Type that the editing copy writes as its outlines (mirrored type, type in
a clip or symbol, type the text document can't hold) put its Text Wrap
Options on the group of glyph outlines. The reader then wrapped area type
around the glyph shapes instead of the type's bounds, so the wrap region
changed when the file was opened.
The group now ends with an unpainted closed rectangle at the type's
geometric bounds, with the type's lock and hidden state, that carries the
wrap options; the group keeps the type's name and has no options of its
own. The `.ai` reader already reads an unpainted path with its own
options (`text_wrap` in crates/eps/src/import/ai/read/document.rs), and
wrap resolution descends into a group without options. The rectangle
counts toward the object limit; without room for it the options are left
out with the limit's warning. Type without finite bounds gets no
rectangle, as it had no wrap region before saving.
A clip path is left out only when it is one rectangle around the whole
picture. Two open figures along the top and bottom of a box enclose nothing,
and the objects under such a clip are no longer imported unclipped.
The Vietnamese row's message column said "all" while the same cell counts messages without a Vietnamese row. It now says dialogs and tooltips are complete and messages are partial, as the Indonesian row does. The row and the evidence line in docs/target-app-parity.md give the catalog's current counts: 1,750 @msg and 49 @msgplural rows, with 1,875 messages in the untranslated report.
Reset Panel records no undo step when it changes nothing, so the character and paragraph tests that reset a second document first give its type a value away from the panel's defaults.
The Indonesian and Bosnian rows now give the catalog counts and the
untranslated-report gaps measured on main at 676f0524cc: Indonesian has
6,825 rows (1,036 @msg) and lacks 60 menu strings and 2,508 scanned
messages; Bosnian has 6,887 rows (1,144 @msg) and lacks 97 menu strings
and 2,416 scanned messages.
Strokes drawn one after another with the same pen join into one path only
when the pen is opaque. A translucent stroke keeps its own opacity, so where
two of them overlap both show.
Items filled with the same pattern brush share its tile, and the tile's
average color is now worked out once per tile instead of once per item.
Finding the blend pixels and settling them onto a side took about 85% of
color quantize on a 16 MP raster, all of it on one thread. The k-means
palette step works on at most 32,768 histogram bins and took under 1%.
- own_colours_among and settled work through bands of about 65,536
pixels with layers::parallel_map (one thread on wasm32) and join the
results in row order. settled reads every pixel's neighbors from the
labels as they were before any change, as before, without copying
them.
- settled_wide settles groups of its 7-column strips apart; each strip
already started from empty buffers.
- A picture with no pixels returns before any band or strip is made.
The output does not change.