Fix nursery crashing during a guest run when the host loop calls start_soon(needs_sniffio)

[gschaffner]

Closes #2183.

[codecov]

Codecov Report

Merging #2462 (ee0bf40) into master (d6454d1) will increase coverage by 0.00%.
The diff coverage is 100.00%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2462   +/-   ##
=======================================
  Coverage   92.45%   92.45%           
=======================================
  Files         118      118           
  Lines       16342    16351    +9     
  Branches     3155     3156    +1     
=======================================
+ Hits        15109    15118    +9     
  Misses       1104     1104           
  Partials      129      129           

Impacted Files	Coverage Δ
trio/_threads.py	100.00% <ø> (ø)
trio/_core/_run.py	98.49% <100.00%> (-0.01%)	⬇️
trio/_core/tests/test_run.py	100.00% <100.00%> (ø)

[gschaffner]

I just force-pushed what should be a fix for the bug in the test.

[gschaffner]

I've just fixed the host_uses_signal_set_wakeup_fd-related test hangs that were occurring on Windows when asyncio.ProactorEventLoop from CPython >= 3.8 was used.

(Curiously, the test only hung forever when using a >= 3.8 ProactorEventLoop and running it from pytest. It ran fine on ProactorEventLoop when called like python -c 'from trio._core.tests.test_guest_mode import test_guest_mode_host_calls_start_soon_sniffio; test_guest_mode_host_calls_start_soon_sniffio()'.)

[gschaffner]

rebased onto master and updated the newsfragment issue number.

[gschaffner]

the CI failure on pypy3.8 nightly on Ubuntu appears to be unrelated to this PR and may just be a flaky failure. the new test that this PR adds did pass on the pypy3.8 nightly on Ubuntu CI. the failure appears to be due to a segfault that occurred while pytest was handling test_dtls.py::test_smoke[ipv6]'s timeout. (note: the same timeout (without a segfault) also occurred on the latest pypy3.8 nightly on Ubuntu CI run on master (https://github.com/python-trio/trio/actions/runs/3684296901/jobs/6233869817#step:5:1073).)

the failure of the new test on Windows+Komodia LSP, however, was occurring before the rebase and is still something that is pretty beyond me—I would be much appreciative if someone familiar with Komodia LSP could help look into it.

[richardsheridan]

The new commit I pushed merges #2574 into this PR, in essentially the state that #2574 would be by applying your review comments to it. Using this roundabout method has the advantage of acknowledging your contribution and inviting you to join the python-trio org! If you acknowledge that you are ok with this, I'll bring it all up to date with master and merge.

[gschaffner]

this looks great to me, thanks @richardsheridan!

also, for what it's worth, i am already a member of the Trio github org, but i still appreciate that this PR acknowledges both my and oremanj's contributions here :)

(i'm also on the trio gitter—not sure what caused trouble when you tried searching for me there!)

[gschaffner]

it might make sense to overwrite 2183.bugfix.rst with this new message instead of keeping both. my message in 2183.bugfix.rst just describes a specific case of oremanj's message here.

[gschaffner]

@richardsheridan it looks like you did git mv newsfragments/{2574,2462}.bugfix.rst, meaning there are still two redundant changelog entries—was that your intention? to clarify, i meant to suggest git mv -f newsfragments/{2574,2183}.bugfix.rst

[richardsheridan]

Oops! I think the commit message phrasing fits better if the link points here to the fix. I'll push a clean renumber commit when CI is working again.

[richardsheridan]

I also added a test for your edge case of a function returning a coroutine!