Skip to content

docs(readme): claims that match the code, and an acknowledgment of gfx-rs/wgpu - #385

Merged
kolkov merged 2 commits into
mainfrom
docs/readme-claims
Nov 7, 2026
Merged

kolkov merged 2 commits into
mainfrom
docs/readme-claims

Conversation

@kolkov

@kolkov kolkov commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

I went through the README claim by claim against the code, the tests and what CI actually runs, and fixed what didn't hold:

  • ErrMapCanceled is the error's real name (the README said ErrMapCancelled).
  • Vulkan on macOS goes through MoltenVK, which the CI runner doesn't have, so it's marked untested. Metal does run in CI, on the macOS runner's virtual GPU.
  • The software rasterizer section now lists what the draw path does. The tile-based parallel rasterizer is only used by tests and benchmarks, and clip-space clipping isn't done yet. The known gaps are listed: no clipping, derivatives return 0, workgroup memory not shared, commands run at record time, no compressed textures.
  • -tags rust with go-webgpu/webgpu v0.5.5 isn't usable for real rendering yet (the 0.34.6 known issue). The README now says so.
  • Dropped the "first" claims, and "full" and "battle-tested" where nothing backs them. StagingBelt no longer claims zero allocation, since no benchmark covers it. The buffer-mapping claim stays: BenchmarkBufferMapReadPrimary shows 0 allocs/op.
  • Added an Acknowledgments section thanking the authors of gfx-rs/wgpu, whose architecture this project follows.

README only, no code changes.

…x-rs/wgpu

Every claim was checked against the code, the tests and CI:
- ErrMapCanceled is the error's name (was ErrMapCancelled).
- Vulkan on macOS is not run in CI (no MoltenVK on the runner): Untested.
  Metal runs on the macOS runner's virtual GPU.
- The software rasterizer lists what the draw path does. The tile-based
  parallel rasterizer is used only by tests and benchmarks, clip-space
  clipping is not done yet, and the known gaps are named.
- The -tags rust build is not usable for real rendering with
  go-webgpu/webgpu v0.5.5 (0.34.6 known issue).
- No 'first' claims, no 'full' or 'battle-tested' without evidence;
  StagingBelt no longer claims zero allocation (no benchmark backs it).
- An Acknowledgments section thanks the authors of gfx-rs/wgpu, whose
  architecture wgpu follows.
… from webgpu

v0.34.8 shipped shared workgroup memory and real barriers on software (P1-SW-025), so it is no longer a known gap. The -tags rust note now says what go-webgpu/webgpu v0.5.5 does on each platform, as its maintainers describe it: an abort at adapter request on Linux amd64 and Intel macOS, and vertex buffer, depth attachment and bind group layout failures on Windows.
@codecov

codecov Bot commented Nov 7, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

@kolkov
kolkov merged commit 1a82059 into main Nov 7, 2026
16 checks passed
@kolkov
kolkov deleted the docs/readme-claims branch November 7, 2026 19:27
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