Threaded type read from a file was flowed through its frames before
the document's composite fonts were registered, so a story set in one
was measured in the fallback font. Frames then held the wrong share of
the story after opening, until a frame was edited. load_with now lends
the document's composite fonts while it lays out its threads.
Object > Rasterize and the redrawn path of Crop Image allowed images up to
65,535 pixels a side, but the renderer draws at most 32,768. A 100 x 5569 pt
rectangle rasterized at 600 ppi came back 32,768 pixels tall, and the bottom
29% of the art was gone. Such an image is now refused like one over the
64 megapixel limit, and the art stays as it was.
The tool-edit sweep sent `{"<key>": "on"}` for every documented key, so
`prefs.export` wrote a preferences file named `on` into crates/engine on
every run, and `folder`, `destinationFolder` and `diffPath` strings from
`junk_params` named locations under the working directory.
- `is_path_key` covers `path`, `diffPath`, `folder` and
`destinationFolder`; `safe_value` sends a temp file for a file key and
a temp folder for a folder key, and both sweeps use it.
- A new test builds both sweeps' cases for every command and fails when
a string under one of those keys is outside the temp folder, without
running any command.
effect.deferredStatus only lists the document's deferred stages, so it joins reviewed::ALLOWED. effect.refreshDeferred runs plug-in effects again, so it joins reviewed::DENIED, which fixes every_command_is_reviewed.
The flattener limited an image's resolution only by its total pixels, while
the renderer draws at most 32,768 pixels a side. A long area, such as a narrow
see-through gradient strip down a 2 m banner at 1200 ppi, came back cut off
at that length and was placed as if whole, so the rest of the strip was lost
in Flatten Transparency and in PDF/X-1a, PDF/X-3 and PDF 1.3 exports. The
resolution is now also limited by the longest side, as raster effects are.
A merge left a second copy of the G9 `.ai`/EPS Missing bullet's opening inside the bullet. That copy says
writing the editing data is out of scope, while the same bullet and file-format-parity.md say a
PDF-compatible `.ai` carries an editing copy of the art. The bullet now has one copy, with the newer
brush wording that matches the 2026-11-05 entry above it.
The graph still stores an opacity or blend mode set on a series mark's row
that has no paint, as it does for transparency.set. The sentence now says
that such a row leaves only the stored paint unchanged.
When text.setGlyphAlternate leaves the text unchanged, it now reads
the glyph from the unchanged text. A character after the first in a run
that holds a glyph choice could be reported with its own glyph instead
of the ligature drawn for it, which text.glyphAlternates reports.
Pasted art whose svg.filter effect named one of the document's SVG filters
but drew another definition kept that name, so editing or deleting the
document's filter rewrote the pasted art's filter. Paste now renames such
a filter to a free name ("Soft 2") when the art comes from another
document or from SVG, and takes the document's filter when the art is
pasted back into the document it was copied from. Symbols, patterns,
brushes and graphic styles on the clipboard follow the same rule.
Category labels beside an axis and labels in pie wedges are centered
on their place by the default leading. A multi-line label that keeps
more leading or paragraph spacing ran down from that place instead.
The graph now puts the middle of such a label's first and last
baselines on its place.
#6090 moved the pinned window width in
the_control_bar_shows_the_paragraph_alignment_of_the_selected_type from
1440 to 1536 pt, since the bar's right end now keeps room for the panel
menu button. The progress line still named 1440 pt.
Taha-Mahmoodi