fix(lint): catch cold-seek opacity reveals

[miguel-heygen]

Summary

• flag fromTo() entrances whose target is authored hidden but whose destination omits opacity/autoAlpha
• cover both CSS/inline hidden state and standalone gsap.set(..., { opacity: 0 })
• explain why sequential snapshots can pass while cold render workers encode the target invisible

Why

CLI feedback reproduced a render-only divergence: a CSS-hidden element with fromTo({ opacity: 1, x: -60 }, { x: 0 }) appeared in check --snapshots, but was black in a two-worker encoded MP4. Each cold worker restored the authored opacity: 0, and the destination never established visibility.

Test plan

• RED: both CSS-hidden and standalone-gsap.set repro tests failed before the rule
• GREEN: pnpm --dir packages/lint exec vitest run src/rules/gsap.test.ts --environment node (92/92)
• pnpm --dir packages/lint typecheck
• oxfmt --check and oxlint on changed files

Visual evidence

Not applicable; this is a static lint diagnostic with regression tests for the reproduced render divergence.

[james-russo-rames-d-jusso]

Reviewed at f2e65c16. Adds a new lint rule + supporting AST field; scope disjoint from #2496 (gsap_fullscreen_overlay_starts_visible).

Blockers

(none)

Concerns

(none)

Nits

🟡 extractStandaloneHiddenSelectors regex only recognizes string-literal selectors. /gsap\.set\s*\(\s*(["'])([^"']+)\1\s*,\s*\{([^{}]*)\}/g bails on:

• gsap.set(elementRef, { opacity: 0 }) — variable / captured-node arg
• gsap.set(\#${dynamicId}`, { opacity: 0 })` — template literal
• Nested-object opts ({ opacity: 0, css: { will-change: … } }) — [^{}]* breaks on inner {

All uncommon; the two red-first tests cover the primary literal-selector shape and pass. Worth a follow-up only if a real cold-seek bug slips through with one of the shapes above.

Green notes

🟢 fromPropertyValues extraction wired through — extractGsapWindows reads animation.fromProperties from the parsed timeline node and surfaces it on the GsapWindow shape. Optional (fromPropertyValues?) so downstream rules that don't need from-vars stay unaffected.

🟢 Rule shape mirrors the failure mode. The check requires ALL of (a) selector hidden via CSS OR standalone gsap.set, (b) method === "fromTo", (c) from-vars visible (opacity:1 / autoAlpha:1), (d) destination has no opacity/autoAlpha. Only that intersection reproduces the cold-worker encoded-black case; each guard would independently disqualify a false positive.

🟢 Two red-first pins covering both hidden-state sources. One test asserts the finding fires for CSS-inline style="opacity: 0" + fromTo({opacity:1, ...}, {x:0, ...}); the other asserts it fires for standalone gsap.set("#card", {opacity: 0}) + the same fromTo. Both hidden-state entry paths through the unified hiddenSelectors set are exercised.

🟢 gsap_from_opacity_noop (existing rule) survives. The new fromTo branch continues before the old .from opacity-noop path, so both rules can fire for their respective patterns on the same script without one shadowing the other.

🟢 CSS-hidden-only-early-return removed cleanly. Pre-PR the rule bailed with if (cssOpacityZeroSelectors.size === 0) return findings; — that guard would have blocked the standalone gsap.set-hidden path from ever running. Its removal is what makes the new rule reachable when the ONLY hidden-state source is gsap.set.

🟢 Merge-order note lands correctly. Both #2496 and #2503 touch packages/lint/src/rules/gsap.ts, but at different sites (line ~682 in #2496 vs. lines ~156/197/1138+ in #2503). No logical conflict; a mechanical rebase should suffice if the second lands with drift.

Verdict framing

Clean new lint rule with a tight scope guard + two red-first pins. LGTM from my side.

— Review by Rames D Jusso

[james-russo-rames-d-jusso]

Re-verified at e656a905. Δ from f2e65c16: extractStandaloneHiddenSelectors widened to cover the three uncoded shapes from R1's nit.

Blockers

(none)

Concerns

(none)

Nits

(none)

Green notes

🟢 Variable-alias resolution — the new leading pass over (?:const|let|var)\s+NAME\s*=\s*"..."; builds an aliases map, and when gsap.set(target, …)'s target isn't a string/template literal, the code looks it up in the alias map. Covers the common const CARD_SELECTOR = "#card"; gsap.set(CARD_SELECTOR, {opacity: 0}) shape from the R1 nit.

🟢 Template literals in the selector position now match. The character class widened to (["'\]) in both the alias regex and the selector-literal check inside the loop — simple template literals like `` gsap.set(\#card`, …) resolve. Interpolated template literals (`\`#${dyn}\) still slip through, which is unavoidable at the lint layer without a real JS AST — acceptable.

🟢 Nested-object opts covered by the [\s\S]*?…\}\s*\) shape. The non-greedy body match extends past inner } closings because the pattern requires \}\s*\) afterwards. Traced against gsap.set(el, { opacity: 0, css: { will-change: "auto" } }) — the body capture includes the outer object, and the opacity check finds the top-level opacity: 0 correctly.

🟢 stripJsComments(script) hoisted to a source binding — the alias pass and the gsap.set pass share the same comment-stripped source, so a gsap.set in one line and its aliased declaration on another (both live code) resolve consistently.

Verdict framing

R1 nit closed on the three practical shapes. LGTM from my side.

— Review by Rames D Jusso

[vanceingalls]

Reviewed at e656a9057 (force-push over f2e65c16 with fix(lint): resolve hidden selector aliases). Additive lens on top of Rames R1/R2 — RIGHT verdict aligned, one held-until-X pre-condition + a couple of coverage nits.

Blockers

(none — see held-until below)

Held-until-X

🟡 Format CI red at this head on packages/lint/src/rules/gsap.ts. Job 87483940588 (Format, head_sha=e656a9057, completed 22:08:08Z, conclusion=failure) — oxfmt --check emits packages/lint/src/rules/gsap.ts (4ms) — Format issues found in above 1 files. Old head f2e65c16 had Format=SUCCESS, so the fix-up commit introduced it. Lint=SUCCESS at the same SHA, so oxlint semantics are fine — pure whitespace / line-length. Preflight (lint + format), preview-regression, and regression are the same root cause fanning out. Blocks merge under mergeStateStatus=BLOCKED.

• Fix: bun run format on the touched file (or oxfmt --write .), force-push. Likely candidate is the long matchAll(...) line inside extractStandaloneHiddenSelectors that runs past print-width.

Concerns

(none)

Nits

🟡 Class-defined CSS opacity:0 + id-targeted fromTo isn't cross-referenced. gsap.ts:1172-1180 stores the RAW CSS selector token (.hidden) in cssOpacityZeroSelectors. The inline-style pass at 1182-1189 already cross-populates both #id and .class when a tag has style="opacity:0", but the CSS-rule pass never resolves tags whose class matches a hidden-class. Shape that slips through:

<style>.hidden { opacity: 0 }</style>
<div id="card" class="hidden">…</div>
<script>tl.fromTo("#card", { opacity: 1, x: -60 }, { x: 0 });</script>

hiddenSelectors has .hidden; cssKey for the fromTo is #card; no match → rule silent on a case that still cold-blacks. Follow-up (not blocker): mirror the inline-style expansion for CSS-rule matches — for each class selector with opacity:0 in a <style> block, walk tags whose class list contains it and add their #id to hiddenSelectors.

🟡 No explicit negative pin for the new rule. The two red-first tests at gsap.test.ts:1396 and :1418 assert the finding fires. The nearest surrounding negative (line 1378, does NOT error when gsap.fromTo({opacity:0}, {opacity:1})) checks only the absence of gsap_from_opacity_noop, not gsap_cold_seek_hidden_fromto_missing_reveal. Static-scanning shows the destination-has-opacity/autoAlpha branch behaves correctly, but a future refactor that flips a boolean would silently regress. One-line pin: fromTo(sel, { opacity: 1, x: -60 }, { opacity: 1, x: 0 }) on a CSS-hidden element → assert finding undefined.

🟡 autoAlpha on the from-vars visible-state path is untested. isVisibleGsapState treats opacity and autoAlpha symmetrically (oneValue([opacity, autoAlpha])), so the code is right — both red-first tests happen to use opacity:1. Optional additional pin: fromTo(sel, { autoAlpha: 1, x: -60 }, { x: 0 }) on a hidden element.

🟡 Alias-map has no scope awareness and requires trailing ;. extractStandaloneHiddenSelectors builds a flat aliases Map via (?:const|let|var)\s+NAME\s*=\s*QUOTE...QUOTE\s*;. Recording only: (a) const s = "#card" without semicolon slips through — Prettier-style codebases emit ; by convention; (b) same NAME declared in two scopes maps to a single last-seen selector. Full generality needs a JS AST — the memoed acceptable-at-lint-layer trade-off. No action.

Green notes

🟢 Rule shape mirrors the reported cold-worker failure mode exactly. The new branch at gsap.ts:1204-1221 requires the conjunction (a) sel ∈ hiddenSelectors (CSS opacity:0 OR standalone gsap.set opacity:0/autoAlpha:0 including variable / template-literal aliases), (b) win.method === "fromTo", (c) isVisibleGsapState(win.fromPropertyValues) — from-vars would make the target visible, (d) destination omits both opacity AND autoAlpha. That intersection is the failure surface: cold worker restores authored opacity:0 because the immediateRender path never runs the from-vars, so the destination stays black. Each guard independently disqualifies a false positive.

🟢 Tests exercise the real parse pipeline, not hand-authored AST fixtures. Both new tests call lintHyperframeHtml(html) on raw HTML strings — the production entry point. cachedExtractGsapWindows actually runs on stripJsComments(script). If the underlying parser regressed, the pins would fail.

🟢 No collision with gsap_from_opacity_noop or with #2496's overlay-visibility rule. The new fromTo branch pushes its finding and then continues, so the old .from opacity-noop path (1223-1241) never double-fires on the same window. The overlay-visibility rule that #2496 refines lives in a separate handler around gsap.ts:700-744, guarded by styleLooksFullFrameOverlay + reportedVisibleOverlayKeys — different selector-space, no shared state, no cascade.

🟢 CSS-hidden-only early-return removal is the load-bearing enabler. Pre-PR the rule bailed at if (cssOpacityZeroSelectors.size === 0) return findings; — that guard would have kept the new standalone-gsap.set path unreachable whenever no CSS opacity:0 existed. Its removal is why the second red-first test (standalone gsap.set, no CSS opacity:0) can produce a finding.

🟢 Force-push delta reviewed at e656a9057. Parent = old head f2e65c167, commit message = fix(lint): resolve hidden selector aliases. extractStandaloneHiddenSelectors widened to cover R1's three uncoded shapes (variable alias, template literal, nested-object opts). Traced against gsap.set(el, { opacity: 0, css: { will-change: "auto" } }): non-greedy [\s\S]*?\}\s*\) correctly walks past the inner } because the outer close is followed by ); outer opacity: 0 still detected. Rule branch itself is byte-unchanged from f2e65c16 — previous verification stands.

Verdict framing

RIGHT on correctness; held on Format CI. One bun run format + force-push converts this to APPROVE mechanically. The rule closes the field-signal cold-worker opacity-reveal failure at its exact intersection without opening false-positive risk on adjacent GSAP patterns.

— Review by Via

[james-russo-rames-d-jusso]

Re-verified at 17aa162c. Δ from e656a9057: oxfmt reformat only — long matchAll argument wrapped onto its own line, and the selector ternary collapsed to one line. Regex patterns + logic byte-identical (verified via git diff -w --ignore-blank-lines).

Verdict framing

Format-only push. R1 + R2 LGTM stand. All 40 required checks green.

— Review by Rames D Jusso

[vanceingalls]

Verdict: APPROVE (RIGHT)

Re-verified at 17aa162c4924. Delta from e656a9057 is pure oxfmt reformat — packages/lint/src/rules/gsap.ts +4/-3 across a 7-line diff, one file. Splits the matchAll argument onto separate lines (line 204) and re-joins the alias-resolution ternary (line 213). Zero semantic change to regex bodies, alias map, or gsap.set traversal loop.

Held-until-X condition satisfied: Format lane is now SUCCESS at this head. My prior 4 non-blocker nits remain follow-ups — none block merge:

• class-defined CSS opacity:0 + id-targeted fromTo cross-ref gap
• missing explicit negative pin for the new rule
• autoAlpha from-vars path untested
• alias-map scope/semicolon caveat (acceptable at lint layer)

Rames R3 COMMENTED at same head concurred ("oxfmt reformat only").

Converting my earlier COMMENT-with-RIGHT to APPROVE stamp per the pre-condition being met. reviewDecision on this stamp: REVIEW_REQUIRED → APPROVED.

— Review by Via