Clear, path.deleteAnchors, Cut, Copy and Cut Path scanned every selected
anchor or segment of a path once for each of its subpaths, so a path with
thousands of subpaths took time that grew with the square of their count.
The selection sets are ordered by subpath, so a new helper, subpath_refs,
reads one subpath's entries with a range lookup instead. The results are
unchanged.
With one selected interior anchor on each of 8,000 open subpaths of one
path, Clear took a median of 74 ms of thread CPU time before and 3 ms
after (release build, five runs each; the timings are noisy because the
CPU was shared and loaded).
Tests in delete_path_pieces.rs check exact surviving subpaths and their
order for anchor and segment selections across open and closed subpaths,
an untouched subpath, references to missing subpaths and indexes, one
Undo/Redo step and a second Clear, and Cut Path on several subpaths.
Without a plug-in store, plugin.install read the process grant once
for its later check and again for its approval. A grant another session
recorded between the two reads approved the module; when that session
revoked it before the install, the check saw the grant it first read
and the install recorded the revoked grant again. The approval now comes
from the first read.
plugin.install published the plug-in before it wrote the module and its
grant to the Plug-ins folder. A plugin.remove in another session in
between removed the plug-in, its file and its grant, and the install then
wrote both back, so the plug-in loaded with its grant at the next launch.
A session attaching the folder in between loaded the old version, which
kept running while the folder held the new one. plugin.remove, loads of
the stored plug-ins and loads of a plug-ins folder now wait until an
install has written the folder, and an install waits for them.
Docs: docs/plugins.md, docs/target-app-parity.md.
Assigning to a named child counts its list of the matching children, and
assigning through an empty XMLList counts the chain of targets it climbs,
before either changes the tree. copy(), toXMLString(), normalize(), the
`..` selector and the depth check on insertion hold the stack they keep
for each open level of the tree. Inserting an XMLList, and assigning one
to an item of an XMLList, hold their copy of the list's items.
copy() reads attributes in place, and toXMLString() reads attributes and
namespace declarations in place, instead of copying the lists. copy()
charges the declarations it copies before it copies them. Deleting by
name or assigning an attribute compacts the remaining links in place
instead of building a second list.
toXMLString() holds the set of prefixes a namespace lookup passes, and
picks a new nsN prefix in one pass over the declarations in scope
instead of rescanning them for each candidate.
The parser holds its open elements and their names, the text it gathers,
each tag's attribute values, the sets that check declarations and
attribute names, and its one copy of an element's namespaces, and it
charges the strings of the namespace declarations it makes. The
allowance it holds from the source's length covers the copy a value
takes as it becomes a string.
Comparing XML values, converting them to strings, assigning an XMLList
to an attribute, and reading a name, a qualifier, a namespace URI or
embedded text from a string built by joining strings flatten it with
flat_charged. An attribute name read from a string reserves its copy,
and an element name shares the string.
Docs: docs/target-app-parity.md.
Spine::at_length walked the first subpath's segments from the start
until one ended at or past the arc length, so placing every blend step,
spread key, blend widget and Objects on Path attachment cost one
comparison per earlier segment. It now searches the anchors' arc
lengths, which never decrease, and picks the same segment: the first
whose end is at or past the length, so an exact anchor length and
repeated lengths from zero-length segments keep their segment. The
point and tangent evaluation is unchanged.
Tests compare the search with the old walk on open and closed spines,
one segment, curves, zero-length segments, every anchor, both ends,
out-of-range and non-finite lengths, and check that a 4,096-segment
spine needs at most 13 comparisons.
193 Hindi rows outside the plug-in messages said something other than
their English: "a number of points" read as "many digits", swatches as
clocks, crop as a harvest, resolution as a resolve, and saving used
words the glossary replaces. Others left English prose in place,
dropped a clause ("to keep them all", "what follows it") or ran the
sentence backwards.
- Rows use सेव, स्वॉच, क्रॉप, रिज़ॉल्यूशन, पाथ and टाइप (for a text
object) as docs/localization-hi.md sets them.
- Values a command reads (enum values, preset names, OpenType tags)
stay as the English writes them.
- A test checks those terms and the point unit in every Hindi row, and
another pins rows whose quantity or clause was lost.
- docs/localization-hi.md, docs/localization-parity.md and
docs/target-app-parity.md carry the new counts and a wording note.
The previous commit took the spans from how far each piece bends away
from the split whenever both handles there were within ONE_CURVE of the
anchor. A curve that barely moves at the split but doesn't bend there
(a straight one, say) has short handles and no bend, so that gave 0/0
and the side fell back to the polygon shares. Now the spans come from
whichever pair is the longer: the handles, or the bends (which go with
the squares of the spans). The check of every piece against the
rebuilt curve is unchanged.
Test: a_split_curve_keeps_the_top_object_patch_where_it_barely_moves_and_runs_straight.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
$.fileName gave the file being run, so a function defined in a file
that #include inserted, in a file $.evalFile ran, or in an earlier run
of a named engine reported the calling file instead of its own.
- vectorcraft_js: HostItem::SiteGetter gives a getter the CallSite
(source tag, line, column) of the code that reads it.
- $.fileName maps that place through the run's source maps, or the
maps a named engine kept, to the file the code is in. It is empty in
code from a script given as source and in kept code whose map was
dropped. Code that eval or Function compiled still gets the file
being run.
Docs: docs/mcp.md, docs/target-app-parity.md.
Save Actions… writes version 6 when a View menu step sits inside a
`command.batch` step, since playback runs that step on its own, and Load
Actions… refuses such a nested step whose params are not an object.
affinity-corpus.yml ran a local job; it now calls craft-common-actions
gate.yml with the same toolchain (1.95), environment and commands, and a
cache of its own. release.yml's git-sources copy is removed: the shared
release.yml runs the same check before any build.
The sweep filtered by transparency kept every open box in one tree, so
a box without transparency met every open box without transparency it
overlapped, though such pairs never join. Both sweeps closed an open
box early only when a later box held it, so in a nest of boxes that
share their left and top edges and get smaller to the right and
bottom, each box met every earlier box around it. One see-through
object inside 512 nested opaque objects took 130,817 meets in the
unfiltered sweep and 131,328 in the filtered one.
The filtered sweep now keeps open see-through boxes and open opaque
boxes in separate trees. An opaque box looks only in the see-through
tree; a see-through box looks in both.
In both sweeps, a box inside an open box it met, ending no further
right, is not opened when that box joins everything it overlaps (any
box without the filter, a see-through box with it). Whatever later
overlaps the inner box overlaps the outer one.
The nest above now takes 512 meets in each sweep, and a nest of 512
boxes alone takes 511 in the unfiltered sweep where it took 130,816.
The groups are unchanged.
font_menu::tests::the_scroll_bar_leaves_the_stars_clear starred every
family in the process-wide font database and listed that database, which
tests running alongside add fonts and composite fonts to. In full
parallel runs it sometimes failed with "8 families, no row stars drawn".
With the four bundled families alone it never reached the scroll bar
check, which ran only for seven or more families.
- The menu's header takes the font database it lists and hands it to
the rows and the name-width measurement. The app passes the
process-wide database as before.
- Each FontDb has an id of its own (FontDb::id), and the widest family
name the menu keeps is stored with that id as well as the generation,
so a width measured for one database is not used for another at the
same generation.
- The test lists a database of the bundled fonts, stars its four
families, opens on Inter, and runs the frames the scroll bar takes to
appear. It checks that the rows overflow the list's scroll area, that
a row star is drawn inside that area, and that each star stays left
of the scroll bar's column (#2277).
With the list's scroll bar set to floating, the test fails with a star
under the bar. Without the id in the widest name's key,
each_font_database_has_its_own_widest_name fails.
The Brush libraries row named four libraries. It now names the nine in
BUILTIN order (crates/engine/src/cmd/brushlib.rs), as the Window ›
Brush Libraries menu shows them, and says nine built-in libraries.
A double-click on a paragraph style's row and the panel menu both open
Paragraph Style Options (ui.paraStyleOptions, paraStyle.edit) with its
seven sections. The type-parity and ui-parity rows still called it the
rename dialog and named ui.paramDialog with paraStyle.rename; they now
name the dialog, its sections and its tests.
The target-app-parity entry said each section has a Reset Panel. Reset
Panel is unavailable in General, and the entry now states that.
An image placed several times in an EPS file (the pieces of an envelope
or a Warp effect, copies of one image, a symbol's instances) wrote its
whole ASCII85 data at every placement: 200 copies for one envelope at
the default fidelity, enough for a large photo to pass the 256 MB page
limit and lose pieces.
The samples of such an image are now written once in the setup as
VCim<i>, and each placement reads them from there:
- Level 3: a ReusableStreamDecode stream, rewound with resetfile and
read through a new FlateDecode filter at each placement.
- Level 2: an array of strings of at most 49,152 compressed bytes, fed
to RunLengthDecode by a procedure.
An image placed once is written where it is drawn, byte for byte as
before. The page limit counts a repeated image's samples once, and a
colorized image is reused per ink.
- With art, the bar is 10 pt tall and the Edit tab leaves out the gaps
above the wheel and above the sliders. The dialog is as tall on the
Edit tab as on the Assign tab, and all of it but the color chips fits
a 1024 by 576 screen.
- recolor.weigh counts a use for the first row that is not excluded
when an excluded row lists the same color first.
A paragraph format given to the empty paragraph after a frame's last
paragraph break was lost when the story flowed again: the paragraph is
shared with the next frame's first paragraph, and the next frame's old
style replaced it, in that frame and in the frames after it.
- The thread reflow also notes which frames' last paragraph changed
style, and story_paras_restyled keeps that style against the next
frame's unedited first paragraph, across unedited empty frames too.
When the same edit formats the next frame's first paragraph, that
format applies, as before.
- story_paras keeps its behavior for its other callers.
- Tests cover area type and type on a path on either side, a next frame
holding text, three frames, undo and redo, a format that changes
nothing, and numbered list counts.
plugin.install wrote the module and its grant to the Plug-ins folder,
attached the plug-in's settings and cache and turned a disabled plug-in
on before the registry's last approval check. When another session
revoked the grant, or recorded one for other bytes, after the command
recorded its own grant and before that check, the install was refused
but left the new module and grant in the folder for the next launch.
A plug-in another session installed under the same id between the
command's checks and the install took no part in the decision, so the
install replaced it without `replace: true`.
- registry::install_checked records the grant (or forgets it) and
installs the plug-in as one step, after a caller check that sees the
plug-in installed under the id and its grant; a refusal changes
nothing.
- plugin.install commits through it before any store write, refusing
with the existing messages when another plug-in now needs `replace`
or the grant its approval came from changed. The store writes, the
put-back on a failed write and the disabled-list update now run after
the commit; their logic is unchanged.
- The grant call-site audit also counts registry::install_checked.
Docs: docs/plugins.md, docs/target-app-parity.md.
Insert Menu Item… now lists the Object › Align items and the View menu
commands the app runs itself (Outline, Pixel Preview, Trim View,
Presentation Mode, Screen Mode, zoom, fit, Rotate View, and the show,
hide and snap items).
- An Align step holds the params its click sends at insertion, so a
later Align To choice does not change it. A step that aligns to the
key object stops the playback when the selection has no key.
- A View menu step plays in its place at every speed and adds no undo
step. At Accelerated the steps on each side of it are one undo step
each. Playing it while recording records it.
- Batch… refuses an action with an included View menu step.
- An action set file with a View menu step is version 6.
- The steps before a stop now play as saved steps, as other playback
does, so a key that an older version recorded no longer fails there.
Docs: docs/mcp.md, docs/target-app-parity.md.
- recolor.weigh drops another row's entry for a color that moved, so
recolor.apply recolors it with the row it moved to and the shares
describe the recolor. A color that the first color of another row
with a new color also lists stays where it is.
- Linked swatch colors are grouped through a map, so a row with many
unlinked colors no longer compares each one with every group.
- The Prominent Colors bar reads the shares again when a row gains or
loses a new color, since that changes which row takes a shared color.
BFlat