Attempt to fix PyPy CI

[A5rocks]

This should in theory enable the already defined PyPy runs on MacOS and Windows. They never worked!!

This also bumps the @actions/setup-python version we use (the old one errored for nightly PyPy versions & gave us a warning in action runs due to using node 12 vs 16)

[codecov]

Codecov Report

Merging #2536 (393e97b) into master (4b6117b) will increase coverage by 0.00%.
The diff coverage is 75.00%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2536   +/-   ##
=======================================
  Coverage   92.45%   92.45%           
=======================================
  Files         118      118           
  Lines       16351    16352    +1     
  Branches     3156     3156           
=======================================
+ Hits        15118    15119    +1     
  Misses       1104     1104           
  Partials      129      129           

Impacted Files	Coverage Δ
trio/_core/tests/test_windows.py	17.82% <0.00%> (ø)
trio/socket.py	90.32% <ø> (ø)
trio/_core/tests/tutil.py	86.15% <100.00%> (ø)
trio/tests/test_dtls.py	99.79% <100.00%> (+<0.01%)	⬆️

[webknjaz]

3332 Segmentation fault (core dumped)

I wonder if this is happening consistently.. Does faulthandler work under PyPy? In some cases, I've found that pytest was crashing due to an old version of its py dependency and pytest-cov was hiding that (the problem only became apparent after disabling it).

[webknjaz]

By the way, I've just looked into ci.sh and was puzzled after seeing it still uses the deprecated setup.py call for making sdists. On a related note, I've recently made an action that allows running tests from sdist instead of relying on the Git checkout which may miss including important bits needed for testing: https://twitter.com/webknjaz/status/1609252309749440512 — maybe somebody will get to integrating it before I do.

Oh, and it's not clear why not use PyPy that GHA ships for testing...

[webknjaz]

Another flaky test: https://github.com/python-trio/trio/actions/runs/3945507801/jobs/6754580357#step:4:1771.

[webknjaz]

I wonder if this is happening consistently..

retriggered to verify: https://github.com/python-trio/trio/actions/runs/3945507679/jobs/6766090599.

[A5rocks]

Oh, and it's not clear why not use PyPy that GHA ships for testing...

I was simply updating ci.sh -- now that I look at it, I see that @actions/python-setup can install a nightly version of pypy! I'll change that out soon.

[webknjaz]

3332 Segmentation fault (core dumped)

I wonder if this is happening consistently..

I just saw it outside this PR.

[webknjaz]

On a related note, I've recently made an action that allows running tests from sdist instead of relying on the Git checkout which may miss including important bits needed for testing: twitter.com/webknjaz/status/1609252309749440512 — maybe somebody will get to integrating it before I do.

I've taken a stab at it #2541

[A5rocks]

It looks like the failure in trio/_core/tests/test_windows.py::test_readinto_overlapped should happen. We pass an immutable bytes object to something that eventually calls ffi.from_buffer(buffer, require_writable=True) (which certainly sounds like it won't accept an immutable bytes object). Why hasn't this errored before?

I will look at the rest of the errors at a later time.

[A5rocks]

OK I think this is ready for another review. The workflows will error but they shouldn't block required checks. We can figure those out at a later date.

[A5rocks]

Looks like all that needs to happen is macos needs its timeout bumped to like 20 minutes (just like windows) https://github.com/python-trio/trio/actions/runs/4169791936/jobs/7218120369

[A5rocks]

Both MacOS tests timed out on test_handshake_over_terrible_network. I don't really have a really accessible macos device but I might be able to try to reproduce this...

Although: it also fails here (kinda): https://github.com/python-trio/trio/actions/runs/4221075009/jobs/7327960276#step:5:1100

[A5rocks]

I filed an issue about our GC behavior over at https://foss.heptapod.net/pypy/pypy/-/issues/3900; hopefully they can fix our problem or tell us what we're missing.