Type on a path sits at fractions of its path's length. Scaling the path
non-uniformly changes where a fraction lands (CodeRabbit's case: M0 0 L100 0
L100 100 under diag(2, 1) moves the start bracket from (200, 0) to (150, 0)).
Path type now bakes the vertical scale into the character sizes and scales
its path uniformly by it, which keeps every fraction where it was; the
horizontal stretch stays in the transform.
Baking an object's scale into the character attributes removed that scale
from the type's transform, but the area frame (and a path type's path), which
is drawn through that transform too, kept its unscaled size. A 120 x 80 frame
scaled 200% shrank back to 120 x 80 on the page, so the 24 pt type broke into
two lines and the rest became overset.
The frame and the path now take the scale the transform loses, so they, the
line breaks and every glyph stay where they were.
Fixes#8297
With Transparency off, Optimize to File Size's Auto Select counts the
colors of the pixels blended over the matte, which are the colors the
file is written with, so art whose blended pixels need more than 256
colors starts with JPEG instead of GIF.
Rich, RGB, Lab, spot-linked and lighter blacks are not marked to
overprint, and paints already set to overprint keep that setting. The
docs, params text and progress line said these blacks knock out.
docs/mcp.md also says that a symbol instance print leaves unplaced, past
16,000,000 objects of printed art or 128 drawn levels, is judged by its
symbol's own art.
Lossy WebP is now written without a note, so the WebP case of
export_selection_reports_the_encoders_warnings asks for interlaced,
which WebP files don't read, and checks that the selection export and
the whole-document export both name it.
Select > Same, Select by Color (select.color) and document.find looked up
the parent of every object they checked by its id, through
Document::parent_of and Document::node, to tell members of a compound
path and faces of a Live Paint group from other objects. The walk
already stands on that parent, so it now hands it to the check with the
object. walk_selectable keeps its signature and visits the same objects
in the same order.
The selected objects, their order and the count reply are the same as
before, and so is the Reselect state of Select > Same. Tests compare the
walk and the parents it hands out with the old walk and
Document::parent_of, and the results of six Same commands and
select.color with the old per-id loop, in four isolation scopes, from
rectangles at the top level, in a group and in a sublayer, compound path
members, Live Paint faces and a compound path. In a damaged document
that holds an id twice, the parent read is the one that holds the
object being checked.
The gain shows on large documents. With 220,482 objects, the search of
Select > Same > Fill Color takes a median of 24 ms of CPU time instead
of 83 ms, and of Stroke Color 22 ms instead of 60 ms (release build on
Linux, seven alternating runs of both loops in one process; an ignored
test, select_same_perf, times them).