Review follow-up: a red/blue regression for the centre-coverage pass. The
shared diagonal sits at x + y = 50.25 in device space (`cm` moves the mesh
down a quarter point), so pixel (20, 30) has its corner in the red triangle
and its centre in the blue one. It must stay red with either triangle
first; replacing `or_insert_with` with `insert` turns it blue when the red
triangle comes first, and the test fails.
Rebased onto main; the vendored hayro-interpret patch is renumbered 36
(41 before; 35 is taken by PR #2977).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gv2WwdgjmBz33GWkwgNn3Q
The mesh sampler took each pixel's colour at its corner, and only when the
corner was strictly inside a triangle. A corner on an edge two triangles
share belonged to neither, so a Type 4-7 shading drew a white seam along
every shared edge that ran through the sample grid, and dropped pixels
along the mesh's own sides (a 30 x 30 square cut along its diagonal:
812 of 900 pixels; MuPDF 900).
Such a pixel now goes to the triangle that covers its centre, edges
included (Triangle::covers), and still takes the colour at its corner, so
every pixel that was drawn keeps its colour. Vendored hayro-interpret
patch 41.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RUNQSPVXx3dDBabRZHAnjz
Review follow-up. A pattern paint keeps the fill or stroke opacity it was
set with (`scn`), so resetting the graphics state's fill alpha did not stop
a pattern-filled stencil mask from being drawn at the opacity squared:
`Pattern::without_opacity` now drops it for the mask's paint (patch 35).
The renderer also painted a pattern-filled stencil mask with the stroke
paint, so an uncoloured tiling pattern drew in its stroke colour (black by
default); it now uses the fill paint.
The stencil test covers a colour fill, an uncoloured tiling pattern and a
shading pattern at ca 1, 0.5 and 0.25; each of the three changes, removed,
makes it fail. Rebased onto main; the vendored hayro-interpret patch is now
number 35 (36 was taken on main).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RUNQSPVXx3dDBabRZHAnjz
hayro-interpret draws every image XObject inside a transparency group
that applies the fill opacity (ca). A stencil mask (/ImageMask true)
also takes the fill colour, with that same alpha, as its paint, so the
opacity was applied twice: at ca 0.5 a mask was drawn at 25 %, at
ca 0.25 at about 6 %. Rectangles and raster images were right.
Vendored hayro-interpret patch (36) resets the fill alpha to 1 once
the image's group is pushed, before the stencil's paint is taken; the
saved state is restored after the image. Type 3 glyph and pattern
stencils take their paint from the glyph or pattern and are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RUNQSPVXx3dDBabRZHAnjz
An optional content membership dictionary may carry a visibility
expression (/VE, PDF 1.6): [/And e...], [/Or e...] or [/Not e], each e an
optional content group or a nested expression. It takes precedence
over /OCGs and /P (ISO 32000-2 8.11.2.2). hayro-interpret ignored it, so
content under such a dictionary stayed visible when its layers were
hidden in the Layers panel.
Vendored hayro-interpret patch (40) evaluates the expression, nested
up to 32 deep. A malformed expression (unknown operator, /Not without
exactly one operand, /And or /Or without any, an operand that is
neither a group nor an expression, too deep) falls back to /OCGs and
/P as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RUNQSPVXx3dDBabRZHAnjz
Merges left four hayro-interpret rows; the current one lacked the
transparency-group, Type 3 rendering-mode and Lab white-point patches (each
numbered 25, 27 or 28 on a branch where another patch took that number).
They are now (41), (42) and (43) in the one row, with their tests.
Claude-Session: https://claude.ai/code/session_01Gv2WwdgjmBz33GWkwgNn3Q
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
`path_as_rect` decided a path was a rectangle from its element count and from
the four corners of its bounding box being covered. Its comment names the shape
it means to match -- one MoveTo, three LineTo, one ClosePath -- but `len() != 5`
also admits one MoveTo and four LineTo, a five-point open polyline, and
covering the corners says nothing about the points between them. Both call
sites then paint `device.draw_rect(bbox)`, so such a path was drawn as its own
bounding box.
Rendered through `pdfcraft-cli render`, an open zigzag whose ends reach the
corners of its bounding box
80 38 m 80 21.75 l 60.571 29.875 l 40 21.75 l 40 38 l S
lost its middle peak and gained a straight edge the file does not contain: no
dark pixel near the peak, and 36 along the bounding box's top. A six-point
zigzag and a real rectangle both drew correctly, which is why this shows on
some art and not the rest -- and `s` escapes it because its ClosePath makes six
elements.
A point that is not a corner now rejects the path. A rectangle still matches
whether written `m l l l` and closed or `m l l l l` back to its first corner,
since every one of its points is a corner. The same mistake also discarded a
clip of that shape, through `push_clip_path`, which this fixes too.
Found from craft-libs' ATTRIBUTION note that the shared reader centralizes the
patched Hayro trees from PdfCraft: the same flaw is in craft-libs, sent there
as storytold/craft-libs#175.
A BricsCAD plan page (3370 × 1524 pt, 251,000 single-segment strokes and 310
one-triangle type 4 shadings painted with `sh` at 50% alpha) took 2 s per
1024 px tile at every zoom, and 7.9 s for a fit-to-width render; on a busy
machine the 20 s watchdog gave the page up ("page 57 took longer than 20 s and
was skipped"). A 16 px tile of the same page took 65 ms, so neither parsing nor
the strokes were the cost: `sh` fills the whole canvas with the shading, and
`set_paint` sampled the mesh texture over the whole path, one hash lookup and
one texel per canvas pixel, 310 times per tile.
hayro-interpret patch (35): `EncodedShadingPattern::extent`, the device-space
pixels a mesh's triangles touched when it was encoded (`None` for shadings that
paint everywhere, and for a mesh with a `/Background`). hayro patch (19):
`set_paint` returns `PaintSetup::Nothing` for a mesh with no triangle on the
canvas (an empty extent, tested before it is intersected: `Rect::ZERO` sits at
the canvas origin), which the fill, stroke, rect and stencil callers skip, and
otherwise intersects the texture with the extent and pads instead of repeating it.
Measured with `page_timing` on the page: fit 7864 → 880 ms, a 1024 px tile
2030 → 85 ms at 100% and 2030 → 70 ms at 400% (the 16 px tile's 65 ms is the
page's re-interpretation, paid per tile), with byte-identical output hashes.
The three `mesh_extent_tests` pin what the texture must and must not cover.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
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
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
A simple font without a usable /ToUnicode reads its text from glyph
names, and a name missing from the glyph list dropped the glyph from the
extracted text while the page still drew. An 821-page book set in Type1C
fonts named C39, C40, ... extracted 4 KB of front matter and nothing else.
- glyph_name_to_text applies the Adobe Glyph List Specification's full
mapping: drop from the first '.', split on '_', map each component
through the list or as uniXXXX / uXXXX. f_i, f_l and hyphen.case now
read as fi, fl and '-'. Type 1, CFF, standard and embedded TrueType
simple fonts use it.
- A Type 1 / CFF / standard font name that only numbers the glyph (one
or two letters then digits, ZapfDingbats' aN excepted) reads as the
code's StandardEncoding character, so C174 is the fi ligature and C208
an em dash.
The book now extracts 1.9 MB (pdftotext: 1.85 MB), with fi and em dashes
where pdftotext writes Latin-1 mojibake.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gv2WwdgjmBz33GWkwgNn3Q
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.
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>
- 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>
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>
Seven tests failed at random on slow or busy machines, main included
(FreeBSD run 38834256439 at 5fdd820b). They now count the work or wait
for the state they check; no assertion is loosened and nothing retries.
- spot_colour_images_run_their_tint_transform_once_per_sample_value
counts tint-transform runs (exactly 256, not one per pixel) through
vendored hayro-interpret patch (34), color::tint_transform_runs, a
thread-local counter, instead of a 10 s deadline.
- shadings_with_large_sampled_functions_paint_quickly: no counter tells
the flat table from the hash map (both read every sample per build),
so the build each paint does is timed against reading its samples on
the same thread, fastest of 8 rounds each, and must stay under 10x
(flat table 1.1x, hash map 109x), instead of a 10 s deadline.
- page_contents_and_nested_forms_share_one_budget: the budget is
checked by its warnings; the deadline is only a hang guard, now 600 s.
- The pen-stroke tests hold finished renders back from the view
(DocView::hold_rasters, for tests) while they look at the settling
stroke, then wait by state for its raster.
- files_and_quit_from_the_operating_system waits for the tab before
deleting the file the open worker reads.
- image_pixels_follow_the_frame_while_a_corner_is_dragged waits for the
page raster on screen before each drag.
Product fix (#1858): a released drawing stayed painted only until any
raster of its page arrived. In full screen its edit is applied a frame
later, so a raster already under way could end the preview first. A
settling drawing is now marked committed when its AddAnnotation lands,
only rasters after that end its preview, and a failed edit drops it.
Patch 38 now also replaces the /ColorSpace of a shading that is
DeviceGray, DeviceRGB or DeviceCMYK with the default for it, whether the
shading is painted by sh or as a shading pattern. Poppler and MuPDF draw
these in the default too. The shading's cache key includes the default,
as the image's does.
The forms test also covers a form whose /ColorSpace has no default: it
draws in its page's, as in Poppler and MuPDF.
Notes and file attachments are saved with the NoRotate flag (/F bit 5),
but nothing honoured it: on a page with /Rotate the icon was fitted to
/Rect and turned with the page, and sat away from where it was placed.
The vendored renderer now honours NoRotate (hayro-interpret patch 37):
the appearance fitted to /Rect is turned against the page's rotation
about the rectangle's upper-left corner, which stays put (ISO 32000-2
12.5.3). Form fields and links stay in /Rect, where their click targets
come from. A counterrotated appearance, as stamps use, would be turned
twice by viewers that honour the flag.
Annotation::shown_rect gives the box the note is drawn in. The UI uses
it for hit-testing, outlines, moving, the hover pop-up, the Comments
panel flash and the edit composer; a resize maps it back to /Rect.
comment_add anchors an icon's displayed top-left at `at`, and
comment_list, doc_info and comment_edit use the drawn box.
This also covers the sticky-note part of #1869.
hayro-interpret cached the colour space built from an /ICCBased profile
stream under the hash of the stream's dictionary bytes alone. Two
different profiles with the same /N and /Length (any two uncompressed
RGB profiles of one size) shared whichever was read first, so colours in
the second space were drawn through the first profile for the rest of
the document. Found while fixing #1201.
The space is now keyed by what its profile is read from (IccProfileKey):
the stream object a space names, looked up without reading the stream,
and on a miss the old dictionary key plus the stream's own bytes
(decrypted, not decoded), so byte-for-byte copies in other objects still
share one parsed profile and a lookup never decodes. A stream written
directly in the colour space array reports the object holding the array,
as every other stream written there does; it is recognised by its
dictionary lying within the array's bytes and keyed by content only.
InterpreterCache::icc_profile_reads counts the profiles read, so the
tests can check that the cache still hits. Tests: two synthetic RGB
profiles of one length (linear sRGB, and red and blue swapped) as
objects, a reused object, a copy and direct streams on one page.
A resource dictionary's DefaultGray, DefaultRGB and DefaultCMYK colour
spaces replace the device space content selects (ISO 32000-2 §8.6.5.6).
hayro ignored them: g/G, rg/RG and k/K set the device space directly,
cs/CS resolved the device names before looking at the resources, and
images took their /ColorSpace as written. Pages that calibrate device
colour this way, common in PDF/A, came out in the wrong colours.
The interpreter now looks the defaults up once per content window
(DefaultColorSpaces), in the resources and those they inherit, so a form
XObject without its own draws in its page's. They replace DeviceGray,
DeviceRGB and DeviceCMYK selected by the colour operators, by cs/CS (by
name or through a /ColorSpace resource that is a device space), and as
the /ColorSpace of image XObjects and inline images. A default that is a
Lab, Indexed, Pattern or device space, has another number of components
or is not a colour space is ignored, and device colour is drawn as
before. Device spaces nested in other spaces, shading dictionaries,
masks and transparency group spaces are unchanged. The decoded-image
cache keys a decode by the default it was drawn in, since one image can
be drawn with and without one.
Part of #1201.
* fix(render): sampled functions index a flat sample array (vendored hayro-interpret patch 31)
A Type 0 sampled function kept its samples in a HashMap keyed by each point of
its /Size grid (two SmallVecs per key, SipHash), and the renderer builds a
shading's functions again every time it is painted, so every paint hashed one
insert per sample. A 2,097,152-sample ramp painted 30 times took 38 s; the
pdf.js corpus's bug1721218 (InDesign) spent most of a 1.5 s page on it.
The samples now stay in stream order in one Vec, and a grid point is found by
strides (first input fastest). The interpolation keeps the same recursion and
arithmetic, so results are bit for bit the same, and a point the stream ended
before still gives no value. The repeated-paint page draws in well under a
second. The maximum sample value no longer computes 2^32 in a u32 (a panic with
overflow checks on at 32 bits per sample).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MoEx9Nj8KC4WKDcf5YaokQ
* fix(render): one decode of an image serves every size it is drawn at (vendored hayro-interpret patch 32)
The decoded-image cache (patch 24) keyed each decode by the size the image is
drawn at, but only the JPX filter decodes a reduced resolution for a size:
every other filter decodes the whole image. So a JPEG or Flate scan missed the
cache at every zoom step, was decoded again, and was kept once per zoom level.
The key now carries the target size only when JPX decodes the image, its
/SMask or its /Mask (read from /Filter or /F, erring towards keeping the
target). `InterpreterCache::decoded_image_decodes` counts the cache's decodes;
the new test draws a Flate image at five sizes and sees one decode, and a
copy whose soft mask is JPX-encoded still decodes per size. Both assertions
fail with the old keying.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MoEx9Nj8KC4WKDcf5YaokQ
* fix(render): patch meshes copy their vertices' colours less (vendored hayro-interpret patch 33)
A Coons or tensor-product patch is drawn as 722 triangles over a 20 x 20 grid
of points, and every vertex's colours (a SmallVec) were produced or copied up
to five times: each grid cell interpolated its four corners itself (1,444
interpolations a patch for 400 points), cloned all six vertices of its two
triangles, and the mesh sampler cloned each vertex again to move the triangle
into device space. Profiling page 8 of the annual report in #1584 put a
quarter of its time in those clones.
Each grid point's colours are now interpolated once, a cell's second triangle
takes the vertices its first copied, and the sampler moves a patch's freshly
made triangles instead of copying them (`TriangleSampler::sample_owned`; a
free-form triangle mesh, which the shading keeps, is still copied). The same
inputs give the same values: rendering the first three pages of every pdf.js
corpus file is byte for byte the same. Release build, Linux: report page 8
773 -> 696 ms, personwithdog.pdf 149 -> 129 ms, issue13520.pdf 104 -> 87 ms.
The new test pins the colours near each corner, at the centre and between them
for a Coons and a tensor-product patch (taken before the change); swapping two
vertices' grid points makes it fail.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MoEx9Nj8KC4WKDcf5YaokQ
* docs: roadmap entries for faster sampled functions, image decodes and patch meshes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MoEx9Nj8KC4WKDcf5YaokQ
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
#2305 fixed WinAnsi 0xAD in `crates/fonts`, and closed with two loose ends it named: the
user-visible extraction path goes through hayro, not that table, so it still reported the
invisible character; and the same Annex D.2 note makes 0xA0 the `space` glyph. This finishes
both, and makes the two tables in the product agree about the same byte.
WinAnsiEncoding gives codes 0xA0 and 0xAD the glyph names `space` and `hyphen` -- the names it
also gives 0x20 and 0x2D -- and MacRomanEncoding gives 0xCA `space`. The tables called them
`nbspace` and `sfthyphen`, whose Adobe Glyph List values U+00A0 and U+00AD are invisible
formatting characters, so extraction hid ink the reader can see: "well-known" drawn with the
0xAD byte matched no search and no redaction pattern, and a non-breaking space did not match the
space a reader typed.
Evidence. Two sources transcribe the table the same way (an Annex D implementation's docs, and
the WinAnsi.enc encoding vector R ships, which gives `/space` at octal 240 and `/hyphen` at
255), and a black-box check settles the rest: on a synthetic fixture pdftotext reads the
base-encoded codes as `a-b a b`, and keeps U+00AD for a code an explicit `/Differences` names
`sfthyphen`. We now do both. Over the 989-file pdf.js corpus the soft hyphens we extract fall
from 117 to 13 and the non-breaking spaces from 30 to 26; everything left comes from `/ToUnicode`
maps, which say what they mean and are not touched.
Scope. The base table only, in the text direction only.
- A `/Differences` entry naming `sfthyphen` or `nbspace` is a deliberate statement about one
code and keeps its glyph list value, as pdftotext does.
- Glyph lookup is untouched, so a font program's own `sfthyphen` program is still found first.
In hayro that is why the alias sits in the Unicode path rather than in the generated encoding
table: `code_to_ps_name` is shared with lookup.
- Writing is unaffected: `encode` takes the lowest code showing a character, so ' ' and '-'
still write 0x20 and 0x2D, and a caller holding a literal U+00A0 still writes 0xA0.
Tests. `fonts` pins both tables and the whole decode/encode path including the `/Differences`
case; `render` pins the extraction a reader actually gets, over WinAnsi, MacRoman and
`/Differences`. Each was watched failing with each cell and each match arm reverted in turn, so
no assertion is passing by accident. fonts 70, render 296+4, redact 49, a11y 43 and edit 156 are
green, with fmt, clippy, `xtask layers` (37 crates) and `xtask assets` clean.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* M3: release Ctrl zoom for pixel-only Wayland wheel events
Use the reported scroll source to avoid unterminated finger gestures for wheel, wheel-tilt and continuous input. Exercise the production policy through egui, including modifier release and preserved finger momentum.
Assisted-by: OpenAI Codex
* M3: keep Cargo lockfile dependency ordering canonical
Record the dependency ordering produced by the completed workspace and wasm gates.
Assisted-by: OpenAI Codex
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(render): keep annotations readable under reading colours (#1548)
Reading colours remapped annotation paints as line art: the yellow
highlight fill is luminance-light, so it was mapped toward the dark
background, and Multiply then darkened the remapped-white words under
it into invisibility.
Two changes in hayro:
- The /Annots loop in interpret_page is bracketed by a new
begin_annotations/end_annotations Device signal (patch 26), and the
renderer's reading-colours mapping leaves annotation paints alone.
Reading colours replace document colours, never the colours a
comment was given — the Acrobat model.
- The page background is painted inside the page transparency group
(patch 27), as the initial backdrop, so a Multiply highlight blends
with the background too. With the background outside the group the
band kept its own flat colour and the words vanished into it. White
pages are unaffected: multiplying with white is the identity.
Regression test reading_colors_keep_highlight_annotations_readable
covers both annotation paths (/AP and the no-/AP /QuadPoints
fallback) under WHITE_ON_BLACK; parity entry a11y.replace-page-colors
updated with the behaviour and evidence.
Co-Authored-By: GLM <noreply@z.ai>
* contributors: credit @edurbs as Eduardo Soares
The contributor's own entry, chosen by them (real and display name).
Co-Authored-By: GLM <noreply@z.ai>
* #1557: record the vendored hayro patches in vendor/README.md
Main already uses hayro-interpret patches (26)-(28), so the annotation hooks become
hayro-interpret (29) and the renderer's parts hayro (17) and (18).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: GLM <noreply@z.ai>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
On macOS the tab strip is the title bar: a transparent title bar over a
full-size content view. AppKit asks the view under the pointer whether a
press there can move the window (`mouseDownCanMoveWindow`), and NSView
answers YES for a view that isn't opaque. winit's view didn't override it,
so AppKit moved the whole window for any drag that began in the 28 pt band
before egui saw it, and the tab reordering from #1551 never got the drag.
vendor/winit now answers with the window's `movable_by_window_background`
attribute (NO unless asked). The empty strip still moves the window through
`ViewportCommand::StartDrag`, and double-clicking it still maximizes. Patch
(12) in vendor/README.md, kept in the build by
`tests/winit_macos_title_bar_drag.rs`.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Highlight appearances written by other editors are a transparency group whose
content is filled with `/BM /Multiply`. Every `/Group` form got a layer of its
own, so the Multiply blended with that layer's transparent backdrop and the
layer then covered the page as an opaque block.
Vendored hayro-interpret patch (28): a group that is neither isolated nor
knockout composites its elements with the backdrop (ISO 32000-2 §11.4.7), so
when it is entered at full opacity with no soft mask and Normal blending there
is nothing for a layer to add, and its content is drawn straight onto the page.
`/I` and `/K` are read from the group dictionary. An isolated group keeps its
blank backdrop; a knockout group or one entered at partial opacity, under a
soft mask or with a blend mode keeps its layer.
Test: `raster::tests::a_multiply_highlight_inside_a_transparency_group_keeps_the_text`
(the appearance drawn as an annotation and from page content, with `/I false
/K false` spelled out and with no group at all; an isolated group as a control;
a group entered at half opacity still composited as one). It fails before the
fix with the text pixel yellow instead of black. parity: view.render-transparency
cites it and says what is still approximated.
Claude-Session: https://claude.ai/code/session_01RUNQSPVXx3dDBabRZHAnjz
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* M2: clip intersection starts its row walk at the first shared row
clip::intersect in vello_common, the step every clipped path passes
through, began its walk over strip rows at the earlier of the two paths'
first rows. Each path therefore visited every row above it, although rows
where one path has no strips produce no output. Every page is clipped (the
crop box is a clip), so this ran for every stroke of every render: on the
A1 engineering sample it was the hottest function in a full-page render.
Start at the later first row, as vello_common 0.1 does. The strips and the
pixels are unchanged; a test counts the rows the walk visits and fails
without the change.
vello_common 0.0.8 is vendored with that one-line PdfCraft patch (marked
in the source), its unit test, the vendor/README.md row and an
ATTRIBUTION.toml entry for its reference image. Cargo.lock records it as a
path crate.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
* Refresh catalog hashes after merging main
---------
Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Apply the minimum after page-unit scaling without inflating small raw crops. Retain the existing unit policy and nearly-zero fallback, with regression coverage for origins, rotations and reciprocal rounding.
Co-authored-by: sanyamsingh001 <314051937+sanyamsingh001@users.noreply.github.com>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
The vendored hayro-syntax passed each filter an output limit, but
ASCII85 ignored it: every `z` expands one byte to four, so a Flate
stage at its limit could be grown another 4x past every budget.
ASCII85 now writes only what fits under the limit, stops with a
warning once it is reached (like RunLength, LZW and Flate), and caps
its starting capacity at the limit.
Test: decompression_bombs_stop_at_the_stream_limit gets a
[/FlateDecode /ASCII85Decode] bomb; it decoded 301,989,892 bytes
before and stops at MAX_DECODED_STREAM now.
Co-authored-by: skylar taylor-barrick <hi@skylar.cc>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
ISO 32000-2 Table 11 defines /Rows 0 (the default) as "height not
predetermined", but the vendored hayro-syntax decoder took it as zero
rows, so scanned pages written that way rendered blank. Treat 0 like an
absent /Rows and fall back to the image height.
Adds a synthetic regression test: an 8x4 all-black Group 4 stream,
hand-encoded from ITU-T T.6, rendered with /Rows absent, 4 and 0. It
fails without the patch.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(render): convert spot-colour images once per pixel value (vendored hayro-interpret patch 28)
An image whose pixels hold at most 16 bits of samples (an 8-bit Separation or
DeviceN image, a 1-, 2- or 4-bit gray or indexed one) is now converted to RGB
through a table of every possible pixel value instead of pixel by pixel.
A Separation or DeviceN tint transform is a PDF function, often a PostScript
calculator program, and the image decoder ran it once per pixel. Fillable
character sheets draw their background art as an 8-bit DeviceN image: one with
a 3304 x 4105 background took 3.8 s and 430 MB to draw each page (Poppler: 0.27 s),
when opening it and again after every field was filled. It now draws in 0.11 s
and 100 MB, pixel for pixel the same.
The table decodes each sample through the same `decode_sample` as the per-pixel
path, keeps `fix_image_length`'s rules for short and overlong data (the pixels
past the data are converted as before), and is used only when the image has
more pixels than the table has entries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KnNgjPEDfKJM1wj17Q8Za8
* fix(render): place widgets without /P in one pass over the pages
Opening a document, the inspector finds the page of a form field whose widget
has no /P by searching every page's /Annots for it, once per field: O(fields x
annotations). A form of 8,000 such fields took 1.17 s to open instead of
0.27 s. The pages' /Annots are now read once, on the first such field, into a
map from annotation to page.
A widget that several pages list is now placed on the first of them, as
pdfcraft-forms places it; the search took whichever page a HashMap visited
first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KnNgjPEDfKJM1wj17Q8Za8
* fix(ui): the Fields panel keeps up with large forms
Every frame the Fields panel searched the form for each row by name, once to
sort the rows into tab order and again to find what a click selects: O(fields²)
a frame. It also laid out, drew and built the tooltip of every row, scrolled
into view or not. With 8,000 fields the app used 13.4 s of CPU for 20 pointer
moves with the panel open (0.2 s with it closed).
The rows are now worked out by `field_rows`, which pairs each row with the
first form field of its name through one map a frame, as the engine pairs
them, and a row out of view is allocated and keeps its clicks, keyboard focus
and accessibility label but isn't drawn and builds no tooltip. The same 20
moves now take 0.86 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KnNgjPEDfKJM1wj17Q8Za8
* docs: roadmap entries for faster form-heavy documents
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KnNgjPEDfKJM1wj17Q8Za8
* fix(render): convert each lookup-table entry on its own (vendored hayro-interpret patch 28)
Patch 28 converted the whole table of possible pixel values in one batch.
Separation and DeviceN `convert_f32` concatenate each tint result, and a Type 4
program that returns the wrong number of components is only warned about, so a
malformed result for a value the image never uses shifted every later entry. A
32 x 32 image whose only sample (255) is red drew black because tint 0 returned
one component (PR #1691 review).
`convert_each` now converts the table one entry at a time, so a malformed entry
changes only itself. The largest table (65,536 entries, a two-ink DeviceN image
over DeviceCMYK) takes about 2 ms longer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* fix(text): read a TrueType font without an encoding as WinAnsi (fixes#1468)
* fix(text): describe the WinAnsi reading by pdftotext's output
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
