GeezerVison

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.44%. Comparing base (38a3184) to head (5235d44).
⚠️ Report is 10 commits behind head on master.

@@            Coverage Diff             @@
##           master    #1793      +/-   ##
==========================================
+ Coverage   93.41%   93.44%   +0.02%     
==========================================
  Files         159      160       +1     
  Lines       10113    10113              
  Branches     1043     1045       +2     
==========================================
+ Hits         9447     9450       +3     
+ Misses        472      471       -1     
+ Partials      194      192       -2     

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

Looks good. One thing to fix before merging, plus a pile of updates I ran through while I was at it.

The fix: the 3.15 jobs are actually running 3.15.0rc3 (Using CPython 3.15.0rc3 in the logs). setup-uv gets uv from uv.lock, which has 0.12.23, and that version doesn't know 3.15 final yet. uv lock --upgrade-package uv --upgrade-package aiohttp, plus the uv-pre-commit rev set to 0.13.5, gets you 3.15.1. aiohttp 3.14.5 is the first release with cp315 wheels; CI is compiling 3.14.3 right now. The push also reruns CI against current master, since these checks predate #1798/#1808.

Worth doing soon: gh-action-pypi-publish v1.14.2 has a critical advisory (GHSA-5xjx-3w67-w9x9, a bad root CA in the trust store Trusted Publishing uses). v1.14.3 fixes it, so it's worth bumping before the next release.

Everything else, all tested together on this PR plus master on Linux. Tests pass on 3.11 through 3.15.1 with the same warnings as now, all pre-commit hooks pass on 3.15.1, and the build and docs are fine.

• mypy 1.20.2 → 2.4.0 (mypy~=1.0 → ~=2.4). 1.x isn't compiled for 3.15, so the hook takes 24 s instead of 4 s. 2.x raises two errors, and they have to land together with the bump since tests need lint:
  • kasa/json.py:30: strict-bytes is on by default now. Declare loads: Callable[[str | bytes | bytearray], Any] before the try.
  • kasa/discover.py:572: newer typeshed types getaddrinfo addresses as str | int. ip = str(adrrinfo[0][4][0]) fixes it.
  • Optionally set python_version = "3.11" in [tool.mypy]. Lint now type-checks against 3.15, so a 3.12+-only stdlib API would slip through. The code passes with it today.
• Workflows:
  • publish.yml: 3.13 → 3.15. Same wheel; the sdist only differs because 3.15's gzip compresses differently.
  • release-draft.yml: has no python-version, so it runs on the runner's 3.12. Add "3.15".
• Lock (uv lock --upgrade). Runtime deps are already at their latest versions apart from aiohttp. The rest:
  • orjson 3.13.0, multidict 7.2.1, python-dateutil 3.0.0, coverage 7.17.0, pytest-mock 3.17.0, pre-commit 4.7.0
  • ruff 0.17.2: no new findings, no formatting changes
  • docs: Sphinx 9.x, myst-parser 5.1.0, docutils 0.22.4. This needs sphinx_rtd_theme~=2.0 → ~=3.1, since 2.0 caps Sphinx below 8.
• Hooks: ruff-pre-commit v0.16.10 → v0.17.2. Optionally rename the hook id ruff → ruff-check, since ruff is now a legacy alias.
• Actions: none of the breaking changes affect us.
  • checkout v7.0.1, setup-uv v10.3.2 (publish.yml stops caching on release events), cache v6.1.0
  • codecov v7.1.1 (new signing-key account), codeql v4.39.1, stale v11.0.0
  • packaging 26.3 in build-packages
  • Read the Docs can go to ubuntu-24.04; it has no 3.15 yet, so keep python: "3".

Per version:

• 3.11/3.12: Sphinx resolves to 9.0.4/9.1.1. Two docs warnings come from Python's own int.from_bytes docstring; they're there today too, and RTD builds on 3.14 anyway.
• 3.13: required checks are the 3.13 jobs, so nothing changes there.
• 3.14: RTD stays here.
• 3.15:
  • kasa-crypt has no cp315 wheels. Without a compiler, speedups quietly falls back to pure Python.
  • orjson's cp315 Linux wheels need glibc 2.39+.

FYI, not from this PR:

• 3.14 on Windows hangs until the 6 h timeout now and then: 12 of 177 runs, always in test_features_all. It just happened again on master. 3.15 Windows may do the same, and a timeout-minutes would at least fail it fast.
• lazygroup.py:67 uses the deprecated click.BaseCommand. That's where all the test_cli warnings come from, and it breaks in Click 9; click.Command works.

Not tested: macOS and Windows, and the action bumps beyond reading their release notes.