mirror of
https://github.com/hanzoai/ml.git
synced 2026-11-20 23:08:29 +00:00
The Rustfmt job runs `cargo fmt --all -- --check` and reported 653 diffs across
43 files. This is that command's output, nothing else — no logic is touched.
Worth noting why it surfaced only now: the previous run on this repo was a
`schedule` trigger, which fires gpu-tests alone. Rustfmt, Clippy and the test
suite belong to the push-triggered workflow and had not run in that window, so
they were not newly broken — they were newly VISIBLE.
Scope: hanzo-flash-attn-v3, hanzo-kernel, hanzo-kernel-macros,
hanzo-metal-kernels, hanzo-ml, hanzo-nn, hanzo-research. The insertion count is
much larger than the deletion count because rustfmt rewraps long lines rather
than merely reindenting them, which is also why `git diff -w` does not collapse
the change — whitespace-blind diff is not a purity test for a formatter.
Evidence it is safe, given `cargo check --workspace` cannot run here (onig_sys
fails to build its vendored oniguruma against this macOS SDK, unrelated to
formatting):
- rustfmt is idempotent on the result. Pass 1 wrote 43 files; pass 2 re-read
and re-parsed every one and produced zero further change. rustfmt fails
outright on unparseable input, so a clean second pass means the emitted code
is valid Rust.
- `cargo check -p` is clean (0 errors) for hanzo-kernel, hanzo-kernel-macros
and hanzo-research — the changed crates that pull no C dependencies.
- `cargo fmt --all -- --check`, the exact command CI runs, now exits 0.
No rustfmt.toml exists and no toolchain is pinned, so this is default stable
style — the same thing dtolnay/rust-toolchain@stable gives the runner.
Clippy is NOT addressed here. Its findings are in hanzo-ml/src/quantized/ and
are real lint work, not formatting.
Co-authored-by: Hanzo Dev <[email protected]>