fix(import): carry Figma winding rule so multi-subpath icon holes render#6416
Merged
Conversation
…entric
Importing "A Channel FX" (.fig) rendered every knob's value-ring arc
visibly off-center of its body — up-and-left, worst on the medium knobs
(~1.2px diagonal on a 3.6px groove). The source is concentric: the body
ELLIPSE at (7.51, 7.51, 22.53²) and the ring arc's circle center at
(18.78, 18.78) share one center only at fractional precision.
Two independent layers were rounding that precision away:
1. The .fig decoder's styleFor (scene.mjs) emitted Math.round()ed
left/top/width/height for box-model nodes while the VECTOR_LIKE lane
kept round2 (2-decimal) placement — the ellipse center drifted
+0.72px/axis relative to its baked-path sibling. Both lanes now emit
round2.
2. Yoga's default pointScaleFactor of 1.0 re-rounded every laid-out box
to the whole-pixel grid, destroying the decoder's precision again.
New opt-in: View::set_subpixel_layout(true) — discovered pass-wide in
yoga_layout.cpp (one pass has one YGConfig) — sets pointScaleFactor 0
for that pass. Imported designs opt in via build_native_view_tree
(native lane) and a typeof-guarded setSubpixelLayout('', true) the JS
codegen emits (bridge lane); stock flowed UIs keep pixel-grid
rounding, pinned by test.
Verified against the .fig's own solved geometry: post-fix laid-out boxes
match Figma to 0.01px, ring/body cross-section asymmetry drops from
0.8px to 0.05px, all 27 ring-knob offsets collapse to the measurement
floor, and thumb_parity improves designers-pick from 77 to 39 bad
blocks (A Channel FX stays at its 3 marginal color blocks).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014e8kbMn8Sf4mF8ozsy1Jot
Hotspot-Grow: core/view/src/design_codegen.cpp reason="emit typeof-guarded setSubpixelLayout opt-in so imported designs keep sub-pixel geometry (knob-ring concentricity fix)"
Hotspot-Grow: core/view/include/pulp/view/view.hpp reason="add set_subpixel_layout opt-in API for import-time sub-pixel layout"
Hotspot-Grow: test/test_design_import_codegen.cpp reason="new test pinning the emitted setSubpixelLayout opt-in line"
Hotspot-Grow: core/view/src/design_import_native_common.cpp reason="set subpixel_layout on the materialized import root so ring geometry stays concentric"
A .fig vector's fillGeometry is a LIST of per-region entries, each with its own windingRule, and the decoder dropped the field entirely — every imported path filled under SvgPathWidget's nonzero default. Figma does not direction-correct the contours of its ODD-declared regions, so a subtracted icon (designers-pick's "Sub" speaker cabinet: five same-direction subpaths declared [NONZERO, ODD, ODD]) rendered as a solid slab instead of a hollow woofer — the fidelity sweep's worst dE2000 (~25), and fully silent. Wire the rule end to end: - paths.mjs: read windingRule per geometry entry; when entries disagree, evenodd wins (Figma direction-corrects NONZERO regions, which fill identically under either rule; ODD regions require evenodd). Expose fillRule / mixedWinding / subpathCount from geometryToPath. - scene.mjs: emit fillRule: 'evenodd' on the envelope vector (nonzero stays implicit); new warning diagnostic vector-fill-rule-approximated for mixed rules and for multi-subpath vectors with no declared rule, so this class can never be silent. - design_ir_json: capture fillRule/fill_rule/fill-rule into attributes["svg_fill_rule"] (evenodd|nonzero only). - design_codegen: emit setSvgFillRule(id,'evenodd') (bridge + widget already supported it; nothing emitted it). - design_cpp_codegen: emit ->set_fill_rule(FillRule::evenodd) at the svg_path call site (SvgRect/SvgLine share emit_svg_paint but have no fill rule). - design_import_native_common: apply_svg_paint(SvgPathWidget&) reads svg_fill_rule plus raw fill-rule/fillRule. Verified on designers-pick 1:486: both Sub icons regain their hollow woofers; thumb_parity's two worst blocks (dE 25.7/24.4 at design 482,461/419,461) collapse below threshold, 77 -> 73 visible-diff blocks against a same-build baseline, and Tom/HiHat/WoodBlock icons pick up their holes too. Tests: fig-lane node tests for the rule pick and the envelope contract, IR round-trip + JS codegen emission, baked-C++ emission, native materializer application, and a raster donut proof that evenodd opens a hole nonzero fills solid (renders compared by difference — the screenshot backend composites onto an opaque background at 2x, so absolute-pixel hole checks always read opaque). Hotspot-Grow: core/view/src/design_ir_json.cpp reason="fillRule capture in the svg attribute block (+14)" Hotspot-Grow: core/view/src/design_codegen.cpp reason="setSvgFillRule emission (+6)" Hotspot-Grow: core/view/src/design_cpp_codegen.cpp reason="set_fill_rule emission at svg_path call site (+12)" Hotspot-Grow: core/view/src/design_import_native_common.cpp reason="fill-rule application in apply_svg_paint (+12)" Hotspot-Grow: test/test_design_import_ir.cpp reason="fill-rule IR round-trip + codegen emission test (+56)" Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014e8kbMn8Sf4mF8ozsy1Jot
…g vectors keep Figma's flex slot A .fig transform's m02/m12 is where the local ORIGIN lands, and a mirror (m00 = -1) puts the box on the other side of it: the visual box spans [m02 - w, m02]. Reading m02 as "left" made the geometry sidecar report the Triaz Env/Reverse chip's icon (Frame 274) one full icon-width right of where Figma draws it, so layout_parity cried dx=-10.4 in all three Triaz files at a node the flex pass had placed exactly right. Three fixes, one rule (mirrorAwareOrigin): - The geometry sidecar (node-box and vector-ink entries) and styleFor's plain absolute placement report the mirror-aware min corner. Scoped to TRUE mirrors (exactly one negative axis): both-negative is a 180deg rotation whose raw-m02 placement the slider-fill lesson pins. - The emitted tree drops a container's mirror, so a vector under a net-mirrored ancestor chain bakes that mirror into its normalized path (geometryToPath's new mirror arg). The icon frame's flip cancels its child Union's own baked flip in Figma; without the re-bake our "reverse" arrow rendered pointing forwards. - A flowing vector's flex slot is its NODE box, not its stroke-inflated ink: the env chip's 12px arrow carries a 2px stroke, and its 14.14px ink pushed the "env" label 2.14px right. The decoder reconciles ink to node box with negative layout margins (gated to overhangs within strokeWeight + 1 so instance-scale gaps are never "corrected"), and the JS codegen now lowers layout.margin_* — it silently dropped them. layout_parity on the three Triaz files: Frame 274 dx=-10.4 -> 0.00 (six instances across A Channel FX / designers-pick / Sequencer), env text dx=+2.14 -> 0.00, Sequencer now fully clean; matched-node counts unchanged (1248/1283/1497) and no new findings. Hotspot-Grow: core/view/src/design_codegen.cpp reason="margin_* lowering in emit_js_layout_constraints; the IR carried margins the JS lane dropped" Hotspot-Grow: test/test_design_import_codegen.cpp reason="regression test pinning setFlex margin lowering" Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014e8kbMn8Sf4mF8ozsy1Jot
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Jul 21, 2026
# Conflicts: # tools/import-design/fig/scene.mjs
# Conflicts: # tools/import-design/fig/paths.mjs
# Conflicts: # tools/import-design/fig/scene.mjs
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
github-merge-queue
Bot
removed this pull request from the merge queue due to invalid changes in the merge commit
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
shipyard-local
Bot
removed this pull request from the merge queue due to a manual request
Jul 23, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Item #1 of the Figma-import hardening plan — the worst + silent gap from the sweep.
Multi-subpath ODD-winding icons (e.g. designers-pick "Sub" speaker cabinets with a hollow woofer) rendered as solid slabs because the decoder never carried Figma's per-path winding rule (fill-rule), so everything filled nonzero.
Root cause + fix
Figma's
fillGeometryis a list of per-region entries, each with its ownwindingRule. The Sub icon concatenates [NONZERO dot, ODD ring, ODD box-with-hole] into one path of same-direction contours. Correct merge is evenodd-wins (Figma direction-corrects its nonzero regions, so they fill identically either way, while ODD same-direction holes fill solid under nonzero). Wired end to end:paths.mjsreads per-entry winding →scene.mjsemitsfillRule→design_ir_json.cpp→setSvgFillRulein web codegen +->set_fill_rulein cpp codegen + native materializer. Newvector-fill-rule-approximateddiagnostic (warning) fires on mixed-rule or multi-subpath-no-rule vectors so the class can never go silent again.Verification
thumb_parityon designers-pick (same-build baseline): 77→73 bad blocks, both Sub blocks eliminated (worst dropped from dE 25.7 to 12.65). channel-fx 3→3 (13 evenodd emissions); sequencer 0 (no evenodd vectors, untouched).ctest -R "svg|path|import|design|native"1056/1056 — the one exclusion is a pre-existing, unrelated PulpSampler SEGFAULT on main (heritage-kit merges feat: complete sampler heritage kit #6396/Harden sampler streaming, interpolation, and heritage pipeline #6332; none of this PR's files in its closure — flagged for a separate issue).🤖 Generated with Claude Code
🔎 Provenance
claudem5w1~/Code/pulp-harden1fd0e374f-d5dc-4e2a-bb33-6fa8b66124a0Resume
Jump to this tab
Relaunch (any agent)
Restore URL — https://claude.ai/code/session_014e8kbMn8Sf4mF8ozsy1Jot
stamped 2026-07-20 23:05 UTC