Skip to content

CI runs the renderer's unit tests; stale binding counts in the docs corrected - #66

Merged
yujiaxu-ops merged 1 commit into
mainfrom
ci-tests-docs
Nov 13, 2026
Merged

yujiaxu-ops merged 1 commit into
mainfrom
ci-tests-docs

Conversation

@yujiaxu-ops

@yujiaxu-ops yujiaxu-ops commented Nov 13, 2026 •

Copy link
Copy Markdown
Collaborator

Step 1 of the order agreed after the review of the renderer: the small things first.

CI runs the renderer's unit tests. The renderer job only built the renderer. None of its ~130 unit tests ran in CI, though all but seven need no GPU: scene, world, script, voxel, physics, the merged layout's naga checks, the kernel's source switches. It now runs cargo test --release --locked after the build. The seven tests that open a device print "skipped" and return when there is no adapter. Checked on this Mac with no adapter (WGPU_BACKEND=vulkan, which macOS lacks): 124 passed, those seven among them. The job is renamed "renderer: release build, unit tests"; no check is required by name. Expect the job to take longer, because the test profile is compiled too.

Docs my PRs left stale. #63 (reservoirs removed) and #65 (the live passes' 28 storage buffers → 24, one bind group) did not update these:

  • docs/RUNTIME.md still said: the kernel binds 20, the live passes 28, seven group-1 buffers including the reservoirs, the history's two bind groups;
  • merged.rs:657 and gpu.rs:1450 comments had the old counts.

Known issues. Under software traversal (the browser's, and a device without ray queries) 24 tests fail on main. 23 were already recorded; the new one is wind/pivot-sway-live (9.241 for 10; hardware traversal passes it). WORKFLOW.md now says that software traversal is in no gate.

Tier D plus a CI change: no renderer code changed (two comments).

🤖 Generated with Claude Code


Note

Low Risk
CI and documentation only; no renderer logic changes beyond comments.

Overview
Adds cargo test --release --locked to the CI renderer job (after the existing release build), renaming it to renderer: release build, unit tests so ~130 non-GPU unit tests run on every renderer-affecting PR; device-dependent tests self-skip on the GPU-less Ubuntu runner.

Updates documentation and inline comments to match the current merged binding layout (19 kernel / 24 live storage buffers, six group-1 buffers, no reservoirs / single live history layout). Records a new software-traversal failure (wind/pivot-sway-live, 24 total) and clarifies in WORKFLOW that render gates use hardware traversal only.

Reviewed by Cursor Bugbot for commit ac5a0b6. Bugbot is set up for automated code reviews on this repo. Configure here.

…t stale in the docs corrected

CI's renderer job built the renderer and ran none of its ~130 unit tests, though all but a few need no GPU. It now
runs `cargo test --release --locked` after the build: on the runner, which has no GPU, the seven tests that open a
device say "skipped" and return (checked on a Mac with no adapter: 124 pass, those seven among them). The job is
named for it; no check is required by name.

The cleanup (#63, the reservoirs out) and the buffer layout (#65, 28 -> 24 live storage buffers, one bind group)
left docs/RUNTIME.md describing the kernel's 20 and the live passes' 28 storage buffers, seven group-1 buffers with
the reservoirs, the history's two bind groups; merged.rs and gpu.rs comments had the old counts.

ROADMAP "Known issues": under software traversal 24 tests fail on main now, the 23 recorded and
wind/pivot-sway-live (9.241 for 10; hardware traversal passes it). WORKFLOW.md says software traversal is in no gate.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@yujiaxu-ops
yujiaxu-ops merged commit ef00aa4 into main Nov 13, 2026
9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant