Skip to content

docs: correct what a review found drifting, and record D74 - #168

Merged
with-developer merged 1 commit into
mainfrom
docs/drift-fixes
Nov 18, 2026
Merged

with-developer merged 1 commit into
mainfrom
docs/drift-fixes

Conversation

@with-developer

Copy link
Copy Markdown
Owner

Purpose

A docs-drift review read every name, number, link and decision reference in the docs against the code, the CI and crates.io. This corrects what it found that was wrong or misleading, docs only, and records the crates.io publication as D74.

Changes

  • CONTRIBUTING: THIRD_PARTY_NOTICES.md ships in the release archives (D51), not "wherever LICENSE does": the crates on crates.io carry LICENSE alone.
  • README, README.ko and the 1.0 checklist: "about thirty" wgpu-typed public items (a scan gives 25 plus the re-export, plus the hooks trait and TexturePicture); issue v1.0 decisions: versioned library name and how wgpu shows in the Rust API #165 corrected the same way.
  • ARCHITECTURE memory table: the packed planes are 12.1 MiB at 4K with the decoder's padding (11.9 was the unpadded size), the player's saving 43.8 MiB; a sentence that said two tests and listed three.
  • README and BENCHMARKS: the 0.11.0 "about as long on the Mac at 60 frames a second" now says what the 0.18.1 runs beside a display loop found; BENCHMARKS' "everything else is hand-written", the CPU encoder's "quarter of a second" (a fifth), perf: at a stream's pace, emberwave's decode takes 1.3 to 1.5 times upstream's #86 as open, the chroma exceptions, the iOS README's "three minutes", the FFI README's "not run on an iPhone".
  • Cargo.toml comment and D5: the two crates are published; new D74 records it (what was published, what was not, how the token is used, Trusted Publishing as the way to avoid a long-lived token, and the open question below).

For the maintainer

D74's "Not done": THIRD_PARTY_NOTICES.md is not in the packages on crates.io (cargo package --list shows LICENSE and no notices file), where the release archives carry both. Whether a crate has to carry PyroWave's notice is a licensing judgement that is not made here.

Testing

Docs, a comment in Cargo.toml and two comments in scripts/bench.sh only: cargo metadata --locked still works, no code changes. The reviewer's other findings are small and were checked against the code before the sentences changed.

A docs-drift review (every name, number, link and decision reference in the
docs against the code, the CI and crates.io) found statements that were no
longer true:
- CONTRIBUTING said THIRD_PARTY_NOTICES.md ships wherever LICENSE does; the
  crates on crates.io carry LICENSE alone (the notices ship in the release
  archives, D51). Whether the crates should carry them is left to the
  maintainer.
- The README and the 1.0 checklist said 26 wgpu-typed public items; the scan
  gives 25 plus the re-export, and D70 said about thirty: now "about thirty".
- ARCHITECTURE's memory table gave the packed planes as 11.9 MiB at 4K; with
  the decoder's padding they are 12.1 MiB, the player's saving 43.8 MiB.
- The README and BENCHMARKS said emberwave and HEVC take about as long on the
  Mac at 60 frames a second, from the 0.11.0 runs; they now say what the
  0.18.1 runs beside a display loop found.
- BENCHMARKS' claims that everything outside the results is hand-written, that
  the CPU encoder takes a quarter of a second (it takes a fifth), that #86 is
  open, the chroma exceptions, and the iOS README's three minutes; the FFI
  README's "not run on an iPhone".
- Cargo.toml and D5 still said the crates are not published: D74 records the
  publication of emberwave-core and emberwave-cli.

Claude-Session: https://claude.ai/code/session_019fJGoCTvP5e7n599q3ZQ42
@with-developer
with-developer merged commit 0754b7b into main Nov 18, 2026
7 checks passed
@with-developer
with-developer deleted the docs/drift-fixes branch November 18, 2026 11:34
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