Headless runs (`figcraft-cli export`, `render`, `convert`, `run`, and the headless MCP server) never indexed the installed fonts, so a text layer in an installed family (Segoe UI, Inter installed per user, ...) was measured and drawn with a fallback face, while the desktop app used the real one. Auto layout sizes from MCP agents therefore differed from what users see in the app.
- `figcraft_text::scan_system_fonts_on_demand()`: the font database indexes the system fonts the first time it is used (first text layout, font lookup or font listing). `figcraft-cli` turns it on at startup for every subcommand, so a run that touches no text never pays for the scan. The desktop app keeps its background scan.
- `scan_system_fonts()` now runs once per process (a `OnceLock`): racing callers wait for the one directory walk instead of each walking the folders.
- Opt-out for reproducible renders: `--no-system-fonts` on any subcommand, or `FIGCRAFT_NO_SYSTEM_FONTS=1` (honoured by the desktop app too). Text then uses only the built-in fonts, as before.
- The scan is cheaper: it reads each file's headers, table directories and `name` tables instead of whole files (about 460 MB on this Windows machine), and takes file metadata from the directory listing instead of a lookup per file. This also speeds up the desktop app's background scan.
- Browser/wasm build unchanged (the scan is a no-op there).
- Docs: `docs/cli.md` (new Fonts section) and `docs/mcp.md` (Fonts under the backends table); CLI usage text mentions the flag.
Clicks on a boolean group test its computed result, so a click in a subtracted hole reaches what is behind it. Layers inside a group, boolean group or instance stay in their parent when dragged (one shared rule). View > Frame outlines and Slices now draw. The idle test's flake came from the render worker clearing its in-flight flag before waking the UI; jobs now carry sequence numbers.
Strokes can use a brush: eight presets drawn in code (four stretched along the path, four scattered with spacing, size, rotation and offset jitter from a fixed seed), drawn by the shared stroke-outline code in canvas, PNG, SVG, PDF and outline stroke, with caps on copies and steps. Pointer events carry pen pressure; the Pencil records it as the stroke's width profile, which brushes follow (Windows pens and browser styluses; constant width elsewhere).
Boolean operands are editable in place with live results (per-operand operations, cached prefixes so a drag recomputes from the edited operand up); a Design panel Arc section; the shape builder takes strokes and open paths as outlined areas; variable width points can be wider on one side.
* M3: components at scale; Alt-drag to swap instances
- Synthetic design system (engine `synthetic_ds`): ~300 components (icons; button, input,
card, nav and badge sets of size x state x type x theme variants), boolean/text/instance
swap/slot properties, instances nested four deep, text and paint styles, light/dark
variables bound across it, and 2,000 instances with overrides at every level.
- Propagation: value, size and added/removed-layer edits of mains reach their copies
directly; other edits rebuild each affected instance once, inner mains first; rebuilds keep
unchanged layers shared; bound variables no longer touch unchanged layers; auto layout
re-flows only where sizes or places changed.
- Fixes: values from variables (theme switches, bound radii) are no longer recorded as
overrides; a detached variable on an instance layer is an override that survives main
edits; exposed nested instances stay exposed; swapped nested instances keep their layer
names and the ids/overrides of layers their own main adds; renamed nested instances keep
their name when detached out.
- Instance swap properties have preferred values. `instance.swapAt` / `library.swapAt` and
Alt-drop from the Assets panel onto an instance swap it (highlighted while dragging).
- Instances pasted into other files follow their mains (library copies shared, published
components used from their library, other mains left behind as a missing main).
- Assets panel lays out only visible tiles and renders thumbnails lazily with a cache
invalidated per changed component; instance panel lists components once per revision.
- `figcraft-cli perf --instances`, UI frame bench and tests in components_scale_tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* M3: the swap perf case drops on a glyph the hit test reaches
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Merge origin/main into feat/components-scale; weighted depth 79%
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Canvas previews zoomed out to 50% or less draw levels of detail
(crates/render/src/lod.rs): subtrees under 2 device px are splatted as
their averaged paint colours (box filter, no paths or effects), text
under 4 px is greeked into bars, small solid rectangles are splatted and
rectangular clip layers are composited without the blend pipeline.
Image paints and masks keep the exact path. Exports, thumbnails and
every other render stay exact (Options::lod is off by default); render
bands are assembled with a row copy instead of a pixmap draw.
.figcraft documents are serialized straight into the compressor and
parsed straight from the text (a header pass reads version and image
index without building a value tree); layer and text fields at their
default values are left out (63 -> 22 MB of JSON at 50k layers, older
files still load); JSON members use deflate level 3.
At 50k layers: zoomed-to-fit render ~58 -> ~8 ms, save ~1.0 s -> ~0.09 s,
load ~0.86 s -> ~0.16 s.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Large documents stay interactive: the canvas re-renders only what a pan uncovers or an edit
touches, the layers panel lays out only the rows in view, and per-frame work no longer walks the
whole page.
- Benchmark: `cargo xtask bench [SIZES]` builds synthetic pages (`figcraft_engine::synthetic`:
screens with an auto layout header, a card group, button instances and shapes/text) at
1k/10k/50k layers and prints engine/render timings (`figcraft-cli perf --layers 1000,10000`) and
headless whole-UI frame timings (an ignored ui-egui test).
- Retained canvas rendering (`figcraft_render::Retained`, used by the render worker): views snap
to whole device pixels, so a pan shifts the previous image in place and renders only the
uncovered strips; an edit renders only where changed layers were and are
(`Document::changed_ids` diffs the two versions skipping shared structure; ancestors that paint
from their children's pixels - booleans, effects, masks - count whole; background blur reach
and noise are allowed for). Zoom, page, options, font or page-level changes, or large changes,
render the whole view as before. `ui.inspect` reports `canvas.renderWork`.
- The texture image is converted on the render thread, not the UI thread.
- Layers panel: only visible rows are laid out (stable ids while scrolling); rows are cached per
document version, page, expanded set and filter.
- Frame labels and layout grids of off-screen top-level frames are skipped; the Prototype tab's
connection sources are cached per document version.
- Text: the layout cache keys by a cheap fingerprint confirmed by `TextData` equality (no Debug
formatting per lookup, 32k entries) and hands out shared layouts; a layout generation marks
renders stale when fonts arrive. Overflowing text is culled by its glyphs, so it no longer
vanishes while its box is off screen.
Before -> after (release, Windows, 1440x900 view, ms; 10k / 50k layers):
- move one layer, re-render @100%: 7.65 / 6.49 -> 0.18 / 0.81
- move one layer, re-render zoomed out: 11.95 / 24.10 -> 0.16 / 0.77
- pan 40 px @100%: 7.53 / 6.82 -> 0.69 / 0.81
- pan 40 px zoomed out: 12.07 / 24.30 -> 1.62 / 4.17
- full render @100% / zoomed to fit: 7.78 / 44.22 -> 6.56 / 38.83 (10k @100%, 50k fit)
- UI frame, every screen expanded in layers: 6.16 / 39.07 -> 0.24 / 0.57
- UI frame, scrolling that panel: 6.11 / 40.18 -> 0.43 / 0.40
- UI frame, layers filtered by name: 1.03 / 6.92 -> 0.17 / 0.36
- UI frame, Prototype tab: 0.89 / 6.45 -> 0.18 / 0.80
- UI frame, idle: 0.31 / 1.43 -> 0.11 / 0.39
Unchanged and within budget: single-layer edits through the engine ~0.01 ms, normalize after a
one-node edit < 0.01 ms, hit test 26 / 93 us. Left: save/load at 50k (~0.7 s each), zoomed-out
renders at 50k (~40 ms).
Tests: retained renders match full renders after pans, moves, recolours, hide/remove/add, undo,
masks, overflowing text, edits under a background blur and an inner-shadow/noise layer, a drop-
shadowed group, and a blur panned in from the view's edge; zoom/page/options/page background
render full; junk views don't panic. Layer-row and prototype-source caches follow edits, undo,
page switches, expansion and the filter. `changed_ids`, the text cache, and the perf CLI.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
