Shortcuts that give the same answers as the lookups they skip:
- is_rtl: nothing before U+0590 is right to left, so Latin, Greek and
Cyrillic text skips the bidi class search.
- strong_rtl and is_mark: ASCII letters are the only strong ASCII
characters, and no ASCII character is a mark.
- cluster_starts: in ASCII text every character starts a grapheme except LF
after CR, and no ASCII character joins a cluster.
- display lines: a glyph run's vectors and text are sized up front, and the
symbol-font check is made once per run instead of once per cluster.
- PDF glyph runs: the baseline (the y most glyphs share) is only worked
out for italic runs, the only ones that use it, and a run on one baseline
returns it without counting.
Tests check each shortcut against the full lookup (every code point for the
bidi classes, every ASCII pair for the clusters).
PDF export of the export_bench document, interleaved with main: median
293 ms -> 241 ms (-18 % with the previous commit). The PDF bytes are
identical.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A paragraph with keep with next measures the paragraph after it before it is
placed (next_first_line). That measurement called layout_para directly, so
every paragraph following a heading was laid out twice in each layout pass:
once to measure it and once, through the cache, when it was placed.
The measurement now goes through the layout cache under the key
para_labelled uses (Ctx::para_ahead) when its inputs are the same: no
proofing, no typed word, not inside a text box, no inline text boxes that
resize, the body's tracked-deletion view. Otherwise it is laid out on its
own as before. It counts nothing (list labels, equation numbers and the
page-total flag are restored), so pagination is unchanged.
Adds crates/engine/examples/export_bench.rs, which saves a long generated
document (the sample document 200 times, 259 pages) through io::save_bytes
and prints wall times:
cargo run --release -p wordcraft-engine --example export_bench -- pdf 200 9
PDF export on that document, interleaved with main (Ryzen, 16 threads):
median 286 ms -> 256 ms (-10.6 %). The PDF bytes are identical.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dpkruse's Genuine Steps Certificate: a bulleted text paragraph closes a
two-column continuous section, then an empty bulleted one closes another.
Word draws no bullet for the empty one; it is now quiet (takes no room) and
draws no label, still counted. A stacked mark that doesn't fit keeps its
line for the caret but draws nothing below the margin (CodeRabbit).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CodeRabbit (local): keep-lines compared its height with the room below the
mid-page section start, so a paragraph that fits on a new page split instead.
It now compares with the next column, or the next page's body when this is the
last column. A stacked mark whose page break before starts a page no longer
gets the previous page's bottom back.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CX1hMKX28764Lrx5moxs1o
A continuous section moves the column top down below the text above. at_top()
then held at every new section start, so a paragraph that did not fit was
placed anyway, as on an empty page. A run of one-line continuous sections ran
off the bottom of page 1 (100 of them reached 2,480 pt down a 792 pt page).
The overflow checks for paragraphs now use fresh(): at the top with nothing
above on this page. at_top() still drives the spacing rules, unchanged.
Empty paragraphs that close a continuous section right after another
section's mark took no room. Measured in Word on synthetic files ("Start", N
such marks, "End"): 10 put "End" about 9 cm down page 1; 50, 100 and 200 all
put it on the first line of page 2, the same with 24 pt marks or no paragraph
spacing. So each takes its line while it fits, then stays at the page's
bottom. A mark right after text stays quiet, as measured before.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CX1hMKX28764Lrx5moxs1o
Noto Sans Thai and its siblings have no Latin letters or digits, so on a
machine whose only Thai fonts are Noto no face covers the test run and
fallback_for_run rightly returns None. The guard now asks whether the Thai
substitute covers the whole run, which doesn't go through fallback_for_run.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CX1hMKX28764Lrx5moxs1o
- proof: pass PoisonError::into_inner instead of an identity closure (S1612).
- layout tests: replace glob super::* imports with explicit names (S2208).
- packaging/macos/import-cert.sh: use [[ for the status test (S7688).
Alphas keys decoded pictures by their bytes' address and length, which is
sound only while the document is borrowed for the layout pass; that is now
said where the key is. LayoutCache's outlines outlive the documents they were
made for and use the same key, so each entry now keeps a Weak reference to
the bytes: their allocation can't be freed and the address reused by another
picture of the same length while the entry is kept. MAX_FLOW_WIDTHS spells
out which layout a box past the 16-width cap reuses.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CX1hMKX28764Lrx5moxs1o
Stacked text is laid out at STACK_WIDTH whatever the frame, so its flow key now is too.
The nearest-layout fallback only picks layouts of the box's TextVert; the first box in a
direction is laid out (at most a few more layouts).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CX1hMKX28764Lrx5moxs1o
Crops alternating between pictures decoded each again and again, spending the budget on
pictures that fit in it. Every picture charged to the budget keeps its alpha (bounded by
the same 2^28-pixel budget).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CX1hMKX28764Lrx5moxs1o
Ultrareview finding 6: boxes past a chain's 16th distinct width (compared
by exact float equality) showed no text, which resized boxes hit easily.
Widths are now told apart to a twip, and past the cap a box takes the
nearest layout already made (same direction, widest that fits, else the
narrowest). Work stays bounded at 16 story layouts.
Ultrareview finding 7 (also raised by CodeRabbit): opaque non-JPEG
pictures were decoded and charged to the per-layout pixel budget though
they give no outline, and a budget refusal was cached in LayoutCache as a
permanent "no outline". The header (read via into_decoder, PNG tRNS
included) now says whether a picture can be transparent before anything is
charged; Alphas::outline reports OverBudget distinctly from Ok(None), and
Ctx::wrap_outline (via picture_outline) doesn't keep a refusal in
LayoutCache::outlines. Within one layout a refused picture is remembered
so its header isn't re-read.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
out any crop's outline decoded the whole shared picture again, so many
small picture references with distinct crops multiplied full-image
decodes without repeating the encoded image. Layout now keeps the last
picture's transparency (one alpha byte per pixel) and reuses it for
every crop of it, and charges each decode's pixels against
MAX_DECODED_PIXELS (2^28) per layout; past it pictures wrap as their
rectangles. Tests: tests_budget::a_picture_cropped_many_ways_is_decoded_once,
tests_budget::decoding_pictures_stops_at_the_layouts_budget.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Audit finding 11: place_table cloned every item of the header rows onto
each page a table spills onto, but its cap (200 000) counted rows. An
exact-height header row keeps all its overflowing paragraphs without
growing, so one header row of thousands of items over hundreds of
pages placed millions of copies. The cap now counts placed items
(MAX_HEADER_ITEMS = 500 000, a row costing at least one) and the
header set repeats whole or not at all; past it later pages carry no
header rows and pagination still finishes. Test:
tests_budget::repeated_header_items_are_capped_across_the_table.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Audit finding 7: flow() laid the head box's whole story out again at
every distinct box line length in the chain and kept each layout while
splitting, so a chain of many one-line boxes of different widths
multiplied the story's layout work and memory by the chain length.
Past MAX_FLOW_WIDTHS (16) distinct lengths the text stops: the box that
would need another layout and the rest of the chain show nothing, as
overflow past the last box is hidden. Work is now at most 16x the
story. Test: tests_budget::a_linked_chain_lays_its_story_out_at_a_bounded_number_of_widths.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Audit finding 35: Table::cols() added a deserialized row's grid_after
(u32) unchecked, and layout_table sized its column vectors from it, so a
small native JSON document could ask for a 16 GB grid (and overflow the
sum on 32-bit wasm). Row::grid_after() now clamps to MAX_COLS as
first_grid() does and the importer does; cols() sums with saturating
adds; layout reads the bounded value. Test:
tests_budget::a_huge_grid_after_is_bounded_like_the_docx_readers.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The layout's linear-time tests (#3866) asserted CPU-time ratios, which a busy
CI machine can skew. Each walk that once ran again for every item it served
(a line's clusters, a page's items, the revision authors, the sections, a
table's rows, a run's glyphs, the clipped cells) now counts the elements it
looks at in a test-only counter (crate::work, nothing outside tests), and the
tests assert 8x the input takes at most 12x the steps (quadratic: ~64x).
proof_text, whose walk is the document model's, stays timed, and a ratio
over the limit is measured twice more before it fails.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CX1hMKX28764Lrx5moxs1o
Finding 3: the table's rows carry no cells, only a column with a width skipped before and after,
so each row adds two spanning entries for next to no layout and counting the grid per entry
dominates (old code 75x for 8x the rows, fixed code 4-8x).
Finding 13: sections that link to the previous section's headers and footers are laid out
against the same sections naming them (title page on, so first-page parts are looked up too);
old code 6.1x the time, fixed code 1.0x, limit 2x.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 30: author_index rebuilt the list of revision authors for every tracked run and paragraph mark drawn, and balloons rebuilt the comment-author list for every comment (quadratic in revisions or comments). display::author_order ranks authors by first appearance once; page_display, page_layers and object_display build the revision order lazily once per call and pass it down, and balloons builds the comment order once per layout. Same colours.
Tests: drawing_tracked_changes_by_many_authors_takes_linear_time, tracked_changes_colour_by_author_in_order_of_appearance, balloons_for_many_authors_lay_out_in_linear_time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 1: cell_clip searched every clipped cell on the page for each line drawn (quadratic in a page of many exact-height cells). clipped_cells now indexes them by story and content path, keeping each one's order; a line looks up only its path's prefixes and takes the earliest, the same cell the scan found first.
Test: drawing_many_clipped_cells_takes_linear_time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 28: column_segments and selection_rects called pieces() per paragraph, and for a header, footer or note each call scanned the whole page (pieces_on). PartPages indexes a page's lines of the story by paragraph the first time the page is looked at; the page order (hint first, continued notes) is unchanged.
Test: selecting_a_long_header_takes_linear_time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 13: linked() and defining_width() walked back through the sections before each section, per section and again per page (layout_pass, headers_footers, footer_table_wraps, header_table_areas). HfLinks resolves every section's linked header and footer sets incrementally once (each inherits the previous one's for the parts it leaves out) and keeps, per part id, the sections naming it, so the defining section is a binary search. Same sets and widths as before.
Test: headers_inherited_through_many_sections_lay_out_in_linear_time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 3: column_content called Table::cols (a walk of every row) for each spanning cell, and there can be one in every row. Hoisted out of the loop; the end is also added with saturating_add.
Test: autofit_tables_with_a_spanning_cell_in_every_row_lay_out_in_linear_time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 23: Ctx::glyphs found each character's first glyph by scanning the shaped glyphs from the start (quadratic in a long run). The first glyph at or past a boundary never moves back as the boundary grows, so one walk gives the same cuts.
Test: math::tests::a_long_text_run_lays_out_in_linear_time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 5: off_at_x and visual_step on a bidirectional line called x_of for every caret offset, each searching the clusters from the line's start (quadratic in a line of many zero-width offsets, e.g. hidden text). ParaLayout::line_carets walks them with a cursor; the first matching cluster never moves back for non-decreasing offsets, so the x's are the same.
Tests: bidi_clicks_and_caret_steps_take_linear_time, line_carets_match_x_of.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Finding 29: proof_text looked up each character's run with props_of_char, walking the runs from the first for every character (quadratic in a paragraph of many runs). A run cursor advancing with the text gives the same props in one pass.
Adds layout::tests_linear (sharing tests.rs's cost_ratio helpers) with proof_text_walks_the_runs_once.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Dylan M. Taylor