Since #8549 the glyphs of lines finished past MAX_PARKED were kept
until the next stroke and, past 4096 of them, matched every stroke.
A `2 Tr` word after a dense page of filled-only type then came in as a
filled object and a stroked one.
- flush_text forgets them: nothing is drawn between an operator's fill
pass and its stroke pass. end_text keeps them, since it also runs
mid fill pass for a glyph kept as an outline.
- LostFills keeps the first 4096 glyphs and matches only those; there
is no overflow that matches every stroke.
- replayed counts the lines its `left` check looks at toward
MAX_REPLAY_WORK.
* SVG export: preserve document unit in viewBox and width/height (#864)
When exporting SVG documents with physical units (mm, cm, in, pc), emit
the physical unit suffix on the root <svg width=... height=...>
attributes while keeping viewBox numbers and inner coordinates in 1:1 points
per SVG 1.1 coords §7.10. This avoids the 75% scaling mismatch when opening
drawings in external tools such as Inkscape or LightBurn, and preserves
the document unit setting across save/open round trips. Unitless documents
(points, pixels) continue to emit unitless dimensions.
Tested via export_physical_units_writes_unit_suffixes in vectorcraft-svg
and full round-trip svg_round_trip_preserves_document_units in vectorcraft-engine.
Ran cargo xtask ci (fmt, clippy, tests, xtask gates).
* SVG export: preserve nonzero width and viewBox on extreme small dimensions
The PDF writer dates each file and makes its IDs from the clock, so two
sheet exports on either side of a second boundary differed there and the
test failed on main's coverage run.
Each <textPath> with an href walked the whole document to find the element it names, so a file with many texts on paths took time proportional to the number of texts times the number of elements. The importer now builds a map of element ids at the first such lookup, keeping the first element in document order for each id, and looks each href up there. Files whose texts name no element do not build the map. The imported document is the same: a test compares the whole document and warnings with the per-id walk, including duplicate, missing and unusable ids.
Clicking between app windows (floating panels and the canvas) passes through one frame where none of them has focus. The palette levels counted the app inactive on that frame and lowered every floating panel's OS window. The panels flickered, could end up disappearing behind the main window until the next repaint, and a click on one panel's bar could hand the keyboard to another.
The app now counts as inactive once none of its windows has had focus for 0.25 s, counted from the first frame that saw none. craft-ui's docking waits as long.
Fixes#6986
- The French script error line now has the catalog's no-break spaces before its colons.
- es.tsv drops the old file.info control character row a merge brought back.
Two PRs merged tonight each added the same row, so bundled_catalogs_are_consistent fails on main
(gu: duplicate "" "Panel"). Both rows have the same translation.