fix(render): an RGB colour in an isolated DeviceCMYK group blends in CMYK, as its ink (#7031)

* fix(render): an RGB colour in an isolated DeviceCMYK group blends in CMYK, as its ink

Part of #4887. An isolated DeviceCMYK group blended in CMYK only when every
colour in it had process tints (DeviceCMYK, DeviceGray); one RGB fill sent
the whole group back to RGB blending.

Vendored hayro-interpret patch (104): Color::ink_tints gives DeviceRGB,
CalRGB, Lab and three-component ICCBased colours the ink that draws back to
them through the bundled CMYK profile (new cmyk_inverse module: black
0.4*k0^2 + 0.6*k0^8 with k0 = 1 - max(r, g, b), matched to Acrobat on
synthetic pages; then cyan, magenta and yellow by at most 12 Newton steps
against the profile; one-entry per-thread cache). ICCBased CMYK still has no
tints.

Test an_rgb_colour_in_an_isolated_cmyk_group_blends_as_its_ink: 50% grey
under Difference with black within 10 levels of Acrobat's 137 112 106;
in-gamut colours draw as themselves; red within 8 of Acrobat's 238 32 37.
With the additive arms reverted the grey draws 93 97 96 and it fails.
#5228's control now uses an RGB image, which still has no tints.

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

* fix(render): an RGB colour with a non-finite component has no CMYK ink

`to_rgba` turns a non-finite component into some finite colour, so the
inverse could not see it and gave that colour's ink. `ink_from_srgb` now
declines such a colour and the group draws it as before.

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

---------

Co-authored-by: KenM76 <KenM76>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

crates/render/src/raster/tests/isolated_cmyk_groups.rs

@@ -88,11 +88,12 @@ endstream endobj
         assert!(distance(&group, &square) <= 3, "{mode}: group {group:?}, square {square:?}");
     }
 
-    // A group whose contents have no ink tints (here the second element is RGB) is
-    // composited in RGB exactly as before: the same as a group blending in DeviceRGB.
-    let rgb = |_: [f32; 4]| "1 0 0 rg".to_string();
-    let cmyk_group = render("/CS /DeviceCMYK /I true", &rgb);
-    let rgb_group = render("/CS /DeviceRGB /I true", &rgb);
+    // A group whose contents have no ink tints (here the second element draws an RGB image
+    // over the page) is composited in RGB exactly as before: the same as a group blending
+    // in DeviceRGB.
+    let image = |_: [f32; 4]| "q 370 0 0 60 0 0 cm BI /W 1 /H 1 /CS /DeviceRGB /BPC 8 /F /AHx ID ff0000> EI Q".to_string();
+    let cmyk_group = render("/CS /DeviceCMYK /I true", &image);
+    let rgb_group = render("/CS /DeviceRGB /I true", &image);
     assert_eq!(cmyk_group.rgba, rgb_group.rgba, "contents without tints are drawn as before");
 }
 
@@ -247,6 +248,32 @@ trailer << /Root 1 0 R >>
     p.rgba[((y * p.width + x) * 4) as usize..][..3].to_vec()
 }
 
+/// An RGB colour painted in an isolated `DeviceCMYK` group is blended as the process ink that
+/// draws back to it through the CMYK profile (hayro-interpret patch (104)), instead of
+/// sending the whole group back to RGB blending.
+///
+/// Under `/BM /Difference` with black ink, a colour's cyan, magenta and yellow are
+/// complemented and its black kept (ISO 32000-2 §11.3.4), so 50% grey, mostly cyan, magenta
+/// and yellow ink, comes out a warm grey: neither the neutral grey of RGB blending nor the
+/// near-black of a grey made from black ink alone. Acrobat draws the same blend on a CMYK page
+/// as (137, 112, 106). Painted without a blend, a colour inside the profile's gamut draws as
+/// itself, and pure red is clipped to the profile's red, as in Acrobat.
+#[test]
+fn an_rgb_colour_in_an_isolated_cmyk_group_blends_as_its_ink() {
+    let resources = "/ExtGState << /D << /BM /Difference >> >>";
+    let pixel = |group: &str| cmyk_group_pixel([100, 100], "/G Do", [20, 20, 80, 80], group, resources, (50, 50));
+    let near = |a: &[u8], b: [u8; 3], within: u8| a.iter().zip(b).all(|(a, b)| a.abs_diff(b) <= within);
+
+    let blended = pixel("0.5 0.5 0.5 rg 20 20 60 60 re f /D gs 0 0 0 1 k 20 20 60 60 re f");
+    assert!(near(&blended, [137, 112, 106], 10), "grey under Difference with black: {blended:?}");
+
+    // (colour, expected pixel, how far from it)
+    for (rgb, expected, within) in [("0.8 0.4 0.2", [204, 102, 51], 3), ("0.3 0.7 0.4", [77, 179, 102], 3), ("1 0 0", [238, 32, 37], 8)] {
+        let drawn = pixel(&format!("{rgb} rg 20 20 60 60 re f"));
+        assert!(near(&drawn, expected, within), "{rgb} drew as {drawn:?}, expected {expected:?}");
+    }
+}
+
 /// From the review of #5228: a shading pattern painted before the group left its pattern
 /// transform on the context, and the finished group was drawn through it, so the cyan half
 /// of the group showed magenta. The group must look the same whether or not a pattern was
@@ -325,3 +352,15 @@ trailer << /Root 1 0 R >>
         assert!(rgb[1] > 120 && rgb[0] < 60 && rgb[2] < 120, "blended in CMYK, green: {rgb:?}");
     }
 }
+
+/// An RGB colour with a non-finite component has no tints: converting it to RGB first would
+/// give some finite colour that is not its ink.
+#[test]
+fn an_rgb_colour_with_a_non_finite_component_has_no_tints() {
+    use hayro::hayro_interpret::color::AlphaColor;
+    use hayro::hayro_interpret::color::Color;
+    let rgb = |c: [f32; 3]| Color::from_rgba(AlphaColor::new([c[0], c[1], c[2], 1.0])).ink_tints();
+    assert!(rgb([0.5, 0.5, 0.5]).is_some());
+    assert_eq!(rgb([f32::NAN, 0.5, 0.5]), None);
+    assert_eq!(rgb([0.5, f32::INFINITY, 0.5]), None);
+}

vendor/README.md

@@ -55,7 +55,7 @@ Other resource categories retain their existing parent-chain lookup.
 
 | Crate | Version | Licence | Patches | Test |
 |---|---|---|---|---|
-| hayro-interpret | 0.7.0 | Apache-2.0 OR MIT | (1) select `/AP /N` appearance states via `/AS` (checkboxes, radios, stateful icons) and honour the NoView flag; (2) Type3 Unicode fallback to `/Encoding` glyph names, then printable ASCII; (3) `Type3Glyph::advance_width`; (4) `InterpreterSettings::ocg_overrides` for viewer layer toggles; (5) soft masks whose group lacks `/CS` were dropped (Chrome gradient text); (6) paint nesting cap `MAX_PAINT_NESTING` for tiling patterns and Type 3 glyphs (self-referencing content overflowed the stack); (7) CID width ranges in `/W` and `/W2` clamped to `MAX_CID` (a `0 4294967295 w` range hung); (8) `InterpreterSettings::hide_comments` skips markup annotations but keeps widgets and links (View ▸ Hide all comments); (9) `image_size_ok`: images and masks over 2^28 pixels are not decoded (a fuzzed `/W 4294967295` stencil mask allocated 4 GiB); (10) `MAX_NESTED_PAINTS`: nested paints (forms, Type 3 glyphs and tiling patterns inside one another) share a per-page budget of 50,000 (a glyph showing eight glyphs of its own font, sixteen deep, hung); (11) depends on skrifa 0.47 instead of 0.42 (upstream read-fonts fixed a Type 1 tokenizer that recursed through runs of numbers: a fuzzed font hung or overflowed the stack), with the one API change (`Charset::glyph_id` returns `Option`); (12) annotation appearances whose `/BBox` or `/Matrix` collapses to a line or a point are skipped (mapping them onto `/Rect` divided by zero and handed the device infinite or NaN matrices); (13) a Highlight annotation without a usable normal appearance is drawn from its `/QuadPoints`, `/C` and `/CA` with Multiply blending, as pdfcraft-annot draws highlights (it was not drawn at all); (14) simple-font `/Widths` (`read_widths`): entries only for codes 0–255 (one entry per code below `/FirstChar` was allocated: `4294967295` asked for 34 GB), an entry that isn't a number leaves its code missing instead of ending the array, and a `/LastChar` below `/FirstChar` gives no widths instead of discarding the font for a standard one; (15) the appearance box is mapped onto `/Rect` corner to corner: the translation now accounts for the scale (an appearance whose box didn't start at the origin and differed in size from `/Rect` was drawn shifted); (16) a clip set by `W`/`W*` takes effect after whichever path-painting operator ends the path (`S`, `f`, `B`, … as well as `n`; ISO 32000-2 §8.5.4): it was applied only on `n`, so `re W* S` clipped nothing and the clip stayed pending for a later, unrelated `n`; (17) `InterpreterSettings::cancelled`: once set, interpretation stops before the next content operator (a render pool sets it for every worker when dropped, after an edit or on closing, and for one worker when the watchdog gives up on its render, so renders whose answers nobody can receive stop instead of running to the end); (18) `MAX_PAGE_CONTENT`: the content a page holds at once (its contents, and the forms, tiling patterns and Type 3 glyph procedures being painted) stays within 512 MiB (`decode_content`; once it runs out, `InterpreterWarning::ContentTruncated` is reported (once per page by pdfcraft-render) so a truncated page is not a clean render; each level gives its bytes back when it finishes): forms nest fifty deep and each level kept its content while the next ran, so a form painting itself around a stream inflating to 256 MiB would have held over 12 GB; and Type 0 sampled functions take at most `MAX_SAMPLE_ENTRIES` (2^21) table points and `MAX_SAMPLE_VALUES` (2^23) samples, read no further than `/Size` asks (256 MiB of 1-bit samples asked for over 8 GB); (19) mesh shadings (types 4–7) are sampled only inside the area being drawn (`ShadingPattern::encode_within`, which the vendored hayro renderer calls with that area plus a pixel), within 8 visits per pixel of it plus 4 per triangle (`TriangleSampler`), and patches are triangulated one at a time: points far off the page were sampled over their whole bounding box, up to 65536 × 65536 entries (a fuzzed 147-byte patch stream; a synthetic page passed 6 GB in six seconds), ten thousand overlapping triangles ran for minutes, and every patch became 722 triangles before any limit; meshes hold at most `MAX_MESH_PATCHES` (2^16) patches and `MAX_MESH_TRIANGLES` (2^20) triangles, a lattice needs two to 2^19 vertices per row (`/VerticesPerRow 0` pushed empty rows for ever; one huge row read the whole stream), and a page samples at most `MAX_PAGE_MESH_TRIANGLES` (2^25) mesh triangles in all (a mesh painted a thousand times was triangulated a thousand times); each cut logs a warning; (20) PDF function construction is limited to 64 levels and 10,000 total nodes with cycle detection; calculator parsing and evaluation share procedure depth/work budgets, integer division/bitshift/rotation/index are panic-free, and fixed-size stack errors propagate; (21) a Type 0 font drawn with a substitute that isn't CID-keyed (it isn't embedded, or its program doesn't load) draws `.nLine truncated
+| hayro-interpret | 0.7.0 | Apache-2.0 OR MIT | (1) select `/AP /N` appearance states via `/AS` (checkboxes, radios, stateful icons) and honour the NoView flag; (2) Type3 Unicode fallback to `/Encoding` glyph names, then printable ASCII; (3) `Type3Glyph::advance_width`; (4) `InterpreterSettings::ocg_overrides` for viewer layer toggles; (5) soft masks whose group lacks `/CS` were dropped (Chrome gradient text); (6) paint nesting cap `MAX_PAINT_NESTING` for tiling patterns and Type 3 glyphs (self-referencing content overflowed the stack); (7) CID width ranges in `/W` and `/W2` clamped to `MAX_CID` (a `0 4294967295 w` range hung); (8) `InterpreterSettings::hide_comments` skips markup annotations but keeps widgets and links (View ▸ Hide all comments); (9) `image_size_ok`: images and masks over 2^28 pixels are not decoded (a fuzzed `/W 4294967295` stencil mask allocated 4 GiB); (10) `MAX_NESTED_PAINTS`: nested paints (forms, Type 3 glyphs and tiling patterns inside one another) share a per-page budget of 50,000 (a glyph showing eight glyphs of its own font, sixteen deep, hung); (11) depends on skrifa 0.47 instead of 0.42 (upstream read-fonts fixed a Type 1 tokenizer that recursed through runs of numbers: a fuzzed font hung or overflowed the stack), with the one API change (`Charset::glyph_id` returns `Option`); (12) annotation appearances whose `/BBox` or `/Matrix` collapses to a line or a point are skipped (mapping them onto `/Rect` divided by zero and handed the device infinite or NaN matrices); (13) a Highlight annotation without a usable normal appearance is drawn from its `/QuadPoints`, `/C` and `/CA` with Multiply blending, as pdfcraft-annot draws highlights (it was not drawn at all); (14) simple-font `/Widths` (`read_widths`): entries only for codes 0–255 (one entry per code below `/FirstChar` was allocated: `4294967295` asked for 34 GB), an entry that isn't a number leaves its code missing instead of ending the array, and a `/LastChar` below `/FirstChar` gives no widths instead of discarding the font for a standard one; (15) the appearance box is mapped onto `/Rect` corner to corner: the translation now accounts for the scale (an appearance whose box didn't start at the origin and differed in size from `/Rect` was drawn shifted); (16) a clip set by `W`/`W*` takes effect after whichever path-painting operator ends the path (`S`, `f`, `B`, … as well as `n`; ISO 32000-2 §8.5.4): it was applied only on `n`, so `re W* S` clipped nothing and the clip stayed pending for a later, unrelated `n`; (17) `InterpreterSettings::cancelled`: once set, interpretation stops before the next content operator (a render pool sets it for every worker when dropped, after an edit or on closing, and for one worker when the watchdog gives up on its render, so renders whose answers nobody can receive stop instead of running to the end); (18) `MAX_PAGE_CONTENT`: the content a page holds at once (its contents, and the forms, tiling patterns and Type 3 glyph procedures being painted) stays within 512 MiB (`decode_content`; once it runs out, `InterpreterWarning::ContentTruncated` is reported (once per page by pdfcraft-render) so a truncated page is not a clean render; each level gives its bytes back when it finishes): forms nest fifty deep and each level kept its content while the next ran, so a form painting itself around a stream inflating to 256 MiB would have held over 12 GB; and Type 0 sampled functions take at most `MAX_SAMPLE_ENTRIES` (2^21) table points and `MAX_SAMPLE_VALUES` (2^23) samples, read no further than `/Size` asks (256 MiB of 1-bit samples asked for over 8 GB); (19) mesh shadings (types 4–7) are sampled only inside the area being drawn (`ShadingPattern::encode_within`, which the vendored hayro renderer calls with that area plus a pixel), within 8 visits per pixel of it plus 4 per triangle (`TriangleSampler`), and patches are triangulated one at a time: points far off the page were sampled over their whole bounding box, up to 65536 × 65536 entries (a fuzzed 147-byte patch stream; a synthetic page passed 6 GB in six seconds), ten thousand overlapping triangles ran for minutes, and every patch became 722 triangles before any limit; meshes hold at most `MAX_MESH_PATCHES` (2^16) patches and `MAX_MESH_TRIANGLES` (2^20) triangles, a lattice needs two to 2^19 vertices per row (`/VerticesPerRow 0` pushed empty rows for ever; one huge row read the whole stream), and a page samples at most `MAX_PAGE_MESH_TRIANGLES` (2^25) mesh triangles in all (a mesh painted a thousand times was triangulated a thousand times); each cut logs a warning; (20) PDF function construction is limited to 64 levels and 10,000 total nodes with cycle detection; calculator parsing and evaluation share procedure depth/work budgets, integer division/bitshift/rotation/index are panic-free, and fixed-size stack errors propagate; (21) a Type 0 font drawn with a substitute that isn't CID-keyed (it isn't embedded, or its program doesn't load) draws `.nLine truncated
 | hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour (since (21), only inside Type 3 glyphs), planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option<ColorReplacement>` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk image that is interpolated (or, since (21), area-averaged) and whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle (halves round up, so a tile snaps an edge where the whole page does) and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroLine truncated
 | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap and page bitmap, and 2^25 cells per halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one, and a 4-byte value per cell); (2) at most as many symbol pixels per symbol dictionary as the largest page bitmap holds, 2^28 (a fuzzed dictionary decoded for minutes; each symbol now pays for its pixels out of the stream's decoding work and is held in the retained bitmaps, and at the earlier 2^26 the soft mask of a scanned drawing, one dictionary of 7,559 symbols, was refused: #6722), and a refusal by one of the decoder's limits is told apart from damaged data (`DecodeError::is_limit`, `OverflowError::BitmapLimit`); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer); (4) a segment header's long-form referred-to segment count is capped at `u16::MAX` and a larger count is refused before its retention or referred-to bytes are read: the 29-bit count went straight to `Vec::with_capacity`, a reservation of about 2 GiB whose failure aborts the process (from storytold/craft-libs#176); (5) what a page's segments hold for later ones (intermediate regions, pattern and symbol dictionaries with their headers and retained contexts, and Huffman tables) is at most 64 MiB, two square bitmaps of 16,384 a side (`MAX_RETAINED_WORDS`), and a dictionary's new and exported symbols are counted against it as they are made: each was bounded on its own, but a few dozen bytes an intermediate region held 33 MB each, and 8 MiB symbols exported again and again killed the process, so a few kilobytes could exhaust memory; (6) a segment may refer to at most 2^20 symbols, checked before any are collected: references may repeat, and 65,535 of them to one dictionary, in a 78 KB stream, asked for 17 GB; (7) a pattern dictionary keeps at most 16 MiB of shifted copies, which only speed up drawing, and none that wouldn't fit what the page has left (`MAX_SHIFTED_WORDS`); (8) a custom Huffman table has at most 2^16 lines, and a stream with its globals at most 2^20 segments and 2^22 references, checked as they are read: both grew before anything counted them, and PDFs of a kilobyte or so reached 5 GB and more behind nested Flate; (9) a page past the bounds of (1) is refused when the stream is opened: a caller sizes its output from the header before the page bitmap is made, and an Indexed image in a 743-byte PDF asked for 137 GB, which aborts; (10) a stream may ask for at most 2^30 units of decoding work, four of the largest pages, about 6 s to 17 s (`MAX_WORK`): decoding a bitmap costs a pixel each and 4 a row, copying one a word each and 1 a row, and each integer an item reads costs 38, what one takes at worst, so one with no pixels pays too, before it is decoded: each bitmap was bounded, but nothing bounded how many there were, and a 795-byte PDF took 265 s; (11) `Image::decode_within` spends from what a run of images may still do, starting with its page's pixels, and `charge_reading` charges reading a stream and its globals before anything is parsed, so an operation over a document can bound what `MAX_WORK` bounds for one image, and images that fail or decode nothing add up too: a file of thousands of images made to reach that limit took as many times as long; (12) adding to a location, a reference offset, a symbol's corner or a Huffman range line near the end of an `i32`, and negating a decoded integer of 2^31, overflowed: a panic wherever overflow is checked. The refinement context, a text region's corners, the integer decoder and the Huffman table wrap, as a release build did, and `Bitmap::combine` saturates, which comes to the same, as a location that overflows is past the end of any page. What is drawn doesn't change; (13) a bitmap filled with a default pixel of 1 (a page, or an intermediate text or halftone region) had 1s past its width in the last word of each row, and a refinement, which reads its reference a word at a time, read them as pixels where the standard has 0 (6.3.5.2): refining such a reference not a multiple of 32 wide decoded to other pixels than were coded, release builds too. `Bitmap::clear_padding` clears those bits when a bitmap is filled with 1s, as cutting symbols and patterns from a collective bitmap already did. Ghostscript reads the test's page and text streams as they decode now; (14) a region's location is unsigned (5.4.4), so one past `i32::MAX` is off the page, where the decoder took it for a signed one: a region there was drawn back onto the page from the left or the top (at 0xFFFFFFFC, four pixels left of it), and a refinement there, of the page or of a region at the same place (7.4.7.5), was refused as an overflow, and Line truncated
 | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary, and an in-file /Root must match its generation (#2365); (7) object parsing builds at most 4,194,304 objects (every array element and dictionary value at any depth, `parser::MAX_OBJECT_NODES`, the same cap as `pdfcraft_cos`) per top-level object, an object stream reads at most 1,000,000 index entries (cos's `/N` clamp), and one load builds at most 8,388,608 objects from all of its object streams together, shared across rayon workers (a decoded stream of `0 0 0 ...` inside the 256 MiB decompression limit built ~120M objects, 10+ GB, and aborted). A lookup parses only the member it wants, so it isn't skipped behind a large member, and a load parses only the members it would add: none its cross-reference table places in another stream (stale copies) and none already loaded as an ordinary object or from an earlier stream. Members the table doesn't list at all are still parsed (a rebuilt table lists no compressed objects) and share the budget, so they can spend it before listed members of the same stream; a load expands its object streams one at a time in object-number order, a stream's members in parallel only when the budget holds all of them and otherwise in index order, so a spent budget skips the same members on every load (#5083) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog`; (7) `inspect::tests::{object_stream_members_over_the_node_cap_are_skipped, object_stream_index_entries_past_the_clamp_are_ignored, object_streams_of_one_load_share_their_node_budget, object_streams_loaded_in_parallel_share_their_node_budget, a_malformed_object_stream_member_refunds_its_reservation, a_tight_budget_builds_object_stream_members_in_index_order, a_lookup_in_an_object_stream_parses_only_the_member_it_wants}` (pdfcraft-render; `NodeBudget` and `ObjectStream::new_within` are public so these can use a small budget) |

vendor/hayro-interpret/src/cmyk_inverse.rs

@@ -0,0 +1,118 @@
+//! PdfCraft patch (104): an sRGB colour as the process ink that draws back to it through the
+//! CMYK profile pages are drawn with, for an additive colour painted into a `DeviceCMYK`
+//! blending space (ISO 32000-2 §11.3.4).
+//!
+//! Black is `K = 0.4 · k₀² + 0.6 · k₀⁸` with `k₀ = 1 − max(r, g, b)`, light in the mid tones
+//! and full for black; cyan, magenta and yellow are then solved for by Newton's method at that
+//! black, clamped to the ink range. A colour inside the profile's gamut draws as itself; one
+//! outside it lands where the clamped solve stops. The curve matches Acrobat's output for the
+//! greys, black and red measured in a CMYK page group, plain and under a blend mode;
+//! §10.4.2.4's black generation and undercolour removal is for a device without a profile.
+
+use moxcms::{ColorProfile, Layout, TransformF32Executor, TransformOptions};
+use std::cell::Cell;
+use std::sync::{Arc, LazyLock};
+
+use crate::color::CMYK_PROFILE;
+
+/// The weight of `k₀²` in the black ink; the rest is on `k₀⁸`.
+const BLACK_GENERATION: f32 = 0.4;
+/// Newton steps; each costs four profile evaluations.
+const MAX_STEPS: usize = 12;
+/// Close enough: well under one 8-bit level.
+const TOLERANCE: f32 = 0.5 / 255.0;
+/// The finite-difference step for the Jacobian.
+const STEP: f32 = 0.02;
+
+static FORWARD: LazyLock<Option<Arc<TransformF32Executor>>> = LazyLock::new(|| {
+    ColorProfile::new_from_slice(CMYK_PROFILE)
+        .ok()?
+        .create_transform_f32(
+            Layout::Rgba,
+            &ColorProfile::new_srgb(),
+            Layout::Rgb,
+            TransformOptions::default(),
+        )
+        .ok()
+});
+
+/// The process ink `[C, M, Y, K]` (0 = no ink) that the CMYK profile draws as `rgb`; for a
+/// colour outside its gamut, where the clamped solve stops. `None` when the
+/// profile can't be evaluated or `rgb` isn't finite.
+///
+/// The last colour converted on this thread is kept, as a run of fills (the glyphs of a
+/// line of text) usually shares one colour.
+pub(crate) fn srgb_to_cmyk(rgb: [f32; 3]) -> Option<[f32; 4]> {
+    if !rgb.iter().all(|v| v.is_finite()) {
+        return None;
+    }
+    let rgb = rgb.map(|v| v.clamp(0.0, 1.0));
+    let key = rgb.map(f32::to_bits);
+    if let Some((last, ink)) = LAST.get()
+        && last == key
+    {
+        return Some(ink);
+    }
+    let ink = solve_ink(rgb)?;
+    LAST.set(Some((key, ink)));
+    Some(ink)
+}
+
+thread_local! {
+    static LAST: Cell<Option<([u32; 3], [f32; 4])>> = const { Cell::new(None) };
+}
+
+fn solve_ink(rgb: [f32; 3]) -> Option<[f32; 4]> {
+    let forward: &TransformF32Executor = FORWARD.as_ref()?.as_ref();
+    let k0 = 1.0 - rgb[0].max(rgb[1]).max(rgb[2]);
+    let k = BLACK_GENERATION * k0 * k0 + (1.0 - BLACK_GENERATION) * k0.powi(8);
+    let eval = |cmy: [f32; 3]| draw(forward, [cmy[0], cmy[1], cmy[2], k]);
+
+    let mut cmy = rgb.map(|v| 1.0 - v);
+    for _ in 0..MAX_STEPS {
+        let drawn = eval(cmy)?;
+        let residual: [f32; 3] = std::array::from_fn(|i| drawn[i] - rgb[i]);
+        if residual.iter().all(|r| r.abs() < TOLERANCE) {
+            break;
+        }
+        // Column `j` of the Jacobian: how the drawn colour moves with ink `j`.
+        let mut columns = [[0.0_f32; 3]; 3];
+        for (j, column) in columns.iter_mut().enumerate() {
+            let step = if cmy[j] <= 1.0 - STEP { STEP } else { -STEP };
+            let mut moved = cmy;
+            moved[j] += step;
+            let shifted = eval(moved)?;
+            *column = std::array::from_fn(|i| (shifted[i] - drawn[i]) / step);
+        }
+        let Some(delta) = solve(columns, residual.map(|r| -r)) else {
+            break;
+        };
+        cmy = std::array::from_fn(|i| (cmy[i] + delta[i]).clamp(0.0, 1.0));
+    }
+    Some([cmy[0], cmy[1], cmy[2], k])
+}
+
+fn draw(forward: &TransformF32Executor, cmyk: [f32; 4]) -> Option<[f32; 3]> {
+    let mut out = [0.0_f32; 3];
+    forward.transform(&cmyk, &mut out).ok()?;
+    out.iter().all(|v| v.is_finite()).then_some(out)
+}
+
+/// `x` with `A x = b`, `A` given by its columns (Cramer's rule). `None` when `A` is
+/// singular or the result isn't finite.
+fn solve(a: [[f32; 3]; 3], b: [f32; 3]) -> Option<[f32; 3]> {
+    let det = |[c0, c1, c2]: [[f32; 3]; 3]| {
+        c0[0] * (c1[1] * c2[2] - c2[1] * c1[2]) - c1[0] * (c0[1] * c2[2] - c2[1] * c0[2])
+            + c2[0] * (c0[1] * c1[2] - c1[1] * c0[2])
+    };
+    let d = det(a);
+    if !d.is_finite() || d.abs() < 1e-9 {
+        return None;
+    }
+    let x: [f32; 3] = std::array::from_fn(|j| {
+        let mut m = a;
+        m[j] = b;
+        det(m) / d
+    });
+    x.iter().all(|v| v.is_finite()).then_some(x)
+}

vendor/hayro-interpret/src/color.rs

@@ -1552,9 +1552,10 @@ impl Color {
     /// its opacity, for blending in a `DeviceCMYK` transparency group (ISO 32000-2 §11.3.4).
     ///
     /// `DeviceCMYK` gives its components. `DeviceGray` is black ink alone, `K = 1 − gray`
-    /// (§10.4.2.2). `None` for every other space, ICCBased CMYK included: such a colour has no
-    /// tints without a colour conversion this crate does not make (an ICCBased colour keeps
-    /// its own profile), and the caller draws it as before.
+    /// (§10.4.2.2). PdfCraft patch (104): an additive colour (`DeviceRGB`, `CalRGB`, `Lab`, a
+    /// three-component ICCBased colour) is the ink that draws back to it through the CMYK
+    /// profile. `None` for every other space, ICCBased CMYK included (it keeps its own
+    /// profile), and the caller draws it as before.
     pub fn ink_tints(&self) -> Option<([f32; 4], f32)> {
         let c = |i: usize| {
             self.components
@@ -1566,11 +1567,27 @@ impl Color {
         let tints = match self.color_space.0.as_ref() {
             ColorSpaceType::DeviceCmyk => [c(0), c(1), c(2), c(3)],
             ColorSpaceType::DeviceGray => [0.0, 0.0, 0.0, 1.0 - c(0)],
+            ColorSpaceType::DeviceRgb | ColorSpaceType::CalRgb(_) | ColorSpaceType::Lab(_) => {
+                self.ink_from_srgb()?
+            }
+            ColorSpaceType::ICCBased(icc) if icc.0.number_components == 3 => {
+                self.ink_from_srgb()?
+            }
             _ => return None,
         };
         Some((tints, self.opacity.clamp(0.0, 1.0)))
     }
 
+    fn ink_from_srgb(&self) -> Option<[f32; 4]> {
+        // `to_rgba` turns a non-finite component into some finite colour, which is not the
+        // colour's ink, so such a colour has no tints.
+        if self.components.iter().any(|c| !c.is_finite()) {
+            return None;
+        }
+        let [r, g, b, _] = self.to_rgba().components();
+        crate::cmyk_inverse::srgb_to_cmyk([r, g, b])
+    }
+
     /// Create a color from RGBA.
     pub fn from_rgba(rgba: AlphaColor) -> Self {
         let c = rgba.components();
@@ -1592,9 +1609,11 @@ pub fn cmyk_to_srgb8(input: &[u8], output: &mut [u8]) -> Option<()> {
     CMYK_TRANSFORM.convert_u8(input, output)
 }
 
-static CMYK_TRANSFORM: LazyLock<ICCProfile> = LazyLock::new(|| {
-    ICCProfile::new(include_bytes!("../assets/CGATS001Compat-v2-micro.icc"), 4).unwrap()
-});
+/// The CMYK profile pages are drawn with.
+pub(crate) const CMYK_PROFILE: &[u8] = include_bytes!("../assets/CGATS001Compat-v2-micro.icc");
+
+static CMYK_TRANSFORM: LazyLock<ICCProfile> =
+    LazyLock::new(|| ICCProfile::new(CMYK_PROFILE, 4).unwrap());
 
 pub(crate) trait ToRgb {
     fn convert_sample(&self, input: &[f32], output: &mut [u8], manual_scale: bool) -> Option<()> {

vendor/hayro-interpret/src/lib.rs

@@ -33,6 +33,7 @@ mod log;
 
 mod cache;
 mod cmyk_group;
+mod cmyk_inverse;
 mod context;
 mod convert;
 mod device;

vendor/hayro/src/renderer.rs

@@ -922,8 +922,8 @@ impl TilingCells {
 #[derive(Clone, Copy, Default)]
 struct InkState {
     pass: Option<InkPass>,
-    /// Something in the group has no ink tints (an RGB colour, a pattern, an image): the
-    /// group is drawn as before instead.
+    /// Something in the group has no ink tints (a pattern, an image, an ICCBased CMYK colour):
+    /// the group is drawn as before instead.
     unsupported: bool,
 }