Files
pdfcraft/crates/automation
+21 f74c19de87 Combine files: move and remove single pages before combining (3 of 3, closes #611) (#790)
* Combine files: a thumbnail grid, card sizes and card actions (refs #611)

Combine files shows each file as a card (the default; Grid/List kept in
the settings, --combine-view): the first page it adds, stacked when it
adds more, its name shortened in the middle. Cards select (click, Ctrl,
Shift, arrows by card and by row), drag to reorder (one undo step; the
grid scrolls at its edges) and share the list's toolbar, selection and
history. Locked, unreadable, bad-range and unmeasurable files say why.

Thumbnails render off the UI thread, only for cards in view: at most
three render threads at once, counted until they exit (no watchdog
replacements), a 32 MB texture budget, keys that never show another
page, password or file, and one heavy step per frame on the web.

Cards are sized from 60% to 200% (toolbar slider and buttons, Ctrl
+/-/0, Ctrl+wheel or pinch over the grid, kept in the settings,
--combine-zoom), keeping the card under the pointer in place. Hovering
a card shows its actions over the top of its page: remove that file,
and a magnifier that shows it large, flipping through the pages it
adds; the same on a right click, and Space by keyboard. The toolbar
wraps between groups at narrow widths.

RenderPool gains retire(), without_replacements() and threads() so a
caller can bound the render threads that exist at once.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Combine files: expand a file into its pages (refs #611)

A file that adds more than one page (up to 2000) expands into a card
per page it adds, in its range's order, each labelled with its page and
the file's name, on a shared tinted band: from its card's hover bar, a
right click, or Expand all / Collapse all in the toolbar. A page card's
magnifier opens the preview on that page; Up and Down step rows of
cards. Expanding is a view: not an undo step, kept across views and
zoom, and the combined file is the same either way.

Thumbnails are now kept per file and page, looked up on demand (the
cost follows the files, not the pages of an expanded file), and a
render's document stays open for two seconds so the next page of the
same file isn't read again.

Adds the Lucide minimize-2 icon (ISC, recorded in ATTRIBUTION).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Combine files: move and remove single pages before combining (closes #611)

With a file expanded, single pages are selected (click, Ctrl, Shift),
dragged anywhere and removed (a page card's trash, the toolbar, Delete),
each one undo step; the files themselves are never changed. A page moved
between another file's pages splits its file into parts that share its
bytes: the list shows each part as a row with its own page range, and
parts that end up side by side join again. A split file still counts,
unlocks and combines as one file.

organize gains combine_grouped (and Session::combine_grouped): each file
is copied once and its pages put in the order shown, so it keeps one
bookmark at its first page, its nested bookmarks, links between its own
pages, its fields and one copy of its attachments. A page listed twice
now gets its own copies of its annotations, and a form field's widget on
it joins the same field, so the field keeps one value. The existing
combine is unchanged; doc_combine gains an optional groups
array (entries with the same number are one file split into parts), so
the same combine is reachable headless.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Progress docs: target-app parity, gaps, alpha gate, readiness by audience (#873)

Progress docs per craftrules standards/progress-docs.md: target-app parity, gaps, roadmap with alpha gate, readiness by audience (full / mainstream / essentials) with Opus 5.5 hours, and feature, format, PDF-spec, hardware, UI and localization parity docs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Measure: snap to the implicit closing edge of filled paths (#803) (#862)

* fix(organize): reject rotation angles that are not multiples of 90 (#801) (#833)

Page rotation only turns in quarter steps, but the degree value was read with an integer division by 90 that threw the remainder away. Asking for 45 degrees saved an unrotated page, 91 and 135 saved a 90 degree turn, and the command still reported success.

Reject the value before touching the document instead, so every caller gets an actionable error and the pages are left alone. The automation tool already rejected it; the single edit command did not.

Fixes #801.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* Autosave: snapshot a new unsaved document before its first edit (#815) (#850)

* OCR: Recognize text in multiple files refuses paths outside the root (#795)

ocr_recognize_files turned a path outside the automation root into a per-file
error and returned success, so no_tool_escapes_the_root reported 9 leaks once
the OCR models were installed. Without them, loading the models failed first
and hid it, and CI never installs them. Every path is now checked before any
work starts, and one outside the root refuses the whole call, like the other
tools. A missing file inside the root stays a per-file error. Nothing was ever
read from outside the root.

ocr_recognize_files_refuses_paths_outside_the_root checks this without the
models.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(text): match phrases across repeated whitespace (#814) (#831)

The phrase search collapsed runs of whitespace in the query but not in the page text it was matching against, so a document drawn with two space glyphs in a row like the issue example never matched the phrase. Single words still matched, which is why it looked like only some files were affected.

Collapse whitespace while the text is flattened, the same way the query is normalised. The regression test fails on the old code because the phrase find returns an empty range, and it also covers tabs and single word queries.

Fixes #814.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* fix(edit): keep a comment whose appearance is missing when flattening (#805) (#838)

Viewers draw notes, FreeText, ink and stamps from their dictionary when the file stores no /AP /N. flatten only drew annotations that already had an appearance and then removed the rest, so a supported note disappeared and the saved page came back blank.

Build the appearance a viewer would show for every selected, visible, non-widget annotation that has none, then draw it as usual. A subtype this build cannot render still fails to build and is handled exactly as before.

The new synthetic test draws the note icon into the page and asserts it is baked in; without the fix flatten reports 0 drawn and the icon is lost.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* fix(comments): keep a reply to a reply in the comment listing (#806) (#839)

comment_list collected only the replies whose /IRT pointed at the root comment, so a reply to a reply was listed nowhere at all. Roots and the root count were already right; the nested reply was simply dropped from its thread.

Follow the /IRT chain to any depth, with a bound, so every reply under a comment is listed in document order. A reply to a reply now stays discoverable through the tool that promises every comment and its replies.

The new test builds a root note, a direct reply and a reply to that reply, then checks both texts are listed and the two notes stay roots. It listed only the direct reply before the fix.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* Issue #774: keep the catalog out of object streams in encrypted files (Acrobat) (#836)

Acrobat reports an encrypted PDF as "damaged and could not be repaired", before
asking for the password, when the document catalog sits in an object stream.
Every other reader tried (qpdf, pypdf, poppler, PDFKit, PdfCraft) accepts such
files, and qpdf keeps the catalog standalone too. Full saves of encrypted
documents now write the catalog as its own object; unencrypted output is
unchanged. Verified with Acrobat DC on macOS: the same file fails with the
catalog packed and opens with it standalone, in AES-256 and RC4-128.


Claude-Session: https://claude.ai/code/session_01REz7wTfQgnopyvN9eQJJVu

Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

* fix(edit): refuse a header start number that would wrap the next page (#804) (#834)

Page numbering adds the start number to the position of the page in the run. At the top of the u32 range that addition wrapped, so a release build saved zero for the second page and still reported success, and a debug build hit a caught overflow error instead.

Check the whole run before drawing anything. When the last page number would pass the u32 limit the operation returns an actionable error and the document keeps no header, rather than a half applied run.

Fixes #804.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* OCR: a scanned euro sign reads as €, not E (#796)

The ocrs crate's default alphabet has an E where its recognition model was
trained with € (ocrs-models DEFAULT_ALPHABET), so a scanned "€250" was
recognised as "E250". pdfcraft-ocr now gives ocrs the alphabet the model was
trained with.

the_alphabet_is_the_trained_one checks the alphabet without the models;
recognize_text_reads_the_euro_sign reads a scan end to end when they are
installed.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(comments): include replies to replies in the comment summary (#820) (#840)

The comment summary gathered a comment's direct replies only, matching /IRT against the comment itself. A reply to a reply points at the direct reply, so its text never reached the generated summary even though the stored discussion kept it.

Replies are now walked to any depth, with a bound, and indented one level per reply. The existing output for direct replies is unchanged.

The new test builds a rectangle, a reply and a reply to that reply, then checks the summary carries both texts with the nested one indented under its parent. It omitted the nested reply before the fix.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* fix(xfdf): drop the replaced note's popup when re-importing a comment (#807) (#860)

* fix(view): check Shift-Command-G before Command-G for find navigation (#810) (#835)

* view: check Shift-Command-G before Command-G for find navigation (#810)

* Tests: the find shortcut test waits for find by a deadline

(cherry picked from commit eaf3180bd8ebb353b3d8c5ca7b9ef7548ed55c37)

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* fix(editing): recognize paragraph drafts as unsaved work (#811) (#843)

* view: check Shift-Command-G before Command-G for find navigation (#810)

* fix: recognize paragraph drafts as unsaved work (Issue #811)

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* fix(automation): retain batch results with identical filenames (#877)

* fix(automation): retain batch results with identical filenames

* Format

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M13: preserve desktop Action Wizard batch results (#886)

* M13: preserve desktop Action Wizard batch results

Co-Authored-By: Codex <noreply@openai.com>

* ROADMAP: date the desktop action outputs entry today

---------

Co-authored-by: Tevinhead <173904641+Tevinhead@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Fit to width fits the shown page in single-page view (#808) (#859)

* fix(links): find URLs when lowercasing changes the text length (#816) (#830)

find_urls case-folded a copy of the page text and then sliced the original
characters with offsets taken from that copy. Case folding can change a
string's length — U+0130 (İ) lowercases to "i" plus a combining dot — so the
offsets drifted past the end of the source and `chars[i..j]` panicked. The CLI
exited 101 before it could create the link; MCP reported an internal tool error
but kept serving.

Match the ASCII prefixes case-insensitively against the original characters
instead, so every offset stays in the source's own index space, and use get()
so no lookup can index past the end.

A new test reproduces the reported panic on the old code (range start index 41
out of range for slice of length 39) and passes with the fix. pdfcraft-annot's
tests, clippy and rustfmt are green.

Fixes #816.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* organize: deleting the opening page drops the catalog OpenAction (#821) (#855)

* fix(text): Turkish dotted and dotless i match in case-insensitive search (#892)

Case-insensitive search folded the query and the page with
str::to_lowercase, which follows no locale: U+0130 (İ) becomes "i" plus a
combining dot above (two characters) and U+0131 (ı) has no case pair. So
"istanbul" never found "İSTANBUL", and "ırmak"/"kış" never found
"IRMAK"/"KIŞ" (or the reverse). The same PageText::find serves Find,
Search & Redact (phrases and word lists) and the automation text search,
so Turkish words could also be left unredacted.

Fold I, i, İ and ı to "i" and every other character with its own
lowercase, one character for one character. Case-sensitive search is
unchanged.

* XFDF keeps polygon and polyline vertices (#809) (#848)

Export wrote polygon and polyline comments without their points and import
had nowhere to read them from, so an XFDF round trip left the comments with
no /Vertices and no appearance. Export now writes the points as XFDF
<vertices>x,y;x,y;…</vertices>, and import reads them back (whole x,y pairs
only) before the appearance is drawn.

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* XFDF keeps a callout's leader, text-box inset, intent and arrowhead (#819) (#854)

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Web: warn before reloading a document with unsaved changes (#812) (#856)

* Issue #813: a cancelled recovery password no longer attaches to the next unlocked document (#858)

Recovering an encrypted snapshot parked its metadata in `pending_recovered`
until the password prompt was answered. Cancel left it there, and the next
encrypted document unlocked in the foreground took the snapshot's original
path and recovery key: Save would overwrite the other file and delete its
snapshot.

The association now travels with the prompt (`PasswordPrompt::recovered`).
Cancel puts the entry back on offer and leaves the store alone; a wrong
password keeps it on the re-raised prompt; the correct one finishes the
recovery of the tab that was just opened. `pending_recovered` is gone.

Test: crash with an encrypted snapshot of A, Recover, Cancel, open encrypted
B with its password. B keeps its own path and is clean, A's snapshot stays
in the store, also after closing B. Fails on main at "B keeps its own path".

Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

* Viewer: touchpad scrolling tracks the finger 1:1 (#759) (#868)

A precision touchpad sends pixel-precise `MouseWheel` events. On Windows
they arrive with no `TouchPhase::Start`, so egui never takes its
no-smoothing "in touch" path and eases any delta of 8 px or more over
several frames, leaving the page a frame behind the finger.

The page view now applies precise (Point-unit) wheel input itself, 1:1,
through the same un-eased scroll path as keyboard and middle-button
auto-scroll, and suppresses egui's eased copy for a short window so its
internal easing tail neither doubles nor trails the motion. A coarse
mouse wheel (Line/Page units) keeps egui's smoothing, which is pleasant
for chunky notches, and single-page paging still owns the vertical axis.

Regression test: one 200 px precise swipe scrolls the full 200 px at
once (an eased path would ramp there over ~0.1 s) and leaves no drift.

Co-authored-by: Justin John <babyjjbabyjj549@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* Keep every part when bookmark titles repeat in doc_split (#800) (#863)

* Windows: add reproducible security diagnostics for storytold/pdfcraft#747 (#769)

Add packaging/windows/security-audit.ps1, a read-only maintainer helper that records a
release file's SHA-256, Authenticode status and, with an external YARA binary and rule,
every matched string identifier and offset. Document the procedure in docs/releasing.md
("Scanner detections").

No application, dependency, installer or network behaviour changes: the diff touches only
the helper and the release documentation.

Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>

* Notice bar: the message wraps beside its buttons and lines up with them (#785) (#832)

The notice bar above the pages (security, forms, XFA, repaired file,
warnings, signatures) laid its message out before its buttons. In a
narrow document area (a side panel open, a longer translation) the
message ran under "Security settings" and on into the side panel, and it
wasn't centred on the buttons.

A shared notice_bar helper now lays the buttons out first, right to
left, centred on a 28 px row, and gives the message the width that's
left: it wraps there (cut short with the full text on hover when less
than 120 px is left) and its first line is centred on the buttons' row.

Regression test: editing::the_security_notice_wraps_beside_its_buttons_and_lines_up_with_them
(one line, wrapped and cut-short widths; fails on main).

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* M10.2: scan JPEG metadata beyond frame headers (#764)

Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Issue #878: Combine stores an image once when its colour space is an indirect name (#883)

ImageMagick writes an image's colour space as its own object (`/ColorSpace 10 0 R`,
`10 0 obj /DeviceRGB`). Each source's copy of that object was never merged, since it
isn't a resource, so identical images compared unequal and every copy was kept:
8 copies of an 80 MB ImageMagick PDF combined into 648 MB instead of 81 MB.

Dedupe now compares a reference to a plain value (name, number, boolean or string) as
that value. The value objects themselves are still never merged. Combine, Insert Pages
and Reduce File Size share this code.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* organize: drop named destinations and their links when a page is deleted (#817) (#851)

* JavaScript: limit util.printf widths and output, and array arguments (#770)

util.printf accumulated a conversion's width and precision with unchecked
arithmetic and no maximum, and its output had no bound; the array
arguments of resetForm and the field value/colour setters were read for
as many items as the array's length says, which a sparse array makes
free. Saturate the numbers and refuse widths and precisions above 4096,
output above 1 MiB and arrays longer than 2^20 with a script error,
before anything is allocated from them.

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

* Linux: "Use system setting" asks the desktop, not winit (#881)

winit reports no system theme on Linux — `ActiveEventLoop::system_theme` is
`None` there on every desktop — so `ThemePreference::System` never saw the
desktop's choice and silently fell back to whatever was on screen: PdfCraft
came up light on a dark desktop, and picking "Use system setting" after Dark
kept it dark. The default setting did not follow the system.

`pdfcraft-platform` now reads the XDG Desktop Portal's Settings interface off
the UI thread and the app follows the answer, including a change made while it
runs. The colour scheme is asked for under its GNOME schema name first, which
GNOME's and XFCE's portals both answer, then under the older freedesktop one,
which KDE answers. A desktop that answers neither leaves the user's own choice
alone, exactly as before.

Co-authored-by: GyuriX <gyurix@users.noreply.github.com>

* Extract/Combine: a page repeated in one selection gets its own annotations (#818) (#852)

* M5.7: Fill & Sign tools pick up placed signatures to move and resize them (#822)

A signature placed with Fill & Sign could only be moved with the Select
tool. While a Fill & Sign tool was active (Sign, Initials, Add text,
Check, Date...), hovering the signature showed the placement crosshair,
a drag did nothing and a click stamped another mark on top of it.

Now, as in Acrobat, a Fill & Sign tool picks up the Fill & Sign marks
already on the page: hovering one shows the move cursor, a click selects
it, a drag moves it and the selected mark's handles resize it. Away from
them the tool places as before, including on other comments. Image
signatures keep their live move/resize preview.

Comments now carry a fill_sign flag (pdfcraft_annot::is_fill_sign for
the engine's list, the same rules in the viewer's inspector), so this
also works on reopened files.

Co-authored-by: lloyd <lloyd@stackworx.io>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* Web: opt-in embedding bridge for host pages (?host=parent) (#842)

A page that frames PdfCraft on the same origin can now pass it PDFs and
receive saves through postMessage, instead of Save becoming a browser
download. Built for a Nextcloud integration, but nothing in it is
host-specific.

- ui-egui: HostSaveFn/HostDirtyFn hooks (`host_save`, `host_dirty`),
  None by default. On wasm, Save hands the bytes to `host_save` when
  set; `host_dirty` hears each frame whether any tab has unsaved work.
- pdfcraft-web: `host::Bridge`, enabled only by `?host=parent` inside a
  frame. Messages: pdfcraft:ready / open / save / dirty / failed; only
  the parent window on our own origin is heard or answered.
  `?author=` names comment authors unless the user chose a name.
- Test: host_dirty_hears_when_unsaved_work_appears_and_goes.

Without `?host=parent` nothing changes.


Claude-Session: https://claude.ai/code/session_01G6GtUCaPiCMEQCN4QcdQnp

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Print: a scaled annotation appearance no longer prints larger than its rectangle (#802) (#861)

* Single instance: enable tab handoff in Windows portable mode (#889) (#891)

* Single instance: enable tab handoff in Windows portable mode (#889)

* Single instance: a portable copy's socket stays in %LOCALAPPDATA%, under its own name

(cherry picked from commit c7db1e80a374e80a11504efefbaac680dfa5e872)

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Issue #390: build the DMG with hdiutil create so the app passes codesign --strict (#823)

* Issue #390: build the DMG with hdiutil create so the app passes codesign --strict

makehybrid -hfs wrote a com.apple.FinderInfo (Finder location -1,-1) on every
file in the image, and codesign --verify --deep --strict rejects that as
"Finder information, or similar detritus", on the app in the DMG and on a copy
installed from it. hdiutil create -srcfolder writes the files as they are; it
attaches a device while copying, so it gets up to five tries. A gate mounts
the finished image read-only and verifies the app inside with --strict.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01REz7wTfQgnopyvN9eQJJVu

* Issue #390: retry hdiutil attach and detach in the DMG gate too

As suggested in review: the gate mounts the signed image under set -e, so a
"Resource busy" there would stop a release after the DMG was built. The same
five tries with growing pauses now cover create, attach and detach.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01REz7wTfQgnopyvN9eQJJVu

---------

Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

* M4: make page-label generation lazy across large documents (#845)

* M4: make page-label generation lazy across large documents

Page labels were expanded into one String per page, and the whole sequence was refused once the labels passed 4 MiB in total. A valid document with long labels could therefore neither be read nor edited. Add a PageLabels view over the label ranges: label(page) builds one label, and iter() builds the labels one at a time, so no string is kept per page. Edits validate each range from its largest number (Roman numerals by their thousands) instead of building every label, so the check no longer depends on the page count.

The per-label length limit, the number-tree work and depth limits and the overflow checks are unchanged. page_labels() keeps its signature and returns the same labels as before for ordinary documents.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>

* M4: drop the aggregate page-label cap from document inspection

Document inspection refused every custom label once their total passed 4 MiB, so a valid document with long labels showed physical page numbers and a warning. Labels are now applied to every page. The per-label length limit, the number-tree work, depth and cycle limits and the prefix bounds are unchanged, and labels are still applied only after every range is accepted.

MAX_LABEL_TOTAL_BYTES is removed. The inspector keeps one label per page (PageInfo.label), so its label memory grows with the page count times the label length. The page count is bounded by the objects in the file, and each label by 1,024 bytes.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>

* M4: keep the inspector's 4 MiB page-label cap until its labels are lazy

Commit a096ece removed the aggregate page-label cap from document inspection. Inspector::fill keeps one label String per page on open, and inspection does not read labels from the lazy PageLabels view yet, so a hostile file with a million labelled pages would allocate on the order of a gigabyte when it is opened (measured here at about 1.8 KiB per labelled page). This reverts that commit. The cap stays until inspection reads labels lazily, as a separate change. The organizer change in 5253124 does not depend on it.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>

* lopdf: bound object lookups through compressed containers and object-stream members (#771)

A compressed object's container was resolved with a fresh cycle set and
without checking that it is an ordinary object, and indirect /Length
chains had no depth limit, so self-containing or chained lookups
overflowed the stack. Pass the caller's set on, require an in-file
container and cap the chain at 16 lookups. Object-stream members were
each parsed to the end of the decoded stream, so repeated or nested
offsets kept many copies of one large object: parse each member within
its own extent and skip repeated offsets.

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

* hayro-syntax: cap the nesting of direct arrays and dictionaries (#772)

Skipping a direct array or dictionary, and the lenient repair of a
dictionary that has an object where a key belongs, recursed once per
level with no bound, so deep nesting in a trailer entry or a content
stream overflowed the stack. Count the depth on the Reader and fail
beyond 128 levels like any other malformed object.

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M11: Reduce file size shrinks office scans, and never saves a bigger copy (refs #490) (#849)

Office scanners write the default decode array (/Decode [0 1 0 1 0 1]) on their page
images. Reduce skipped every image with a decode array, so a scan stayed the same
size. The default array is now treated as absent; any other array still skips the
image.

Reduce also saved its copy even when it wasn't smaller, reported as "-N% smaller".
It now saves nothing and says the file is already as small as it can be made;
doc_reduce returns "written": false. Optimize PDF still saves (what it discards
may matter more than the size) but says "not smaller".

Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M11: Reduce file size and Optimize PDF drop images nothing draws, such as a deleted image (#894)

Deleting an image with Edit a PDF removes its Do operation, but the image stays in the
page's /Resources /XObject, so every full rewrite (Reduce file size, Optimize PDF)
carried its data along: a page left blank still weighed the whole picture.

Clean Up now drops image and form XObjects from a page's own resources when their name
is written in no stream the document still uses. Shared and inherited resources are left
alone, and if any stream can't be read the way a viewer would read it, nothing is removed:
viewers look names up through parent resources, so only a name written nowhere is safe to
drop. doc_optimize reports unused_xobjects.

Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* M9: name the Windows store certificates that can't sign (#179) (#761)

A store certificate PdfCraft can't use (no private key, a P-521 /
Brainpool / Ed25519 key, a key CNG can't open) was dropped from the
Sign dialog silently, so a user whose CA-issued identity shows in
Acrobat but not here had nothing to go on. `windows::list` now returns
those under `unusable` with the subject, fingerprint and a reason that
names the Windows error code; the Sign dialog shows them greyed and
unselectable with the reason, and `sign_windows_ids` returns them.
A key that needs a prompt to open (NTE_SILENT_CONTEXT, NTE_UI_REQUIRED
while listing) is listed and prompts at signing instead of being
skipped. Certificates without a private key stay out of the dialog but
are reported by the tool.

Tests: the explained-error table, the store listing's invariants, a
key-less certificate in the ignored store test, the automation output,
and a kittest dialog test for a greyed entry. Verified by hand that a
.p12 imported into the CNG and legacy Microsoft providers lists, and
that a P-521 certificate is explained.

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M12: XFA rows in areas get their instances, and areas stay out of data paths (#853)

A Designer form whose table (a top-to-bottom subform inside an area of a
positioned page) asks for several rows with `<occur min="10"/>` was laid
out with one row, and a FormCalc total looping over the rows stayed
empty.

- layout: a repeating subform inside a top-to-bottom container in
  positioned content is drawn and measured once per instance (occur and
  data), with its own SOM index, as flowed content already was. One
  `instances` helper now serves flowing, table rows and this case.
- data: areas are named in SOM expressions but are not data scopes.
  `data_path(tpl, som)` leaves them out (unknown segments are kept), and
  layout, the script tree, reading and writing the datasets and the
  engine's value and row operations use it. Before, values Adobe's
  viewers saved were not read, what was filled here went under a new
  element named after the area (which Adobe's viewers would most likely
  not read), and scripts counted the rows there: one, so a loop over
  the table's rows failed on the missing ones.
- write_datasets_with / read_values_with take the parsed template when
  the caller has it (the engine's cached one); the old functions parse
  the document's.
- Tests: layout regression, data_path rules, and an engine round trip
  shaped like such a form (rows, Adobe-written values, a FormCalc total
  over every row, the data written back, reopening).

Signed-off-by: Hadi Chokr <hadichokr@icloud.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Exported images keep their transparency, JPEG cut-outs included (#645) (#763)

* M10: exported images keep their transparency, JPEG cut-outs included (#645)

Export all images and Save image as wrote a JPEG with a soft mask as the
bare JPEG and dropped soft masks of another size, JPEG-compressed soft
masks, stencil /Mask images and colour-key /Mask arrays, so cut-out
images came out on a solid white or black background.

- An image with transparency (/SMask, stencil /Mask, colour-key /Mask) is
  written as a PNG with it as alpha, JPEG images included (decoded with
  the image crate's JPEG decoder, already in the tree). JPEGs without
  transparency are still written byte for byte.
- Soft and stencil masks of another size are scaled to the image's size;
  /Decode inversion is honoured and /Matte pre-blending is undone.
- Image sizes are checked (positive, at most 100 Mpx) before use, so a
  negative /Width no longer overflows in Save image as.
- Regression tests: JPEG + soft mask, scaled soft mask, stencil mask,
  colour key, matte, and damaged masks/sizes (all fail or panic on main).

Fixes #645

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* M10: use as_chunks for colour-key pairs (clippy 1.99 chunks_exact_to_as_chunks) (#645)

Rust 1.99's clippy denies chunks_exact with a constant size, which failed
the Clippy step on all CI targets. as_chunks::<2> also drops the indexing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Fill colour for rectangle, oval and polygon comments (#686) (#793)

* M5: fill colour for rectangle, oval and polygon comments (#686)

Rectangles, ovals, polygons and clouds could only be given a border
colour. They can now be filled, or have their fill removed:

- the comment quick bar has a Fill colour control for those tools
  ("No fill" or a swatch), kept with the other per-tool styles;
- Comment Properties > Appearance has a Fill colour row, applied in the
  same undo step as the other changes and carried over by Make Current
  Properties Default;
- comment_edit takes `fill` (a colour, or "none").

pdfcraft-annot gains `set_fill` (writes or removes /IC and redraws the
appearance; lines, ink and locked comments refuse it) and reports
`fillable`/`fill` in the comment properties. The engine adds
Edit::FillAnnotation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019XAJjko29kzKj7ULXoXSnN

* M5: update ATTRIBUTION.toml hashes for the edited i18n catalogs (#686)

The fill-colour strings changed 13 translation catalogs, so
xtask's the_real_manifest_passes failed on ubuntu and windows CI.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M14: localize displayed shortcut key names (#496) (#884)

Co-authored-by: Tevinhead <173904641+Tevinhead@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M3, #740: column select (Alt/Option-drag, or Edit ▸ Column select) (#765)

As in Acrobat, holding Alt (Option on macOS) when a drag starts selects only
the text inside the rectangle drawn, instead of everything between two
points in reading order. It works with the Select tool, the text markup
tools and Redact, so a column can be copied, highlighted or redacted.

- render: PageText::glyphs_in (glyph centre inside the rectangle),
  glyph_rects (one box per run, never covering unselected text) and
  column_text: rows top to bottom, side-by-side pieces separated by a tab,
  right to left in right-to-left rows, reading order for vertical text.
- ui-egui: a selection is a reading-order run or a column rectangle; the
  rectangle is outlined while dragging and the cursor is a crosshair.
  ⇧-click clears a column selection (it has no anchor glyph).
- Edit ▸ Column select (edit.column_select, quick tool "column-select")
  does it with a plain drag. It is the way in on Linux desktops whose
  window manager takes Alt-drag for moving windows (Xfce, Cinnamon, MATE,
  KDE Plasma 5 by default), where the app never sees the drag.
- automation: text_extract takes rect for the same extraction headless.
- "Column select" translated in all 15 catalogs (Arabic included); their ATTRIBUTION.toml
  hashes updated.
- parity: new view.column-select entry with platform notes and gaps
  (no selection across pages, no ⇧-click extension).

Windows: winit hands an Alt release to the system menu loop only when the
window has a native menu, which PdfCraft's does not, so Alt-drag cannot
leave the window in menu mode.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Combine files: a thumbnail grid, card sizes and card actions (refs #611) (#777)

Combine files shows each file as a card (the default; Grid/List kept in
the settings, --combine-view): the first page it adds, stacked when it
adds more, its name shortened in the middle. Cards select (click, Ctrl,
Shift, arrows by card and by row), drag to reorder (one undo step; the
grid scrolls at its edges) and share the list's toolbar, selection and
history. Locked, unreadable, bad-range and unmeasurable files say why.

Thumbnails render off the UI thread, only for cards in view: at most
three render threads at once, counted until they exit (no watchdog
replacements), a 32 MB texture budget, keys that never show another
page, password or file, and one heavy step per frame on the web.

Cards are sized from 60% to 200% (toolbar slider and buttons, Ctrl
+/-/0, Ctrl+wheel or pinch over the grid, kept in the settings,
--combine-zoom), keeping the card under the pointer in place. Hovering
a card shows its actions over the top of its page: remove that file,
and a magnifier that shows it large, flipping through the pages it
adds; the same on a right click, and Space by keyboard. The toolbar
wraps between groups at narrow widths.

RenderPool gains retire(), without_replacements() and threads() so a
caller can bound the render threads that exist at once.

Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Combine files: expand a file into its pages (2 of 3, refs #611) (#788)

* Combine files: a thumbnail grid, card sizes and card actions (refs #611)

Combine files shows each file as a card (the default; Grid/List kept in
the settings, --combine-view): the first page it adds, stacked when it
adds more, its name shortened in the middle. Cards select (click, Ctrl,
Shift, arrows by card and by row), drag to reorder (one undo step; the
grid scrolls at its edges) and share the list's toolbar, selection and
history. Locked, unreadable, bad-range and unmeasurable files say why.

Thumbnails render off the UI thread, only for cards in view: at most
three render threads at once, counted until they exit (no watchdog
replacements), a 32 MB texture budget, keys that never show another
page, password or file, and one heavy step per frame on the web.

Cards are sized from 60% to 200% (toolbar slider and buttons, Ctrl
+/-/0, Ctrl+wheel or pinch over the grid, kept in the settings,
--combine-zoom), keeping the card under the pointer in place. Hovering
a card shows its actions over the top of its page: remove that file,
and a magnifier that shows it large, flipping through the pages it
adds; the same on a right click, and Space by keyboard. The toolbar
wraps between groups at narrow widths.

RenderPool gains retire(), without_replacements() and threads() so a
caller can bound the render threads that exist at once.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Combine files: expand a file into its pages (refs #611)

A file that adds more than one page (up to 2000) expands into a card
per page it adds, in its range's order, each labelled with its page and
the file's name, on a shared tinted band: from its card's hover bar, a
right click, or Expand all / Collapse all in the toolbar. A page card's
magnifier opens the preview on that page; Up and Down step rows of
cards. Expanding is a view: not an undo step, kept across views and
zoom, and the combined file is the same either way.

Thumbnails are now kept per file and page, looked up on demand (the
cost follows the files, not the pages of an expanded file), and a
render's document stays open for two seconds so the next page of the
same file isn't read again.

Adds the Lucide minimize-2 icon (ISC, recorded in ATTRIBUTION).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Progress docs: target-app parity, gaps, alpha gate, readiness by audience (#873)

Progress docs per craftrules standards/progress-docs.md: target-app parity, gaps, roadmap with alpha gate, readiness by audience (full / mainstream / essentials) with Opus 5.5 hours, and feature, format, PDF-spec, hardware, UI and localization parity docs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Measure: snap to the implicit closing edge of filled paths (#803) (#862)

* fix(organize): reject rotation angles that are not multiples of 90 (#801) (#833)

Page rotation only turns in quarter steps, but the degree value was read with an integer division by 90 that threw the remainder away. Asking for 45 degrees saved an unrotated page, 91 and 135 saved a 90 degree turn, and the command still reported success.

Reject the value before touching the document instead, so every caller gets an actionable error and the pages are left alone. The automation tool already rejected it; the single edit command did not.

Fixes #801.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* Autosave: snapshot a new unsaved document before its first edit (#815) (#850)

* OCR: Recognize text in multiple files refuses paths outside the root (#795)

ocr_recognize_files turned a path outside the automation root into a per-file
error and returned success, so no_tool_escapes_the_root reported 9 leaks once
the OCR models were installed. Without them, loading the models failed first
and hid it, and CI never installs them. Every path is now checked before any
work starts, and one outside the root refuses the whole call, like the other
tools. A missing file inside the root stays a per-file error. Nothing was ever
read from outside the root.

ocr_recognize_files_refuses_paths_outside_the_root checks this without the
models.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(text): match phrases across repeated whitespace (#814) (#831)

The phrase search collapsed runs of whitespace in the query but not in the page text it was matching against, so a document drawn with two space glyphs in a row like the issue example never matched the phrase. Single words still matched, which is why it looked like only some files were affected.

Collapse whitespace while the text is flattened, the same way the query is normalised. The regression test fails on the old code because the phrase find returns an empty range, and it also covers tabs and single word queries.

Fixes #814.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* fix(edit): keep a comment whose appearance is missing when flattening (#805) (#838)

Viewers draw notes, FreeText, ink and stamps from their dictionary when the file stores no /AP /N. flatten only drew annotations that already had an appearance and then removed the rest, so a supported note disappeared and the saved page came back blank.

Build the appearance a viewer would show for every selected, visible, non-widget annotation that has none, then draw it as usual. A subtype this build cannot render still fails to build and is handled exactly as before.

The new synthetic test draws the note icon into the page and asserts it is baked in; without the fix flatten reports 0 drawn and the icon is lost.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* fix(comments): keep a reply to a reply in the comment listing (#806) (#839)

comment_list collected only the replies whose /IRT pointed at the root comment, so a reply to a reply was listed nowhere at all. Roots and the root count were already right; the nested reply was simply dropped from its thread.

Follow the /IRT chain to any depth, with a bound, so every reply under a comment is listed in document order. A reply to a reply now stays discoverable through the tool that promises every comment and its replies.

The new test builds a root note, a direct reply and a reply to that reply, then checks both texts are listed and the two notes stay roots. It listed only the direct reply before the fix.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* Issue #774: keep the catalog out of object streams in encrypted files (Acrobat) (#836)

Acrobat reports an encrypted PDF as "damaged and could not be repaired", before
asking for the password, when the document catalog sits in an object stream.
Every other reader tried (qpdf, pypdf, poppler, PDFKit, PdfCraft) accepts such
files, and qpdf keeps the catalog standalone too. Full saves of encrypted
documents now write the catalog as its own object; unencrypted output is
unchanged. Verified with Acrobat DC on macOS: the same file fails with the
catalog packed and opens with it standalone, in AES-256 and RC4-128.


Claude-Session: https://claude.ai/code/session_01REz7wTfQgnopyvN9eQJJVu

Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

* fix(edit): refuse a header start number that would wrap the next page (#804) (#834)

Page numbering adds the start number to the position of the page in the run. At the top of the u32 range that addition wrapped, so a release build saved zero for the second page and still reported success, and a debug build hit a caught overflow error instead.

Check the whole run before drawing anything. When the last page number would pass the u32 limit the operation returns an actionable error and the document keeps no header, rather than a half applied run.

Fixes #804.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* OCR: a scanned euro sign reads as €, not E (#796)

The ocrs crate's default alphabet has an E where its recognition model was
trained with € (ocrs-models DEFAULT_ALPHABET), so a scanned "€250" was
recognised as "E250". pdfcraft-ocr now gives ocrs the alphabet the model was
trained with.

the_alphabet_is_the_trained_one checks the alphabet without the models;
recognize_text_reads_the_euro_sign reads a scan end to end when they are
installed.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(comments): include replies to replies in the comment summary (#820) (#840)

The comment summary gathered a comment's direct replies only, matching /IRT against the comment itself. A reply to a reply points at the direct reply, so its text never reached the generated summary even though the stored discussion kept it.

Replies are now walked to any depth, with a bound, and indented one level per reply. The existing output for direct replies is unchanged.

The new test builds a rectangle, a reply and a reply to that reply, then checks the summary carries both texts with the nested one indented under its parent. It omitted the nested reply before the fix.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* fix(xfdf): drop the replaced note's popup when re-importing a comment (#807) (#860)

* fix(view): check Shift-Command-G before Command-G for find navigation (#810) (#835)

* view: check Shift-Command-G before Command-G for find navigation (#810)

* Tests: the find shortcut test waits for find by a deadline

(cherry picked from commit eaf3180bd8ebb353b3d8c5ca7b9ef7548ed55c37)

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* fix(editing): recognize paragraph drafts as unsaved work (#811) (#843)

* view: check Shift-Command-G before Command-G for find navigation (#810)

* fix: recognize paragraph drafts as unsaved work (Issue #811)

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* fix(automation): retain batch results with identical filenames (#877)

* fix(automation): retain batch results with identical filenames

* Format

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M13: preserve desktop Action Wizard batch results (#886)

* M13: preserve desktop Action Wizard batch results

Co-Authored-By: Codex <noreply@openai.com>

* ROADMAP: date the desktop action outputs entry today

---------

Co-authored-by: Tevinhead <173904641+Tevinhead@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Fit to width fits the shown page in single-page view (#808) (#859)

* fix(links): find URLs when lowercasing changes the text length (#816) (#830)

find_urls case-folded a copy of the page text and then sliced the original
characters with offsets taken from that copy. Case folding can change a
string's length — U+0130 (İ) lowercases to "i" plus a combining dot — so the
offsets drifted past the end of the source and `chars[i..j]` panicked. The CLI
exited 101 before it could create the link; MCP reported an internal tool error
but kept serving.

Match the ASCII prefixes case-insensitively against the original characters
instead, so every offset stays in the source's own index space, and use get()
so no lookup can index past the end.

A new test reproduces the reported panic on the old code (range start index 41
out of range for slice of length 39) and passes with the fix. pdfcraft-annot's
tests, clippy and rustfmt are green.

Fixes #816.

Co-authored-by: Big Pickle <noreply@opencode.ai>

* organize: deleting the opening page drops the catalog OpenAction (#821) (#855)

* fix(text): Turkish dotted and dotless i match in case-insensitive search (#892)

Case-insensitive search folded the query and the page with
str::to_lowercase, which follows no locale: U+0130 (İ) becomes "i" plus a
combining dot above (two characters) and U+0131 (ı) has no case pair. So
"istanbul" never found "İSTANBUL", and "ırmak"/"kış" never found
"IRMAK"/"KIŞ" (or the reverse). The same PageText::find serves Find,
Search & Redact (phrases and word lists) and the automation text search,
so Turkish words could also be left unredacted.

Fold I, i, İ and ı to "i" and every other character with its own
lowercase, one character for one character. Case-sensitive search is
unchanged.

* XFDF keeps polygon and polyline vertices (#809) (#848)

Export wrote polygon and polyline comments without their points and import
had nowhere to read them from, so an XFDF round trip left the comments with
no /Vertices and no appearance. Export now writes the points as XFDF
<vertices>x,y;x,y;…</vertices>, and import reads them back (whole x,y pairs
only) before the appearance is drawn.

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* XFDF keeps a callout's leader, text-box inset, intent and arrowhead (#819) (#854)

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Web: warn before reloading a document with unsaved changes (#812) (#856)

* Issue #813: a cancelled recovery password no longer attaches to the next unlocked document (#858)

Recovering an encrypted snapshot parked its metadata in `pending_recovered`
until the password prompt was answered. Cancel left it there, and the next
encrypted document unlocked in the foreground took the snapshot's original
path and recovery key: Save would overwrite the other file and delete its
snapshot.

The association now travels with the prompt (`PasswordPrompt::recovered`).
Cancel puts the entry back on offer and leaves the store alone; a wrong
password keeps it on the re-raised prompt; the correct one finishes the
recovery of the tab that was just opened. `pending_recovered` is gone.

Test: crash with an encrypted snapshot of A, Recover, Cancel, open encrypted
B with its password. B keeps its own path and is clean, A's snapshot stays
in the store, also after closing B. Fails on main at "B keeps its own path".

Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

* Viewer: touchpad scrolling tracks the finger 1:1 (#759) (#868)

A precision touchpad sends pixel-precise `MouseWheel` events. On Windows
they arrive with no `TouchPhase::Start`, so egui never takes its
no-smoothing "in touch" path and eases any delta of 8 px or more over
several frames, leaving the page a frame behind the finger.

The page view now applies precise (Point-unit) wheel input itself, 1:1,
through the same un-eased scroll path as keyboard and middle-button
auto-scroll, and suppresses egui's eased copy for a short window so its
internal easing tail neither doubles nor trails the motion. A coarse
mouse wheel (Line/Page units) keeps egui's smoothing, which is pleasant
for chunky notches, and single-page paging still owns the vertical axis.

Regression test: one 200 px precise swipe scrolls the full 200 px at
once (an eased path would ramp there over ~0.1 s) and leaves no drift.

Co-authored-by: Justin John <babyjjbabyjj549@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* Keep every part when bookmark titles repeat in doc_split (#800) (#863)

* Windows: add reproducible security diagnostics for storytold/pdfcraft#747 (#769)

Add packaging/windows/security-audit.ps1, a read-only maintainer helper that records a
release file's SHA-256, Authenticode status and, with an external YARA binary and rule,
every matched string identifier and offset. Document the procedure in docs/releasing.md
("Scanner detections").

No application, dependency, installer or network behaviour changes: the diff touches only
the helper and the release documentation.

Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>

* Notice bar: the message wraps beside its buttons and lines up with them (#785) (#832)

The notice bar above the pages (security, forms, XFA, repaired file,
warnings, signatures) laid its message out before its buttons. In a
narrow document area (a side panel open, a longer translation) the
message ran under "Security settings" and on into the side panel, and it
wasn't centred on the buttons.

A shared notice_bar helper now lays the buttons out first, right to
left, centred on a 28 px row, and gives the message the width that's
left: it wraps there (cut short with the full text on hover when less
than 120 px is left) and its first line is centred on the buttons' row.

Regression test: editing::the_security_notice_wraps_beside_its_buttons_and_lines_up_with_them
(one line, wrapped and cut-short widths; fails on main).

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* M10.2: scan JPEG metadata beyond frame headers (#764)

Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Issue #878: Combine stores an image once when its colour space is an indirect name (#883)

ImageMagick writes an image's colour space as its own object (`/ColorSpace 10 0 R`,
`10 0 obj /DeviceRGB`). Each source's copy of that object was never merged, since it
isn't a resource, so identical images compared unequal and every copy was kept:
8 copies of an 80 MB ImageMagick PDF combined into 648 MB instead of 81 MB.

Dedupe now compares a reference to a plain value (name, number, boolean or string) as
that value. The value objects themselves are still never merged. Combine, Insert Pages
and Reduce File Size share this code.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* organize: drop named destinations and their links when a page is deleted (#817) (#851)

* JavaScript: limit util.printf widths and output, and array arguments (#770)

util.printf accumulated a conversion's width and precision with unchecked
arithmetic and no maximum, and its output had no bound; the array
arguments of resetForm and the field value/colour setters were read for
as many items as the array's length says, which a sparse array makes
free. Saturate the numbers and refuse widths and precisions above 4096,
output above 1 MiB and arrays longer than 2^20 with a script error,
before anything is allocated from them.

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

* Linux: "Use system setting" asks the desktop, not winit (#881)

winit reports no system theme on Linux — `ActiveEventLoop::system_theme` is
`None` there on every desktop — so `ThemePreference::System` never saw the
desktop's choice and silently fell back to whatever was on screen: PdfCraft
came up light on a dark desktop, and picking "Use system setting" after Dark
kept it dark. The default setting did not follow the system.

`pdfcraft-platform` now reads the XDG Desktop Portal's Settings interface off
the UI thread and the app follows the answer, including a change made while it
runs. The colour scheme is asked for under its GNOME schema name first, which
GNOME's and XFCE's portals both answer, then under the older freedesktop one,
which KDE answers. A desktop that answers neither leaves the user's own choice
alone, exactly as before.

Co-authored-by: GyuriX <gyurix@users.noreply.github.com>

* Extract/Combine: a page repeated in one selection gets its own annotations (#818) (#852)

* M5.7: Fill & Sign tools pick up placed signatures to move and resize them (#822)

A signature placed with Fill & Sign could only be moved with the Select
tool. While a Fill & Sign tool was active (Sign, Initials, Add text,
Check, Date...), hovering the signature showed the placement crosshair,
a drag did nothing and a click stamped another mark on top of it.

Now, as in Acrobat, a Fill & Sign tool picks up the Fill & Sign marks
already on the page: hovering one shows the move cursor, a click selects
it, a drag moves it and the selected mark's handles resize it. Away from
them the tool places as before, including on other comments. Image
signatures keep their live move/resize preview.

Comments now carry a fill_sign flag (pdfcraft_annot::is_fill_sign for
the engine's list, the same rules in the viewer's inspector), so this
also works on reopened files.

Co-authored-by: lloyd <lloyd@stackworx.io>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* Web: opt-in embedding bridge for host pages (?host=parent) (#842)

A page that frames PdfCraft on the same origin can now pass it PDFs and
receive saves through postMessage, instead of Save becoming a browser
download. Built for a Nextcloud integration, but nothing in it is
host-specific.

- ui-egui: HostSaveFn/HostDirtyFn hooks (`host_save`, `host_dirty`),
  None by default. On wasm, Save hands the bytes to `host_save` when
  set; `host_dirty` hears each frame whether any tab has unsaved work.
- pdfcraft-web: `host::Bridge`, enabled only by `?host=parent` inside a
  frame. Messages: pdfcraft:ready / open / save / dirty / failed; only
  the parent window on our own origin is heard or answered.
  `?author=` names comment authors unless the user chose a name.
- Test: host_dirty_hears_when_unsaved_work_appears_and_goes.

Without `?host=parent` nothing changes.


Claude-Session: https://claude.ai/code/session_01G6GtUCaPiCMEQCN4QcdQnp

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Print: a scaled annotation appearance no longer prints larger than its rectangle (#802) (#861)

* Single instance: enable tab handoff in Windows portable mode (#889) (#891)

* Single instance: enable tab handoff in Windows portable mode (#889)

* Single instance: a portable copy's socket stays in %LOCALAPPDATA%, under its own name

(cherry picked from commit c7db1e80a374e80a11504efefbaac680dfa5e872)

---------

Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Issue #390: build the DMG with hdiutil create so the app passes codesign --strict (#823)

* Issue #390: build the DMG with hdiutil create so the app passes codesign --strict

makehybrid -hfs wrote a com.apple.FinderInfo (Finder location -1,-1) on every
file in the image, and codesign --verify --deep --strict rejects that as
"Finder information, or similar detritus", on the app in the DMG and on a copy
installed from it. hdiutil create -srcfolder writes the files as they are; it
attaches a device while copying, so it gets up to five tries. A gate mounts
the finished image read-only and verifies the app inside with --strict.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01REz7wTfQgnopyvN9eQJJVu

* Issue #390: retry hdiutil attach and detach in the DMG gate too

As suggested in review: the gate mounts the signed image under set -e, so a
"Resource busy" there would stop a release after the DMG was built. The same
five tries with growing pauses now cover create, attach and detach.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01REz7wTfQgnopyvN9eQJJVu

---------

Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

* M4: make page-label generation lazy across large documents (#845)

* M4: make page-label generation lazy across large documents

Page labels were expanded into one String per page, and the whole sequence was refused once the labels passed 4 MiB in total. A valid document with long labels could therefore neither be read nor edited. Add a PageLabels view over the label ranges: label(page) builds one label, and iter() builds the labels one at a time, so no string is kept per page. Edits validate each range from its largest number (Roman numerals by their thousands) instead of building every label, so the check no longer depends on the page count.

The per-label length limit, the number-tree work and depth limits and the overflow checks are unchanged. page_labels() keeps its signature and returns the same labels as before for ordinary documents.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>

* M4: drop the aggregate page-label cap from document inspection

Document inspection refused every custom label once their total passed 4 MiB, so a valid document with long labels showed physical page numbers and a warning. Labels are now applied to every page. The per-label length limit, the number-tree work, depth and cycle limits and the prefix bounds are unchanged, and labels are still applied only after every range is accepted.

MAX_LABEL_TOTAL_BYTES is removed. The inspector keeps one label per page (PageInfo.label), so its label memory grows with the page count times the label length. The page count is bounded by the objects in the file, and each label by 1,024 bytes.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>

* M4: keep the inspector's 4 MiB page-label cap until its labels are lazy

Commit a096ece removed the aggregate page-label cap from document inspection. Inspector::fill keeps one label String per page on open, and inspection does not read labels from the lazy PageLabels view yet, so a hostile file with a million labelled pages would allocate on the order of a gigabyte when it is opened (measured here at about 1.8 KiB per labelled page). This reverts that commit. The cap stays until inspection reads labels lazily, as a separate change. The organizer change in 5253124 does not depend on it.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>

* lopdf: bound object lookups through compressed containers and object-stream members (#771)

A compressed object's container was resolved with a fresh cycle set and
without checking that it is an ordinary object, and indirect /Length
chains had no depth limit, so self-containing or chained lookups
overflowed the stack. Pass the caller's set on, require an in-file
container and cap the chain at 16 lookups. Object-stream members were
each parsed to the end of the decoded stream, so repeated or nested
offsets kept many copies of one large object: parse each member within
its own extent and skip repeated offsets.

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

* hayro-syntax: cap the nesting of direct arrays and dictionaries (#772)

Skipping a direct array or dictionary, and the lenient repair of a
dictionary that has an object where a key belongs, recursed once per
level with no bound, so deep nesting in a trailer entry or a content
stream overflowed the stack. Count the depth on the Reader and fail
beyond 128 levels like any other malformed object.

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M11: Reduce file size shrinks office scans, and never saves a bigger copy (refs #490) (#849)

Office scanners write the default decode array (/Decode [0 1 0 1 0 1]) on their page
images. Reduce skipped every image with a decode array, so a scan stayed the same
size. The default array is now treated as absent; any other array still skips the
image.

Reduce also saved its copy even when it wasn't smaller, reported as "-N% smaller".
It now saves nothing and says the file is already as small as it can be made;
doc_reduce returns "written": false. Optimize PDF still saves (what it discards
may matter more than the size) but says "not smaller".

Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M11: Reduce file size and Optimize PDF drop images nothing draws, such as a deleted image (#894)

Deleting an image with Edit a PDF removes its Do operation, but the image stays in the
page's /Resources /XObject, so every full rewrite (Reduce file size, Optimize PDF)
carried its data along: a page left blank still weighed the whole picture.

Clean Up now drops image and form XObjects from a page's own resources when their name
is written in no stream the document still uses. Shared and inherited resources are left
alone, and if any stream can't be read the way a viewer would read it, nothing is removed:
viewers look names up through parent resources, so only a name written nowhere is safe to
drop. doc_optimize reports unused_xobjects.

Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* M9: name the Windows store certificates that can't sign (#179) (#761)

A store certificate PdfCraft can't use (no private key, a P-521 /
Brainpool / Ed25519 key, a key CNG can't open) was dropped from the
Sign dialog silently, so a user whose CA-issued identity shows in
Acrobat but not here had nothing to go on. `windows::list` now returns
those under `unusable` with the subject, fingerprint and a reason that
names the Windows error code; the Sign dialog shows them greyed and
unselectable with the reason, and `sign_windows_ids` returns them.
A key that needs a prompt to open (NTE_SILENT_CONTEXT, NTE_UI_REQUIRED
while listing) is listed and prompts at signing instead of being
skipped. Certificates without a private key stay out of the dialog but
are reported by the tool.

Tests: the explained-error table, the store listing's invariants, a
key-less certificate in the ignored store test, the automation output,
and a kittest dialog test for a greyed entry. Verified by hand that a
.p12 imported into the CNG and legacy Microsoft providers lists, and
that a P-521 certificate is explained.

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M12: XFA rows in areas get their instances, and areas stay out of data paths (#853)

A Designer form whose table (a top-to-bottom subform inside an area of a
positioned page) asks for several rows with `<occur min="10"/>` was laid
out with one row, and a FormCalc total looping over the rows stayed
empty.

- layout: a repeating subform inside a top-to-bottom container in
  positioned content is drawn and measured once per instance (occur and
  data), with its own SOM index, as flowed content already was. One
  `instances` helper now serves flowing, table rows and this case.
- data: areas are named in SOM expressions but are not data scopes.
  `data_path(tpl, som)` leaves them out (unknown segments are kept), and
  layout, the script tree, reading and writing the datasets and the
  engine's value and row operations use it. Before, values Adobe's
  viewers saved were not read, what was filled here went under a new
  element named after the area (which Adobe's viewers would most likely
  not read), and scripts counted the rows there: one, so a loop over
  the table's rows failed on the missing ones.
- write_datasets_with / read_values_with take the parsed template when
  the caller has it (the engine's cached one); the old functions parse
  the document's.
- Tests: layout regression, data_path rules, and an engine round trip
  shaped like such a form (rows, Adobe-written values, a FormCalc total
  over every row, the data written back, reopening).

Signed-off-by: Hadi Chokr <hadichokr@icloud.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Exported images keep their transparency, JPEG cut-outs included (#645) (#763)

* M10: exported images keep their transparency, JPEG cut-outs included (#645)

Export all images and Save image as wrote a JPEG with a soft mask as the
bare JPEG and dropped soft masks of another size, JPEG-compressed soft
masks, stencil /Mask images and colour-key /Mask arrays, so cut-out
images came out on a solid white or black background.

- An image with transparency (/SMask, stencil /Mask, colour-key /Mask) is
  written as a PNG with it as alpha, JPEG images included (decoded with
  the image crate's JPEG decoder, already in the tree). JPEGs without
  transparency are still written byte for byte.
- Soft and stencil masks of another size are scaled to the image's size;
  /Decode inversion is honoured and /Matte pre-blending is undone.
- Image sizes are checked (positive, at most 100 Mpx) before use, so a
  negative /Width no longer overflows in Save image as.
- Regression tests: JPEG + soft mask, scaled soft mask, stencil mask,
  colour key, matte, and damaged masks/sizes (all fail or panic on main).

Fixes #645

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* M10: use as_chunks for colour-key pairs (clippy 1.99 chunks_exact_to_as_chunks) (#645)

Rust 1.99's clippy denies chunks_exact with a constant size, which failed
the Clippy step on all CI targets. as_chunks::<2> also drops the indexing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Fill colour for rectangle, oval and polygon comments (#686) (#793)

* M5: fill colour for rectangle, oval and polygon comments (#686)

Rectangles, ovals, polygons and clouds could only be given a border
colour. They can now be filled, or have their fill removed:

- the comment quick bar has a Fill colour control for those tools
  ("No fill" or a swatch), kept with the other per-tool styles;
- Comment Properties > Appearance has a Fill colour row, applied in the
  same undo step as the other changes and carried over by Make Current
  Properties Default;
- comment_edit takes `fill` (a colour, or "none").

pdfcraft-annot gains `set_fill` (writes or removes /IC and redraws the
appearance; lines, ink and locked comments refuse it) and reports
`fillable`/`fill` in the comment properties. The engine adds
Edit::FillAnnotation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019XAJjko29kzKj7ULXoXSnN

* M5: update ATTRIBUTION.toml hashes for the edited i18n catalogs (#686)

The fill-colour strings changed 13 translation catalogs, so
xtask's the_real_manifest_passes failed on ubuntu and windows CI.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M14: localize displayed shortcut key names (#496) (#884)

Co-authored-by: Tevinhead <173904641+Tevinhead@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* M3, #740: column select (Alt/Option-drag, or Edit ▸ Column select) (#765)

As in Acrobat, holding Alt (Option on macOS) when a drag starts selects only
the text inside the rectangle drawn, instead of everything between two
points in reading order. It works with the Select tool, the text markup
tools and Redact, so a column can be copied, highlighted or redacted.

- render: PageText::glyphs_in (glyph centre inside the rectangle),
  glyph_rects (one box per run, never covering unselected text) and
  column_text: rows top to bottom, side-by-side pieces separated by a tab,
  right to left in right-to-left rows, reading order for vertical text.
- ui-egui: a selection is a reading-order run or a column rectangle; the
  rectangle is outlined while dragging and the cursor is a crosshair.
  ⇧-click clears a column selection (it has no anchor glyph).
- Edit ▸ Column select (edit.column_select, quick tool "column-select")
  does it with a plain drag. It is the way in on Linux desktops whose
  window manager takes Alt-drag for moving windows (Xfce, Cinnamon, MATE,
  KDE Plasma 5 by default), where the app never sees the drag.
- automation: text_extract takes rect for the same extraction headless.
- "Column select" translated in all 15 catalogs (Arabic included); their ATTRIBUTION.toml
  hashes updated.
- parity: new view.column-select entry with platform notes and gaps
  (no selection across pages, no ⇧-click extension).

Windows: winit hands an Alt release to the system menu loop only when the
window has a native menu, which PdfCraft's does not, so Alt-drag cannot
leave the window in menu mode.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Combine files: a thumbnail grid, card sizes and card actions (refs #611) (#777)

Combine files shows each file as a card (the default; Grid/List kept in
the settings, --combine-view): the first page it adds, stacked when it
adds more, its name shortened in the middle. Cards select (click, Ctrl,
Shift, arrows by card and by row), drag to reorder (one undo step; the
grid scrolls at its edges) and share the list's toolbar, selection and
history. Locked, unreadable, bad-range and unmeasurable files say why.

Thumbnails render off the UI thread, only for cards in view: at most
three render threads at once, counted until they exit (no watchdog
replacements), a 32 MB texture budget, keys that never show another
page, password or file, and one heavy step per frame on the web.

Cards are sized from 60% to 200% (toolbar slider and buttons, Ctrl
+/-/0, Ctrl+wheel or pinch over the grid, kept in the settings,
--combine-zoom), keeping the card under the pointer in place. Hovering
a card shows its actions over the top of its page: remove that file,
and a magnifier that shows it large, flipping through the pages it
adds; the same on a right click, and Space by keyboard. The toolbar
wraps between groups at narrow widths.

RenderPool gains retire(), without_replacements() and threads() so a
caller can bound the render threads that exist at once.

Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>

* Combine files: expand a file into its pages (refs #611)

A file that adds more than one page (up to 2000) expands into a card
per page it adds, in its range's order, each labelled with its page and
the file's name, on a shared tinted band: from its card's hover bar, a
right click, or Expand all / Collapse all in the toolbar. A page card's
magnifier opens the preview on that page; Up and Down step rows of
cards. Expanding is a view: not an undo step, kept across views and
zoom, and the combined file is the same either way.

Thumbnails are now kept per file and page, looked up on demand (the
cost follows the files, not the pages of an expanded file), and a
render's document stays open for two seconds so the next page of the
same file isn't read again.

Adds the Lucide minimize-2 icon (ISC, recorded in ATTRIBUTION).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Signed-off-by: Hadi Chokr <hadichokr@icloud.com>
Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: MarekSpodBiedry <73122968+MarekSpodBiedry@users.noreply.github.com>
Co-authored-by: muraa-p <85794781+muraa-p@users.noreply.github.com>
Co-authored-by: Big Pickle <noreply@opencode.ai>
Co-authored-by: mlkonrad <marlonk@gmail.com>
Co-authored-by: Tristan von Kries <tristanvonkries@gmail.com>
Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Krzysztof Cieślik <132496025+Hathor875@users.noreply.github.com>
Co-authored-by: Nigel Ihza Firdausy <106657905+Ni-zav@users.noreply.github.com>
Co-authored-by: Tevinhead <barrett.wallace@gmail.com>
Co-authored-by: Tevinhead <173904641+Tevinhead@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: snipeTR <31975916+snipeTR@users.noreply.github.com>
Co-authored-by: DWAA1660 <dwatnip123@gmail.com>
Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Justin John <34035011+justinjohn0306@users.noreply.github.com>
Co-authored-by: Justin John <babyjjbabyjj549@gmail.com>
Co-authored-by: cyfung1031 <44498510+cyfung1031@users.noreply.github.com>
Co-authored-by: 才倾梦华V <185041488+CQMHV@users.noreply.github.com>
Co-authored-by: Ikki Nishihara <b25d025@akita-pu.ac.jp>
Co-authored-by: gyurix <barathgy@gmail.com>
Co-authored-by: GyuriX <gyurix@users.noreply.github.com>
Co-authored-by: LloydNicholson <34131158+LloydNicholson@users.noreply.github.com>
Co-authored-by: lloyd <lloyd@stackworx.io>
Co-authored-by: CyberFox-81 <brittondcouchman@gmail.com>
Co-authored-by: Aviral Bansal <94526754+A-viral-dev@users.noreply.github.com>
Co-authored-by: Epsilon-LLC <64039063+EpsilonLLC@users.noreply.github.com>
Co-authored-by: Silver <hadichokr@icloud.com>

* Combine files: move and remove single pages before combining (closes #611)

With a file expanded, single pages are selected (click, Ctrl, Shift),
dragged anywhere and removed (a page card's trash, the toolbar, Delete),
each one undo step; the files themselves are never changed. A page moved
between another file's pages splits its file into parts that share its
bytes: the list shows each part as a row with its own page range, and
parts that end up side by side join again. A split file still counts,
unlocks and combines as one file.

organize gains combine_grouped (and Session::combine_grouped): each file
is copied once and its pages put in the order shown, so it keeps one
bookmark at its first page, its nested bookmarks, links between its own
pages, its fields and one copy of its attachments. A page listed twice
now gets its own copies of its annotations, and a form field's widget on
it joins the same field, so the field keeps one value. The existing
combine is unchanged; doc_combine gains an optional groups
array (entries with the same number are one file split into parts), so
the same combine is reachable headless.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Combine: a repeated widget joins its field's /Kids array object too

* Organize tests: a repeated widget that is its own field stays one object (#790)

---------

Signed-off-by: Hadi Chokr <hadichokr@icloud.com>
Co-authored-by: Andrew Taylor <andrew.taylor@tayeng.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: MarekSpodBiedry <73122968+MarekSpodBiedry@users.noreply.github.com>
Co-authored-by: muraa-p <85794781+muraa-p@users.noreply.github.com>
Co-authored-by: Big Pickle <noreply@opencode.ai>
Co-authored-by: mlkonrad <marlonk@gmail.com>
Co-authored-by: Tristan von Kries <tristanvonkries@gmail.com>
Co-authored-by: Tristan von Kries <tristan@MacBook-Pro-von-Tristan.local>
Co-authored-by: Krzysztof Cieślik <132496025+Hathor875@users.noreply.github.com>
Co-authored-by: Nigel Ihza Firdausy <106657905+Ni-zav@users.noreply.github.com>
Co-authored-by: Tevinhead <barrett.wallace@gmail.com>
Co-authored-by: Tevinhead <173904641+Tevinhead@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: snipeTR <31975916+snipeTR@users.noreply.github.com>
Co-authored-by: DWAA1660 <dwatnip123@gmail.com>
Co-authored-by: DWAA1660 <DWAA1660@users.noreply.github.com>
Co-authored-by: Justin John <34035011+justinjohn0306@users.noreply.github.com>
Co-authored-by: Justin John <babyjjbabyjj549@gmail.com>
Co-authored-by: cyfung1031 <44498510+cyfung1031@users.noreply.github.com>
Co-authored-by: 才倾梦华V <185041488+CQMHV@users.noreply.github.com>
Co-authored-by: Ikki Nishihara <b25d025@akita-pu.ac.jp>
Co-authored-by: gyurix <barathgy@gmail.com>
Co-authored-by: GyuriX <gyurix@users.noreply.github.com>
Co-authored-by: LloydNicholson <34131158+LloydNicholson@users.noreply.github.com>
Co-authored-by: lloyd <lloyd@stackworx.io>
Co-authored-by: CyberFox-81 <brittondcouchman@gmail.com>
Co-authored-by: Aviral Bansal <94526754+A-viral-dev@users.noreply.github.com>
Co-authored-by: Epsilon-LLC <64039063+EpsilonLLC@users.noreply.github.com>
Co-authored-by: Silver <hadichokr@icloud.com>
2026-10-10 17:01:08 -04:00
..

pdfcraft-automation

Agent control for PdfCraft: a headless tool table over the engine, and an opt-in MCP server.

  • Layer: L7 (architecture §13), but headless. It depends on engine, render and organize, never on a UI toolkit.
  • Status: over a hundred tools; pdfcraft-cli tools lists the current set with their schemas (the count isn't repeated here, so it can't go stale). They cover inspection, rendering, text, page edits and organizing, page labels, bookmarks, comments and their review, form filling and authoring (fields, properties, scripts, data exchange, detection), redaction and sanitizing, password protection and signatures, calibrated 2D measurements and CSV export, adding text, images, links and stamps, optimizing, OCR, printing, export, metadata, undo/redo, save, combine, extract and split. Comment tools take geometry in the same top-left-origin points as everything else, and comment_add can mark text by searching for it (find). The MCP server also serves the open documents as resources, page images included. The running app's UI is driven separately, through the ui.* control channel in pdfcraft-ui-egui.

API

let mut a = Automation::new().with_root("/work")?;      // optional: confine all paths
let doc = a.call("doc_open", &json!({ "path": "in.pdf" }))?;   // Vec<Content>
a.write_output("page1.png", &png)?;                      // save a result yourself, under the same root rules
tools() -> Vec<ToolDef>                                  // name, title, description, input_schema, read_only, destructive, command
mcp::McpServer::new(a).serve(stdin, stdout)?             // newline-delimited JSON-RPC 2.0

Content is Json(Value) or Png { data, width, height }. Errors are UnknownTool, InvalidArgs (the call didn't match the schema) or Failed (a readable message for the agent).

Conventions

  • Tool names are snake_case (page_rotate): MCP clients reject dots. ToolDef::command links a tool to the registry id it automates (page.rotate), and command_list reports the link.
  • Pages and positions are 1-based. Rectangles are PDF points with the origin at the top-left of the displayed page.
  • Unknown arguments are rejected, so typos fail loudly.
  • Tools that change a document return its summary (doc, pages, dirty, undo, redo, …).
  • comment_image_preview returns JSON and a PNG: layer: "background" (the default) gives the page without the selected image signature/initials; layer: "image" gives its embedded image with alpha. The metadata supplies its displayed rectangle, document rotation, image_rotation (how far clockwise the image is shown turned: 0 for one placed by PdfCraft, which keeps it upright as displayed) and annotation opacity. Each result can be saved with the CLI's --out. Cache the background and transform the image for live movement/resizing; comment_edit commits the final geometry in one undo step. Previewing never changes the document.
  • For proportional image signature resizing, use the embedded image layer's width/height ratio rather than a previously stretched annotation rectangle. The GUI preserves this ratio at corners with the opposite corner anchored; edge midpoint handles change only their own axis. comment_edit accepts the resulting rectangle.
  • doc_close refuses to drop unsaved changes unless discard_changes: true. doc_save writes atomically, incrementally in place, and in full for a new path.
  • With a root set, every read and write path must resolve inside it (symlinks and .. included). Relative paths resolve inside the root, and .. is resolved by name before the check. Every path outside the root gets the same refusal (<path> is outside the allowed directory <root>), whether or not it exists, so a confined agent can't probe the rest of the disk; on Windows another network share or device path (\\host\share, \\?\UNC\…, \\.\…) is refused without being contacted. "Not found" and other filesystem errors are reported only for paths inside the root. Tools that write several files into a folder name them after the document or input file, with separators, colons and control characters replaced by _, so a name can't lead them out of the folder. The root itself must be a folder.

Shared field properties

form_set_props accepts either field (one name) or fields (1–1000 unique names):

{"doc": 1, "fields": ["first_name", "last_name"], "required": true,
 "appearance": {"width": 3, "border": "#1473E6"}}

The batch is atomic and creates one undo step. A missing or locked field, invalid option, unknown argument or malformed appearance refuses the whole edit. Set locked: false to unlock and change fields together. Only supplied properties change: each field/widget keeps its other colours, fonts, values and unknown PDF dictionary keys. "none" removes a border or fill colour. Names, rectangles and rotations require the single-field form. Results keep field for single edits and return fields for batches. CLI and MCP use the same tool.

MCP server

Opt-in only. Nothing starts it automatically, and it opens no port. It runs while pdfcraft-cli mcp [--root DIR] runs, normally launched by an agent from its MCP configuration. It stops when stdin closes. The CLI's mcp Cargo feature (on by default) compiles it out entirely when disabled.

pdfcraft-cli mcp --compact (or McpServer::with_compact(true)) keeps tools/list short for agents with small context budgets. It lists only the core tools (mcp::COMPACT_CORE_TOOLS: open, info, save, close, render, text extract and find, combine, split, undo) plus the common command/inspection/preview tools and two meta tools: tool_search (optional query and category, the name prefix; or name for one tool's full input_schema) and tool_call (name and arguments, run through the same Automation::call). tools/call by name keeps working for every tool, and the server instructions mention the meta tools. Without the flag nothing changes.

Implemented: initialize (protocol 2025-06-18, 2025-03-26, 2024-11-05), ping, tools/list (with readOnlyHint/destructiveHint annotations), tools/call (JSON results also returned as structuredContent; images as image/png), and resources/list, resources/templates/list and resources/read. The resources expose the open documents read-only: pdfcraft://doc/{doc}/info (JSON), …/text, …/page/{page}/text and …/page/{page}/image{?dpi} (PNG, 1–600 dpi). Tool failures come back as isError: true results, so the agent can read and recover from them.

Adding a tool

  1. Add the engine capability first, with its tests (the tool is a thin adapter).
  2. Add a ToolDef in src/tools.rs: a precise description, the schema, ro()/destructive(), and cmd() if a registry command exists.
  3. Handle it in Automation::call, validating pages and positions with the existing helpers.
  4. Add an end-to-end test in tests/automation.rs. tool_table_is_well_formed checks names, schemas and command links.

Common command tools

Following FilmCraft's MCP conventions, command_list {doc?, filter?, enabled_only?} keeps its {commands: [...]} result and adds each mapped tool's parameter schema. command_run {id, params?} runs that command through its existing headless tool: include doc and other tool arguments in params. Commands without a headless mapping return a tool error. Unknown parameter keys are ignored with a warnings array; required values and paths still go through the original validation.

command_batch {steps: [{id, params?}], stop_on_error?} returns completed, failed, and per-step {ok, result?, error?} entries. The default is to stop at the first error; earlier edits remain. MCP marks a batch with any failed steps as isError, including through compact tool_call. doc_inspect {doc?} returns one document's info, or {documents: [...]} for all open documents. render_preview {doc?, page?, max_side?} returns a PNG (page defaults to 1, maximum side defaults to 1024, allowed 1–4096). Omit doc only when exactly one document is open.

All task tools remain available; compact mode includes the five common tools. Every MCP tool has all four annotation hints. Unknown top-level argument keys return JSON-RPC -32602 with the accepted keys; failures in the invoked tool remain isError results. An escaped tool panic is reported as an internal tool error, and the server continues serving.

Exports complete synchronously. progressToken and cancellation notifications are harmlessly ignored; no background-job or cancellation behavior is added.

Session resources pdfcraft://document and pdfcraft://commands, and the refusal of unknown argument keys, are described in docs/mcp.md.