[miguel-heygen]

What

The linter now parses each <style> block with postcss once per lint, and every rule shares that one tree. Before this change, ten rule passes each parsed the same block again.

Why

gsap.test.ts "scans a closed style block of many unclosed rules in linear time" lints 40 000 unclosed rules and checks a 2 s limit. On CI it took about 2 s and failed by chance, on main as well as on unrelated PRs. The cost came from ten parses and ten throwaway trees per block, which also made the garbage collector work harder. It was not slow regex scanning. Every lint of a large stylesheet paid the same extra cost.

Related work

None.

How

• styleRoots.ts caches each block's parsed tree, or its parse error, in a WeakMap keyed by the block. This follows the per-block cache in scriptPositions.ts. The cache lasts as long as the lint context does.
• The style parse sites in core.ts, fonts.ts, gsap.ts and textures.ts now read the shared tree. No rule modifies it, so sharing it is safe. A comment on styleRoot notes that rules must not modify it.
• The timing tests are unchanged. They still guard against a slow scan with the same limits, and now have about 6x headroom.

Test plan

• Unit tests added: styleRoots.test.ts. One lint calls postcss.parse once for a style block that several rules read; on main it is called 10 times, and that is how the test fails there. A malformed block is still reported once.
• All lint package tests pass 3 runs in a row on Linux (1309 tests). tsc --noEmit passes for the lint package. oxfmt and oxlint are clean on the changed files.
• One lint of 40 000 unclosed rules: about 400 ms before, about 60 ms after. 10 000 to 40 000 rules grew 6.8x before and 1.8x after.
• fails without the fix: packages/lint/src/styleRoots.test.ts › "parses each style block once per lint, however many rules read it". On main it counts 10 postcss.parse calls for the one block instead of 1. Shown by running the test against main's rule files.
• Comments follow CONTRIBUTING.md "Comments"
• Local pre-commit lint, format, typecheck, fallow and commitlint were skipped because this checkout has no local install. The same checks ran on Linux as listed above, and CI runs them in full.

Size

This is a small, standalone lint fix. The flaky timing test fails on main and on any open PR, so it lands on its own rather than inside an unrelated studio change.

[github-actions]

Edit accuracy: accurate 267 (base branch 267), smooth 116 of those

The gate passes.
Smoothness is reported in the artifact, not gated. A case fails only if it fails 2 of 3 runs.

Quarantined, measured but not gated (8)

• seqplainkeys-none-center-r0-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-center-r0-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-center-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-pct-r0-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-pct-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-px-r0-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): fail / pass / pass, passes
• seqplainkeys-none-px-r30-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-px-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run

[heygengenesis]

Verdict: REQUEST CHANGES · 1 blocker · 0 important · 0 nits · head c00b7c4b

• blocker (lane rule): this PR fixes a test that fails on main, so the body needs a fails without the fix: line naming a test. The Test plan describes styleRoots.test.ts failing on main, but not on that line. Add something like fails without the fix: packages/lint/src/styleRoots.test.ts › "parses each style block once per lint, however many rules read it" (it gets 10 postcss.parse calls on main).

I found no code blocker.

Scope

Read the full diff (6 files). Every postcss.parse(style.content) on a lint-context style block now goes through styleRoot. The parse sites still left are separate inputs: inline style="" attributes at core.ts:713 and project CSS files at project.ts:684,752. None of the rules that now share the tree modify it (I searched for remove, replaceWith, append, clone* and .value/.prop/.selector assignments). A parse error is cached, then thrown again to each caller, and every caller still catches it: the syntax rule reports it once, the others skip the block. The WeakMap is keyed by the block object, the same pattern as scriptPositions.ts. The spy test hooks the same postcss.parse the rules call, so it would really count 10 parses on main. CI, CodeQL, regression and Windows are green at this head.

— Genesis

[heygengenesis]

Verdict: APPROVE · 1 resolved, 0 open, 0 new · c00b7c4b → c00b7c4b · ✅ The body now has a fails without the fix: line naming styleRoots.test.ts. The code is unchanged.

— Genesis