feat(sdk): stage 4 — canUndo/canRedo, removeElement GSAP cascade, override-set cleanup

This commit is contained in:
Vance Ingalls
2026-06-16 18:37:17 -07:00
parent 8bf516570b
commit 9c0c28725f
+191
View File
@@ -0,0 +1,191 @@
@hyperframes/sdk — Status Report
June 13, 2026
OVERVIEW
The SDK is a headless, framework-neutral composition editing engine for HyperFrames.
It lets hosts (Studio, AI agents, Pacific/HeyGen) read and mutate compositions through
a typed method layer that emits RFC 6902 patches. Work is split into 9 stages.
Stages 1–3a are complete and merged. Stages 3b and the acorn parser stack are code
complete and pending review. An interactive playground (sdk-playground) now exercises
the full op surface end-to-end against a live preview and file-backed persistence.
WHAT IS MERGED (DONE)
Stage 1 — Design Decisions
All three gate decisions locked:
- Brand representation: CSS variables (not data-brand-* attributes)
- Timeline model: per-element start/duration/hold
- Storage model: content-addressed SHA-256 keys
Stage 2 — Foundation (Phase 1)
Package scaffolding, types, GSAP serializer, PersistAdapter contract suite (T13),
three host-archetype usage examples. Fully shipped.
Stage 3 — Browser-safe Parser (Phase 2)
Replaced recast/Babel (Node-only, 500KB) with acorn + magic-string (15KB, browser-safe).
This was the technical gate for all editing operations. Five PRs in a Graphite stack:
- #1338 lint rule: warn on missing data-no-timeline
- #1368 acorn read path + differential tests
- #1369 acorn write path (magic-string offset-splice)
- #1370 parse-parity suite (recast vs acorn agreement)
- #1392 swap studio-api files.ts from recast to acorn
Stage 3a — Session API (Phase 3a)
Core session: document model, dispatch/patch loop, history coalescing,
PersistAdapter queue, optional history module. Merged PR #1325.
Gesture-to-Keyframes Stack
The prerequisite for Studio migration step 3 (edit ops). PRs #1301 and #1311 merged.
WHAT IS CODE COMPLETE, IN REVIEW (latest first)
Stage 4 (partial) — can() + T4 tests + FsAdapter T13 [June 13, PR #1425/#1426]
can() now returns structured CanResult ({ok:true} | {ok:false, code, message, hint?})
instead of a bare boolean. validateOp() in mutate.ts matches the same shape.
CAN_OK constant + canErr() helper keep callers concise.
T4 dispatch-boundary tests (session.dispatch.test.ts) cover:
- patch events fire after dispatch
- override-set pipeline via setVariable
- can() CanResult shape for valid/invalid ops
- batch() applies all ops and emits a single patch event
- addGsapTween round-trip via session
- custom origin forwarded through patch event
FsAdapter flush() was a silent no-op: in-flight writes could be abandoned before
they settled, causing a torn write on fast save-quit. Fixed with inflightWrites:
Set<Promise<void>> — flush() now awaits all in-flight promises. Also fixed a
same-millisecond version-key collision with a versionCounter monotonic suffix.
FsAdapter is now exercised by the T13 PersistAdapter contract suite, closing the
gap between the stub and the contract.
Stage 3b — GSAP Editing Engine (Phase 3b)
9 new SDK operations shipped in PR #1379:
addGsapTween, setGsapTween, removeGsapTween
addGsapKeyframe, setGsapKeyframe, removeGsapKeyframe
addLabel, removeLabel
setClassStyle
A code review (June 12) found and fixed 6 bugs before the stack merges:
1. Program-scope variable bindings were silently lost (querySelector at top
level was never resolved). Fixed with a null-key fallback in the scope chain.
2. fromTo argument guard was too loose (could read undefined args[2]/args[3]).
3. removeAnimation had a fuzzing fallback that silently deleted the wrong
animation by converting -from- IDs to -to- IDs. Removed.
4. stagger property was handled in addGsapTween but missing from setGsapTween.
5. apply-patches script case had no handler for op=remove.
6. valueToCode had no NaN guard and used a wrong regex for property key safety.
setTiming GSAP-sync fix (June 13, found via playground)
setTiming was a silent no-op for animation timing on GSAP compositions: it
stamped data-start/data-end on the DOM node, but the runtime re-stamps those
attributes FROM GSAP positions on next init, so the edit was overwritten.
handleSetTiming now also rewrites the GSAP script (parseGsapScriptAcornForWrite
+ updateAnimationInScript), flushing both models in one patch pair so they stay
in sync. DAW-style trim in the playground depends on this.
Stack is awaiting re-stamp from reviewers (Miguel on #1379, Rames on #1368/#1370).
Stage 5 — Adapters (partial, in playground)
fs PersistAdapter is now implemented (was a stub): node:fs/promises read/write,
timestamped version history under .hf-versions/<path>/, prune to maxVersions (20),
listVersions/loadFrom. Still needs to be run against the T13 contract suite.
A fetch-based PersistAdapter (browser → Vite dev-server endpoints) provides the
HTTP shape ahead of the real post-Pacific HTTP adapter.
A concrete PreviewAdapter (PlaygroundPreview) implements select/on('selection')
and the draft/commit/cancel stubs — first real impl beyond the contract.
S3 + production HTTP adapters and the headless null PreviewAdapter remain.
Stage 8 — Packaging and DX (partial)
sdk-playground shipped: interactive browser harness over the full op surface
(setStyle/setText/setAttribute/removeElement/setVariableValue/find/selection
proxy/all 9 GSAP+label ops/setClassStyle), live preview iframe with click-select
and drag-to-reposition, DAW-style timeline trim via setTiming, file-backed
persistence with version history, undo/redo/can/getOverrides/flush, raw-HTML
editor modal. README documents built vs planned.
@hyperframes/editor drop-in, CDN bundle, npm create scaffold, and the docs
ladder remain.
WHAT IS NOT STARTED
Stage 4 remainder
removeElement v1 — cascade rules, override-set removal, soft/hard op classification,
inverse patch carrying full serialized subtree.
createHistory — 300ms coalescing, origin-scoped.
Stage 5 remainder — Adapters (Phase 4)
S3 and production HTTP PersistAdapters (post-Pacific).
PreviewAdapter: headless null adapter for agents and CI.
Stage 6 — Advanced (Phase 5)
Sub-composition editing via scoped element IDs.
Soft vs hard op classification (reversible vs destructive).
Testing utilities for hosts building on the SDK.
Stage 7 — Studio Migration
5-step feature-flag migration. Each step can be rolled back independently.
Step 1: persistence
Step 2: selection
Step 3: edit ops (commit handlers become dispatch calls — absorbs the old R5 track)
Step 4: history
Step 5: App.tsx collapse
Canvas, panel, and timeline components land in a new @hyperframes/react package.
This stage is roughly as much work as everything shipped so far.
Stage 8 remainder — Packaging and DX
@hyperframes/editor: 5-line drop-in component plus CDN bundle.
npm create scaffold (playground itself is shipped — see above).
Full documentation ladder (quickstart, guides, generated reference, architecture explanation).
Stage 9 — Pacific / AI Studio Integration
HyperframesElement draft type, draft_to_edit classification.
heygen_video_workflow fan-out to hyperframes_producer_activity (transparent WebM at
graphic-clip level).
Embedded session wiring, word-aligned timing with shared resolver.
SEPARATE TRACKS
Runtime bridge freeze (R6)
Rename window.__* globals to versioned __HYPERFRAMES.* namespace.
Gated on Phase 1 (complete). Not yet scheduled.
Write-path migration (T6f)
The read path now uses acorn everywhere. The write path (convertToKeyframes,
removeAllKeyframes, setArcPath, etc.) still uses recast. Full migration is the
next post-merge task for the parser team.
GSAP plugin support branch
feat/gsap-plugin-support: code complete, 95 tests, no PR ever opened.
Branch is 79 days old and may be superseded by the acorn parser work.
Decision needed: open a PR, archive, or rebase onto current stack.
ROUGH PROGRESS
Stages complete or in review: 1, 2, 3 (all), 3a, 3b, 4 (partial: can/T4/FsAdapter)
Stages partially done: 5 (fs + fetch adapters, PreviewAdapter impl),
8 (playground shipped)
Stages not started: 4 (remainder: removeElement, createHistory), 6, 7, 9
Current stack is approximately 40% of the full roadmap by stage count.
Stage 7 (Studio Migration) and Stage 9 (Pacific) are the largest remaining efforts.
OPEN QUESTIONS FROM PRD
1. commitPreview() location — recommendation: core session API over PreviewAdapter
2. setHold surface — recommendation: dispatch op
3. Render granularity: per-element WebM now, per-scene layers later?
4. Animation markers: expose GSAP labels for host-level sequencing?
5. North star: model the entire Pacific draft as one SDK document?
CONTACTS
Acorn parser / Phase 3b: Vance
Studio migration: TBD
Pacific integration: TBD