Merge #6510: M1: the renderer reads a cut-off last update's objects, as the document model does (core.repair-broken-files)

# Conflicts:
#	vendor/README.md

crates/render/src/raster/tests.rs

@@ -9238,6 +9238,183 @@ fn a_rebuilt_table_finds_an_object_whose_header_names_another_generation() {
     assert_eq!((page.width, page.height), (200, 100));
 }
 
+/// A one-page file: a 200 x 100 page, its offsets right.
+fn one_page_file() -> Vec<u8> {
+    one_page_file_with(false)
+}
+
+/// The same, and with `stale` a 999 x 999 copy of the page after it that the table does not
+/// name: a file read through its table shows 200 x 100, one rebuilt from object headers 999 x 999.
+fn one_page_file_with(stale: bool) -> Vec<u8> {
+    let mut out = b"%PDF-1.4\n".to_vec();
+    let mut offsets = Vec::new();
+    for body in
+        ["<< /Type /Catalog /Pages 2 0 R >>", "<< /Type /Pages /Kids [3 0 R] /Count 1 >>", "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 100] >>"]
+    {
+        offsets.push(out.len());
+        out.extend_from_slice(format!("{} 0 obj\n{body}\nendobj\n", offsets.len()).as_bytes());
+    }
+    if stale {
+        out.extend_from_slice(b"3 0 obj\n<< /Type /Page /Parent 2 0 R /MediaBox [0 0 999 999] >>\nendobj\n");
+    }
+    let xref = out.len();
+    out.extend_from_slice(b"xref\n0 4\n0000000000 65535 f \n");
+    for o in offsets {
+        out.extend_from_slice(format!("{o:010} 00000 n \n").as_bytes());
+    }
+    out.extend_from_slice(format!("trailer\n<< /Size 4 /Root 1 0 R >>\nstartxref\n{xref}\n%%EOF\n").as_bytes());
+    out
+}
+
+/// `base` with an incremental update appended: a new 300 x 150 first page whose content stream is
+/// `padding` bytes long, and the page tree redefined to hold both pages. With `footer` the update
+/// ends in its trailer, `startxref` and `%%EOF`; without, it stops after its cross-reference rows,
+/// as a save cut short does.
+fn with_update(base: &[u8], padding: usize, footer: bool) -> Vec<u8> {
+    let prev = base
+        .windows(9)
+        .rposition(|w| w == b"startxref")
+        .map(|p| String::from_utf8_lossy(&base[p + 9..]).split_whitespace().next().unwrap_or("").to_string())
+        .unwrap();
+    let mut out = base.to_vec();
+    let mut offsets = vec![(4, out.len())];
+    out.extend_from_slice(b"4 0 obj\n<< /Type /Page /Parent 2 0 R /MediaBox [0 0 300 150] /Contents 5 0 R >>\nendobj\n");
+    offsets.push((5, out.len()));
+    out.extend_from_slice(format!("5 0 obj\n<< /Length {padding} >>\nstream\n").as_bytes());
+    out.extend(std::iter::repeat_n(b' ', padding));
+    out.extend_from_slice(b"\nendstream\nendobj\n");
+    offsets.push((2, out.len()));
+    out.extend_from_slice(b"2 0 obj\n<< /Type /Pages /Kids [4 0 R 3 0 R] /Count 2 >>\nendobj\n");
+    let xref = out.len();
+    out.extend_from_slice(b"xref\n0 1\n0000000000 65535 f \n");
+    for (num, o) in offsets {
+        out.extend_from_slice(format!("{num} 1\n{o:010} 00000 n \n").as_bytes());
+    }
+    if footer {
+        out.extend_from_slice(format!("trailer\n<< /Size 6 /Root 1 0 R /Prev {prev} >>\nstartxref\n{xref}\n%%EOF\n").as_bytes());
+    }
+    out
+}
+
+/// What the renderer shows of `file` (page count and the first page's size) and what the
+/// document model reads (the page tree's /Count), for comparing the two.
+fn renderer_and_model_pages(file: Vec<u8>) -> ((usize, (u32, u32)), i64) {
+    let bytes = Arc::new(file);
+    let mut r = PageRenderer::new(bytes.clone(), RenderConfig::default());
+    let count = r.page_count();
+    let page = r.render(RenderRequest { page: 0, kind: RequestKind::Pixels, tile: None, scale: 1.0, tag: 0, preview: false });
+    assert!(page.error.is_none(), "{:?}", page.error);
+    let doc = pdfcraft_cos::Document::open(bytes).unwrap();
+    let catalog = doc.get(doc.root().unwrap());
+    let pages = doc.resolve(catalog.as_dict().unwrap().get(b"Pages").unwrap());
+    let model = pages.as_dict().and_then(|d| d.int(b"Count")).unwrap();
+    ((count, (page.width, page.height)), model)
+}
+
+/// PDFBOX-3934, PDFBOX-3787: the last incremental update was cut off after its cross-reference
+/// rows, with no trailer, `startxref` or `%%EOF`, so the last `startxref` in the file is the
+/// older revision's, more than the footer window (4 KiB) from the end, and objects follow it.
+/// The update's objects are the latest revision (ISO 32000-2 §7.5.6); the document model rebuilds
+/// the file from object headers and reads them. The renderer read the older table the last
+/// `startxref` names instead and showed one page where the model has two. It now rebuilds too.
+#[test]
+fn an_update_cut_off_before_its_trailer_shows_its_pages_as_the_model_reads_them() {
+    let file = with_update(&one_page_file(), 5000, false);
+    assert_eq!(model_first_media_box(&file), [0.0, 0.0, 300.0, 150.0], "the model's first page is the update's");
+    let ((count, size), model) = renderer_and_model_pages(file);
+    assert_eq!(model, 2, "the model reads the update");
+    assert_eq!(count, 2, "the renderer shows the update's page tree");
+    assert_eq!(size, (300, 150), "and its new first page");
+}
+
+/// The same update with its footer is read through its table as before: renderer and model both
+/// show its two pages.
+#[test]
+fn a_complete_update_reads_as_before() {
+    let ((count, size), model) = renderer_and_model_pages(with_update(&one_page_file(), 5000, true));
+    assert_eq!((count, size, model), (2, (300, 150), 2));
+}
+
+/// A file with data after `%%EOF` that holds no object header (a web page glued to a served file)
+/// is read through its table by the renderer, as before: the file has a stale copy of its page that
+/// only a rebuild would show. The model, finding no `startxref` in its footer window, rebuilds from
+/// object headers and takes that stale 999 x 999 copy, so here the two still differ, as they did
+/// before this change.
+#[test]
+fn trailing_data_without_objects_reads_as_before() {
+    let mut junk = one_page_file_with(true);
+    junk.extend_from_slice(b"<html><p>1 0 objection: 12 0 endobj, 3 0obj, 4 0 obj_name, 4294967296 0 obj</p>");
+    junk.extend(std::iter::repeat_n(b' ', 5000));
+    junk.extend_from_slice(b"</html>\n");
+    assert_eq!(model_first_media_box(&junk), [0.0, 0.0, 999.0, 999.0], "the model's rebuilt page");
+    let ((count, size), model) = renderer_and_model_pages(junk);
+    assert_eq!((count, size, model), (1, (200, 100), 1));
+}
+
+/// The /MediaBox of the first page in the document model's page tree (its first /Kids entry).
+fn model_first_media_box(file: &[u8]) -> Vec<f64> {
+    let doc = pdfcraft_cos::Document::open(Arc::new(file.to_vec())).unwrap();
+    let catalog = doc.get(doc.root().unwrap());
+    let pages = doc.resolve(catalog.as_dict().unwrap().get(b"Pages").unwrap());
+    let kid = doc.resolve(pages.as_dict().unwrap().get(b"Kids").unwrap().as_array().unwrap().first().unwrap());
+    kid.as_dict().unwrap().get(b"MediaBox").unwrap().as_array().unwrap().iter().filter_map(|x| x.as_f64()).collect()
+}
+
+/// A cut-off update on `one_page_file` whose object headers are written `N G obj{after}<<` in place
+/// of `N G obj\n<<`. Only the update is rewritten, so the base file's table stays at the offset
+/// its `startxref` names and reading it raises no error that would rebuild the file anyway: the
+/// update is found only if the header scan accepts what follows `obj`.
+fn cut_off_update_with_headers(after: &str) -> Vec<u8> {
+    let base = one_page_file();
+    let mut file = with_update(&base, 5000, false);
+    let update = String::from_utf8(file.split_off(base.len())).unwrap();
+    assert_eq!(update.matches(" obj\n<<").count(), 3, "every updated header is rewritten");
+    file.extend_from_slice(update.replace(" obj\n<<", &format!(" obj{after}<<")).as_bytes());
+    file
+}
+
+/// A cut-off update whose object headers run straight into their dictionaries (`4 0 obj<<`) is
+/// still found: a delimiter ends the keyword (ISO 32000-2 §7.2.3, Table 2), and the renderer
+/// shows the update's new first page, as the model does.
+#[test]
+fn a_cut_off_update_with_obj_followed_by_a_delimiter_is_rebuilt() {
+    let file = cut_off_update_with_headers("");
+    assert_eq!(model_first_media_box(&file), [0.0, 0.0, 300.0, 150.0], "the model's first page is the update's");
+    let ((count, size), model) = renderer_and_model_pages(file);
+    assert_eq!((count, size, model), (2, (300, 150), 2));
+}
+
+/// The same with a comment straight after the keyword (`4 0 obj%comment`): `%` is a delimiter
+/// that starts a comment (ISO 32000-2 §7.2.3 Table 2, §7.2.4), so it ends the keyword too.
+#[test]
+fn a_cut_off_update_with_obj_followed_by_a_comment_is_rebuilt() {
+    let file = cut_off_update_with_headers("%comment\n");
+    assert_eq!(model_first_media_box(&file), [0.0, 0.0, 300.0, 150.0], "the model's first page is the update's");
+    let ((count, size), model) = renderer_and_model_pages(file);
+    assert_eq!((count, size, model), (2, (300, 150), 2));
+}
+
+/// A cut-off update that ends within the footer window leaves the older `startxref` near the end,
+/// which the model reads too: both keep the older revision there.
+#[test]
+fn a_short_cut_off_update_reads_as_before() {
+    let ((count, size), model) = renderer_and_model_pages(with_update(&one_page_file(), 100, false));
+    assert_eq!((count, size, model), (1, (200, 100), 1));
+}
+
+/// A file cut short in the middle of its last update, whose new page tree names a page that is
+/// not there: the rebuilt table finds no page, so the older table is read and the file still
+/// opens with the page it had, rather than not at all.
+#[test]
+fn an_update_cut_off_before_its_new_pages_keeps_the_older_table() {
+    let file = String::from_utf8(with_update(&one_page_file(), 5000, false)).unwrap().replace("4 0 obj", "9 0 obj").into_bytes();
+    let mut r = PageRenderer::new(Arc::new(file), RenderConfig::default());
+    assert_eq!(r.page_count(), 1);
+    let page = r.render(RenderRequest { page: 0, kind: RequestKind::Pixels, tile: None, scale: 1.0, tag: 0, preview: false });
+    assert!(page.error.is_none(), "{:?}", page.error);
+    assert_eq!((page.width, page.height), (200, 100));
+}
+
 /// #3109 / #3116: clipping-only text (`7 Tr`) installs a clip for subsequent painting
 /// but draws no pixels itself.
 #[test]
@@ -9780,3 +9957,97 @@ fn a_stream_with_a_crypt_filter_of_its_own_is_drawn() {
         }
     }
 }
+
+/// Data after `%%EOF` that holds an object header, past the footer window, is read the way the
+/// document model reads it: the model rebuilds the file from object headers and takes the later
+/// definition of the page, and the renderer shows that same page, so the two still agree.
+#[test]
+fn trailing_data_with_an_object_header_reads_as_the_model_reads_it() {
+    let mut f = one_page_file();
+    f.extend_from_slice(
+        b"<html><pre>
+3 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 999 999] >>
+endobj
+</pre>",
+    );
+    f.extend(std::iter::repeat_n(b' ', 5000));
+    let doc = pdfcraft_cos::Document::open(Arc::new(f.clone())).unwrap();
+    let page = doc.get(pdfcraft_cos::ObjRef::new(3, 0));
+    let model_box = page.as_dict().and_then(|d| d.get(b"MediaBox").cloned());
+    assert_eq!(model_box, Some(pdfcraft_cos::Object::Array([0, 0, 999, 999].map(pdfcraft_cos::Object::Int).to_vec())), "the model");
+    let ((count, size), model) = renderer_and_model_pages(f);
+    assert_eq!((count, size, model), (1, (999, 999), 1), "the renderer, as the model");
+}
+
+/// An encrypted file whose last update was cut off is still refused with a wrong password, and
+/// opened with the right one.
+#[test]
+fn an_encrypted_file_with_a_cut_off_update_still_needs_its_password() {
+    use pdfcraft_cos::{Algorithm, Document, NewEncryption, SaveOptions, write_full};
+    let mut doc = Document::open(Arc::new(one_page_file())).unwrap();
+    doc.set_encryption(&NewEncryption {
+        algorithm: Algorithm::Aes128,
+        user_password: "user",
+        owner_password: "owner",
+        permissions: -1,
+        encrypt_metadata: true,
+        seed: [7; 32],
+    })
+    .unwrap();
+    let base = write_full(&doc, &SaveOptions { object_streams: false, ..SaveOptions::default() }).unwrap();
+    let file = Arc::new(with_update(&base, 5000, false));
+    let open = |password: &str| {
+        let mut r = PageRenderer::new(file.clone(), RenderConfig { password: Some(Arc::from(password)), ..RenderConfig::default() });
+        let page = r.render(RenderRequest { page: 0, kind: RequestKind::Pixels, tile: None, scale: 1.0, tag: 0, preview: false });
+        (r.page_count(), page.error.is_none(), (page.width, page.height))
+    };
+    let (_, drawn, _) = open("wrong");
+    assert!(!drawn, "a wrong password opens nothing");
+    let (count, drawn, _) = open("user");
+    assert!(count >= 1 && drawn, "the right one opens it");
+}
+
+/// An encrypted file whose cut-off update ends in a trailer dictionary without `/Encrypt`: the rebuild would take that
+/// later dictionary and read the encrypted content as plain bytes, so it is not taken. A wrong password is still
+/// refused, and the right one opens the table the last `startxref` names, its content decrypted and drawn.
+#[test]
+fn a_rebuild_whose_trailer_lacks_the_encryption_dictionary_is_not_taken() {
+    use pdfcraft_cos::{Algorithm, Document, NewEncryption, SaveOptions, write_full};
+    let plain = b"%PDF-1.4
+1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj
+2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj
+3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 200 100] /Contents 4 0 R >> endobj
+4 0 obj << /Length 25 >> stream
+0 0 1 rg 10 10 50 50 re f
+endstream endobj
+trailer << /Root 1 0 R >>
+%%EOF"
+        .to_vec();
+    let mut doc = Document::open(Arc::new(plain)).unwrap();
+    doc.set_encryption(&NewEncryption {
+        algorithm: Algorithm::Aes128,
+        user_password: "user",
+        owner_password: "owner",
+        permissions: -1,
+        encrypt_metadata: true,
+        seed: [7; 32],
+    })
+    .unwrap();
+    let root = doc.root().unwrap();
+    let mut file = write_full(&doc, &SaveOptions { object_streams: false, ..SaveOptions::default() }).unwrap();
+    // The update: one object, past the 4 KiB footer window, and a trailer without `/Encrypt`, cut off before its
+    // `startxref`.
+    file.extend_from_slice(b"900 0 obj\n<< /Length 5000 >>\nstream\n");
+    file.extend(std::iter::repeat_n(b' ', 5000));
+    file.extend_from_slice(format!("\nendstream\nendobj\ntrailer\n<< /Size 901 /Root {} {} R >>\n", root.num, root.generation).as_bytes());
+    let file = Arc::new(file);
+    let open = |password: &str| {
+        let mut r = PageRenderer::new(file.clone(), RenderConfig { password: Some(Arc::from(password)), ..RenderConfig::default() });
+        let page = r.render(RenderRequest { page: 0, kind: RequestKind::Pixels, tile: None, scale: 1.0, tag: 0, preview: false });
+        page.error.is_none().then(|| rgb_at(&page.rgba, page.width as usize, 30, page.height as usize - 30))
+    };
+    assert_eq!(open("wrong"), None, "a wrong password opens nothing");
+    assert_eq!(open(""), None, "nor does none");
+    assert_eq!(open("user"), Some([0, 0, 255]), "the right one draws the decrypted content");
+}

parity/acrobat-features.toml

@@ -46,8 +46,8 @@ milestone = "M1"
 status = "shipped"
 commands = ["file.open"]
 tools = ["doc_open"]
-evidence = ["crates/cos/src/document/tests.rs::damaged_xref_is_reconstructed", "crates/cos/src/parser.rs::indirect_stream_with_wrong_length_recovers", "crates/cos/src/document/tests.rs::self_referential_length_does_not_overflow", "crates/engine/src/tests.rs::a_repaired_file_is_rewritten_by_save_even_without_edits", "crates/automation/tests/it/automation.rs::saving_a_repaired_file_puts_the_repair_on_disk", "crates/engine/src/tests.rs::a_signed_repaired_file_keeps_its_bytes_when_saved_without_edits", "crates/cos/src/document/tests.rs::a_table_numbered_from_1_that_begins_with_object_0_is_read_from_0"]
-notes = "Save and doc_save write a reconstructed file in full even when nothing was edited, so the repair reaches the disk and the saved file opens without one (#1937); a signed one keeps its bytes, so its signatures stay valid. A repair does not mark the document as having unsaved changes, so closing it does not offer to save the repaired file. A classic table numbered 1 n that begins with object 0's free entry is read from 0 (#2676) and logged."
+evidence = ["crates/cos/src/document/tests.rs::damaged_xref_is_reconstructed", "crates/cos/src/parser.rs::indirect_stream_with_wrong_length_recovers", "crates/cos/src/document/tests.rs::self_referential_length_does_not_overflow", "crates/engine/src/tests.rs::a_repaired_file_is_rewritten_by_save_even_without_edits", "crates/automation/tests/it/automation.rs::saving_a_repaired_file_puts_the_repair_on_disk", "crates/engine/src/tests.rs::a_signed_repaired_file_keeps_its_bytes_when_saved_without_edits", "crates/cos/src/document/tests.rs::a_table_numbered_from_1_that_begins_with_object_0_is_read_from_0", "crates/render/src/raster/tests.rs::an_update_cut_off_before_its_trailer_shows_its_pages_as_the_model_reads_them"]
+notes = "Save and doc_save write a reconstructed file in full even when nothing was edited, so the repair reaches the disk and the saved file opens without one (#1937); a signed one keeps its bytes, so its signatures stay valid. A repair does not mark the document as having unsaved changes, so closing it does not offer to save the repaired file. A classic table numbered 1 n that begins with object 0's free entry is read from 0 (#2676) and logged. A last update cut off before its trailer (objects after a last `startxref` more than 4 KiB from the end) is rebuilt from object headers by the renderer as by the document model, so both show its latest objects."
 
 [[feature]]
 id = "core.malformed-input-safety"

vendor/README.md

@@ -43,7 +43,7 @@ Other resource categories retain their existing parent-chain lookup.
 | 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 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer); (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 the page with it. `Bitmap::combine_region` keeps such a location past the page. A refinement of another region refines it where it is, as Table 35 has it (GRREFERENCEDX and GRREFERENCEDY 0), whatever the two locations say; a refinement of the page, see (16); (15) a refinement whose reference is offset right of it (GRREFERENCEDX above 0, which a refined text instance has from a wider bitmap, TaLine 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) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` |
-| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); an EI followed by a delimiter (EI%..., EI[...) ends the image like one followed by white space; only a regular character after EI skips the candidate; (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwLine truncated
+| hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); an EI followed by a delimiter (EI%..., EI[...) ends the image like one followed by white space; only a regular character after EI skips the candidate; (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwLine truncated
 | winit | 0.30.13 | Apache-2.0 | (1) on Windows 11 (build 22000 and later), `WM_DPICHANGED` applies the rect Windows suggests. It used to nudge the window onto the monitor `MonitorFromWindow` names, which during a drag is still the monitor being left; the nudge sent another `WM_DPICHANGED`, so a window dragged to a monitor with another scale factor grew and stayed on the wrong monitor at the wrong scale (#324). Backport of winit master 488c036a ([winit#4341](https://github.com/rust-windowing/winit/pull/4341)), which no 0.30 release has; Windows 10 keeps the old adjustment, as on winit master. (2) on Wayland, a seat's text input is destroyed when the seat goes away, not whenever it loses a capability: losing any capability (a touch screen or tablet unplugged, a remote-desktop session ending) destroyed it, and only an added capability recreated it, so IME stopped working until restart. Port of [winit#4747](https://github.com/rust-windowing/winit/pull/4747); (3) on Windows, an IME composition reports its cursor from `GCS_CURSORPOS` only when `WM_IME_COMPOSITION` says it is set, and at the end of the composition otherwise: the Microsoft Korean IME never sets it, so the cursor read as 0 (egui then drew no caret while composing; it now draws it after the syllable). Port of [winit#4746](https://github.com/rust-windowing/winit/pull/4746); (4) on Windows, a hidden window is not maximized until it is shown, and a hidden window shown while maximized gets one `ShowWindow(SW_MAXIMIZE)`: `SW_MAXIMIZE` shows a hidden window, so PdfCraft, which eframe keeps hidden until its first frame is drawn, appeared unpainted (white) when it restored a maximized window, and could lose its focus. Port of [winit#4587](https://github.com/rust-windowing/winit/pull/4587) (the later, focus-safe form of [winit#4575](https://github.com/rust-windowing/winit/pull/4575)); (5) on X11, when allowing or disallowing IME replaces the focused window's input context (egui does it each time a text field gains or loses focus), the new context is focused: no focus event follows, so it stayed unfocused until the window lost and regained focus. Port of [winit#4727](https://github.com/rust-windowing/winit/pull/4727); (6) on macOS, `insertText` commits through IME when a composition was started (`pending_commit`, set by `setMarkedText`) rather than when marked text is left: input methods that end a composition with `unmarkText` before `insertText` lost the text, and Apple Korean dropped the key typed after a syllable (`한` then `5` or `;`); other `insertText` calls stay `KeyboardInput`; a commit outside a key press (a candidate clicked with the mouse, dictation) clears the marked text and leaves no key to swallow, as in [winit#4748](https://github.com/rust-windowing/winit/pull/4748) (without it the next key, such as Backspace, was lost). `validAttributesForMarkedText` lists the underline and marked-clause attributes, without which some input methods don't compose. Port of [winit#4650](https://github.com/rust-windowing/winit/pull/4650), the 0.30 backport of [winit#4651](https://github.com/rust-windowing/winit/pull/4651) (merged on winit master), with the underline attribute spelled as AppKit's `NSUnderlineStyleAttributeName` (`NSUnderline`); (7) on macOS, disallowing IME with a composition in progress (focus leaving a text field) tells the input method to discard its marked text (`discardMarkedText`): it kept composing it once IME was allowed again, so the abandoned syllable came back (`하`, then `가` typed in the next field, became `학`, then `하` + `가`; Japanese `か` then `さ` became `傘`). Port of [winit#4745](https://github.com/rust-windowing/winit/pull/4745); (8) on macOS, text inserted outside a key press without a composition (the emoji picker) is reported as `Ime::Commit`; it was dropped. Port of the emoji-picker part of [winit#4749](https://github.com/rust-windowing/winit/pull/4749) (0.30 backport of [winit#4748](https://github.com/rust-windowing/winit/pull/4748)); its `selectedRange` change for dictation is left out because it switches on press-and-hold for every held key (an accent menu instead of key repeat, and the chosen accent added after the repeated letters); (9) on macOS, a key whose text was committed through IME isn't also sent as `KeyboardInput` unless AppKit forwards it as a command (Enter): Apple Korean answers Space with no syllable in progress with `setMarkedText(" ")`, `insertText(" ")`, `setMarkedText("")`, and the last call reset winit's state, so the space arrived twice. Same goal as [winit#4478](https://github.com/rust-windowing/winit/pull/4478), whose changes don't reach this call order (macOS 14); to be proposed upstream there; (10) on macOS, when the first key in a fresh app reaches Apple Korean's `insertText` as a raw compatibility jamo instead of starting a composition, that key is interpreted once more, synchronously, before it is sent as `KeyboardInput` (g k s typed ㅎㅏㄴ instead of 한); only an unmodified, Line truncated
 | egui-wgpu | 0.36.2 | MIT OR Apache-2.0 | (1) `winit::surface_fit`: a window's surface is configured within the device's `max_texture_dimension_2d`, both sides shrunk by one factor that egui then draws at (`pixels_per_point × render_scale`), so the whole window is drawn into the surface (on DX12 it is stretched over the window and lines up with the pointer; wgpu's GL backend copies it unscaled into a corner); the longer side becomes exactly the limit; the depth and MSAA textures follow, and a recreated surface starts again from the window's own size. egui-wgpu configured the window's size as is, and `Surface::configure` panics beyond the limit: a window restored at another monitor's scale (opening an Outlook attachment on a mixed-DPI laptop asked an 8192-pixel device for 8600 × 3423, #577) or stretched across monitors closed PdfCraft, before its window appeared when it happened at start-up ([emilk/egui#8361](https://github.com/emilk/egui/issues/8361)). The published crate leaves out egui's licence files; `LICENSE-APACHE` and `LICENSE-MIT` are from the 0.36.2 tag; (2) `winit::catch_errors`: a new window's surface is configured inside wgpu error scopes (validation, out of memory, internal), and a failure is returned as `WgpuError::ConfigureSurface` (for the root window at start-up eframe returns it as `Error::Wgpu`, which PdfCraft retries with OpenGL); nothing is kept for the window. wgpu's default handler panicked instead, and eframe configures the root surface before the app is created, so a driver that can't configure one closed PdfCraft at launch (a 2015 Intel GPU through wgpu's GL backend timed out waiting for the GPU to go idle, `GpuWaitTimeout`, #519). Unchanged, so still a panic: an error in a later reconfigure (resize, a lost surface), a device lost during the first configure (wgpu reports it only to a callback), and a window shown at zero size (first configured on resize) | (1) `pdfcraft` `tests::a_window_larger_than_the_gpu_limit_gets_a_surface_within_it`, `tests::surface_fit_never_empties_or_overflows_a_surface`, and `tests::vendor_provenance_egui_wgpu_surface_fit_identity`, which keeps the patch in the build and in the paths that size a surface; (2) `pdfcraft` `tests::a_gpu_error_while_a_window_is_set_up_is_returned_not_a_panic` (a real device: an error inside `catch_errors` is returned, not a panic) and `tests::vendor_provenance_egui_wgpu_catch_errors_identity`, which keeps it around `add_surface`'s configure |
 | vello_common | 0.0.8 | Apache-2.0 OR MIT | (1) `clip::intersect` (the clip step every clipped path passes through) starts its row walk at the first row both paths have strips in, not the earlier of the two first rows. The walk visited every row from the top of the clip for each path (on the A1 sample's full-page canvas, about half its strip rows per stroke), and rows where one path has no strips produce no output, so the strips are unchanged (vello_common 0.1 starts at the later row too). Its cost grew with the page, and a full-page render of the A1 sample was about 30% faster; the output is byte-identical; (4) `flatten::stroke`: when kurbo 0.13.1's stroker returns an outline containing NaN, which `fill` would drop whole, the path is stroked again from a copy (`robust_for_stroking`, one element out per element in) in which each curve whose control points lie within the stroke tolerance of its chord is a line (the #4907 fix, a nearly straight cubic given a NaN join, kurbo issues #617 and #519; its row here was lost in a merge and is restored with this entry) and each cubic whose second control point is its end point, as `y` builds (ISO 32000-2 §8.5.2.1 Table 58), has that control point moved the tolerance towards its first (`open_zero_end_handle`; the curve moves by at most 4/9 of the tolerance): kurbo's cusp search on such a curve's offset can stop on the end point, where the derivative is zero, and normalising it gives NaN, so at 16, 32 and 63 pt a two-curve `y` path drew nothing and a file in the PDFBox tracker lost its 32 pt "XX" outline. Only an exact zero triggers it (none of 192,000 random strokes with a handle 1e-9 long, nor of 256,000 `v` strokes, came back with NaN; 1,073 of 192,000 random `y`/`v` strokes at 1–64 pt did before and none after). A finite outline takes the old path unchanged | (1) `clip::tests::intersect_walks_only_rows_both_paths_have_strips` counts the rows the walk visits (`cargo test --manifest-path vendor/vello_common/Cargo.toml`) and fails without the patch; the render crate's pixel tests run through the patched crate; (4) `flatten::tests::a_tiny_degenerate_cubic_does_not_drop_the_stroke` and the render crate's `stroke_degenerate_cubic` fail without the near-line replacement, `flatten::tests::a_wide_stroke_of_a_curve_with_a_zero_end_tangent_is_drawn` (widths 1–64, every join and cap, `y` and the mirrored `v`) and the render crate's `stroke_zero_end_tangent` fail without the moved control point; `a_finite_outline_is_stroked_as_before` and `a_path_with_nan_is_still_dropped` pass either way |

vendor/hayro-syntax/src/pdf.rs

@@ -6,7 +6,7 @@ use crate::page::Pages;
 use crate::page::cached::CachedPages;
 use crate::reader::Reader;
 use crate::sync::Arc;
-use crate::xref::{XRef, XRefError, fallback, root_xref};
+use crate::xref::{XRef, XRefError, cut_off_update_follows, fallback, root_xref};
 
 pub use crate::crypto::DecryptionError;
 use crate::metadata::Metadata;
@@ -47,7 +47,36 @@ impl Pdf {
         let data = data.into();
         let password = password.as_bytes();
         let version = find_version(data.as_ref()).unwrap_or(PdfVersion::Pdf10);
-        let xref = match root_xref(data.clone(), password) {
+        // PdfCraft patch: (hayro-syntax (43)) a last update cut off before its trailer is read by
+        // rebuilding the table from object headers, the latest definition of each object
+        // winning, as the document model reads such a file. Only when that finds no page is the
+        // older table the last `startxref` names read instead (a file cut short in the middle of
+        // the update, whose new page tree names pages that are not there).
+        // The table the last `startxref` names is read first either way: a wrong password is
+        // refused through it, and the rebuilt table must be encrypted exactly when it is, so a
+        // rebuild that missed the encryption dictionary is not taken.
+        let root = root_xref(data.clone(), password);
+        if let Err(XRefError::Encryption(e)) = root {
+            return Err(LoadPdfError::Decryption(e));
+        }
+        let root_encrypted = root.as_ref().ok().map(XRef::is_encrypted);
+        if cut_off_update_follows(data.as_ref())
+            && let Some(xref) = fallback(data.clone(), password)
+                .filter(|x| root_encrypted.is_none_or(|e| e == x.is_encrypted()))
+        {
+            let xref = Arc::new(xref);
+            if let Some(pages) = CachedPages::new(xref.clone())
+                && !pages.get().is_empty()
+            {
+                return Ok(Self {
+                    xref,
+                    header_version: version,
+                    pages,
+                    data,
+                });
+            }
+        }
+        let xref = match root {
             Ok(x) => x,
             Err(e) => match e {
                 XRefError::Unknown => {

vendor/hayro-syntax/src/xref.rs

@@ -632,6 +632,15 @@ impl XRef {
         locked.xref_map = xref_map;
     }
 
+    /// PdfCraft patch: (hayro-syntax (43)) whether objects read through this table are
+    /// decrypted.
+    pub(crate) fn is_encrypted(&self) -> bool {
+        match &self.0 {
+            Inner::Dummy => false,
+            Inner::Some(r) => !matches!(r.decryptor.as_ref(), Decryptor::None),
+        }
+    }
+
     #[inline]
     pub(crate) fn needs_decryption(&self, ctx: &ReaderContext<'_>) -> bool {
         match &self.0 {
@@ -874,6 +883,11 @@ pub(crate) enum XRefInput<'a> {
     RootRef(ObjectIdentifier),
 }
 
+/// PdfCraft patch: (hayro-syntax (43)) how far from the end of the file a `startxref` counts as
+/// the file's own footer, as `pdfcraft-cos` judges it (ISO 32000-2 §7.5.5 puts it on the last lines
+/// of the file).
+const STARTXREF_TAIL: usize = 4096;
+
 pub(crate) fn find_last_xref_pos(data: &[u8]) -> Option<usize> {
     let needle = b"startxref";
     let pos = findr_needle(data, needle)?;
@@ -884,6 +898,83 @@ pub(crate) fn find_last_xref_pos(data: &[u8]) -> Option<usize> {
     finder.read_without_context::<i32>()?.try_into().ok()
 }
 
+/// PdfCraft patch: (hayro-syntax (43)) whether the last update of the file was cut off before its
+/// own cross-reference section and trailer: the last `startxref` is not near the end of the file
+/// and objects follow it. That section would supersede the older ones (ISO 32000-2 §7.5.6), so
+/// the update's objects are the document; reading the older table the last `startxref` names
+/// shows a revision the document model (`pdfcraft-cos`, which rebuilds such a file from its
+/// object headers) does not have. A `startxref` near the end, or one followed only by other data
+/// (trailing junk), is no cut-off update.
+pub(crate) fn cut_off_update_follows(data: &[u8]) -> bool {
+    let needle = b"startxref";
+    findr_needle(data, needle).is_some_and(|pos| {
+        pos < data.len().saturating_sub(STARTXREF_TAIL)
+            && data
+                .get(pos.saturating_add(needle.len())..)
+                .is_some_and(has_object_header)
+    })
+}
+
+/// PdfCraft patch: (hayro-syntax (43)) whether `data` holds an `N G obj` header: two runs of
+/// digits and the keyword, separated by white space, at the start of the data or after white
+/// space or `>`, the keyword ended by white space, a delimiter or the end of the data (ISO
+/// 32000-2 §7.2.3, Table 2), as the document model's repair scan reads it: `1 0 obj_name` is
+/// no header, and neither is one whose object number doesn't fit 32 bits (`4294967296 0 obj`),
+/// which that scan skips too. One linear pass, no allocation.
+fn has_object_header(data: &[u8]) -> bool {
+    let digits = |from: usize| {
+        data.get(from..).map_or(0, |rest| {
+            rest.iter().take_while(|b| b.is_ascii_digit()).count()
+        })
+    };
+    let spaces = |from: usize| {
+        data.get(from..).map_or(0, |rest| {
+            rest.iter()
+                .take_while(|&&b| is_white_space_character(b))
+                .count()
+        })
+    };
+    let mut i = 0;
+    while i < data.len() {
+        let at_start = i
+            .checked_sub(1)
+            .and_then(|prev| data.get(prev))
+            .is_none_or(|&b| is_white_space_character(b) || b == b'>');
+        let n = if at_start { digits(i) } else { 0 };
+        if n == 0 {
+            i += 1;
+            continue;
+        }
+        let mut at = i + n;
+        let s1 = spaces(at);
+        at += s1;
+        let g = digits(at);
+        at += g;
+        let s2 = spaces(at);
+        at += s2;
+        // The object number as the model reads it, a 32-bit unsigned integer.
+        let number_fits = data
+            .get(i..i + n)
+            .and_then(|d| core::str::from_utf8(d).ok())
+            .is_some_and(|d| d.parse::<u32>().is_ok());
+        if number_fits
+            && s1 > 0
+            && g > 0
+            && s2 > 0
+            && data.get(at..).is_some_and(|rest| rest.starts_with(b"obj"))
+            && data.get(at + 3).is_none_or(|&b| {
+                is_white_space_character(b)
+                    || matches!(b, b'(' | b')' | b'<' | b'>' | b'[' | b']' | b'{' | b'}' | b'/' | b'%')
+            })
+        {
+            return true;
+        }
+        // Past the digits read: none of them starts a header.
+        i += n;
+    }
+    false
+}
+
 /// A type of xref entry.
 #[derive(Debug, PartialEq, Eq, Clone, Copy)]
 enum EntryType {