A filter plug-in gets 50 M instructions and a minute for each run.
Filters that order a plotter's paths, pack shapes or run a simulation
need minutes, and the app shouldn't freeze while they compute.
A filter's manifest may now ask for "budget": "heavy" (Budget): each run
gets HEAVY_FUEL (1 G instructions) and HEAVY_WALL_TIME_MS (300 s). Only a
filter may ask. plugin.install refuses such a filter without allowHeavy:
true, and plugin.inspect returns the entry a module would have without
installing it. Install Plug-in… and File › Open ask the user and install
on OK; the plug-in store keeps the answer, and folder loads skip heavy
filters with a note, since they ask nobody.
The app runs a heavy filter on a thread of its own (plugin_runs), not in
Background Save's queue, with the status bar and a progress window
(Cancel, Continue in Background) that it shares with Background Export
(export_progress::window). FilterRun splits plugin.run: prepared on the
session, run on any thread, its output handed in, and the next
plugin.run with the same params applies that output instead of running
the plug-in again, so the edit stays one guarded, journaled command. A
result whose document or plug-in changed meanwhile is dropped, and the
status bar says to run it again. A cancelled plug-in run is now
EngineError::Cancelled.
What a filter keeps with the document (#3720) goes with the run: it is
given _context.documentData when the run is prepared, and the
documentData of its output is applied with its objects, in the same undo
step, when the run is handed in (FilterRun::finish).
Tests: plugins/tests/abi.rs (100 M loop steps out of a light budget,
within a heavy one; folders skip heavy filters), manifest.rs, engine
tests_plugins.rs (consent, inspect, the store, hand-in, a handed-in
run's document data, cancel) and
ui-egui tests_plugin_runs.rs (installing asks, no Preview, a background
run as one undo step, a stale result kept out, the progress window).
CRAFT_UI_HEAVY_FILTER_WASM=<folder> writes a test filter to try in the
app. Its strings and messages have rows in each complete catalog
(COMPLETE_MESSAGES).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HhvnnnD5fTFupK8WWq35un
The Shape Builder lines of docs/target-app-parity.md repeated after
recent merges: the Tool Options line three times, one copy still saying
that Gap Detection and the stroke options are not built, and the stroke
split and Pick Color From Artwork lines twice, one artwork copy quoting
the reference app's combine page. One copy of each is left, in the
order Tool Options, stroke split, Pick Color From Artwork, marquee,
without the quotation.
The Pick Color From Artwork line in docs/target-app-parity.md quoted a
sentence from the reference app's Shape Builder combine page. The line
now links the page and describes the rule without quoting it.
Hyphenation... and Justification... already shipped upstream (#2310,
#2755). This ports the part of the archived fork PR #10 that upstream
lacked: a paragraph's Justification > Auto Leading now also sets
- the Leading the Character (and Properties) panel shows for Auto type,
blank where the selected paragraphs differ;
- the leading Alt+Up/Down (type.step) steps from, per paragraph (a run
spanning paragraphs is split so each takes its own);
- the lines Fill with Placeholder Text counts in area type.
ParaStyle::effective_leading mirrors the layout (exactly
CharStyle::effective_leading at the default 120%). Tests in doc, engine
and ui-egui; parity docs no longer list Hyphenation/Justification as
missing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014gCpjL3yUhaH3jRp7ksufh
Pattern tiling and scatter cap each factor on its own: 20,000 tiles a
run, MAX_SUB subdivisions a segment, 20,000 copies a track. Those caps
are per track, and a path has as many tracks as it has subpaths, so
they multiply rather than bound.
The audit called the end-to-end impact unverified, so it was measured.
A pattern brush on parallel lines, each long enough to want more tiles
than a run allows:
1 subpath 20,000 nodes 320,000 anchors 36 ms
4 subpaths 80,000 nodes 1,280,000 anchors 142 ms
16 subpaths 320,000 nodes 5,120,000 anchors 571 ms
Exactly 20,000 nodes a subpath, linear with no ceiling. Scatter is the
same shape.
MAX_ANCHORS bounds the total for one brushed stroke, threaded through
both. It is 2,000,000, the figure the stacked-effects budget already
uses (effects::distort::MAX_COPY_ANCHORS, #3578), and bounded the same
way: per stroke rather than per document. One track at its own cap is
well inside it, so the budget bites only where the caps were
multiplying; sixteen subpaths now stop at the budget in 113 ms.
A piece is taken whole or not at all, so art is never half-drawn. The
walk that counts an art tile's anchors is iterative, and a tree it
cannot finish within MAX_WALK nodes is charged the whole budget rather
than undercounted, so nesting buys no room.
The per-factor caps are untouched: the issue asks for a total, and
changing both at once would hide which one did the bounding.
A Shape Builder drag merged only the regions the pointer passed over,
and the tool did not read Shift. The reference app's Shape Builder page
describes a Shift-drag that draws a rectangular marquee to merge many
paths at once.
With Shift held at the press, the drag now draws a rectangular marquee
from where it started to the pointer. The regions it reaches are shaded
while it is drawn and merge into one shape on release; with Alt also
held at the press, they are deleted. The marquee reaches each region
whose outline crosses it or has an anchor inside it, and the region that
holds the whole marquee. A Shift-click that does not move is a click,
and a drag without Shift works as before. With In Merge Mode, Clicking
Stroke Splits the Path on, a Shift-click on a stroke splits it as a
click does; a marquee splits no stroke, and while it is drawn the
pointer does not show scissors and no stroke is highlighted.
shapeBuilder.merge takes the marquee as rect: [x, y, width, height],
where x and y are where the drag started, and points may then be left
out. For Pick Color From Artwork, the corner where the drag started
counts as the drag's first point, and the opposite corner counts as its
last. With rect, splitStroke does not split a stroke.
Docs: docs/mcp.md, docs/target-app-parity.md.
Cause: with Snap to Point, the point of a moved selection that snaps to a
ruler guide is the grabbed point (#1534). Away from an anchor or center
that was the raw pressed spot (select.rs: `grab` fell back to `p`;
direct.rs: a straight segment's `grab: p`, MoveObject's `start`). A press
on a stroke or within the pick tolerance of an edge lies a few pixels
beside the path, so the pressed spot landed on the guide, "align" showed,
and the path stopped that far beside it: 0.096 pt at 3,891% and 0.0099 pt
at 28,757% in the #3701 screenshots, visible only when zoomed in.
Change: `guides::hold_on_path` finds the nearest point of the path (on a
straight segment exactly on its line). `select::hold_point` uses it for a
press on the stroke or outline, or within 6 px of it. The Selection tool
(`State::Moving::hold`), Direct and Group Selection moves
(`State::MoveObject::hold`) and Direct Selection's straight-segment drag
(`grab`) snap that point. The move delta still runs from the press, so
nothing jumps. A press in the middle of a fill still snaps the pressed
spot, as Snap to Point is documented to.
Verification: tests_prefs::a_move_held_by_a_paths_stroke_lands_the_path_on_a_ruler_guide
and tests_prefs::a_straight_segment_dragged_by_direct_selection_lands_on_a_ruler_guide
fail without the change (edge 0.0257 pt and anchors -0.0052 pt off the
guide) and pass with it (|gap| < 1e-9 pt). A headless egui/wgpu render at
3,891% and 28,757% puts the guide, the selection outline and the centre of
the 1 pt stroke band in the same pixel column. cargo xtask ci: all 10
steps passed; 9429 tests passed, 0 failed.
A filter plug-in had nowhere to keep settings that belong to the whole
document rather than to one object, such as the pens a plotter holds or
a machine profile: it could only ask for them again on every run.
- A filter's output may hold documentData: any JSON of at most 1 MiB
(objects::MAX_DOCUMENT_DATA) that replaces what the plug-in keeps with
the document, null removing it, in the same undo step as its objects
(objects::decode_output). Too much refuses the output and the
document stays as it was.
- A filter gets it back in _context.documentData (null until it keeps
something). A filter reads and writes only its own.
- It is the document's foreign data, Document::unknown["pluginData"] by
plug-in id (vectorcraft_doc::plugin_data), where the perspective grid
and the brushes keep theirs. So the native file needs no section of
its own for it, and hosts from before keep it when they save. An
exporter is given the whole document, so it can read every
plug-in's. Nothing else lists it, as the reference app's Document
Info lists no plug-in data.
A live effect isn't given it: its geometry is worked out from the
object alone wherever it is drawn.
Tests: plugin_data.rs (each plug-in its own entry, forgetting what isn't
kept copies nothing, a damaged value holds nothing), objects/tests.rs
(kept, forgotten, refused over the limit), tests_plugins.rs (one undo
step with the objects, read back, not given to another filter, kept by
the native file among its foreign data, refused over the limit with the
document unchanged, null removes it) and tests/plugin_export.rs (an
exporter finds another plug-in's). The size error has its row in each complete
catalog (COMPLETE_MESSAGES).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HhvnnnD5fTFupK8WWq35un
The planar map found each edge's source input by scanning every input
segment, and the face around each hole or merge point by scanning every
face, so it was quadratic in the segments; the Shape Builder therefore
gave up past 4096 segments (traced or type outlines) and offered whole
shapes. Index outline segments and faces by bounds on both axes
(RectIndex), walk an edge's segments lazily when finding its leaving
angle, and raise SHAPE_BUILDER_MAX_SEGMENTS to 65,536. Divide with open
paths shares the cap; its cap test now follows the constant.
Tests: a 20,000-segment blob cut by two lines, a 300-line hatch, the
index against a brute-force scan, and the existing cap test at the new
cap. Progress docs updated (G11, parity evidence, roadmap).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014gCpjL3yUhaH3jRp7ksufh
With In Merge Mode, Clicking Stroke Splits the Path on, the run a click
split off could run against its path. A click on a rectangle's top side
gave a side from (100, 0) to (0, 0) on a rectangle that runs from
(0, 0) to (100, 0) there, so arrowheads and other settings that follow
a stroke's direction moved to the other end. The run now goes the way
the path goes.
The two paths of a split, and what an Alt-click or a merge leaves of a
path, had a gradient fitted to the path's bounds fitted again to their
own bounds. They now keep it where it was on the path.
A run that left 0.05 pt or less of a path still split it, into the
run and a speck. Such a path now does not split, and the tool shows
no scissors over it.
Docs: docs/mcp.md (Shape Builder Tool Options).
Review follow-up (CodeRabbit): the kept 2026-10-05 assessment still read
"raster effects (~20%)" with no pointer, unlike the other rows of that
table this PR annotated. It now notes the 57/57 filters and the Effect
Gallery (#2883) and points to the area table, leaving the historical
number as it was.
A chain of geometry effects makes at most 2 million anchors of a path (of a
type, symbol or image art's paths together) and about 8 million lines for the
renderer to draw at 1 px/pt, or what the path costs when that is more.
Roughen spaces its points further apart, Zig Zag makes fewer ridges a
segment, Free Distort, Twist and the warps split segments into longer pieces
(map_nonlinear_within) and Transform makes fewer copies where they would pass
it; a result drawn with too many lines is built again with fewer points, and
an effect that still doesn't fit is left out.
Three Roughens of a rectangle in a 25 KB file needed gigabytes of geometry,
and 2 million anchors of them flattened into 141 million lines (7 GB, past
vello's 2^27 lines a path); the file now opens in 0.4 s at about 450 MB.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Illustrator 30.0's Method choice in the Gradient panel
(https://helpx.adobe.com/illustrator/desktop/paint-and-fill/create-and-edit-gradients/control-perceptual-interpolation-in-gradients.html).
- paint.editGradient takes `method: classic|perceptual` (one undo step);
paint.setFill/setStroke take it in `gradient`, and gradient params (swatches,
copied paints) carry it when Perceptual. Anything else is refused without
editing. One edit can set `method` and `dither` together.
- The Gradient panel shows a Method menu (Classic, Perceptual) for linear and
radial gradients, dimmed without a gradient; the Dither box stays under the
stops.
- i18n: "Perceptual" where a complete catalog lacked it, and the `method`
error in all 24 catalogs.
- Docs: G13 (estimate 14–24 h → 13–23 h, after Dither's 15–25 h → 14–24 h),
target-app-parity (row and evidence), ui-parity, mcp.md, plugins.md.
With the Artboard tool, File › Revert kept the active artboard, but a
command.batch with a file.revert step made artboard 1 the active one.
A batch that switched to another document and back, and the Actions
panel menu's Batch… over a folder whose files it closes or over data
sets with a Revert step, did the same. Each of them starts the tool
afresh and ends in the document it started in, and the app moved the
view to the fresh tool's artboard 1. Only a plain file.revert was
handled, by its command id.
The engine now counts the times switching or closing documents, or
replacing the active one, starts the tool afresh (Session::tool_resets),
and records whether the tool's options were set after that
(Session::tool_set_since_reset). When the count changes during a
command and nothing set the tool's options after it, the app starts
the tool on the active artboard, as it does when a command chooses
another document. When a later step chose the tool's artboards (a
Paste or Cut after the revert), the artboard the tool then has active
becomes the active one.
Docs: docs/mcp.md, docs/target-app-parity.md.
With Pick Color From Artwork, a Shape Builder merge took the look of
the front-most object over the first covered region the drag touched,
and the current fill when the drag touched only areas no object
covers. The reference app's combine page gives the look of the object
where the drag starts, else of the object where it ends, else of the
topmost selected object.
shapeBuilder.merge now follows that rule: the merged shape takes the
appearance, opacity and blending mode of the front-most of the objects
it is given (the selection by default) under the first point, else of
the one under the last point, else of the front-most of those objects.
A drag from a hole across a gray bar into a red square now gives a red
shape, where it gave the bar's look. A click in an area no selected
object covers, such as a hole that gray bars enclose or a triangle
that lines enclose, takes the front-most selected object's look
instead of the current fill. Pick Color From Color Swatches works as
before. The tool already sends the press point first and the release
point last.
Docs: docs/mcp.md, docs/target-app-parity.md.
PDF import measured each gap between two glyphs against the larger of
the line's first glyph size and the next glyph's. A word set beside a
much larger one, apart by positioning and no space glyph (a browser's
margin between two sizes), stayed under a fifth of that em: "≤ 4" at
20 pt and "hours" at 8 pt, 3.4 pt apart, came in as "≤ 4hours".
A gap is now measured against the smaller of the two glyphs beside it.
Lines in one size read their gaps as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Illustrator 30.0's gradient Dither, from the public documentation
(https://helpx.adobe.com/illustrator/desktop/paint-and-fill/create-and-edit-gradients/control-dither-in-gradients.html):
a Dither checkbox in the Gradient panel; dither converts the gradient into
pixels with subtle noise, works with gradient fills only, and follows the
Document Raster Effects resolution.
- Gradient gains `dither` (serde default false, written only when on, so
older files open unchanged and new files keep it).
- The renderer draws a dithered linear or radial fill as an image paint: the
gradient sampled a cell per device pixel or finer (quarter-octave steps,
cached per thread), each channel given seeded triangular noise of up to a
level before rounding. Same pixels every run; strokes and undithered fills
draw exactly as before. PNG and other raster exports go through it.
- SVG and PDF write a dithered fill as an image at the document's raster
effects resolution, clipped to the shape (the freeform-gradient path);
characters' fills stay vector gradients in SVG, with a warning.
- paint.editGradient and paint.setFill take `dither`; strokes refuse it.
The Gradient panel's Dither box is dimmed behind the Stroke proxy.
- i18n rows in all 24 catalogs; docs: G13, target-app-parity, ui-parity,
mcp.md, plugins.md.
The File Handling preference Link Placed PDF and .ai Files
(linkPlacedPdf) is removed. File › Place with Link now places a page of
a PDF, .ai or .ait file read as PDF pages as a linked placed document,
as it links a raster image, and link: false embeds the page. Link is on
by default in the Place dialog and in file.place, so a PDF or .ai page
placed from a file, which used to be embedded, is now linked unless
Link is off. A password or textAs: "outlines" still embeds, and so do
EPS and PostScript .ai files.
The Place dialog's Link tooltip describes a PDF or .ai file as a
document.
A .ai file read from its editing data (its PDF part only a placeholder)
finds the files its art places by name, beside it. Package now copies
those files into the same folder as the .ai file, and warns when one has
to be copied under another name.
File › Place with Link kept a link only to raster images and VectorCraft
documents. A PDF, .ai or .ait page was always embedded as a group, so
Update Links never saw a newer version of the file.
A new preference, File Handling › Link Placed PDF and .ai Files
(linkPlacedPdf), is off by default. With it on, Link places the page as
a placed document, as it places a VectorCraft document: the object
keeps the file's bytes, a preview, the page number and the page's box
(crop, art, trim, bleed or media, stored as page_box). Links check,
update and relink it, Break Link and Object › Expand give the group
placing it without Link gives, every output draws the page's vectors,
and saving and reopening keep the link. With the preference off, or
with a password or textAs: "outlines", the page is embedded as before.
EPS and PostScript .ai files are still embedded.
Package copies a placed PDF or .ai file as it is. The Place dialog's
Link tooltip describes a PDF or .ai file as a document while the
preference is on. Relinking a placed document to a file that is not a
VectorCraft document, a PDF or a PDF-compatible .ai file says so.
Docs: docs/mcp.md, docs/target-app-parity.md.
Review follow-up (CodeRabbit): the two older G2 bullets counting 46 and 49
filters were superseded by Conté Crayon (57/57) and are folded into one
current bullet; the G2 row in the ranked table and the entry's estimate now
both say 4–6 h, as ROADMAP.md and roadmap.md do. ROADMAP.md's revision
history gains its 2026-10-22 row.
