- inspect.code {language: "tailwind"}: a layer tree as a React component
styled with Tailwind CSS v4 utilities, no stylesheet. Auto layout maps to
flex utilities on the spacing scale; fixed-width auto layout frames, text
and wide layers become fluid (w-full max-w-[…]); instances are inlined
with their overrides, except those connected to code (codeConnect), which
are their snippet with one import each; bound variables are theme
utilities (bg-accent, rounded-card).
- mobile: <frame id> (same layer names): mobile-first code, desktop classes
that differ behind lg:, resets for desktop defaults, padding compared per
side; layers only on desktop are hidden below lg, those only on mobile are
listed in the header.
- variable.export {format: "tailwind"}: @import "tailwindcss" and an @theme
block (--color-*, --radius-*, --text-*, --spacing-*, --font-*), names
unique per file and shared with page code.
Builds on the token roles of the CSS export px fix (#1008).
- data.fill {id, rows, gap?, formats?, dir?, format?, scale?}: repeats a
frame once per row in one undo step; each row key names layers in the
copy (text, or {text?, fill?, visible?, fit?}). `formats` makes each row
once per size (feed, story, square), the copy resized so its auto layout
reflows. With `dir`, each copy is exported under a unique file name; a
failed export is reported in `errors` and keeps the copies.
- text.fit {ids?, maxLines?, minSize?, orphans?}: the largest font size,
down to a floor, at which fixed-width text stays within N lines and its
box; avoids a one-word last line; `fits` comes from the final text.
- Plugins can't call data.fill: it runs other commands inside, past the
plugin policy, like command.batch.
A radius exported as `--radius: 12` was invalid where CSS reads it as a
length (`border-radius: var(--radius)` ignores a bare number). Number
values now carry px unless the layers bound to them, their scopes, or the
variables aliasing them make them an opacity or a font weight, which stay
bare. The roles come from a small token_roles module.
Updates the variables_io_tests expectation for the radius (4 → 4px,
6.5 → 6.5px), and adds a test for a bound radius, a bound opacity and an
opacity alias's target.
Cause: in the browser, exporting several layers started one download
per file. Chrome holds back every download after the first until the
site is allowed automatic downloads, so the other files didn't arrive.
Fix: file.exportLayers takes `zip: true` (without `dir`) and returns the
files together as one base64 zip `data`, written by a new
figcraft_format::zip_files (the workspace's existing zip dependency).
The UI asks for it when the platform saves by downloads and hands over
one `<document name>.zip`.
Test: ui-egui several_layers_download_as_one_zip (one download, a zip
holding both layers). It fails on the previous commit (two downloads).
component.addProperty {type: "boolean", default: false, layer} hid the
bound layer in new instances but left it showing in the main component,
so the main didn't show its own default.
On the main, a boolean property's value is now the visibility of the
layers bound to it:
- addProperty, bindProperty {field: "visible"} and editProperty {default}
show or hide the main's bound layers to match the default (every
variant of a set for editProperty). Without a default, addProperty
takes the bound layer's current visibility instead of always `true`.
- Showing or hiding a bound layer on the main sets the default, and the
property's other bound layers (in every variant) follow. Edits inside
instances stay overrides.
- Existing instances keep their stored values; documents open as saved
(no normalisation on load).
Fixes#993
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
node.set took no `path`, so redrawing a vector meant deleting and recreating it and
losing its id, place in the layer list, paints, styles, bindings and step references.
`path` now takes SVG path data for vector layers and replaces the geometry through
the same reader node.create uses (path_network, shared with create). width/height in
the same call scale it as create does. The path's own box goes at the layer's current
origin, so rotation and flips stay; with x/y in the same call it goes at x/y plus the
box's offset, which is where create puts it.
On several layers, layers that aren't vectors are skipped with a warning, as auto
layout keys already are (the skip logic is shared now). A single non-vector is
refused, and so is a vector inside an instance: there is no geometry override, so
the next sync would undo it. Data that doesn't parse, draws nothing or has
non-finite bounds is refused cleanly; create now refuses non-finite bounds too.
Tests cover keeping the layer, placement and scaling against create, rotation,
undo/redo, invalid data, non-vectors, instances, boolean operands, edit mode, and a
junk-path sweep. The engine-wide junk sweep has node.set path entries.
Fixes#1001
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A batch step can name itself with "as" and later steps refer to it with
{"$ref": "name"}, alongside positional {"$step": n}. "field" now takes a
slash path into the result as well as a key, so {"$ref": "copy",
"field": "ids/0"} reaches the first id edit.duplicate returned (the
convention variable.bind uses for fills/0/color; a leading "/" as in
JSON Pointer also works, and a key holding "/" still matches first).
Resolution moves to crates/engine/src/batch.rs (BatchStep, BatchScope),
shared by command.batch, the CLI script runner, the control channel and
the MCP batch tool, which now passes "as" through. Duplicate names fail
before any step runs; unknown, same-step or later names, bad paths and
out-of-range indexes fail the batch with the step, name and full path.
$ref is reserved like $step; nested batches keep their own names.
Fixes#994.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`node.reparent {ids, parent, index}` read `index` against the parent's
children as they were before the move, so within the same parent moving a
layer up one place did nothing and asking for two places moved it one.
`index` now counts the parent's children after the move, the way
`node.reorder` already does, in the same parent or a new one; several
layers still land together in their stacking order.
The layers panel was the only caller passing an index other than 0. It
drops on a gap between rows counted before the move, so it now subtracts
the dragged layers below that gap and drags behave as before.
Documents `index` in the command's params and in docs/mcp.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: in the web build no file has a path, so Save goes to Save as, and
Save as and every Export go through Services::pick_save / pick_folder.
WebServices keeps the trait's defaults there, which return no path, so
the callback never ran: no download, no message. Only Variables ▸ Export
used the platform's downloads (#368).
Fix: where services.downloads() is true, those routes run their command
without a path and hand the bytes to services.download():
- engine: file.saveAs takes `data: true` and returns the .figcraft file
as base64 `data` with its `name`, marking it saved (no path, nothing
remembered). Programmatic Save as without either still asks for a path.
- ui: FigApp::download decodes a result's `data` and says when nothing
arrived; Save as, one-off export (PNG/JPG/WebP/SVG/PDF), .fig export,
frames to PDF and multi-layer export (one download per file) use it.
Tests: engine save_as_data_hands_back_the_file_and_marks_it_saved;
ui-egui web_save_tests (save, export, refused download) with the fake
browser Services from variables_export_ui_tests. The UI tests fail
without the fix.
JIS X 4051's line-start (行頭禁則) and line-end (行末禁則) sets were mostly
there — closing punctuation and brackets, small kana, the prolonged sound
mark, iteration marks and opening brackets already never start or end a line —
but a few categories fell through:
- the halfwidth small kana ｧ–ｯ (U+FF67..U+FF6F) and the halfwidth voicing
marks ﾞﾟ (U+FF9E/U+FF9F) were not line-start prohibited, so a wrapped line
could begin with them;
- the spacing (semi-)voiced sound marks ゛゜ (U+309B/U+309C) and the kana
iteration marks 〱〲〳〴〵 (U+3031..U+3035) likewise;
- the halfwidth parentheses ｟｠ were missing from the line-end/line-start
sets.
Add them (ranges in `class`, the two parentheses in the OPEN and CLOSE
strings), and pin every JIS category in the behaviour suite: the existing 22
cases stay and 9 join them (halfwidth small kana and small tsu, halfwidth and
spacing voicing marks, kana and vertical kana iteration marks, both halfwidth
parentheses), plus a wrap case where a halfwidth small kana moves down with
its base (キｬ) and a case showing a manual newline may still start a line with
a prohibited character (禁則 is for automatic wrapping only).
`break_opportunities` fails on the old sets at the first new halfwidth
small-kana case.
Latin wrapping, ideograph breaks and the rest of the behaviour suite are
unchanged, and the headline numbers stay as they are.
Verified: cargo test -p figcraft-text --lib 92 passed; cargo clippy
-p figcraft-text --all-targets -- -D warnings clean; rustfmt clean on the
touched files.
Docs: docs/target-app-parity.md text behaviour row and ROADMAP progress log,
revision history and header.
A design-system .fig file (54 pages) stopped at 244 248 layers, the rest
of its pages skipped, and most of its instances came in as frames "of
components that aren't in the file", with the cap note shown twice.
- The layer budget follows the compressed canvas at 8 bytes a layer, not
32: design systems are mostly instances, which pack tightly. Files under
about 1.6 MB keep the 200 000 floor and the 2 million ceiling stays
(measured on synthetic files: about 5.4 KB and 19 µs a layer).
- The library copies the pages use (the file's hidden internal canvas) are
mapped before the pages: at the cap the pages' last layers are skipped,
not the mains their instances link to. The REST stage keeps building
pages first (its ids are first-wins) and allows for the wrapper frame of
each library copy, the one layer the mapper doesn't count; that layer
tripped the REST stage's own cap and repeated the note.
- Instances and swap defaults whose main is in the file but past the cap
say so instead of "aren't in the file" (en and es catalogs).
- The .fig import gives a note the mapper already gave only once.
Tests in fig/layer_cap_tests.rs: the budget for a #981-sized canvas, library
mains ahead of the pages at the cap, the note for a capped local main, and
a whole 210 000-layer file with one cap note and its library links kept.
All four fail on the previous code.
Plain `-` zoomed out but plain `=` (the `+` key without Shift) did nothing,
so a canvas zoomed out with `-` couldn't be zoomed back in the same way;
held down, `-` reaches the 2% floor and `=` leaves you there. Ctrl+Shift+=
zoomed in while Ctrl+Shift+- did nothing either.
Found with a trace of the input on GNOME Wayland: Ctrl+wheel, Ctrl+=,
Ctrl+-, Ctrl+Shift+= and `-` all reach the canvas and zoom; `=` and
Ctrl+Shift+- arrive too but no binding takes them.
`view.zoomIn` gains `=` and `view.zoomOut` gains `Shift+Cmd+-`, pinned in
the shortcut parity table; neither key was bound elsewhere and the menus
still show Ctrl/Cmd and `+`/`-`. A keymap test checks each zoom-in key
has its zoom-out twin on both platforms.
Panel dropdowns (prototype device, Preferences, blend modes, versions…)
draw their rows with `right::menu_row` on the theme's popup fill, which is
white in the light theme. Since 5a4b7c27 those rows painted their labels in
`menu_text`, white in both themes because it belongs to the dark menu bar,
so the options were invisible until hovered.
Rows now take the theme's text (`t.text`, shortcuts `t.text_secondary`) and
white only on the green hover highlight, as before that commit. The menu
bar, tool flyout and style/variable pickers paint their own rows on
`t.menu` and are unchanged.
Regression test: a dropdown opened headless in both themes keeps its labels
at 4.5:1 and shortcuts at 3:1 against the popup fill.
The colour picker's Window area is keyed by egui as Id::new(Some(title)), but
the style picker looked it up as Id::new("color_picker"), so the lookup always
missed and any click in the picker counted as an outside click that closed the
editor (and the picker with it). Give the window a fixed id and query it through
colorpicker::window_id(), with a regression test.
Preserve existing Korean translations and exclude unused keys and keys already proposed in open PRs.
Static catalog, key, placeholder, ellipsis, shortcut and whitespace checks passed. Cargo fmt, clippy, tests, xtasks and native/web UI validation were not run: Rust tools are unavailable and installation awaits approval. See docs/korean-translation-validation.md.
Include document UID in pages_seen_page to correctly handle page selection and scroll tracking when switching between documents. Record page as seen only after the active row renders so that pages filtered out by search scroll into view when the filter is cleared.
Found while testing crates/format/src/lib.rs:256 (to_json/write_json/save_bytes) and :323 (parse_json)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/svg/src/lib.rs:589 (also :604 and :611)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The threaded test asserted a fixed glass scene renders in over 50 ms,
which would flake on a fast machine or once the glass speedups land.
It now grows the scene until the full render takes at least twice the
threshold plus the preview's own time, so the watchdog's preview comes
first with a margin wherever it runs.
Refs #295
Whether a zoom got a preview rested on the previous render from
scratch, so the first render, another page or file, and a zoom from a
quick view to a heavy one came without one: the page colour showed
until the full render was done.
Rendering every such preview up front would cost 8-56% of a full render
(bench pages of 1k-50k layers at 2880x1800; per-layer costs dominate
large pages). Instead the render thread arms a watchdog thread as it
starts a render from scratch without a preview; once that render has
run past 12 ms the watchdog renders the preview beside it, on one core
while the full render keeps the others, and sends it unless the full
image is done. Quick renders cost nothing and never flash a preview.
Warm slow zooms keep the preview first, which keeps up with a gesture.
With two threads sending, finished images carry their request's
number: a poll takes the newest request's, and of one request the
refined image; the canvas never shows a preview over the refined image
of its own or a newer request.
Refs #295
inspect_document sends a shallower tree when the document's is too large
and points at get_node for the rest, so get_node was the way out of that
limit -- and it had none of its own. Measured against a document with a
frame of 15,000 children, through figcraft-cli mcp:
inspect_document depth 64, allPages 14.29 MiB (8.00 MiB cap not
applied before #915)
get_node depth 64 10.51 MiB
Both are over the 10 MiB buffer the report is about, and for get_node no
depth makes it smaller: the width is at depth 1, so depth 1 is already
9.12 MiB of text and depth 0 has no children at all.
get_node now takes the same maxBytes, cuts the subtree by depth as the
document's tree is cut, and then takes a window over the children when
even depth 1 is too large. The marker says how many children were shown
of how many, and the offset to ask for next; `offset` is a new argument
for reading the rest. After this:
get_node depth 64 8.00 MiB, children 11427 of
15000 from 0, next 11427
The hint in inspect_document's marker no longer promises that get_node
has no size limit, since that was the hole: it now says a wide node
answers with a window and an offset.
The test pages through a windowed node and asserts every child is
reported exactly once, in order, with every page inside the limit.
Unbounded, the first page is 23,003 bytes against a 7,667-byte limit.
A .fig import's notes were deduplicated and left uncounted, so each kind
was named once with no sign of how much of the file it applied to: a
file whose every text layer lost a setting read exactly like one where a
single layer did. That is what made the loss in #467 invisible -- the
reporter saw one line while three kinds of data went, one of them on all
83 text layers.
The mapper now collects its notes through the same `Warnings` the REST
importer already uses, which keeps the first-seen order and the cap and
appends `(×n)` where a note was met more than once. Nothing else
changes: the note text stays the catalog key it was, and the import
report already reads a `(×n)` suffix off before translating the base
(`ui-egui`'s `note_text`), so a counted note is still translated.
`figcraft-cli info` prints these notes, so it now names each kind with
its count.
With the count dropped again, the test reports the two notes plain:
["connector layers import as static frame/vector snapshots ...",
"connector label was skipped: depth limit prevents nested ..."]
Part of #467.
Two things the limit got wrong, both reported by CodeRabbit on #915.
maxBytes was compared with the escaped text block alone, but what goes
on the wire is that block inside a content object, inside the JSON-RPC
result, with the request id and the newline the transport writes: a
block measured at exactly maxBytes left as a larger message than the
caller said it could read. A kilobyte of the caller's figure is now kept
for that wrapping. About ninety bytes of it is fixed and the rest is the
request id, so the figure holds for any id up to roughly 900 bytes --
every id in practice, a number or a UUID. Sizing the real envelope here
would mean handing the transport's request id to every tool, and the
assumption is written down instead, in the schema and at the constant.
A depth: 0 request over the limit returned early, so the one response
that cannot be made any smaller was the one sent without the truncated
marker -- while a bisection that walked down to depth 0 sent the same
tree with it. Depth 0 now takes the same path as the bisection: the tree
goes out, over the limit, with the marker saying so.
Without the first fix the test reports a limit at the exact size of the
tree as no truncation; without the second the depth-0 marker is absent
(left: (None, None), right: (Some(0), Some(0))).
- Add Polish command labels for arrange.reorder, file.moveTab, pref.startOnHome,
pref.uiScale, text.loadFont, text.refit, and text.rescanFonts.
- Ensures 100% command coverage with zero fallback to English in the Actions palette.
Found while testing crates/collab/src/code.rs:~150 (Local::from_link, u64::from_str_radix on the token)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- Mark Polish catalog as reviewed and approved by native speaker (@lukasz-sklad)
- Replace "Lekki" with "Jasny" for Light mode
- Translate disambiguated context keys (Fill|sizing, Join|stroke, Light|glass, Open|comments)
- Catch up with 15 UI strings added in recent main commits
- Update docs/translating.md to list Polish as reviewed
vetlucasmartins
Ridzzz0Alam
KameronandClaude Opus 5.5
midasdf
Nicek
kolla
BFlat
AKY
AganzeFelicite
NetterSilverandClaude Haiku 5.5
Siade Lucas
tylerjenningsw
Hamza
YarilaFox
Lucas