WORK-522
ID:WORK-522Status:done

Migrate frame and substrate chrome to postAssemble facets

Move SPEC-086 frame chrome (step 1g) and SPEC-087 substrate fills (step 1h) into facets. Both resolve early but apply late, and to a target chosen at resolve time — the rune root or the [data-section="media"] zone, which does not exist until assembly. They are the canonical two-phase facets, and the second real test of postAssemble after cover.

Priority:mediumComplexity:complexMilestone:v0.30.1Source:SPEC-124
main View source

Criteria completion

Criteria completion: 9 of 9 (100%) checked; tracking started on Aug 27, no incremental history yet0%25%50%75%100%Aug 27Sep 6

Tracking started Aug 27 — check back for trends.

Relationships

Implements 1
Related 7
Branches 3
main current done
claude/transform-package-refactor-7mxi8a doneclaude/spec-124-facet-registry pendingmain done

Acceptance Criteria

  • frame and substrate are facets; steps 1g and 1h and their *MetaProps / *RootDataAttrs locals are gone from engine.ts
  • Target selection (self vs media, including config.frameTarget / config.substrateTarget and the per-instance substrate override) resolves in resolve and applies in postAssemble
  • self-target chrome still contributes its custom properties to the root's inline style, in the same position
  • The {% ref "SPEC-116" /%} frame-overflow="bleed" inert-marker strip on a clip host is preserved, with its warn
  • warnFrameNoTarget, warnFrameOverflowClip and warnSubstrateNoMedia become FacetWarnings with dedupe keys and unchanged messages
  • FRAME_FACET_META moves with the frame facet; SUBSTRATE_CELL and SUBSTRATE_OPACITY move with the substrate facet and are inventoried as hard-coded theme values
  • All 630 pre-existing transform tests pass unmodified, including frames.test.ts, substrate.test.ts and guest-fit.test.ts
  • Unit tests cover target resolution for both facets and the overflow-marker strip
  • npm run build, the full repo suite, and refrakt contracts --check pass

Approach

resolveFrameChrome and resolveSubstrate are already extracted functions returning a chrome bundle, so the resolve half is mostly wiring. The work is in the apply half: today the engine holds frameTargetKind and substrateTargetKind as locals and branches on them at two later points. In the facet model the target is resolution state and the application is postAssemble.

This item will likely want postAssemble to report contributions rather than only mutate — self-target chrome contributes root styles, which the engine currently reads from the chrome bundle after the fact. Decide whether postAssemble gains a return value or whether self-target styles are emitted from resolve (they can be: the root exists at resolve time). Prefer the latter if it works, to keep the second phase mutation-only.

Blocked by

  • WORK-518

Blocks

  • WORK-524

References

  • SPEC-124 — facet registry (the spec this work item realizes)
  • SPEC-086 — surface chrome: elevation and frame presets
  • SPEC-087 — surface fills: substrate patterns
  • SPEC-116 — frame-overflow and the clip-host inert-marker strip

Resolution

Completed: 2026-08-27

Branch: claude/transform-package-refactor-7mxi8a

What was done

  • packages/transform/src/facets/chrome.ts — the shared two-phase chrome contract: Chrome (dataAttrs + styles + consumes), ChromeTarget (self | media | null), ChromeCarry, applyChromeToTag, hasMediaSection.
  • packages/transform/src/facets/frame.ts — the SPEC-086 axis, owning FRAME_FACET_META, preset extends resolution, the guest-fit displace-mode default, and the SPEC-116 overflow-marker strip.
  • packages/transform/src/facets/substrate.ts — the SPEC-087 axis, owning SUBSTRATE_CELL and SUBSTRATE_OPACITY and the target-override precedence.
  • packages/transform/src/engine.ts — steps 1g, 1h and 6c removed along with resolveFrameChrome, resolveSubstrate, applyChromeToTag, the FrameChrome/SubstrateChrome interfaces and three *_WARNED sets. Down to 1889 lines from 2223 before the facet work began.
  • packages/transform/src/helpers.ts — findMediaZone moved here; engine and facets now share one implementation.
  • packages/transform/test/facets/{frame,substrate}.test.ts — 42 new tests covering target resolution for both axes, the overflow-marker strip, both phases, and the carry channel.

Notes

Sequenced ahead of tint deliberately, as the item most likely to push on the interface. It did, three times:

  1. FacetContext.theme — facets had no access to theme preset registries. frame resolves a named preset against the theme's frame registry; tints and backgrounds are declared alongside because the next two axes need the same kind of lookup.
  2. FacetResult.carry — private scratch from a facet's resolve to its own postAssemble. The chrome bundle and its target are neither emitted nor expressible as a string, and state is Record<string, string>. Re-resolving in the second phase would duplicate the work and let the two phases disagree.
  3. postAssemble returns FacetWarning[] | void instead of void — a media-target chrome finding no media zone is only detectable against the assembled tree. An earlier attempt added an ad-hoc ctx.warn() for this and was backed out: it made the two phases inconsistent with each other for no gain, and would have bypassed the collector's dedupe.

Also removed the duplicate findMediaZone — the engine kept its own while helpers gained one during WORK-518, which is exactly the duplication that item's resolution warned about.

Self-target chrome is emitted from resolve (the root exists then) rather than applied in postAssemble, keeping the second phase mutation-only as the item's approach proposed.

All 630 pre-existing transform tests pass unmodified; only the facet tests were updated, for the extended interface.