leaves_an_unlisted_number_null() { + let base = scanned_page(); + let at = String::from_utf8_lossy(&base).rfind("\nxref\n").unwrap() + 1; + let entries = table_entries(&base); + let mut offset_zero = vec!["0 6".to_string()]; + offset_zero.extend(entries.iter().cloned()); + offset_zero[5] = "0000000000 00000 n ".into(); + let mut off_by_one = vec!["0 5".to_string()]; + off_by_one.extend(entries[1..6].iter().cloned()); + for (name, table, found) in [("offset 0", offset_zero, 4), ("off by one", off_by_one, 5)] { + let bytes = [&base[..at], b"7 0 obj\n(stale)\nendobj\n", &base[at..]].concat(); + let doc = Document::open(Arc::new(with_table(&bytes, &table, 9))).unwrap(); + assert!(matches!(&*doc.get(ObjRef::new(found, 0)), Object::Stream(_)), "{name}: the gap is filled"); + assert_eq!(*doc.get(ObjRef::new(7, 0)), Object::Null, "{name}: {:?}", doc.repair_log()); + assert_eq!(doc.xref_entry(7), None, "{name}"); + } + } + + /// The last object of a subsection numbered one too low is filled from that subsection, + /// whatever later happens to the row before it: an incremental save that rewrites object 4, + /// or object 4's own row at offset 0. + #[test] + fn the_last_object_of_a_shifted_subsection_is_filled_whatever_its_neighbour_says() { + let base = scanned_page(); + let entries = table_entries(&base); + let mut off_by_one = vec!["0 5".to_string()]; + off_by_one.extend(entries[1..6].iter().cloned()); + let shifted = with_table(&base, &off_by_one, 5); + let prev = String::from_utf8_lossy(&shifted).rfind("\nxref\n").unwrap() + 1; + let mut updated = shifted.clone(); + let body = updated.len(); + updated.extend_from_slice(b"4 0 obj\n<< /Length 21 >>\nstream\n10 0 0 10 0 0 cm /Im0 Do\nendstream\nendobj\n"); + let section = updated.len(); + updated.extend_from_slice( + format!("xref\n4 1\n{body:010} 00000 n \ntrailer\n<< /Size 6 /Root 1 0 R /Prev {prev} >>\nstartxref\n{section}\n%%EOF\n").as_bytes(), + ); + let mut neighbour_at_zero = off_by_one.clone(); + neighbour_at_zero[5] = "0000000000 00000 n ".into(); + for (name, bytes) in [("updated", updated), ("row 4 at offset 0", with_table(&base, &neighbour_at_zero, 5))] { + let doc = Document::open(Arc::new(bytes)).unwrap(); + let image = doc.get(ObjRef::new(5, 0)); + assert_eq!(image.as_dict().and_then(|d| d.name(b"Subtype")), Some(&b"Image"[..]), "{name}: {:?}", doc.repair_log()); + } + } + /// The image the first page draws as `/Im1`, followed through the page tree. fn first_page_image(doc: &Document) -> Arc { let catalog = doc.get(doc.root().unwrap()); From 79d843524cf6b2f0f01f4f7649e24f6c8e8db41e Mon Sep 17 00:00:00 2001 From: aiongg Date: Thu, 22 Oct 2026 01:46:03 +0000 Subject: [PATCH 31/34] Integration: #2220 look up set-aside object numbers in a set fill_xref_gaps scanned every xref oddity for each object header; a file with many offset-0 rows and many headers took quadratic time. --- crates/cos/src/document.rs | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/crates/cos/src/document.rs b/crates/cos/src/document.rs index b24bea637..6d39d53db 100644 --- a/crates/cos/src/document.rs +++ b/crates/cos/src/document.rs @@ -1968,9 +1968,10 @@ impl Document { } else { HashSet::new() }; + let odd: HashSet = self.xref_oddities.iter().map(|(_, n)| *n).collect(); let mut found = 0usize; for (&num, &(offset, generation)) in &located { - let left_out = past_shift.contains(&num) || self.xref_oddities.iter().any(|(_, n)| *n == num); + let left_out = past_shift.contains(&num) || odd.contains(&num); if num != 0 && !filled.contains_key(&num) && left_out { filled.insert(num, XrefEntry::InFile { offset, generation }); found += 1; From 69c87d2f3e3353a204b09f91a8fc9198c3e1fdca Mon Sep 17 00:00:00 2001 From: aiongg Date: Thu, 22 Oct 2026 01:48:30 +0000 Subject: [PATCH 32/34] Integration: #3195 keep CMap entries read before a stream ends inside a block MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit A begin… block cut off at the end of the stream returned None through ? and discarded the whole map; it now ends the scan like any other unparseable tail. Test: stream_cut_off_inside_a_block_keeps_the_entries_before_it. --- vendor/hayro-cmap/src/lib.rs | 7 ++++++ vendor/hayro-cmap/src/parse.rs | 41 ++++++++++++++++++++++++++-------- 2 files changed, 39 insertions(+), 9 deletions(-) diff --git a/vendor/hayro-cmap/src/lib.rs b/vendor/hayro-cmap/src/lib.rs index 4786e75e5..fb42ff37e 100644 --- a/vendor/hayro-cmap/src/lib.rs +++ b/vendor/hayro-cmap/src/lib.rs @@ -1117,6 +1117,13 @@ endbfchar ); } + #[test] + fn stream_cut_off_inside_a_block_keeps_the_entries_before_it() { + let cmap = parse_with_preamble(b"1 beginbfchar <41> <0058>\n<42> <00"); + assert_eq!(cmap.lookup_bf_string(0x41), Some(BfString::Char('X'))); + assert_eq!(cmap.lookup_bf_string(0x42), None); + } + #[test] fn truncated_stream_with_no_entries_reports_failure() { // The tolerant parser keeps collected entries, but a stream that diff --git a/vendor/hayro-cmap/src/parse.rs b/vendor/hayro-cmap/src/parse.rs index 1137ab078..a93620927 100644 --- a/vendor/hayro-cmap/src/parse.rs +++ b/vendor/hayro-cmap/src/parse.rs @@ -50,8 +50,10 @@ pub(crate) fn parse_inner<'a>( let mut last_name: Option> = None; // PdfCraft patch: track whether the scan was cut short. A CMap with an - // unparseable tail keeps the entries read so far instead of being - // discarded wholesale; one that yielded nothing still reports failure. + // unparseable tail, including one that ends inside a `begin…` block, + // keeps the entries read so far (the block parsers push into the shared + // lists as they go) instead of being discarded wholesale; one that + // yielded nothing still reports failure. let mut truncated = false; // PdfCraft patch: track a `usecmap` reference that could not be resolved @@ -94,25 +96,46 @@ pub(crate) fn parse_inner<'a>( } else { match name.as_str() { Some("begincodespacerange") => { - parse_codespace_range(&mut scanner, &mut _codespace_ranges, &mut ctx)?; + if parse_codespace_range(&mut scanner, &mut _codespace_ranges, &mut ctx).is_none() { + truncated = true; + break; + } } Some("begincidrange") => { - parse_range(&mut scanner, &mut ranges, &mut ctx, "endcidrange")?; + if parse_range(&mut scanner, &mut ranges, &mut ctx, "endcidrange").is_none() { + truncated = true; + break; + } } Some("begincidchar") => { - parse_char(&mut scanner, &mut ranges, &mut ctx, "endcidchar")?; + if parse_char(&mut scanner, &mut ranges, &mut ctx, "endcidchar").is_none() { + truncated = true; + break; + } } Some("beginnotdefrange") => { - parse_range(&mut scanner, &mut notdef_ranges, &mut ctx, "endnotdefrange")?; + if parse_range(&mut scanner, &mut notdef_ranges, &mut ctx, "endnotdefrange").is_none() { + truncated = true; + break; + } } Some("beginnotdefchar") => { - parse_char(&mut scanner, &mut notdef_ranges, &mut ctx, "endnotdefchar")?; + if parse_char(&mut scanner, &mut notdef_ranges, &mut ctx, "endnotdefchar").is_none() { + truncated = true; + break; + } } Some("beginbfchar") => { - parse_bf_char(&mut scanner, &mut bf_entries, &mut ctx)?; + if parse_bf_char(&mut scanner, &mut bf_entries, &mut ctx).is_none() { + truncated = true; + break; + } } Some("beginbfrange") => { - parse_bf_range(&mut scanner, &mut bf_entries, &mut ctx)?; + if parse_bf_range(&mut scanner, &mut bf_entries, &mut ctx).is_none() { + truncated = true; + break; + } } Some("usecmap") => { // PdfCraft patch: an unresolvable or unparseable `usecmap` From c4fa66ecbd6e5c7ebf55d1718e4e3933eb9f51f9 Mon Sep 17 00:00:00 2001 From: aiongg Date: Thu, 22 Oct 2026 01:48:59 +0000 Subject: [PATCH 33/34] Integration: #3668 vendor/README.md: describe the delimiter after EI in hayro-syntax patch (4) Adds content::tests::inline_image_ei_may_be_followed_by_a_delimiter to its test list. --- vendor/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vendor/README.md b/vendor/README.md index 3054d7853..228764340 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -17,7 +17,7 @@ In `hayro-interpret`, Type 3 glyph bounding boxes are transformed to device spac | hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour, planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk, interpolated, axis-aligned image is snapped to whole device pixels: shrunk once to its rounded device rectangle and drawn 1:1 at the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat); (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group with its background painted inside it as the initial backdrop, so a Multiply highlight blends with a dark reading-colours background instead of staying an opaque block (identity on white pages); (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349) | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_samples_the_same_columns`, `raster::tests::non_integer_minification_samples_colour_and_mask_from_the_same_columns`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::interpolated_minified_image_edges_snap_to_whole_pixels` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path` | | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap, page bitmap, halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer) | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding` | -| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306); (22) an object stream's index of `objnum offset` pairs is parsed once per stream and kept beside its decoded data (`ObjectStreamIndexes`, `XRef::with_object_stream_indexes`), and an offset past `usize` is malformed instead of wrapping: `XRef::get_with` parsed the whole index again for every object it fetched from the stream, so reading all of a stream of N objects took N² steps (a Firefox page whose 65,542 objects share one stream took 37 s to draw and 40 s to extract its text); (23) the inline-image end-of-data look-ahead no longer rejects a candidate `EI` because a byte above 0x7F lies between it and the next one: when the candidate is the end, those bytes are the content stream that follows the image, and its strings carry such bytes (WinAnsi `(Grüße) Tj`, the two-byte codes of any Type 0 font), so the operator check that follows decides alone, bounded as in (4) (a scanned page whose OCR text held an accented word and a later capital "EI" — EINGANG, BEI, KLEIN — drew and extracted nothing after the image: 40 of 275 pages across five real documents); (24) `xref::prev_offset`: a trailer `/Prev` written as a reference `N G R` is read as the offset that object holds when the section's own entries place it (iText's `indirectPrev.pdf`; each section's entries are read before its `Prev` is followed), and otherwise as offset N (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366), as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (22) `raster::tests::objects_in_one_large_object_stream_are_read_in_linear_time`; (23) `text::tests::text_after_an_inline_image_survives_a_non_ascii_byte_before_a_later_ei`, `content::tests::an_inline_image_ends_at_its_own_ei_when_later_text_holds_a_non_ascii_byte` (hayro-syntax); (24) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `xref::tests::prev_as_reference_to_an_object_holding_the_offset_reads_that_offset`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections`, `document::tests::a_prev_reference_to_an_object_holding_the_offset_reads_that_offset` +| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); an EI followed by a delimiter (EI%..., EI[...) ends the image like one followed by white space; only a regular character after EI skips the candidate; (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306); (22) an object stream's index of `objnum offset` pairs is parsed once per stream and kept beside its decoded data (`ObjectStreamIndexes`, `XRef::with_object_stream_indexes`), and an offset past `usize` is malformed instead of wrapping: `XRef::get_with` parsed the whole index again for every object it fetched from the stream, so reading all of a stream of N objects took N² steps (a Firefox page whose 65,542 objects share one stream took 37 s to draw and 40 s to extract its text); (23) the inline-image end-of-data look-ahead no longer rejects a candidate `EI` because a byte above 0x7F lies between it and the next one: when the candidate is the end, those bytes are the content stream that follows the image, and its strings carry such bytes (WinAnsi `(Grüße) Tj`, the two-byte codes of any Type 0 font), so the operator check that follows decides alone, bounded as in (4) (a scanned page whose OCR text held an accented word and a later capital "EI" — EINGANG, BEI, KLEIN — drew and extracted nothing after the image: 40 of 275 pages across five real documents); (24) `xref::prev_offset`: a trailer `/Prev` written as a reference `N G R` is read as the offset that object holds when the section's own entries place it (iText's `indirectPrev.pdf`; each section's entries are read before its `Prev` is followed), and otherwise as offset N (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366), as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`, `content::tests::inline_image_ei_may_be_followed_by_a_delimiter`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (22) `raster::tests::objects_in_one_large_object_stream_are_read_in_linear_time`; (23) `text::tests::text_after_an_inline_image_survives_a_non_ascii_byte_before_a_later_ei`, `content::tests::an_inline_image_ends_at_its_own_ei_when_later_text_holds_a_non_ascii_byte` (hayro-syntax); (24) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `xref::tests::prev_as_reference_to_an_object_holding_the_offset_reads_that_offset`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections`, `document::tests::a_prev_reference_to_an_object_holding_the_offset_reads_that_offset` | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it. | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed` | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary, and an in-file /Root must match its generation (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW and RunLength never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it` | From aef42bc92a92dc944d5d2af9f774af67fdf055ae Mon Sep 17 00:00:00 2001 From: aiongg Date: Thu, 22 Oct 2026 01:50:23 +0000 Subject: [PATCH 34/34] Integration: cargo fmt --- xtask/src/assets.rs | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/xtask/src/assets.rs b/xtask/src/assets.rs index 21dc6dea7..abd5f070c 100644 --- a/xtask/src/assets.rs +++ b/xtask/src/assets.rs @@ -47,10 +47,7 @@ pub const ADOBE_DATA_ALLOWED: &[(&str, &str)] = &[("hayro-cmap", "assets/cmaps.b /// technical data we hold ourselves rather than through a dependency. Each still needs an /// `[[asset]]` entry with `adobe_data = true`, so its SHA-256 is checked and any change to the /// table has to be re-attested deliberately. -pub const ADOBE_DATA_FILES: &[&str] = &[ - "crates/fonts/src/std14_metrics.rs", - "vendor/hayro-cmap/assets/cmaps.brotli", -]; +pub const ADOBE_DATA_FILES: &[&str] = &["crates/fonts/src/std14_metrics.rs", "vendor/hayro-cmap/assets/cmaps.brotli"]; /// File extensions that count as assets wherever they appear in the repository. const ASSET_EXTENSIONS: &[&str] = &[