egui-winit swallows the Cmd/Ctrl+V press and sends a paste event only
when the clipboard holds text, so with an image on the clipboard only
the key's release arrives and the image fallback, which waited for the
press, never ran. Paste from a V release whose press never came and
brought no paste event; the press and paste event are tracked from the
raw events at the start of each frame.
On Wayland arboard read the X11 clipboard, which the compositor does
not fill for a native Wayland window, so every system paste found it
empty. Enable arboard's wayland-data-control: it uses the Wayland
clipboard when WAYLAND_DISPLAY is set and falls back to X11 otherwise.
Right-click > Paste here now follows the same newest-wins rule, so a
newer image beats copied layers there too, and lands at the point.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Shift on a selected layer took it out of the selection on press, so the
drag that followed had nothing to move, and with several layers selected
the grabbed one stayed behind. It now leaves the selection on release,
and only when the press didn't become a drag. Where no move can start
(a version being viewed), it still leaves on press.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A log whose generation isn't its snapshot's was skipped as a leftover of a
compaction cut short, but neither generation is under a checksum: one flipped
bit made the server serve the older document, report nothing and overwrite the
log. Replay such a log over the snapshot instead: replay skips what the
snapshot already holds, so a genuine leftover changes nothing, while edits the
snapshot lacks are recovered and the damage is reported.
Found while testing crates/engine/src/cmd/edit.rs:499 (copy_to_clipboard, called from copy at :536 and cut at :542)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
`node.create` read its `parent` as `id(p, "parent").unwrap_or(page)`, so a `parent` that isn't a
layer id — a `{"$step": n}` nothing had resolved, a name, a negative number — quietly meant "no
parent" and the layer landed on the page, with the call reporting success. The same shape sat on
`image.place` and `instance.create`, on `index` (a string index appended where it was asked to
insert), and on the `ids` of every command that targets layers (an entry that wasn't an id was
dropped and the rest ran). A number that names nothing was reported; anything that isn't a number at
all was not.
`params::opt_id` and `params::opt_index` now say what the commands mean: missing and `null` are
"none", and anything else has to be a layer id (or a whole, non-negative number) and is named when it
isn't. `req_targets` refuses an `ids` entry that isn't an id, before falling back to the selection. A
number that names no layer still reports `no such node`, and a boolean is still a number to `f64_of`,
as it is for `x`, `width` and the rest.
This turns a silently wrong result into an error on purpose, so a caller that passed something that
isn't an id and relied on the fallback now hears about it.
`figcraft-cli run` ran each step of a script on its own, so a script's `{"$step": n}` reached
`parent` unresolved and the layer landed on the page while every step reported success — the same
commands through `command.batch` (the MCP `batch` tool, the control channel, plugins) put it inside
the frame. One `edit.undo` took back one step of a script rather than the script, and a failing step
left the earlier ones applied.
`Session::batch_step` is now the one path both take — one step of a batch, its `$step` references
resolved against the results before it — and `run_script` wraps its steps in `Session::grouped`: the
whole script is one undo step, and a step that fails takes the script back. The result lines come out
once the script is through, so none of them reports work a later failure took back; a failure prints
one line with `rolledBack: true` and exits 1. `docs/cli.md` documents the reference and the output.
Duplicate repeats the last duplicate's offset, but of the ways to move the copy
before the second Cmd+D only `node.set` and Alt-drag were covered. Dragging it
with the mouse is how almost everyone does it, and is what #567 describes; an
arrow-key nudge is how a small offset is usually made exact. Both work, because
the step is measured from where the copy ended up rather than recorded from the
gesture, but nothing held them there.
The drag test also asserts the drag moved the copy rather than making another
one, left it selected, and left the original in place: the ways the feature
could break without the step arithmetic breaking.
Both fail if the step lookup in `edit.duplicate` is stubbed out.
`cargo xtask ci` was red on a clean main on macOS: the test's last shutdown
runs after the server has answered an invalid request and closed the
connection, and `shutdown(2)` on a socket whose peer has already closed is
ENOTCONN on macOS and the BSDs where Linux returns success.
The test is right about what it checks; it checked a return value that is not
portable. Closing the client is what the worker thread waits for, and dropping
the socket does that whether or not shutdown reported anything. The file's
other six shutdowns run while the peer is still open, which is well defined on
both platforms, so they stay checked.
With no CI running the tests on pull requests, the local gate is the only
signal a contributor gets, and it could not be green on macOS.
- Code generation gains a Flutter target (ours; the incumbent generates CSS, iOS and Android only), which makes Dart the seventh language: `language: flutter` (with `dart` as an alias, so `CodeLang::from_id` maps both, `id()` is `flutter` and the label `Flutter`).
- `inspect.code` writes a Dart widget tree for the layer: a `Container` for a box, with its size, padding and a `BoxDecoration` (colour, radius, border, shadows); `Row` / `Column` for auto layout, with `spacing`, `crossAxisAlignment` and `textBaseline`, and `stretch` wherever a child fills the cross axis; `Expanded` for a child that fills the main axis; `Stack` with `Positioned` for free children; `Text` with a `TextStyle` (family, size, weight, style, height, letter spacing, decoration and alignment); `Image.asset` for images; `Opacity` for layer opacity; `SizedBox` for a capped subtree. Colours are `Color(0xAARRGGBB)` and `$` is escaped in string literals, as in Compose.
- `inspect.componentCode` writes a `StatelessWidget` for a component or set: properties become constructor parameters with their defaults, a variant property becomes a Dart enum (lowerCamel members, never a keyword), a boolean a `bool`, a text a `String`, a swap a `Widget` and a slot a `List<Widget>?`. Instances are calls that pass only what they change, and an instance's snippet puts the `import 'package:flutter/material.dart';` line before the call, since Dart's imports come first.
- Code connect takes Flutter too: `codeConnect.set {language: flutter}` stores an import path, a name and a template, and the snippet carries a Dart `import '…';` line. The default template passes named parameters (`Card(size: {size})`), as SwiftUI does, and import paths are checked as Dart imports (letters, digits, `_ . / : -`).
- The Inspect panel's language picker iterates `CodeLang::ALL`, so Flutter appears there and in the Connect to code note; the 13 UI catalogs and the component playground's language list follow.
- Docs: README, ROADMAP, docs/mcp.md, docs/gaps.md, docs/components-parity.md, docs/file-format-parity.md, docs/target-app-parity.md and docs/dev-mode-parity.md.
- Tests: Flutter cases in the codegen, component codegen and inspect behaviour suites.
Node lookups in the document's B-tree were still a third of a render on a
900 000-layer file: a small subtree was walked for its bounds, again to
see whether it could be splatted, and again to splat it, and each child
of a drawn layer was looked up by paint_runs, by the mask check and by
draw_node.
- The render's table also keeps the layers it looked up, keyed with a
multiply hash instead of SipHash (ids are counters), and so do the
bounds it keeps.
- draw_children looks each child up once, works out the paint runs from
those (Document::paint_runs_of) and draws the child it has.
- Whether a layer can be splatted is kept like its bounds: layers with
children only, answers the nesting cap decided aren't kept, and kept
answers are read only where their levels fit under the cap.
Ctx gets a second lifetime, for the document the kept layers come from.
Single-threaded cycles per 1440x900 render fitted to the page: 0.39 ->
0.34 G on a page of 246 204 layers, 0.26 -> 0.22 G on one of 118 358.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YQxjkNSUJhTkTfY74aNeki
Culling a layer asked for its paint bounds, which walk its whole subtree,
and then did the same for each child: every layer's subtree was walked
again at each level above it, and once more in each parallel band. The
bounds worked out during a render are now kept for it (layers with
children only; bounds the 64-level cap cut short are not kept, and kept
ones are read only where their levels still fit under the cap, so culling
is exactly as before).
Every render also looked through every layer of the document for the
largest background blur, to size the overlap of its bands: 900 000 layers
on one large file, on the main thread before the bands start. The
document now works it out when first asked after a change, like whether
it has masks.
A 1440x900 render on a 900 000-layer file, page of 118 358 layers: at
100% 13.0 -> 0.6 ms, at 25% 16.5 -> 3.4 ms, fitted to the page 56 -> 31 ms;
page of 246 204 layers: 12.6 -> 2.5 ms, 12.1 -> 2.0 ms, 69 -> 62 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YQxjkNSUJhTkTfY74aNeki
Hovering a family in the Typography font dropdown previews it on the
canvas via text.preview; closing without picking restores the original,
clicking applies it as one undo step.
Fonts activated through Adobe Fonts rendered in the fallback font without
a warning: Creative Cloud keeps them in its own folder as hidden files
named by number, which the font scan never read. The macOS scan now
includes ~/Library/Application Support/Adobe/CoreSync/plugins/livetype,
and files without an extension are indexed when they start with a font
signature. Fonts are read in place.
Closes#298
In vector edit mode a point dragged onto another point of its path
snapped there but stayed a separate point, so a path couldn't be closed
or joined by dragging. A single dragged point dropped within the point
hit radius (6 px) of another point now merges into it, in the same undo
step as the move, and that point stays selected.
Closes#401
Copies written by plugins and tools carry their image bytes in the
message: a paint's image.dataBlob indexes the message's blobs. A bare
canvas (what the clipboard holds, see #647) has no images/, so these
fills came in as grey placeholders. The importer now falls back to the
embedded bytes, keyed by image hash, after the archive's images/.
Blobs over the image size cap and indexes past the blobs are skipped and
stay placeholders. Tests: a bare canvas carrying its image in a blob, and
a dataBlob past the blobs.
Found while testing crates/engine/src/cmd/edit.rs:1180
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The pages list took a fixed 40% of the left panel (#272). The line under
it now drags to make the list taller or shorter: at least three rows, and
always leaving the layers their header and three rows. A double-click puts
back the default. The main window keeps the height in the preferences, as
it does the panel widths, saved once the drag is over.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YQxjkNSUJhTkTfY74aNeki
Flatten (Cmd+E, Alt+Shift+F) and Join selection (Cmd+J) built the result
from each layer's fill outline, so open paths, which have none, were
dropped: two pen paths flattened to "nothing to flatten", and next to a
closed shape they vanished. Vectors now bring their own network (open
paths, unfilled loops and each region's winding stay), mapped into the
parent's space; other layers still flatten to their outline.
Closes#377
ReadOne replies with a variant body (signature v); deserializing it
straight to u32 (signature u) is a signature mismatch that zvariant
rejects, so the initial colour scheme never arrived and system mode
kept its light fallback. Take the Value and read the uint32 out of it.
`document.json` was read under the 512 MB cap for any archive member, so a
large imported document saved but didn't open again: 900 000 layers save
as a 1.48 GB document.json, packed into 244 MB. It may now unpack to 16
times its packed size (counted up to the archive's own size), within
512 MB and 4 GB; images and other members keep 512 MB. A small archive
still can't unpack into gigabytes.
That saved file reopens in 13 s with a 5.9 GB peak.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YQxjkNSUJhTkTfY74aNeki
A real 705 MB .fig file (401 644 nodes, 894 004 layers once instances are
expanded) didn't open: "message: too many values". Lifting that cap alone
made it import only its first 200 000 layers, and lifting that one took
41 GB of memory.
The global caps stay as they are. The .fig import gets budgets from the
size of its compressed canvas instead (counted up to the file's own size),
not from the message it unpacks to, since chunks inflate up to 512 MB:
- the kiwi value budget is one value per 2 bytes of the canvas, at least
the old 20 million (kiwi is only read for .fig files);
- the layer cap is one layer per 32 bytes of the canvas, at least the REST
import's 200 000 and at most 2 000 000. A small or crafted file can't
expand its instances into millions of layers, however far it unpacks; the
REST JSON import keeps 200 000.
Memory:
- The mapper no longer passes the document through `json!`, which serialises
its arguments: the whole tree was deep-copied twice.
- Vector networks are built once per distinct network and handed to the REST
importer beside the JSON (`vectorNetworkRef`), not written into it as JSON
objects for every instance that shows them.
- Decoded kiwi objects are trimmed, and each import stage's input is dropped
as soon as the next stage is built.
On that file: peak footprint 41 GB -> 11.3 GB, 67 s -> 25 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YQxjkNSUJhTkTfY74aNeki
A library file is cached by its length and modification time. File systems keep the time in coarse ticks, so a same-length rewrite right after a read can keep the stamp, and library.updates then reports no change. A stamp now proves the file unchanged only once its modification time lies at least two seconds before the read; otherwise the file is read again.
Figma puts copied layers in the clipboard's HTML as a base64 .fig canvas
between (figma) markers. edit.pasteFromFigma reads it with the .fig import,
turns the layers into the session clipboard and pastes them as edit.paste
does (selected frame, in place or view centre), with the import's warnings.
The desktop app reads the clipboard's HTML (arboard) and Paste prefers a
Figma copy; the browser's paste event carries plain text only.
Tested on copies written by our own .fig export; not yet on real copies.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ROADMAP.md, docs/localization-parity.md and docs/translating.md said
Swedish had 120 of the 588 command labels and menu wordings; with the
previous commit it has all of them. Worded without counts, so the docs
stay right as commands are added.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The [commands] and [menu] rows held their command id as the text
("align.bottom" = "align.bottom"), which the menus and the command
palette would show as is. They now translate the English labels the
code gives those ids (spec! and m! in crates/engine), e.g.
"align.bottom" = "Justera nederkant". Also the interface strings that
are new on upstream main (23bdebd).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Straight connectors saved with a manual routing snapshot (connector and derived-frame version 1) now come in as an editable vector plus their label text layers, instead of an empty frame. Label placement composes the saved background and text transforms, and label paints and text styles use the existing import path. Connectors of other versions or line styles keep the existing fallback with their compatibility notes. At the layer depth or node limit the saved vector is kept without extra label layers, and skipped labels are reported once. Live endpoint bindings and routing are not retained.
A filled area of a vector network can now carry its own paint stack. A region without paints of its own keeps drawing with the layer's fills, and a region with an explicitly empty list draws nothing, so the two cases stay distinguishable. The canvas, SVG and PDF output, boolean and hit-test geometry, vector editing and region operations all honour the per-region paints and keep them when regions are split, merged or reordered.
The .fig importer resolves the individual paint stacks of a layer's filled areas through its normal paint mapping and hands them to the document builder as private mapper data. Public REST vectorNetwork JSON never consumes that data and its region paints are dropped, so only validated .fig input can set them. Invalid entries are skipped with an import note. Live style and variable links on individual regions and writing region paints in the experimental .fig exporter are not part of this change.
Text layers, text styles and instance overrides in .fig files can carry list spacing, hanging list, hanging punctuation and leading trim. The importer now reads them from the highest layer with a valid value, so an invalid instance override row falls through to the valid value below it, while explicit zero, false and NONE values stay resets. Both text layers and text styles share one mapping. The native text data stores these options once per text node, so differing values on separate runs are not represented, and the experimental .fig exporter keeps its existing note that these settings are left out.
Text layers, styled runs and text styles in .fig files carry a skip-ink boolean for their decoration. The importer now reads it into the native decoration look: an omitted run value inherits the layer's, a local value overrides a linked shared text style without changing the style itself, and instance overrides keep an explicit false over a true main. A malformed value counts as unset. The exporter writes the actual value of the layer and of each styled run instead of a constant true, and a decoration look that differs from the default only in skip ink no longer triggers the unsupported-decoration note, so default native text exports its real false value.
Supported TEXT_PATH layers import as editable native text on their vector path, including mixed text styles, a start inside a later segment and reversed text direction. The importer converts the segment parameter stored in the file to the native distance along the complete path.
The mapping accepts a connected, unbranched path with one recognized start-segment marker and the default placement above the path. Damaged or unsupported path data keeps the text readable off its path and adds an import note. Other vertical placements, multiline text and ambiguous or disconnected networks are not converted. Underline and strikethrough remain a native path-rendering limitation and are reported. Public REST TEXT_PATH layers and the experimental .fig writer are unchanged.
The vector network decoder now keeps each segment's style id. Import reports a segment style that names a row of the layer's style table as metadata that was not interpreted, and one that names no row as a reference to a missing style entry, each at most once per file. The earlier note claimed that stroke styles were left out, which overstated what is known about the metadata. Segment geometry, paint and width are unchanged, and the style ids are not kept in the native vector model.
