[iwoplaza]

Screen.Recording.2026-09-06.at.17.31.51.mov

[github-actions]

pkg.pr.new

No packages changed since release

[github-actions]

Bundle size comparison (import * as ... in PR vs import * as ... in target):

🟢 Decreased	➖ Unchanged	🔴 Increased	❔ Unknown
0	356	0	0

No notable changes.

If you wish to run a comparison for other, slower bundlers, run the 'Tree-shake test' from the GitHub Actions menu.

[github-actions]

Resolution Time Benchmark

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Random Branching (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.72, 1.47, 3.30, 5.70, 5.87, 9.57, 19.84, 22.33]
  line [0.85, 1.64, 3.66, 5.47, 6.12, 8.68, 20.84, 20.97]
  line [0.83, 1.68, 3.50, 5.64, 5.84, 10.47, 21.56, 22.56]

Loading

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Linear Recursion (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.19, 0.36, 0.55, 0.66, 0.95, 0.97, 1.14, 1.22]
  line [0.19, 0.38, 0.52, 0.66, 0.92, 0.96, 1.18, 1.30]
  line [0.23, 0.43, 0.58, 0.72, 0.93, 0.91, 1.10, 1.25]

Loading

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Full Tree (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.69, 2.04, 3.04, 6.31, 10.78, 22.64, 48.36, 96.44]
  line [0.83, 1.99, 3.43, 6.00, 10.98, 22.02, 48.02, 96.52]
  line [0.79, 1.91, 3.27, 6.17, 11.25, 23.93, 47.29, 97.52]

Loading

[Copilot]

Copilot review overview

🟡 Changes recommended

Quality state races and incorrect device-motion axis mapping cause observable behavior errors.

Review effort: Balanced
Findings: 3

Open (3)

• Initial quality display does not match rendering resolution · New
• Stale benchmark results overwrite newer quality selections · New
• Device orientation axes are mapped incorrectly · New

What changed in this PR

Adds an advanced “Jelly Knob” WebGPU example featuring ray-marched rendering, spring physics, motion input, and configurable quality.

Changes:

• Implements the knob’s rendering, materials, TAA, camera, and physics.
• Adds interaction, phone motion, quality controls, and resource cleanup.
• Registers the example and adds lifecycle tests.

File	Description
apps/​typegpu-docs/​tests/​individual-example-tests/​jelly-knob-lifecycle.test.ts	Tests spring behavior and resource cleanup.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​utils.ts	Adds rendering and texture helpers.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​taa.ts	Implements temporal anti-aliasing.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​spring.ts	Implements scalar and 3D springs.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​sdfs.ts	Defines scene SDF geometry.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​raymarchers.ts	Implements ray-marched shading.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​motion.ts	Handles phone motion input.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​meta.json	Provides example metadata.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​material.ts	Defines jelly reflection behavior.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​knob.ts	Manages knob state and physics.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​index.ts	Integrates rendering, controls, input, and cleanup.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​index.html	Adds the example canvas.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​dataTypes.ts	Defines GPU data structures and layouts.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​constants.ts	Defines rendering and physics constants.
apps/​typegpu-docs/​src/​examples/​rendering/​jelly-knob/​camera.ts	Implements camera projection and jitter.
apps/​typegpu-docs/​src/​components/​ExampleShowcase.tsx	Registers advanced showcase examples.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[pullfrog]

ℹ️ No blocking issues — one minor robustness suggestion inline. Two of the existing Copilot findings appear to be false positives.

Reviewed changes

This run read the full 17-file diff of the new jelly-knob example, ran astro check (0 errors), oxlint --type-aware (clean), and the new lifecycle test (3/3 pass), and validated each open Copilot finding.

• New example — jelly-knob/ ray-marched knob: SDF scene, TAA + blue-noise dither, spring physics, phone-motion input, and controls, registered as an advanced example.
• Lifecycle — handleResize destroys and recreates render + TAA textures and bind groups; onCleanup tears down the root and removes its window listener.
• Tests — jelly-knob-lifecycle.test.ts pins spring convergence/substep parity and texture/listener cleanup; it also exercises shader generation indirectly via import + render.
• Showcase list — registers rendering--jelly-knob and, unrelated to this example, rendering--triplanar-mapping.

ℹ️ Two of the prior review's findings don't hold up

Copilot's review raised three points; two are incorrect and shouldn't drive changes.

• Initial quality vs. rendering scale — the controls runner eagerly invokes each control's handler with its initial value on mount (initializeParam in apps/typegpu-docs/src/utils/examples/exampleRunner.ts:7, called from executeExample). So onSelectChange('Ultra') runs handleResize() with qualityScale = 1.0 right after import, and the displayed selector matches the render resolution.
• rotationRate axes — per W3C Device Orientation and Motion §6.3.2, the DeviceMotionEventRotationRate getters are alpha → x-axis, beta → y-axis, gamma → z-axis ("The alpha getter steps are to return the value of this's x axis rotation rate"). motion.ts maps them in exactly that order, and its comment is correct. Copilot's alpha = Z mapping is the DeviceOrientationEvent Euler-angle convention, which does not apply to rotationRate.

Technical details

# Prior-review findings assessment

## Affected sites
- apps/typegpu-docs/src/utils/examples/exampleRunner.ts:7-26 — `initializeParam` dispatches `onSelectChange`/`onToggleChange`/... with `param.initial`; `executeExample` calls it for every control.
- apps/typegpu-docs/src/examples/rendering/jelly-knob/motion.ts:77-85 — passes `alpha, beta, gamma` as x, y, z.

## Required outcome
- Do not change the `Quality` initial value or the `motionToWorld(alpha, beta, gamma, ...)` ordering on account of these two findings.

( ◦°^°◦) 🔍

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using DeepSeek Flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

This run re-reviewed the commits pushed since the prior Pullfrog review (c9b87ccf → f2a48984): a cancellation guard for the Auto-quality benchmark, touch-cancel handling, a shared camera-position constant, and a far-plane fix.

• Superseded Auto benchmark is cancelled — Quality.onSelectChange tags each selection with a monotonic qualityRequestId and a disposed flag; a late autoSetQuality result neither writes qualityScale nor calls handleResize(), and the benchmark loop breaks early once cancelled. This closes the prior inline finding.
• Touch cancel handled — touchend and touchcancel now share handleTouchEnd, so a cancelled gesture releases the knob.
• Camera position centralised — a new CAMERA_POSITION constant replaces the duplicated (0, 2.7, 0.8) literal in index.ts and motion.ts, which now derives its basis from it (same math).
• Projection far aligned — CameraController.updateProjection default changed 100 → 10, matching the constructor and MAX_DIST = 10.
• Unrelated showcase entry dropped — the branch no longer adds rendering--triplanar-mapping to ADVANCED_EXAMPLE_KEYS.
• Tests extended — the lifecycle test now covers the superseded-Auto and cleanup-cancel paths and restores globals via vi.unstubAllGlobals().

( ◦°^°◦) 🏴☠️

Ahoy! I re-reviewed the fresh commits on the Jelly Knob PR: the Auto-quality race be fixed with a request-id guard, touch-cancel be handled proper, and the camera position be shared twixt modules. No new leaks nor wrecks spotted — she be ready to sail, matey!

  ^{｜ View workflow run ｜ Using DeepSeek Flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

This run re-reviewed the commit pushed since the prior Pullfrog review (f2a48984 → 583043e0). The only authored delta is the default Quality selection; the remaining 17 files are byte-identical to the previously reviewed version.

• Default quality is now Auto — Quality.initial changed from 'Ultra' to 'Auto' in jelly-knob/index.ts. On mount the controls runner eagerly invokes onSelectChange('Auto'), which starts the adaptive benchmark and applies its scale (or the 0.5 fallback when timestamp-query is unavailable). The existing qualityRequestId/disposed guard already covers this async completion path, so no lifecycle regression.

Verification: jelly-knob-lifecycle.test.ts 3/3 pass; oxlint --type-aware on the changed files is clean.

( ◦°^°◦) 🏴‍☠️

Ahoy! I re-reviewed the fresh commit on the Jelly Knob PR: the default quality be now Auto, settin' the adaptive benchmark to run on mount. The old request-id guard still be holdin' her steady — no new leaks nor wrecks spotted, so I gave her the green flag. The prior review's one thread already be resolved, matey!

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

This run re-reviewed the commits since the prior Pullfrog review (583043e → c6f9d0d). The one authored code change swaps a precomputed blue-noise texture for a GPU-side implementation; the rest of the range is a rebase onto main.

• GPU-side blue noise replaces the precomputed texture — new jelly-knob/blueNoise.ts ports hash12 (Fi-hash white noise) and blue12 (high-pass over the 3×3 neighborhood, which simplifies to center - avg_neighbors); index.ts now dithers with blue12(std.floor(input.position.xy)) instead of sampling a 64×64 blueNoise2d texture.
• @typegpu/noise no longer needed for dither — the blueNoise2d import is gone (the package is still used for perlin3d in raymarchers.ts), so the example no longer depends on the blue-noise work from the base branch.
• Rebase onto main — the example is byte-identical to the prior review (jelly-knob/ tree dfd3025d, lifecycle-test blob 33c230e, and the rendering--jelly-knob registration all unchanged); only blueNoise.ts and index.ts differ.

Verification: jelly-knob-lifecycle.test.ts 3/3 pass; oxlint --type-aware on the two changed files is clean; resolving blue12 through tgpu.resolve emits valid WGSL.

ℹ️ docs/jelly-knob no longer descends from its base branch

docs/jelly-knob was rebased onto main, so it no longer contains the PR base feat/typegpu-noise-blue-noise (755f9a082). GitHub now renders this PR as 329 changed files / +20,716 — essentially all of main since the merge-base (743c3b31) plus the example — even though the PR only authors the jelly-knob example. Since the example no longer uses blueNoise2d from the base branch, retargeting the PR to main (or rebasing onto the current base tip) would keep the diff scoped to the example.

Technical details

# Base branch is no longer an ancestor of the head

## Affected sites
- PR metadata: base `feat/typegpu-noise-blue-noise` (`755f9a082`), head `docs/jelly-knob` (`c6f9d0d`).
- `git merge-base 755f9a082 c6f9d0d` → `743c3b31`; `git rev-list c6f9d0d` does not contain `755f9a082`.
- `apps/typegpu-docs/src/examples/rendering/jelly-knob/` tree `dfd3025d` is identical at `583043e` and `9ad34b1c`; only `blueNoise.ts` and `index.ts` differ at head.

## Required outcome
- The PR diff should reflect only the example's changes, not the intervening `main` history.

## Suggested approach (optional)
- If the example is meant to land independently, change the PR base to `main`.
- Otherwise rebase `docs/jelly-knob` onto the current `feat/typegpu-noise-blue-noise` tip.

## Open questions for the human
- Is the base retarget intended, or was the rebase onto `main` incidental?

( ◦°^°◦) 🏴‍☠️

Ahoy! I re-reviewed the fresh commit on the Jelly Knob PR: the blue-noise texture be swapped for a GPU-side hash and she compiles clean with the tests passin' — but the branch be rebased off main, so the PR shows the whole fleet o' unrelated files; check yer base, matey!

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[aleksanderkatan]

On my phone, the example still does not work

Edit: I'm looking for the exact place that is problematic, I'll update once I find it.

[pullfrog]

ℹ️ No critical issues — one minor note inline.

Reviewed changes

This run re-reviewed the commit pushed since the prior Pullfrog review (709105cb → b1edfc3a); the rest of the PR is unchanged.

• Phone-motion button is now mobile-only — index.ts computes isMobile from DeviceMotionEvent availability plus matchMedia('(pointer: coarse)') and short-circuits the 'Enable Phone Motion' control to false elsewhere, so desktop control panels no longer show a button that cannot do anything. This answers the reviewer's request to hide it on PC.
• Uses the existing short-circuit idiom — the false control value is already supported by defineControls and skipped by exampleRunner.addParameters, matching hasSubgroups && {...} and hdr && {...} in other examples.

( ◦°^°◦) 🏴‍☠️

Ahoy! I re-reviewed the newest commit on the Jelly Knob PR and spied that the phone-motion button now be hidden on PC as the matey asked — but she still shows on Android where the button does naught, so I noted that one small thing.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using deepseek-v4.1-flash-1017 (free via Pullfrog for OSS) ｜ 𝕏}

[aleksanderkatan]

I found one place that results in an error, however it seems like there are multiple problematic spots, and it's starting to take too much time. I'll just give up on it working on my phone, and try to report the repro somewhere

[aleksanderkatan]

https://issues.chromium.org/issues/583927742

[pullfrog]

✅ No new issues found.

Reviewed changes

This run re-reviewed the commit pushed since the prior Pullfrog review (b1edfc3a → 35baec24); the rest of the PR is unchanged.

• Phone-motion button now gated on the permission API — 'Enable Phone Motion' uses the new PhoneMotion.needsPermission getter (no active listener and a requestPermission API present) alongside the coarse-pointer check, so the control renders only on iOS where a tap is actually required. On Android and desktop the constructor has already started listening, so needsPermission is false and the button stays hidden.
• Prior Android no-op-button thread retired — the inline note from the previous review is addressed by the same change and the thread was resolved.

( ◦°^°◦) 🏴‍☠️

Ahoy! I re-reviewed the newest commit on the Jelly Knob PR and the phone-motion button now only shows where the waters be askin' for a tap — no dead buttons on Android, matey!

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash-1017 (free via Pullfrog for OSS) ｜ 𝕏}