ANIM-b slice 2: -webkit- transition/animation aliases, @-webkit-keyframes, and web frame fixtures for easings, direction and fill backwards

[thejackshelton]

ANIM-b slice 2 (T065), web side: the -webkit- transition and animation aliases, @-webkit-keyframes, and Chrome frame fixtures that prove the easing, direction and fill-mode values the web profile was missing. It stays out of the state-program, the driver and state.ts (the native animation lane, slice 1, owns those).

What changed

• Aliases. packages/dragon/src/css/aliases/animation.ts adds 14 aliases, one sorted line in aliases.ts. These are Chrome 145's alias_for entries (css_properties.json5 lines 9700-9733 and 9928-9945):

  • -webkit-animation and -webkit-animation-{delay, direction, duration, fill-mode, iteration-count, name, play-state, timing-function};
  • -webkit-transition and -webkit-transition-{delay, duration, property, timing-function}.

  None is parsed with UseAliasParsing: in css_parsing_utils.cc, shorthands_custom.cc and longhands_custom.cc it is used only for background, mask and perspective. So each alias parses exactly as its property. They use ALIAS: legacy -webkit- aliases of supported properties resolve as in Chrome (44 aliases) #172's mechanism. aliases.ts now also accepts the transition and animation list properties as alias targets, since those stay out of LONGHANDS (T065 option B).

• @-webkit-keyframes is registered as keyframesAtRule, so it parses as @Keyframes, as Chrome's ConsumeKeyframesRule(webkit_prefixed) does (css_parser_impl.cc 895-898 and 1531-1581, a new ports.json entry).

  • KeyframesRule.prefixed records which form was written.
  • keyframesOverride in analysis/animations.ts applies Chrome's KeyframeStyleShouldOverride: an unprefixed rule beats a prefixed one of the same name in either order, and otherwise the later rule wins. Dragon refuses @layer, so all rules share one layer.
  • It has its own feature, at-rule:@-webkit-keyframes, so the row names the fixture that proves it.
  • Refusals name the rule as written, for example @-webkit-keyframes inside @media is not supported (package MQ-R).

• Frame fixtures (group animations, picked up by the anim-* glob, ltr and rtl at 400x300, captured from Chrome by pnpm regen's anim-capture):

  • anim-easings: transitions with ease-in, ease-out, ease-in-out, step-start and step-end, plus the longhand form. The state is set, then cleared, so each runs both ways.
  • anim-timing: animations with every easing: linear, ease, ease-in, ease-out, ease-in-out, an overshooting cubic-bezier(), steps(4, jump-none), steps(3, jump-both), step-start and step-end, plus the longhand form.
  • anim-direction: alternate, reverse, and fill-mode backwards with positive delays. This includes reverse backwards, where the before-phase holds the to value, and alternate backwards with three iterations.
  • anim-webkit: every alias, an alias and its property in the same rule in both orders, and the @-webkit-keyframes precedence. That covers an unprefixed rule before and after a prefixed one, and two prefixed rules.

Rows promoted (web, animation context, all exact)

• animation-direction:alternate, animation-direction:reverse
• animation-fill-mode:backwards
• animation-timing-function:cubic-bezier(), :ease-in, :ease-out, :ease-in-out, :steps()
• transition-timing-function:ease-in, :ease-out, :ease-in-out
• at-rule:@-webkit-keyframes

That is 12 new rows. Existing rows gain the new fixtures as further proof cases, and no status changes. step-start and step-end key steps(), which anim-steps already proved for transitions; anim-easings and anim-timing add cases for them. iOS and Android rows wait for the device run (the driver's job).

Music-player north-star delta

None: examples/music-player/dragon/north-star-check.json did not change in the regen.

What passed

• pnpm typecheck.
• pnpm regen via /tmp/job.sh animb2-regen: fixed point after 4 passes. I then ran pnpm notices:gen for the new ports entry.
• vitest, 115/115 pass across 10 files: dragon aliases, seams, registry-claims, keyframes, anim-program, animations and animation-kinds; parity anim-frames (the host frame lanes and chrome-dual against the new captures), chrome-ports and registry-claims.
• The full pnpm test was not run, per the lane contract.

Changed tests, and why

• dragon/test/aliases.test.ts:
  • The alias count goes from 44 to 58.
  • Sample values were added for the 14 animation targets.
  • The parse comparison now also compares Declaration.animation, since an animation declaration's longhands are empty.
  • The fixture-coverage check also reads anim-webkit/view.css for the animation aliases.
  • New tests for @-webkit-keyframes: it parses as @Keyframes in any case, it is refused where @Keyframes is, and it follows the AddKeyframeStyle precedence.
• dragon/test/registry-claims.test.ts: an alias family must still be a property family. The one exception is animation, whose own properties are ANIM_LIST_PROPERTIES plus the ANIM_SHORTHANDS keys. The no-alias-is-a-property check now covers those too, so the check is wider, not looser.
• dragon/test/seams.test.ts:
  • -webkit-keyframes joins keyframes as accepted.
  • The byte-identical pin drops -webkit-keyframes from its list, as ANIM-b1 dropped keyframes. The new hash, a0302641…, is what the base a36ce22 gives for the same list. I ran it on the base, so every other at-rule's refusal is unchanged.
• dragon/test/keyframes.test.ts: "@-webkit-keyframes stays refused" becomes "a valid one is accepted". A nameless one (none) and one inside @layer are still refused.
• dragon/test/animations.test.ts: the band-gate test used ease-in-out, which this PR proves. It now uses transition-delay: inherit, which is still unproven, and keeps its intent.

Outside the spec

• docs/ports.json: css_parser_impl.cc entry (and THIRD_PARTY_NOTICES.md).
• packages/dragon/src/css/at-rules.ts: one registry line (MQ-R1 1/2 (T067): @media bands for native in the compiler, the band runtime reference and Chrome resize traces #140 also edits this file, in another line).
• packages/dragon/src/analysis/animations.ts: the keyframes map and the feature key (INL1a 5/5 (C2): lowering inline boxes and <br>s, native runtime emit, 7 inline fixtures and lanes #91, a stacked PR, also edits this file).

Reviewed diff: about 52 KB. Everything else is regen output under .macroscope/ignore.md shapes.

🤖 Generated with Claude Code

[thejackshelton]

Fixed the landing-proof failure in packages/parity/test/device-vectors.test.ts ('expected 26 to be 18'): the animator suite's declared count is now read from packages/layout/rt-vectors/animator/cases.json (26 with this PR's four frame fixtures; it stays >= 18), and a short animator run is checked to fail, alongside the existing undeclared-suite check. device-vectors.test.ts passes (8/8). Test-only change, no regen needed.

[thejackshelton]

Landing stopped at step test (pnpm land)

packages/parity/test/device-vectors.test.ts, rerun alone on a quiet machine, exited 1 (log /tmp/land-201-test-solo-1.log)
 ❯ packages/parity/test/device-vectors.test.ts:52:86
     50|     expect(hit).toBeGreaterThanOrEqual(p1Floor.p1.counts.hit);
     51|     // ANIM-b1 3b: the animator suite is declared with one case per fr…
     52|     expect(declared.find((d) => d.corpus === 'p1' && d.suite === 'anim…
       |                                                                                      ^
     53|     const shortHit = whole.map((s) => (s.corpus === 'p1' && s.name ===…
     54|     expect(judgeDeviceVectors(t, shortHit, digests, host)).toMatchObje…

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/2]⎯


 Test Files  1 failed (1)
      Tests  2 failed | 6 passed (8)
   Start at  23:53:12
   Duration  1.75s (transform 1.12s, setup 0ms, import 1.67s, tests 7ms, environment 0ms)

Found by bisecting the batch #140 #201 #197 (3 proofs): master with #140 passes, adding #201 fails.

Fix the cause, then hand #201 back to the landing queue with the new clean head.