Tracking: repository modernization #277

Activity

1. jaysonsantos

   added

   modernizationRepo modernization effortRepo modernization effort

   on Sep 8, 2026
2. jaysonsantos commented on Sep 8, 2026

   

   jaysonsantos

   on Sep 8, 2026

   OwnerAuthor

   More actions

   Seven pull requests are open for review. None is merged. No issue is closed by
   hand. Each pull request body holds a Closes #NNN line, which closes the issue
   on merge.

   The pull requests

   Pull request	Base branch	Closes	Gate result
   #278 — set the Python 3.10 floor and remove the Python 2 support layer	main	#265, #266, #267	261 passed, flake8 0 errors
   #279 — migrate packaging to PEP 621 and replace flake8 with ruff	#278	#268, #269, #270	261 passed, ruff check clean, ruff format --check clean
   #280 — modernize the tests-and-lint workflow and add Renovate	#279	#271	261 passed, ruff check clean, ruff format --check clean
   #281 — gate publish.yml on tags and move PyPI to Trusted Publishing	main	#272	261 passed, flake8 0 errors
   #282 — replace assert guards, add an exception message, stop a password leak	main	#273	268 passed, flake8 0 errors
   #285 — harden the conftest fixtures and close the IPv6 and SASL gaps	main	#274	271 passed, flake8 0 errors
   #286 — remove the per-call struct format strings from the response parse path	#278	#276	261 passed, flake8 0 errors

   The baseline on main is 261 passed. #282 adds 7 tests. #285 adds 10 tests.
   No pull request removes a test.

   Every gate ran in the Nix dev shell:

   nix develop --command bash -c 'pytest -q'
   

   Merge order

   #281, #282, and #285 are independent. Merge them in any order.

   #278 first, then #279, then #280. Each stacks on the one before, so a reader
   sees the incremental diff only. #286 also stacks on #278 and can merge as soon
   as #278 lands.

   Two pull requests need an action before merge

   #281 has a release hazard. It removes the PYPI_API_TOKEN password input.
   Register the trusted publisher on pypi.org and create the pypi environment in
   GitHub settings before you merge. A merge without those steps breaks the next
   tag push. The exact steps are at the top of the pull request body. I did not
   touch PyPI settings, GitHub environment settings, or the PYPI_API_TOKEN
   secret.

   #280 needs the Renovate GitHub App. renovate.json alone does nothing.
   Install the app at https://github.com/apps/renovate and grant it access to this
   repository. Two settings in that file need your preference: timezone is
   Europe/Berlin, and schedule runs once per week.

   Deferred

   #275 (py.typed and annotations) is not implemented. It depends on #278,
   #279, and #282, all of which are open. I left a comment on #275 that states
   this and records two facts that changed under it: the two # type: ignore
   comments it names are already gone, and the lone comment-style hint at
   bmemcached/client/mixin.py:153 now reads # type: (str, int) -> bool.

   Nothing was blocked

   Every issue in scope for this run is fully implemented. Two items came close.

   The SASL integration test in #285 needed a memcached built with SASL, which
   nixpkgs does not provide by default. #285 adds --enable-sasl to the
   flake.nix memcached override, in the same way the flake already adds
   --enable-tls, and adds cyrus_sasl.bin for saslpasswd2. The test runs and
   passes in the dev shell. It skips in CI, because .ci-before-script.sh
   installs neither sasl2-bin nor a SASL-enabled memcached. Adding those belongs
   to #271, not to #274, so I did not fold it in.

   The bytes() copy in _read_socket, the second item in #276, stays. #286
   holds the measurement and the reason: removing it changes public return types,
   because a bytearray slice is a bytearray. The acceptance criteria allow the
   copy to stay with a stated reason.

   Two new issues

   Both came out of #274 and are linked from #285. Neither is folded into a pull
   request.

   • Map memcached SASL status 0x20 to InvalidCredentials #283 — memcached answers a failed SASL auth with status 0x20. The
     client maps only 0x08 to InvalidCredentials, so a wrong password raises
     the base MemcachedException. The mocked test in test/test_auth.py hid
     this, because it feeds 0x08 by hand. The new SASL integration test asserts
     the real behavior and names this issue in its docstring.
   • Replace the deprecated pytest.yield_fixture in test/test_tls.py #284 — test/test_tls.py:16 still uses the deprecated
     pytest.yield_fixture. It is outside the file list of Harden test/conftest.py and close the IPv6 and SASL test gaps #274.

   Notes on three decisions

   flake.nix was updated in the same pull request that removed each dependency,
   so no later gate runs against a stale dev shell. #278 drops six and mock.
   #279 drops flake8 and adds pkgs.ruff, plus twine so the packaging gate
   runs there. #285 adds SASL to the memcached build.

   Three pin versions differ from the issue text, because the issue text is now
   out of date. .pre-commit-config.yaml in #279 uses commitizen v4.18.0 and
   pre-commit-hooks v6.0.0 in place of the proposed v3.29.1 and v4.6.0.
   The ruff hook rev is v0.16.5, which matches the ruff version in the dev
   shell, so local runs and hook runs agree.

   actions/upload-artifact in #280 stays at v4, as the issue text specifies.
   The current release is v7.0.1. Renovate will propose the major bump once it
   runs.

   Three findings in the "Out of scope" section of this issue are untouched, as
   intended: the missing GitHub releases for v0.31.3, v0.31.4, and v0.32.0;
   the unittest.TestCase style across eight test files; and the stale
   intersphinx_mapping in docs/conf.py. The manual version copy at
   flake.nix:68 is also untouched, and #279 states this in its body.

   On the related report #250, which #274 asked me to check: #285 turns the most
   likely cause, a missing or slow memcached, into one clear skip message
   instead of a wall of ConnectionRefusedError. It does not confirm the original
   diagnosis, so I did not close that issue.