Merging this PR will not alter performance

✅ 6 untouched benchmarks

Comparing eshattow:workflow-ci-riscv64 (b3feba3) with master (bc8ec8d)

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.76%. Comparing base (bc8ec8d) to head (b3feba3).
⚠️ Report is 104 commits behind head on master.

@@           Coverage Diff           @@
##           master    #1666   +/-   ##
=======================================
  Coverage   99.76%   99.76%           
=======================================
  Files          33       33           
  Lines        3401     3401           
  Branches      461      461           
=======================================
  Hits         3393     3393           
  Misses          5        5           
  Partials        3        3           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

force pushed with requested changes rolled into single commit, resolves linter reported errors in commit subject formatting

I don't know what to suggest to fix the more general workflow problems on this repository for external contributions, tracking issue for that is #1668

@bluetoothbot review

PR Review — ci: ci.yml introduce riscv64 linux and musllinux to wheels build

Mechanically the change is correct and mirrors the established armv7l QEMU-split pattern, just with riscv64 as the target arch. The cp39 entries from the earlier revision are gone, addressing @bdraco's review comments. Two non-blocking notes: (1) the new entries use YAML flow style while the surrounding armv7l entries use block style — worth normalising for grep/diff ergonomics; (2) cp314t on emulated riscv64 is the riskiest cell and worth watching on the first green run before relying on it. No correctness, security, or build-correctness concerns with the matrix itself (cibuildwheel 3.4.1 supports riscv64 manylinux + musllinux, and the existing tonistiigi/binfmt:qemu-v8.1.5 image registers riscv64).

🟢 Suggestions

The existing armv7l entries above (lines 197-260) use multi-line block style:

- os: ubuntu-latest
  qemu: armv7l
  musl: "musllinux"
  pyver: cp310

The new riscv64 entries use YAML flow style ({os: ubuntu-latest, qemu: riscv64, ...}). Mixing the two styles in the same include: list makes future diffs and edits noisier (e.g. adding a new key would have to be done twice in two different shapes). Consider expanding the riscv64 entries to the same block form so the matrix stays uniform.

- {os: ubuntu-latest, qemu: riscv64, musl: "musllinux", pyver: cp310}

Free-threaded CPython under QEMU user-mode emulation is the slowest combination in this matrix (no GIL fast paths, all-emulated). The PR description cites ~55 min wall time for the full split — the cp314t-riscv64 cell is likely to be the long-tail straggler and the one most prone to timeouts as cibuildwheel pulls in the free-threaded toolchain in an emulated rootfs. It's worth verifying the cp314t-riscv64-musllinux and cp314t-riscv64-manylinux cells actually green on the first run before declaring this matrix stable; if they routinely timeout, consider dropping just those two entries (cp310–cp314 still gives RISC-V coverage) rather than letting them block the wheel job.

- {os: ubuntu-latest, qemu: riscv64, musl: "musllinux", pyver: cp314t}
- {os: ubuntu-latest, qemu: riscv64, musl: "", pyver: cp314t}

Checklist

• No hardcoded secrets or credentials
• Matches existing matrix entry style — suggestion ＃1
• cp39 entries removed per prior review
• QEMU image / cibuildwheel versions support riscv64
• Free-threaded (cp314t) coverage considered — suggestion ＃2

Summary

Mechanically the change is correct and mirrors the established armv7l QEMU-split pattern, just with riscv64 as the target arch. The cp39 entries from the earlier revision are gone, addressing @bdraco's review comments. Two non-blocking notes: (1) the new entries use YAML flow style while the surrounding armv7l entries use block style — worth normalising for grep/diff ergonomics; (2) cp314t on emulated riscv64 is the riskiest cell and worth watching on the first green run before relying on it. No correctness, security, or build-correctness concerns with the matrix itself (cibuildwheel 3.4.1 supports riscv64 manylinux + musllinux, and the existing tonistiigi/binfmt:qemu-v8.1.5 image registers riscv64).

Automated review by Kōan8e703fa
b3feba3