wheel_build_env = compiled: .pkg-compiled points every pyXY-compiled test env at one shared packaging env. .pkg-compiled has no basepython set, so it uses whichever interpreter runs tox. The resulting wheel is ABI-tagged for that one Python version (e.g. cp312-cp312-…whl) and cannot be installed into py310-compiled / py311-compiled / etc.

The test plan only validates py312-compiled locally, which is exactly the cell that would hide this — CI will exercise the rest of the matrix and likely fail (or silently rebuild and waste time, depending on tox/pyproject_api behaviour).

The aiosignal layout the maintainer pointed at is pure-Python, so it doesn't model this case. Look at a compiled aio-libs/ansible project (e.g. ansible/pylibssh, which webknjaz also referenced) for the pattern — usually a factor-parameterised .pkg-cpython-X.Y-compiled per Python version, or set_env/base_python overrides per factor. Worth confirming tox -e py311-compiled,py312-compiled builds two distinct wheels before declaring this done.

wheel_build_env =
    pure: .pkg-pure
    compiled: .pkg-compiled

The PR body and AGENTS.md text describe a cython env ("regenerate the *.c siblings of *.pyx") but the actual tox.ini ships no [testenv:cython]. The Makefile also no longer exposes a cythonize target.

This is consistent with the maintainer's review ("make cythonize is not used anywhere. No need to wrap it or keep it in any way.") — so dropping the env is intentional. But the PR description and the test-plan checkbox tox -e cython — regenerates _helpers_c.c cleanly are now lies. Update the PR description and the news fragment so they match what actually ships, otherwise reviewers will scroll the diff looking for code that isn't there.

Also double-check the AGENTS.md claim that the .c files are "regenerated as a side-effect of the compiled test envs" — that's only true if the PEP 517 backend's maybe_prebuild_c_extensions path runs during each compiled wheel build and writes the regenerated .c back into src/propcache/. If the backend writes into a build temp tree instead, the side-effect promise is misleading and there's no longer any developer-facing way to refresh the committed-equivalent .c files.

The previous Makefile chained doc: doctest doc-spelling so make doc was a one-shot "build docs + run doctest + spell check". The new wrapper only invokes tox -e build-docs, dropping the bundled doctest and spelling runs.

For habitual contributors this is a silent behaviour change — they will run make doc, see it pass, push, and then get a CI failure on spellcheck-docs or doctest. If the intent is to make each step independent, that's reasonable but worth calling out explicitly in AGENTS.md (the current text says "tox -e build-docs (or make doc) builds the docs locally" which implies parity, not a behaviour change).

Either restore the dependency (doc: doctest doc-spelling build-docs style) or document the behaviour change in the news fragment so contributors know to run all three.

.PHONY: doc
doc:
	$(TOX) -e build-docs

HOME = {envtmpdir}/home forces every test env to run with a throwaway HOME. That's good for hermeticity, but combined with the SSH_AUTH_SOCK pass-through it sends mixed signals: SSH agent forwarding is preserved, but the ~/.ssh/config / ~/.gitconfig / ~/.netrc that an integration test might rely on are not.

If no test currently reads from HOME, this is fine. If any future test (or pre-commit hook running under tox, see line 100) tries to load ~/.cache/<tool> or ~/.config/<tool>, it will silently get a fresh state on every invocation. Worth a comment explaining the intent ("strict isolation; if you need a real HOME, add it to pass_env") so the next maintainer doesn't spend an afternoon debugging why their git-based test "works locally".

    # Make sure tests cannot accidentally rely on resources outside the
    # working tree by isolating ``HOME`` per env.
    HOME = {envtmpdir}/home

Maintainer suggested tox-uv in a follow-up. Adding requires = tox-uv >= 1.x (or tox-uv >= 1.x + runner = uv-venv-lock-runner) to the [tox] section will dramatically speed up env creation for the matrix (especially the pure/compiled × 5 Pythons fan-out). The aiosignal/pylibssh/cheroot tox.inis the maintainer pointed at all use this — and since the maintainer asked you to borrow that pattern, this is probably the single most important alignment to do.

If you want to keep classic venv as a fallback, leave the requires opt-in and let tox -e py312-compiled work without uv installed.

The pre-command nukes build/* so a leftover build/lib.linux-cpython-310/ from a previous interpreter doesn't pollute the current run. Good.

But the in-tree PEP 517 backend writes src/propcache/_helpers_c.c and (depending on how inplace is wired) potentially src/propcache/_helpers_c.*.so. Stale variants of either can be picked up by import-time discovery during py3XY-pure test runs (which explicitly want the pure Python fallback), giving you a false "compiled is faster" / silent compiled-fast-path crossover.

If this isn't an issue today (because the backend writes only into build/ and the wheel install is fully isolated), a one-line comment to that effect would save the next reader the same investigation. If it could be an issue, extend the cleanup to src/propcache/*.so and src/propcache/__pycache__.

commands_pre =
    {envpython} -B -c \
        "import pathlib, shutil; \
        [shutil.rmtree(p, ignore_errors=True) for p in pathlib.Path('build').glob('*')]"

• Already-solved check: negative (confidence=high, reasoning=No commit on master introduces tox.ini or migrates make targets to tox wrappers.)
• Rebased koan/implement-51 onto upstream/master
• Applied review feedback
• Pre-push CI check: previous run #26054822006 failed
• Applied pre-push CI fix
• Pre-push CI fix applied
• Force-pushed koan/implement-51 to origin
• CI check enqueued in ## CI (async)

@@           Coverage Diff           @@
##           master     #242   +/-   ##
=======================================
  Coverage   99.54%   99.54%           
=======================================
  Files          10       10           
  Lines         439      439           
  Branches       17       17           
=======================================
  Hits          437      437           
  Misses          2        2           

Flags with carried forward coverage won't be shown. Click here to find out more.

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

• Already-solved check: negative (confidence=high, reasoning=No commit on master introduces tox.ini or migrates make targets to tox wrappers.)
• Rebased koan/implement-51 onto upstream/master
• Pre-push CI check: previous run still pending
• Force-pushed koan/implement-51 to origin
• CI check enqueued in ## CI (async)