feat: blob-tracking layer foundation — CPU tracker + inner-effects modules (plan 008 S1.2/S2.1)#101
Closed
git-chad wants to merge 1 commit into
Closed
Conversation
The two pure, browser-independent modules plan 008 isolates as its independently-landable foundation, with full unit-test coverage: - src/lib/blob-tracking/tracker.ts — CPU blob tracker (S1.2), zero DOM/three imports so it runs under bun test and mirrors into the package unchanged: motion/luminance binarization with auto fallback, 4-connected component labeling, top-N by area, persistent greedy nearest-neighbor matching with EMA smoothing, grace despawn, monotonic ids, position-history ring, reset(), and a once-per-distinct-time step guard for export determinism - src/lib/blob-tracking/inner-effects.ts — inner-effect set (S2.1: effect types minus blur) with tolerant parse/serialize of parameter overrides - 19 tests covering the full S1.2 and S2.1 case lists Both modules are intentionally unwired: no blob-tracking layer is registered, so editor behavior is unchanged and no unverified option appears in the picker. The GPU/DOM stages (BlobTrackingPass, decorations, export determinism, package mirror) are deferred — they need a WebGPU browser this run had no access to. See plans/008-artifacts/status.md, which also records the S1.1 readback-API spike result (API present in three@0.183.2; no STOP). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
git-chad
added a commit
that referenced
this pull request
Jul 7, 2026
* chore: fix pre-existing lint errors Biome lint fails on the unmodified tree with 4 errors and 7 warnings, which blocks every CI/verification gate that requires a clean lint run: - convert type-only `three/webgpu` namespace imports to `import type` in four package pass files (useImportType errors) - remove unused biome-ignore suppression comments in both three-tsl.d.ts ambient declaration files (suppressions/unused) - comment the two intentionally empty callbacks (noEmptyBlockStatements) - simplify the splat() guard expression in the fluid runtime (useSimplifiedLogicExpression; boolean-algebra equivalent) The two useExhaustiveDependencies warnings in properties-sidebar.tsx are left as-is: the "extra" dependencies are intentional effect triggers and removing them would change behavior. Warnings do not fail the lint gate. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * test: add baseline tests for project-file, parameter-schema, timeline and composition The repo had zero test files while `bun test` has been wired up in package.json through its entire history. This adds the first suite over the pure logic that user data flows through: - parseLabProjectFile: valid v1/v2 parse, clone semantics, and every validation error message - parameter-schema: clone semantics, default building, equality, and signature stability/order-independence - evaluateTimelineForLayers: linear/step/color/vec2 interpolation, clamping outside the keyframe range, disabled/unknown-layer tracks, property bindings and multi-track merging - composition crop math: centered crops, degenerate canvases, and frame intersection (including the 1x1 clamp for disjoint frames) 50 tests, no source changes. Plan 001 (#84). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * ci: validate editor app lint, types and tests in CI CI previously only typechecked and built packages/shader-lab-react — the ~230-file editor app was never linted, typechecked, or tested. Add the three app validation steps after the package build (the app's tsconfig resolves @basementstudio/shader-lab from the package's dist/, so the build must run first). Uses typecheck:tsc (stock TypeScript) rather than typecheck (tsgo preview build) so CI gates on the stable toolchain. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix: close video encoder when export fails or is cancelled Add an idempotent close() to VideoExportEncoder and call it from exportVideo's finally block whenever finalize() did not complete, so aborted or failed exports release the WebCodecs hardware encoder instead of leaking it. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix: time out stalled video texture loads createVideoTexture now rejects after 30s if neither the playing event nor an error settles the promise, detaching the error handler and releasing the video element. Resolve/reject are wrapped in settle helpers that clear the timer and guard against double settlement. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix: only invoke onRuntimeError with null after a previously reported error ShaderLabComposition called the consumer onRuntimeError callback with null after every successful renderer initialization (and after every successful shader recompile via the internal error channel), because null was used as a success/cleared signal on the same path as real errors. Consumers following the README example logged "null" on every clean init, and apps mirroring the message into UI state showed a permanent false error. The composition now tracks the last message actually delivered to the consumer in a ref that survives config-change re-inits: - null on a clean init is a no-op toward the consumer (internal error overlay state is still cleared) - after a real error was delivered, resolving it calls the consumer with null exactly once so error UIs can recover - consecutive identical errors and repeated clears are deduplicated Also documents the callback semantics in the onRuntimeError JSDoc and the README error-handling example, and adds a patch changeset. Fixes #81 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * perf: lazy-load export dialog and video muxers Cut the video-export stack out of the editor route's first load: - editor-topbar.tsx: load EditorExportDialog via next/dynamic (ssr: false) and only mount it after the export dialog has been opened once, so the ~46 KB dialog chunk is fetched on demand. - video-export-encoder.ts: drop the static mp4-muxer/webm-muxer imports; createMuxer is now async and dynamically imports only the muxer for the chosen format (per-format loading), deferring ~58 KB more until an export actually runs. First Load JS for /tools/shader-lab: 2656.0 KB -> 2556.1 KB (measured by summing script chunks referenced by the prerendered route HTML; Next 16 Turbopack no longer prints size columns). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix: validate .lab imports with zod Replace the manual shape checks in parseLabProjectFile with a zod v4 schema that validates layers (discriminated by kind, enum-checked type, param value types), timeline tracks/keyframes, assets and composition. Unknown keys pass through via looseObject so files with extra data keep importing, and fields added to BaseLayer after the .lab format shipped (maskConfig, fluidInteractionEvents, runtimeError) stay optional for legacy files. Schema issues map to the exact pre-existing user-facing error strings. Also adds hasImportedCustomShaderCode, which detects custom-shader layers whose sourceCode differs from the built-in starters, plus four new validation test cases. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat: require confirmation before running imported custom shader code Gate .lab project import behind a window.confirm when the parsed file contains custom-shader layers with non-starter sourceCode; cancelling aborts the whole import before applyLabProjectFile touches any store. window.confirm is a stopgap noted for a design follow-up. Also cap parseSvgRasterResolution at 8192 to close the unbounded-canvas DoS from untrusted svgRasterResolution params. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * chore: add renderer drift inventory and consolidation design (plan 006) Deliverables for the renderer-deduplication spike: - plans/006-artifacts/drift-inventory.md — machine-generated classification of all 90 files across the four trees (script committed alongside); normalization covers both import-specifier style and biome formatting so only logic drift counts - plans/006-artifacts/consolidation-design.md — source-of-truth decision (package), consumption mechanism (fluid-style source re-export shims), type-unification analysis, phased migration order, and the CI guardrail The one-pass pilot (step 3) hit the plan's STOP condition and was reverted: the app's ambient module declaration types every package import as a silent `any` (probed empirically), and the package PassNode lacks three methods the app pipeline calls on every pass every frame — a package-sourced pass type-checks vacuously and throws at runtime. The design doc's Phase 0 (base-class reconciliation) is the recommended follow-up plan; after it the pilot pass becomes a two-line shim. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * perf: replace localeCompare in parameter signatures parameterValuesSignature runs per layer per frame in the render loop's change-detection path. Parameter keys are plain ASCII identifiers, so lexicographic comparison is equally deterministic and injective while avoiding ICU collation on every comparison. Signatures are ephemeral (compared only against the previous frame, never persisted) and the only callers are pipeline-manager.ts and tests, so the ordering change is safe. The comparator is extracted to a compareKeys helper because an inline nested ternary violates the repo's noNestedTernary rule. This is step 2 of plan 007 (#90) — the unconditional micro-fix. The structural identity fast-paths (step 4) are profile-gated and were NOT applied: this run had no WebGPU-capable browser to produce the required measurements or the stale-canvas regression matrix. See plans/007-artifacts/profile.md for the gate rationale and the follow-up procedure. A stop after step 2 is an explicitly valid plan outcome. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat: add blob-tracking tracker and inner-effects foundation (plan 008) The two pure, browser-independent modules plan 008 isolates as its independently-landable foundation, with full unit-test coverage: - src/lib/blob-tracking/tracker.ts — CPU blob tracker (S1.2), zero DOM/three imports so it runs under bun test and mirrors into the package unchanged: motion/luminance binarization with auto fallback, 4-connected component labeling, top-N by area, persistent greedy nearest-neighbor matching with EMA smoothing, grace despawn, monotonic ids, position-history ring, reset(), and a once-per-distinct-time step guard for export determinism - src/lib/blob-tracking/inner-effects.ts — inner-effect set (S2.1: effect types minus blur) with tolerant parse/serialize of parameter overrides - 19 tests covering the full S1.2 and S2.1 case lists Both modules are intentionally unwired: no blob-tracking layer is registered, so editor behavior is unchanged and no unverified option appears in the picker. The GPU/DOM stages (BlobTrackingPass, decorations, export determinism, package mirror) are deferred — they need a WebGPU browser this run had no access to. See plans/008-artifacts/status.md, which also records the S1.1 readback-API spike result (API present in three@0.183.2; no STOP). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Collaborator
Author
|
Consolidated into #104. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Addresses #91 (Plan 008). This lands the pure, fully unit-tested foundation the plan deliberately isolates as independently shippable; the GPU/DOM stages are scoped as a documented follow-up (rationale below).
What's in
src/lib/blob-tracking/tracker.ts(S1.2) — the CPU tracker: zero DOM/three imports (runs underbun test, mirrors into the package unchanged). Motion/luminance binarization withautofallback (30 static steps → luminance, re-arms on motion), 4-connected component labeling, top-N by area, persistent greedy-NN matching + EMA smoothing + 10-frame grace despawn + monotonic ids + 16-entry position-history ring,reset(), and the once-per-distinct-timestep guard that makes export prewarm idempotent.src/lib/blob-tracking/inner-effects.ts(S2.1) — the inner-effect set (EFFECT_LAYER_TYPESminusblur) and tolerantparseInnerEffectParams/serializeInnerEffectParams(bad JSON → defaults, unknown keys dropped, type-mismatched values ignored).Both modules are unwired — no
blob-trackinglayer is registered, so editor behavior is unchanged and no half-working option appears in the picker.S1.1 readback spike
readRenderTargetPixelsAsync(rt, x, y, w, h)is present in the installedthree@0.183.2(Renderer.js:2884, returns a Promise viacopyTextureToBuffer) — matches the plan's assumption, STOP condition does not trigger.What's deferred, and why
S1.3–S1.5 (registration +
BlobTrackingPassGPU pass + export-determinism), S2.2–S2.4 (child pass + sidebar + 22-type matrix), S3 (decoration overlay), S4 (package mirror + changeset + README + preview). These are GPU/DOM-bound and unverifiable in this automated run — no WebGPU browser is reachable (chrome tooling pinned to Chromestable, not installed). Shipping a readback pass, temporal export stepping, and a decoration canvas unverified risks precisely the failure modes the plan flags (nondeterministic export, decorations leaking into the mask, child-pass GPU leaks). The tracker API (step/getBlobs/reset) is shaped to be exactly whatBlobTrackingPasswill consume, so this is genuine unblocking work, not throwaway. Full deferral map inplans/008-artifacts/status.md.Verification
bun test→ 71 pass (19 new) ·bun run typecheck:tsc,bun run lint→ exit 0App-only, no layer registered → no changeset. Stacked on #93 (Plan 001) → #92. Merge order: #92 → #93 → this.
🤖 Generated with Claude Code