CI changes

[A5rocks]

Fixing existing runs:

• Don't run Jedi on PyPy. (see Export test fixes #2887 and Enable mypy and disable jedi on PyPy #3075) (need to update test_static_tool_sees_class_members too)
• Only run autodeps in python-trio/trio (see Try to prevent autodeps from running in forked repos #2892)
• Add a gc.collect() to the top of test_for_leaking_fds (this seems to address test flakiness I've seen elsewhere) (see Address PyPy flakiness in test_for_leaking_fds #2888)
• Start running mypy on PyPy. (see Enable mypy and disable jedi on PyPy #3075)
• switch macos to run on m1 (see Make sure we're running on M1 MacOS #2953)

Adding more runs:

• Add an anaconda run
• Alpine run (see Add Alpine to CI #2933)
• Try out qemu for a freebsd
• Cython run (see Add Cython to CI #2942)
  • Run full test suite on Cython
• Update MacOS and Windows Python matrix to support 3.11 and 3.12. (see Add Python 3.11 and 3.12 to Windows and MacOS matrices #3017)
• Add CPython 3.13 (see Fixes for 3.13 #3005)
• ... Android run??? (https://github.com/marketplace/actions/android-emulator-runner ??)
• pypy nightly 3.11 (not possible)
• add pypy runs for macos (see Test PyPy 3.10 on both Windows and MacOS #3074)
• add pypy runs for windows (see Error in CI with Windows pypy-3.9-nightly on master #2776 and pypy-3.10 support #2678) (see Test PyPy 3.10 on both Windows and MacOS #3074)

Removing runs:

• Don't run PyPy nightlies less than 3.11 (see Remove PyPy nightly runs #2952)

These feel like they will take more motivation than I currently have :(

[jakkdl]

A more fun change: Add python 3.13

[TeamSpen210]

We also aren't running 3.11/3.12 on Windows.

[jakkdl]

weird CI fail for macOS 3.8: https://github.com/python-trio/trio/actions/runs/7232062095/job/19705878765?pr=2891

[jakkdl]

Add back Alpine Linux for testing musl libc.
#2917

[jakkdl]

Sleep timing fail on MacOS pypy3.10 https://github.com/python-trio/trio/actions/runs/7448486091/job/20262980216?pr=2886

FAILED ../../../_temp/116e3acf-3f1d-45b1-afa8-48d74a477716/pypy-c-jit-184208-0c87577e022f-macos_x86_64/lib/pypy3.10/site-packages/trio/_tests/test_timeouts.py::test_sleep - assert (1.70860500699996 / 1.0) < 1.5

Not especially weird, and only problematic if it's encountered regularly.

[jakkdl]

Is it possible to trigger actions on "push or pull request" rather than "push and pull request"? I started #2942 on a branch in trio/ to be able to run tests w/o spamming a PR, but once I opened the PR I'm getting double runs.

[TeamSpen210]

A few possible ways:

• Using concurrency groups - create a group name including the commit hash, then it'd only allow one of the pair to run at a time. Then we'd set "cancel-in-progress" to true only on the PR run, causing the push to be cancelled. Downside is that the push events are still run and cancelled, wasting time and cluttering the runs list.
• We could set the push event to only apply to master, so you'd need to open a PR to get CI. May not be helpful.
• We could add a job to the workflow that runs before everything else on push requests. It'd fetch all PRs via the API, get their hash, figure out if we match and then skip the rest. Shouldn't be too complex?

[A5rocks]

I think the 3rd is the best solution. Just a gh pr view on push and if there's a PR then don't run the next steps. Maybe a if: in the actual CI workflows that checks input and otherwise skips?

The 1st isn't great cause it'll clutter notifications too

[jakkdl]

Edited list in OP to refer to #2776 and #2678 wrt to pypy+windows (and split up mac&windows at the same time)

[jakkdl]

A more fun change: Add python 3.13

with a first beta released this should probably be done sooner rather than later.

[jakkdl]

windows 3.13 x64 hang: https://github.com/python-trio/trio/actions/runs/12391464668/job/34588610030

[jakkdl]

We should move test_static_tool_sees_all_symbols from the tests run on all platforms to our check.sh/"check formatting" run - it's often veery slow (20+ seconds for the worst matrix entry)

[A5rocks]

We should move test_static_tool_sees_all_symbols from the tests run on all platforms to our check.sh/"check formatting" run - it's often veery slow (20+ seconds for the worst matrix entry)

I agree. My first thought is maybe just use pytest -k? That should help.

[jakkdl]

We should move test_static_tool_sees_all_symbols from the tests run on all platforms to our check.sh/"check formatting" run - it's often veery slow (20+ seconds for the worst matrix entry)

I agree. My first thought is maybe just use pytest -k? That should help.

definitely not the prettiest solution to do pytest -k "not test_static_tool" in ci.sh and pytest -k test_static_tool in check.sh, but would be a quick fix that could be prettified later.

actually scratch that, let's just take all of src/trio/_tests/test_exports and move it somewhere like tests/static_tests/. _trio_check_attrs_aliases.py is there already anyway