With drop shadows the backdrop was read twice: before the content and
again after the shadows. The first read is only what backdrop effects
inside the content see, so it now runs only when the layer has such
descendants, and the content is copied for the shadow caster only when it
gets a layer blur, noise or texture. The output is unchanged.
A layer with a background blur (or glass) drew its blurred backdrop into
its own pixels before casting its drop shadows from them, so a translucent
fill cast a shadow as dark as an opaque one, and with "show behind
transparent areas" the shadow was cast from the backdrop's pixels. The
backdrop was also read before the shadows were drawn, so a shadow showing
behind was never seen through the blur.
When a layer has both, it now casts its shadows from what it paints (with
the layer blur, noise and texture it will get), draws them, and only then
reads the backdrop, which goes under its content; the content still sees
the first backdrop under it while it's drawn. Layers without both draw as
before.
Colour emoji came out as the placeholder rectangles their outlines carry
(Apple Color Emoji) or as .notdef when the face has no outline at all:
nothing read sbix, CBDT/EBDT or COLR. A glyph now asks its face for
colour ink before its outline: an embedded bitmap strike decoded to
premultiplied RGBA and placed by the strike's bearings the way Skia (and
CoreText) places it, scaled to the size the run asks for, or a COLR v0
glyph's layers filled from the font's palette.
Faces that carry only colour tables are indexed again: Apple Color Emoji
is no longer skipped for being 190 MB (the scan still reads headers and
name tables only), and "Segoe UI Emoji" and "Noto Color Emoji" join
"Apple Color Emoji" as colour fallbacks, read only for a text that
nothing else covers. The png crate decodes the strikes; skrifa keeps
shaping, metrics and outlines.
Left for later: COLR v1 paint graphs, alpha-only strikes and the SVG and
PDF exports (text is outlines there).
Tests: 8 cases in the text crate (placement, PNG and raw-BGRA decoding,
scaling, COLR layers and bitmaps of the machine's own colour font) and a
render case gated on that font; the font-free ones run everywhere.
Verified on Windows with Segoe UI Emoji (COLR) and with Noto Color Emoji
(CBDT) read from a downloaded file; the sbix path runs through the same
code and needs a macOS run to be seen.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
Boolean operands restarted subtree_outline at depth 0, so booleans nested
in booleans bypassed the 64-level limit and could overflow the stack.
Co-authored-by: Justin John <babyjjbabyjj549@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
SVG and PDF exports check the visible subtree's depth before bounds or drawing and refuse, with an actionable message, a document nested past 256 levels instead of silently leaving the deep layers out; nothing partial is written, and an inner layer can still be exported. The CLI and MCP report the same error. Fixes#208.
Also: on Windows, figcraft and figcraft-cli link with an 8 MB main-thread stack (Windows gives 1 MB where Linux and macOS give 8), so an export at the depth limit no longer overflows it.
Co-authored-by: bflatastic <bflatastic@users.noreply.github.com>
- Cursors: the engine names a cursor for every tool, modifier and hover
target (rotate turned to its corner, pen states, eyedropper, comment,
copy on Alt, not-allowed while an instance refuses the edit); the UI shows
the system's or draws its own; guides and handles keep their cursors;
ui.inspect reports it.
- Panels: each edge drags its panel within limits, a double-click resets,
widths are per window and kept in the preferences (view.panelWidths); a
small window keeps its canvas.
- Outline mode: Include hidden layers and Include object bounds
(view.outlinesHiddenLayers, view.outlinesObjectBounds), drawn and
clickable; light lines on a dark page; check marks in the View menu.
- A file or page made in the dark theme has a #1E1E1E page (sessions
without an interface keep #F5F5F5); text drawn on a dark page is white.
- Tests for Enter and Escape on a selected text.
Vector edit mode can now edit several vector layers at once, and the rest of the vector edit tail
from the public documentation landed. A behaviour suite came first
(`crates/engine/src/vector_multi_behaviour_tests.rs`, 5 behaviour tests with about 30 table rows and
steps, plus a junk-parameter sweep of every `vector.*` command run in multi-edit mode).
- **Multi-edit.** Enter, a double-click or `vector.edit.begin {ids}` with several shapes selected
puts them all in edit mode, converting them in one undo step. The clicked or first one is the
active layer. Pressing another layer's point, handle or segment makes that layer active. Shift-click,
marquee and lasso select points in any layer. Dragging a selected point, the arrows, Delete,
delete and heal, mirroring, smoothing, corner radius and the new cap and join commands reach every
layer's selected points in one undo step. The pen, bend, paint bucket, cut, eraser, join and
position work in the active layer. A click away from every edited layer leaves edit mode.
`vector.edit.state` now reports `layers`, `others` and `handles`.
- **Several handles.** Shift-click Bezier handles to select more than one. Dragging one moves the
others by the same amount. Also available as `vector.edit.select {handles}` and
`vector.edit.moveHandles`. Selected handles are drawn filled.
- **Rounding next to curves.** A sharp point with a corner radius rounds even when one of its
segments is curved. The curve is cut where the arc meets it and keeps the rest of its shape, and
the arc meets each side smoothly (`VectorNetwork::rounded`).
- **A cap and a join per point.** `Vertex` gets optional `cap` and `join` values. When unset, the
layer's apply. The commands are `vector.edit.strokeCap` and `vector.edit.strokeJoin` (`null`
clears). The Design panel's Point section gets End point and Join dropdowns, each with a "Same as
layer" option. The stroke is drawn run by run and each point's own join is drawn over the break
(`figcraft_pathops::stroke_outline_of`, new `points.rs`). This applies to the canvas, PNG and PDF,
and to SVG as outlines. Outline stroke follows it, render bounds count arrowheads from points, and
the stroke cache key includes the points' styles.
- **Pen outside edit mode.** Clicking one of the path's own points (other than the first or last)
connects there and ends the path, as a vector network. With an open vector selected, clicking one
of its end points continues that path in edit mode.
- MCP: headless `key` Enter on shapes now enters vector edit mode, as in the desktop app. The new
commands have Spanish labels, and the new UI string has en and es entries.
- Docs: target-app-parity (area 4: 98%/72% to 99%/73%, 2-4 h to 1-3 h; area 20: 80%/55% to
88%/60%, 2-5 h to 1-3 h), ROADMAP, gaps (gap 9 is down to tablet pressure on macOS and Linux; the
gap 13 note is updated), ui-parity and mcp.md. Totals are recomputed by the doc's method, because
rows added since the last recompute had left them stale. Feature breadth is now ~83% and ready for
real work ~52%. Mainstream stays at ~49% (49.7%, rounded down). Full parity goes from
~306-595 h to ~294-572 h.
The tail of the text area, from the public documentation of the target app's text settings, with a table-driven behaviour suite (each case derived in one line; "silent" cases pin today's behaviour).
- **Lists per paragraph** (`TextData::paragraphs`, one `ParaList {list, indent}` per paragraph; empty when the layer has one list style at level 0, so old files load and save unchanged). The list buttons, Shift+Cmd+8 / Shift+Cmd+7, `text.listStyle` and Type settings act on the paragraphs the selection touches while editing (the whole layer otherwise). The per-paragraph list follows edits: a new paragraph copies the one it split from, joined paragraphs keep the first one's.
- **Nesting** up to 8 levels: Tab nests a list item, Shift+Tab moves it out; Enter on an empty item and Backspace at an item's start move it a level out, then out of the list. Typing `- `, `* ` or `1. ` at the start of a plain paragraph starts a list. Numbers count per level (1. a. i., then round again), bullets go • ◦ ▪; a plain paragraph, a shallower item or the other list kind starts the count again. Each level indents by one marker gutter; a nested bullet a font lacks comes from a fallback font.
- **List spacing** (space between two items of a list, in place of paragraph spacing), **hanging lists** (the outer gutter outside the box, item text lined up with plain paragraphs) and **hanging punctuation** (opening quotes and brackets outside the start edge of left-aligned and justified lines, closing punctuation outside the end edge of right-aligned and justified ones; hung punctuation takes no room when lines wrap).
- **Decoration look** per character (`TextStyle::decoration_look`, boxed, None by default): style (solid, dotted, wavy), thickness and underline offset (auto, px or % of the font size), skip ink (underlines break around descenders) and colour (its own style index in the layout, so canvas, PNG and PDF paint it without renderer changes). Decorations are drawn from `Decoration::path()` (bar, dots or wave) on the canvas, in PNG, PDF and outlined SVG.
- `node.set` keys: `listStyle` and `indentation` (range-aware with start/end), `listSpacing`, `hangingPunctuation`, `hangingList`, `textDecorationStyle`, `textDecorationThickness`, `textDecorationOffset`, `textDecorationSkipInk`, `textDecorationColor`; documented in `KEYS` and docs/mcp.md. `text.edit.key` documents Tab.
- Inspect: `listStyle`, or `lineTypes` and `lineIndentations` per paragraph; `listSpacing`, `hangingPunctuation`, `hangingList`, the decoration keys. CSS: `text-decoration-style`, `-thickness`, `-color`, `text-underline-offset`, `text-decoration-skip-ink: none` (browsers skip ink by default). Dev Mode rows: decoration style, list spacing.
- REST import: `lineTypes` with `lineIndentations` per paragraph (they collapsed to one list style for the layer before), `listSpacing`, `hangingPunctuation`, `hangingList`, the decoration style, thickness, offset, skip ink and colour; instance override names for them.
- Type settings ▸ Basics: decoration style, thickness and offset fields (with Auto), Skip ink, decoration colour; List spacing; Hanging punctuation and Hanging list checkboxes; the list control acts on the edited paragraphs. New strings in en/es.
- Components carry the new text fields in the characters and text-style override groups; scaling scales list spacing and px decoration lengths.
- `TextData::path` is boxed (`Option<Box<TextPath>>`) to keep `NodeKind` under clippy's enum size limit after the new fields.
**A paint and effects behaviour suite written from the public documentation, and fixes for every divergence it found.** Each documented rule (fill stacking and opacity, paint vs layer opacity, every blend mode per paint and per layer, pass-through and Normal containers, the four gradient kinds and their handles, stops, image scale modes, rotation and adjustments, the colour models and hex entry, the document colour profile, drop and inner shadows, showing shadows behind transparent areas, layer and background blurs, effects on groups, effect order, what exports include) is a table-driven case with a one-line derivation; where the documentation is silent the case says so and pins today's behaviour. Scenes are small and synthetic; pixel cases use tolerances (blurs are checked against the normal distribution with sigma = radius / 2).
**Fixes (13 render cases and 8 command/export cases fail on the old code)**
- **Pass-through and Normal containers**: a group or frame set to Normal now blends its children among themselves first (it never isolated), and a **clipping frame no longer isolates** its children: they were drawn on a fresh layer, so a Multiply child inside a frame (frames clip by default) met neither the frame's fill nor what is behind it. The clipped layer now starts as a copy of what lies under it and goes back in its place through the clip (a memcpy for rectangular clips, so no slower); backdrop effects see the right composite, also for a clipping frame inside a faded group. Same in exports: SVG gives Normal groups `isolation:isolate` and puts a clipping frame's clip on what blends instead of a clipped `<g>` (a clipped group is a stacking context, which isolates); PDF wraps Normal groups in an isolated group (`Node::isolates_children`, shared).
- **Plus darker** (linear burn) is drawn by its formula on the canvas and in pixel exports for paints, layers, shadows and noise (it was multiply); tiny-skia lacks the mode, so `fx::plus_darker` composites it and `paint::composite` routes every layer, shadow and noise composite through it.
- **Shadows hidden behind the layer**: a drop shadow that doesn't show behind transparent areas is now hidden behind what the layer covers, translucent and 0% fills included (it was cut out by the layer's alpha, so it showed through a 50% fill anyway); groups and layers whose pixels a layer blur or texture moves still hide it behind their pixels. One `shadow::Caster::new` serves the canvas, PNG and the PDF's shadow rasters. **SVG** never cut shadows out: its filters now composite them `out` of an image of the layer's outline (or `SourceAlpha` for groups), and the filter region reaches as far as the effects (a fixed 200% box cut off a small layer's long shadow).
- **Fill images cover the box to its edges**: an upscaled Fill (or covering Crop) image faded to transparent over a source pixel at the box's edge.
- **"Normal" meant pass-through** when named: `BlendMode::from_name("normal")` found pass-through first (its CSS keyword is also `normal`), so `node.set {blend: "normal"}` and the blend menu's Normal set pass-through. Labels now win, and the menu sends labels.
**Added**
- A blend mode menu for each paint in the colour picker (no pass-through for paints), sharing `right::blend_menu` with the Design panel's layer menu; strings already translated.
- Inspect (`node.get`, MCP) reports a solid paint's `blend` when it isn't normal.
A 64-case behaviour suite derived from public documentation (sizing, padding, strokes, gaps, alignment, wrap, absolute position, grid, stacking) plus command-level tests. Fixes: frames never shrink below their padding, only the frame's own inside stroke counts when strokes are included (now the default for new auto layout), auto gaps stop at 0, grid fr tracks never go below their content, and canvas stacking (reverseZ) now drives drawing, exports and hit testing. New settings: evenly/around auto gaps, vertical wrap with a column gap, per-child grid cell alignment, strokes included/excluded and canvas stacking in the Design panel.
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).
Text layers can follow a path (start offset, flip, alignment; edited in place; canvas, PNG, SVG, PDF; SVG textPath imports as real text on a path). A new pattern paint tiles a source layer in grid or hex rows with scale, spacing and alignment, live as the source changes (SVG pattern, PDF vector tiling pattern).
Animated GIF image fills keep their frames (first frame on the canvas and in exports; the presenter and preview play them, capped frames and memory, idle when not playing). TIFF images open, import, place and drop (pure-Rust decoder already in the tree); SVG exports embed TIFF and BMP as PNG. HEIC and video stay out (non-Rust decoders).
The last three catalogued menu items got commands: File > Open recent (recent files kept in preferences, capped, missing files greyed and dropped when picked, Clear recent), outlines for the selected layers only, and mask outlines. Menu breadth 100%.
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.
* M8: retained canvas correctness
Drags now show live on the canvas. The canvas used to drop every finished
render whose request had been superseded, and it asks for the next render
before it polls. So while an edit lands every frame (any drag), no image was
ever shown until the pointer was released. Holding the button still made it
worse: the same move ran the gesture again each frame, built a new document
and asked for a new render, which kept the canvas stuck on the pre-drag image.
- Canvas: show every finished render (renders finish in order; the image is
re-projected to the current view). The generation filter is gone.
- Engine: Session::pointer skips a drag move identical to the last one, and
Session::preview keeps the document when the preview changes nothing (a
handle held still).
- Render: clip and layer rectangles snap edges that are within float noise of
a pixel boundary before rounding outwards. An edge exactly on a boundary came
out a hair to either side depending on the render's origin, so a clip let a
whole column through in some renders and not others.
- Tests: a seeded random walk compares the retained render with a full render
after every step: edits, undo/redo, reparenting, masks, groups, effects,
strokes, text, auto layout, instances, variables, styles, pans and zooms. It
found the clip-edge divergence; a minimised regression test covers it. Also
a headless UI test for a live drag, and engine tests for skipped repeated
moves and unchanged previews.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* M8: retained canvas tests follow the new blur effect type
Layer blurs are `LayerBlur(Blur)` since the progressive blur landed; the random
walk now also tries progressive layer blurs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- Inner shadows paint over the fills and under the children and strokes, as documented (canvas,
PNG, PDF and SVG alike; text fills every run before the inner shadow and the strokes).
- Layer and background blur gain a progressive type: the radius ramps linearly from a start radius
at a start point to the end radius at an end point (unit-box points, like gradient handles).
Rendered by blending a few geometric blur levels, each blurred only around the pixels that use
it; bands and partial renders match one piece. Design panel type menu and Start/End fields,
on-canvas handles while the settings are open, effect.add/effect.set params, startRadius
variable bindings, scale tool, REST import (blurType, startRadius, startOffset, endOffset),
Inspect row; CSS leaves progressive blurs out.
- SVG export embeds renders for what filters can't match: noise grains over the vector layer
(crisp, blended), texture and progressive layer blur whole, background blur and glass from what
lies under the layer among the exported layers. PDF and SVG share the renderer's helpers.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* M2: auto layout keyboard reordering, nested auto layout fixes, section hide contents and dev status
- Arrow keys move auto layout children one place: along the flow (past hidden and
absolute siblings), between wrapped rows, to the neighbouring grid cell, and into
the neighbouring nested auto layout frame; no undo step when nothing moves.
- Sections: section.toggleContents (hide contents), object.wrapInSection, frames in
sections are top-level frames (labels, flows, navigation, PDF export), navigating
to a section shows its frame shown last.
- Ready for dev turns "Changed" when contents are edited; dev.resolveChange with a reason.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* M2: README and ROADMAP for auto layout arrow keys and sections
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* M7: roadmap refresh after #68–#77
Parity table rows for canvas, selection and transform, layers, effects,
text, prototyping and comments now reflect the merged work; weighted
feature depth ~76%; README command count and hand-off features.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* M2: the dev change reason resolves only on Enter or Resolve
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- Background blur sees the whole backdrop under the layer: the renderer keeps the stack of
isolated surfaces it draws into (clipping frames, rounded clips, faded groups, masks) and
flattens it for the layer's area, reading past the layer's own box. A blur inside a clipping
frame now shows the frame's fill instead of the sharp layers below, on the canvas and in PNG
exports. The PDF export draws background blur and glass through the same code from a render of
what lies under the node (`backdrop::render_beneath`, which the PDF's backdrop now shares), and
its images match the PNG's pixels.
- Shadow spread offsets the outline: the shape filled and stroked twice the spread wide with
mitred joins (grow) or cut away (shrink), the stroked bands cached per outline. Ellipses stay
ellipses, stars keep sharp points, text keeps its letterforms, rounded corners grow their
radius. Inner shadows shrink the outline the same way. Canvas, PNG and PDF share the shadow
code; SVG writes spread shadows as the same geometry and now exports inner shadows.
- Glass effect (light angle and intensity, refraction, depth, dispersion, frost, splay): the
frosted backdrop bent towards the inside within `depth` of the edge, colour fringes, light on
the rim, from an exact distance transform with smoothed normals.
- Noise: mono, duo and multi colour modes, grain size, density, opacity and blend mode; grains
live in the layer's coordinates (stable when panning and zooming). Texture is now drawn
(pixels scattered by the radius, grain size, clip to shape).
- Design panel rows for noise mode/colours/opacity, texture clip to shape and every glass value;
`effect.add` takes `glass`; REST import of NOISE, TEXTURE and GLASS; Inspect rows and CSS;
the scale tool scales grain sizes, texture radius, glass depth and frost.
- Tests in new modules: render backdrop/spread/grain tests, PDF backdrop tests, SVG, doc,
engine and REST effect tests.
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>
A stroke can carry a width profile: widths at positions along the path
(each subpath by length), interpolated smoothly (monotone cubic).
- doc: WidthProfile / WidthPoint / WidthPreset (uniform, wedge, taper both
ends, eye, rounded, pinch) on StrokeProps, serde-defaulted so older files
load; paint reach and hit testing follow the widest point.
- pathops: path measuring by length and the variable outline (offset along
the normals, round/miter/bevel joins, caps, dashes keep their width,
Inside/Outside clip as before; bounded samples and dashes, uniform
fallback). Canvas, PNG, SVG (filled outline), PDF and outline stroke use it.
- engine: stroke.widthProfile {preset | points}, node.set widthProfile,
inspect output; the Variable width tool (Shift+W) in vector edit mode:
drag the path to add a width point, slide points, drag their handles,
Delete removes the selected one; vector.edit.widthPoint add/set/remove.
Every gesture is one undo step.
- UI: width profile dropdown with previews drawn in code and a flip button
in the stroke settings; the tool in the edit toolbar (new icon drawn in
code); width handles and a hover preview on the canvas; Position/Width
fields for the selected width point.
- Tests: profile maths, outlines, render regressions, tool gestures,
commands, exports and junk sweeps. ROADMAP, README, docs updated.
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>
The canvas renders large views in parallel horizontal bands, and each band crops a layer to its
own rows. Three effects went wrong at those crops:
- Inner shadow (the reported bug): the padded shadow buffer treated everything beyond the cropped
layer as outside the shape, so every band edge cast shadow inwards, drawing dark bands across
the layer (any shape and fill; also at the view's edge when a layer was partly off screen). The
shape itself is now rasterised over the padded buffer at the shadow's offset (sub-pixel
accurate), so only its real edges cast shadow.
- Layers starting above or left of the pixmap (inside the blur margin) were sized from the
clamped origin and came out short by the margin: a drop-shadowed or background-blurred layer
crossing a band edge lost a strip at its far side.
- Background blur read a backdrop that stopped at the band's edge (streaks at band edges). Bands
now overlap by what background blurs read (3 sigma + box rounding) and keep only their own
rows; a blur wider than a band renders the view in one piece.
Regression tests (crates/render/src/band_tests.rs): an image-filled ellipse and a rotated solid
rectangle with inner shadows (spread, offsets) under a background-blurred panel, at 1x and 2x,
rendered in 2, 3 and 7 bands, must match the one-piece render row by row; the ellipse's interior
far from its edge must equal the render without the shadow on every row; a huge backdrop blur
falls back to one piece.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- Layers gain `maskType` (alpha by default, so files are unchanged): a vector mask clips by the
mask layer's outline whatever its fills and opacity; a luminance mask shows content where the
mask is bright and hides it where it is dark.
- `node.set {maskType}` (validated), `node.get` reports it, and the Appearance section shows a
Mask dropdown when the selection is masks.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
bflatastic