GameCult/Ratatoskr
Watch
9
0
Fork
You've already forked Ratatoskr
 0
Code Issues Pull requests 2 Projects Releases Packages Wiki Activity Actions 1

scene: ridge scene as the default background, no WebGL (replaces #13's default) #21

Closed
nibu wants to merge 0 commits from nibu/ridge-scene into main
pull from: nibu/ridge-scene
merge into: GameCult:main
Conversation 1 Commits - Files changed -
nibu commented 2026-11-06 18:34:28 +00:00
Member
Copy link

Replaces the Voxel Space default (#13) with a scene that does no per-frame march and uses no WebGL. Meta's Redmi readout on #13 was ~100 ms frames with heavy aliasing between layers.

Base: main (14d9fa62, checked just now). #13 is still open against main and is not a base; this branch does not contain it. #13 and this PR both rewrite index.ts/tier.ts (style door) and the startFormat line, so they conflict: merge one. Void's src/sky/sun.js is PR #20, not landed, so sky colours are a placeholder behind one function.

What it is

  • Same createScene contract: createRidgeScene({ canvas, tier, ... }) returns { login, dispose } or null (CSS ground kept). Tier flag still defaults to off. startFormat on Scene becomes optional (the ridge scene has no prepass target).
  • 4 ridge layers generated once on the CPU: a clean-room port of CultMath's cultmath_hash/cultmath_value_noise and fbm (periodic, so the drift loops). Each is painted once to a plain integer-size 2D canvas, two periods wide (1x, or 2x at DPR>=2), over a baked sky gradient on the host canvas. No resampling between layers.
  • Per frame: nothing in script. Each layer slides by a Web Animation of translate3d (150-420 s per period). Paused when hidden; under prefers-reduced-motion painted but never played. tier: 'still' flattens to the host canvas immediately.
  • Cadence guard: after 5 warm-up frames, rAF intervals are sampled for 2 s; p95 > 33 ms flattens to a still image and removes the layers. ?probe=1 shows n / median / p95 and keeps sampling.
  • ridge/skyColour.ts skyColours() is the single replaceable function for sky colour (intended: Void's light()).
  • The ray-march scene stays opt-in via backgroundScene.style: 'raymarch'; its existing tests now pass style: 'raymarch'.

Verified

  • vitest, scene dir: 89 tests pass (26 new: noise wrap, profile periodicity and integer lattice, layer sizes integer, sampler maths, fallback, reduced motion, hidden pause, probe, dispose, door). Six hand mutants of the guard logic; five killed, one survivor (warm-up skip) fixed with a new test that now kills it.
  • Headless Chromium (software, this host, no GPU), 3.5 s: probe reads median 16.7 ms, p95 16.7 ms (the 60 Hz vsync). That says the main thread is idle; it says nothing about a phone.
  • tsc on the new files: only the pre-existing ?raw import errors from running outside Vite.

Not measured (plainly)

  • Anything on the Redmi: frame cadence, p95, whether compositing four layers holds 60 fps, memory (about 4 layers x 6 MB at DPR 2, plus sky), and thermal behaviour over time.
  • Whether the 33 ms guard fires correctly on a real 30 Hz power-saver display (it would read p95 ~33.3 and fall to still).
  • Visual aliasing: sub-pixel translate of canvas layers is resampled by the compositor, which is the same class of artefact Meta saw, though with silhouettes only. Not seen on the phone.
  • Light/dark and the real sky colours (placeholder until #20).
  • Eslint and the repo's tsc (no repo toolchain here); prettier 2.8.1 was run.
  • Plateau-like flat tops on the second and third layers in one screenshot; not investigated.

Preview: https://minds.gamecult.org/nibu/ratatoskr-ridge-b48685b9/ (add ?probe=1; ?tier=still). Merge nothing, deploy nothing.

🤖 Generated with Claude Code

Replaces the Voxel Space default (#13) with a scene that does no per-frame march and uses no WebGL. Meta's Redmi readout on #13 was ~100 ms frames with heavy aliasing between layers. **Base: `main`** (14d9fa62, checked just now). #13 is still open against `main` and is not a base; this branch does not contain it. #13 and this PR both rewrite `index.ts`/`tier.ts` (style door) and the `startFormat` line, so they conflict: merge one. Void's `src/sky/sun.js` is PR #20, not landed, so sky colours are a placeholder behind one function. **What it is** - Same `createScene` contract: `createRidgeScene({ canvas, tier, ... })` returns `{ login, dispose }` or `null` (CSS ground kept). Tier flag still defaults to `off`. `startFormat` on `Scene` becomes optional (the ridge scene has no prepass target). - 4 ridge layers generated **once** on the CPU: a clean-room port of CultMath's `cultmath_hash`/`cultmath_value_noise` and fbm (periodic, so the drift loops). Each is painted once to a plain integer-size 2D canvas, two periods wide (1x, or 2x at DPR>=2), over a baked sky gradient on the host canvas. No resampling between layers. - Per frame: nothing in script. Each layer slides by a Web Animation of `translate3d` (150-420 s per period). Paused when hidden; under `prefers-reduced-motion` painted but never played. `tier: 'still'` flattens to the host canvas immediately. - Cadence guard: after 5 warm-up frames, rAF intervals are sampled for 2 s; p95 > 33 ms flattens to a still image and removes the layers. `?probe=1` shows `n / median / p95` and keeps sampling. - `ridge/skyColour.ts` `skyColours()` is the single replaceable function for sky colour (intended: Void's `light()`). - The ray-march scene stays opt-in via `backgroundScene.style: 'raymarch'`; its existing tests now pass `style: 'raymarch'`. **Verified** - vitest, scene dir: 89 tests pass (26 new: noise wrap, profile periodicity and integer lattice, layer sizes integer, sampler maths, fallback, reduced motion, hidden pause, probe, dispose, door). Six hand mutants of the guard logic; five killed, one survivor (warm-up skip) fixed with a new test that now kills it. - Headless Chromium (software, this host, no GPU), 3.5 s: probe reads median 16.7 ms, p95 16.7 ms (the 60 Hz vsync). That says the main thread is idle; it says nothing about a phone. - tsc on the new files: only the pre-existing `?raw` import errors from running outside Vite. **Not measured (plainly)** - Anything on the Redmi: frame cadence, p95, whether compositing four layers holds 60 fps, memory (about 4 layers x 6 MB at DPR 2, plus sky), and thermal behaviour over time. - Whether the 33 ms guard fires correctly on a real 30 Hz power-saver display (it would read p95 ~33.3 and fall to still). - Visual aliasing: sub-pixel translate of canvas layers is resampled by the compositor, which is the same class of artefact Meta saw, though with silhouettes only. Not seen on the phone. - Light/dark and the real sky colours (placeholder until #20). - Eslint and the repo's `tsc` (no repo toolchain here); prettier 2.8.1 was run. - Plateau-like flat tops on the second and third layers in one screenshot; not investigated. Preview: https://minds.gamecult.org/nibu/ratatoskr-ridge-b48685b9/ (add `?probe=1`; `?tier=still`). Merge nothing, deploy nothing. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
scene: ridge scene as the default background, no WebGL
Some checks failed
CLA Assistant / CLAssistant (pull_request_target) Has been cancelled
Details
Build pull request / Build pull request (pull_request) Has been cancelled
Details
Check PR title / lint (pull_request_target) Has been cancelled
Details
b48685b986 
Four ridge layers generated once on the CPU from CultMath value-noise fbm, each painted
once to an integer-size 2D canvas over a baked sky; CSS translate3d drift only. Pauses
hidden / reduced motion; falls back to a still image if rAF p95 > 33 ms over 2 s;
?probe=1 shows median/p95. Same createScene contract, tier flag still defaults off.
Ray-march scene stays behind backgroundScene.style = 'raymarch'.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
scene: analytic aerial perspective on the baked ridge layers
Some checks failed
CLA Assistant / CLAssistant (pull_request_target) Has been cancelled
Details
Build pull request / Build pull request (pull_request) Has been cancelled
Details
69d8ff3fa1 
Per-layer Beer-Lambert transmittance exp(-k d) mixes each layer toward the haze colour, with
desaturation, 0-2 px edge softening (2D canvas filter, where supported) and an optional faint
warm sun-side in-scatter growing with distance. All at bake time; no per-frame work. Sky colours
still come only from skyColours() (gains optional sunAzimuth).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
nibu commented 2026-11-10 22:01:14 +00:00
Author
Member
Copy link

Closed: superseded by #81, which landed the ridge scene as the default on dev.

Closed: superseded by #81, which landed the ridge scene as the default on dev.
nibu closed this pull request 2026-11-10 22:01:14 +00:00
Some checks failed
CLA Assistant / CLAssistant (pull_request_target) Has been cancelled
Details
Build pull request / Build pull request (pull_request) Has been cancelled
Details

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
GameCult/Ratatoskr!21
No description provided.