Add rsloop as backend for asyncio tests. by tapetersen · Pull Request #1199 · agronholm/anyio

tapetersen · 2026-07-01T12:35:01Z

Changes

rsloop as an implementation of asyncio's AbstractEveltLoop has reached full compatibility required by AnyIO's tests.
This PR adds it as an alternative backend to test against along uvloop

Checklist

You've added tests (in tests/) which would fail without your patch
You've updated the documentation (in docs/), in case of behavior changes or new
features
You've added a new changelog entry (in docs/versionhistory.rst).

Vizonex

@tapetersen I love this because it's another low level event loop that has windows support. besides just winloop and the selector system pywepoll, (both of which I wrote) I'm glad to see libraries with less neglect for windows support (excluding rloop & uvloop) this had always been a reoccurring pet-peve of mine but I'm glad to see something new that could provide a positive impact for windows users. I approve this pull request due to that fact and I hope rsloop receives positive attention.

tapetersen · 2026-07-08T13:00:30Z

@Vizonex I mostly completed what was needed to get the test-suite to pass in rsloop for curiousity. I havn't used it for something real yet and don't have any idea for how ready it is. If nothing else it would be good to see it tries in more projects if you have any.

There is no need for this PR to use it just pass loop_factory=rsloop.new_event_loop to anyio.run

anyio.run(main, backend="asyncio", backend_options={"loop_factory": rsloop.new_event_loop})

Vizonex · 2026-07-08T15:53:08Z

@Vizonex I mostly completed what was needed to get the test-suite to pass in rsloop for curiousity. I havn't used it for something real yet and don't have any idea for how ready it is. If nothing else it would be good to see it tries in more projects if you have any.

There is no need for this PR to use it just pass loop_factory=rsloop.new_event_loop to anyio.run
anyio.run(main, backend="asyncio", backend_options={"loop_factory": rsloop.new_event_loop})

@tapetersen I already added it to cyares Vizonex/cyares#91 It might be added to more places if I can think of anywhere to have it placed.

agronholm · 2026-07-10T09:12:49Z

Can you figure out if the timeouts happen for a reason or if the test suite needs more time to the increased number of tests?

tapetersen · 2026-07-10T11:18:24Z

I have gotten hold of a remote windows VM and can confirm they look like real timeouts, not just slow tests. I'll see if I can figure out why.

agronholm · 2026-07-10T11:58:09Z

I have a real Windows machine here so I can also do this when time allows.

tapetersen · 2026-07-10T12:10:56Z

I tried adding a fail_after around the connect_tcp call in the test_connection_refused case and it fails and moves on. after that the rest of the socket tests passes. So something is preventing rsloop from just failing the connection attempt.

agronholm · 2026-07-10T12:15:32Z

Confirmed locally. test_connection_refused hangs with rsloop.

tapetersen · 2026-07-10T12:26:22Z

Yeah, it's probably a rsloop bug of not failing when connecting to a closed socket on windows. I'll see if I can distill it and add an issue there instead.

Skipping these let's the whole suite through otherwise though. Should we just mark them as xfail or wait until it's fixed upstream (which I'll try to do anyway)?

tapetersen · 2026-07-10T13:29:41Z

Upstream issue here for the windows issue

RustedBytes/rsloop#60

For the macos timeout I don't have a better way to test. I may report it anyway.

agronholm · 2026-07-10T21:31:06Z

Skipping these let's the whole suite through otherwise though. Should we just mark them as xfail or wait until it's fixed upstream (which I'll try to do anyway)?

I don't have a strong opinion on this, but this seems like a fairly serious bug in rsloop. I'm leaning towards waiting until the test suite passes cleanly. This is what we did with winloop too.

egorsmkv · 2026-07-15T09:46:54Z

I have fixed that bug on Windows in 0.1.22 version. can we re-run CI checks?

tapetersen · 2026-07-15T10:33:41Z

I think I managed to re-run them. There is still some error on macOS as well that I don't have a a good way to debug. It may be ok to exclude that though as long as it works on other platforms.

agronholm · 2026-07-15T20:31:25Z

Given that both macOS jobs failed with a BusError, it sounds like there is genuinely something still wrong.

egorsmkv · 2026-07-17T14:06:40Z

I will try to figure out what's wrong with those CI runs today.

egorsmkv · 2026-07-18T11:52:38Z

@tapetersen can you re-run tests again here, please?

agronholm · 2026-07-18T16:30:13Z

Not sure what you expected to happen, but even more test jobs failed.

egorsmkv · 2026-07-18T19:10:14Z

@agronholm can you re-run again? in my local test - it works now with v0.1.25

agronholm · 2026-07-19T10:21:46Z

Something is still off, as some tests fail due to the lack of an exception where one was expected. I confirmed that 0.1.25 is installed before the test run.

egorsmkv · 2026-07-19T10:38:43Z

Something is still off, as some tests fail due to the lack of an exception where one was expected. I confirmed that 0.1.25 is installed before the test run.

Thank you. I will work on the remaining three failed tests. Before about five tests failed.

egorsmkv · 2026-07-19T11:59:01Z

@agronholm can you bump the version to 0.1.26, please?

tapetersen · 2026-07-19T12:03:00Z

@egorsmkv I can so it but if you want to continue testing it may be better to just create your own PR that will be triggered when you push. I created this one to iron out the Unix bugs and see how far I came to complete. Since I'm a collaborator now I can even push to your branch if you check that flag so it may be easier to just handle that way.

egorsmkv · 2026-07-19T12:13:37Z

@tapetersen now all checks are passed so there is no need or it

agronholm · 2026-07-19T15:28:23Z

@tapetersen is this good for merging? I see it's still marked as a draft.

tapetersen · 2026-07-19T17:54:07Z

@agronholm There is still the one refcycle leak on happy-eyeballs I marked as skip since it wasn't fixed and I didn't know if we eanted to take on extra test-burden when I created it but if the test-integration looks ok for you I don't mind merging it.

agronholm · 2026-07-19T21:32:14Z

+        # These below may be sent with explicit default arguments and so must accept
+        # these literally
+        cwd: StrOrBytesPath | None = None,
+        env: Mapping[str, str] | None = None,
+        startupinfo: Any = None,
+        creationflags: int = 0,
+        start_new_session: bool = False,
+        pass_fds: Sequence[int] = (),


Hm, can you help me understand this change? It doesn't seem directly related to the subject of the PR.

While trying to initially get rsloop to pass one of the failures was that it didn't accept these arguments (at least cwd and env) explicitly as None while it handled them correctly if omitted (treating them as if None was passed).

Since anyio actually passes None explicitly see https://github.com/agronholm/anyio/blob/master/src/anyio/_core/_subprocesses.py#L184 a backend is required to accept that interface (as **kwargs: Any doesn't put any requirements on them).

The bug is since fixed in rsloop so it's not required but was related. I like making these kind of things explicit in the types if possible but can also just remove it or put it in a separate MR if you prefer.

But as long as we keep forwarding **kwargs:Any it won't really catch that much of anything with typing so I can very well see the case for removing it.

So basically rsloop would be fine with open_process(...) and do the correct default thing for cwd and env if ommitted but not accept the explicit defaults as in open_process(..., cwd=None, env=None) as it didn't like the None type.

agronholm · 2026-07-20T23:01:37Z

I don't think this needs a changelog entry as it's not a user-facing change. Thoughts?

tapetersen · 2026-07-21T17:35:56Z

@agronholm I agree and remove it. How do you want to do with the extra argument annotations for open_process?

They are not necessary and are a matter of taste but I tend to prefer them.

Also are you ok with the skipped ref-leak test for happy eyeballs or should I unskip it and hold the PR until it's fixed in rsloop?

agronholm · 2026-07-21T18:10:58Z

Also are you ok with the skipped ref-leak test for happy eyeballs or should I unskip it and hold the PR until it's fixed in rsloop?

@egorsmkv What's your take on this? A refleak isn't that serious a problem, but if you are planning to fix that soonish, then I'd rather wait for that.

egorsmkv · 2026-07-21T19:23:59Z

Also are you ok with the skipped ref-leak test for happy eyeballs or should I unskip it and hold the PR until it's fixed in rsloop?

@egorsmkv What's your take on this? A refleak isn't that serious a problem, but if you are planning to fix that soonish, then I'd rather wait for that.

I will try to fix that tomorrow

This reverts commit 5494d02.

Remove the skipped test as it should be fixed upstream before merge.

tapetersen · 2026-07-22T14:57:19Z

@egorsmkv I've removed the skip in the code so it should start failing on CI again.

egorsmkv · 2026-07-23T10:05:44Z

okay, let's test with v0.1.27

tapetersen · 2026-07-23T10:49:06Z

@egorsmkv I've retriggered it. I think it should pick up the latest if published since we don't pin any dependencies here but of not I'll push a change with a higher bound.

egorsmkv · 2026-07-24T05:47:56Z

@tapetersen can you update to v0.1.30? i tested it in #1238 and it now passed tests

tapetersen · 2026-07-24T20:00:49Z

@egorsmkv Done

egorsmkv · 2026-07-25T07:14:55Z

@agronholm

agronholm

Thanks!

Add rsloop as backend for asyncio tests.

d5cd831

Vizonex approved these changes Jul 7, 2026

View reviewed changes

Merge branch 'master' into test-rustloop

da97856

egorsmkv mentioned this pull request Jul 18, 2026

Fix trio on python 3.10 RustedBytes/rsloop#63

Closed

Merge branch 'master' into test-rustloop

83e1172

Bump the minimum rsloop version

a4350e4

Bump rsloop requirement

94f3692

Merge branch 'master' into test-rustloop

b84aa25

tapetersen marked this pull request as ready for review July 19, 2026 17:54

agronholm reviewed Jul 19, 2026

View reviewed changes

tapetersen added 3 commits July 20, 2026 01:13

Add complete argument list to AsyncBackend.open_process

ef006c4

Add versionhistory line.

5494d02

Merge remote-tracking branch 'upstream/master' into test-rustloop

2ea9936

tapetersen added 2 commits July 22, 2026 16:53

Revert "Add versionhistory line."

7603c2f

This reverts commit 5494d02.

Remove skipped test on rsloop.

37b4629

Remove the skipped test as it should be fixed upstream before merge.

Bump rsloop dependency.

56138a2

tapetersen added 2 commits July 24, 2026 21:59

Bump rsloop in test.

8284534

Merge remote-tracking branch 'upstream/master' into test-rustloop

4e16584

Merge branch 'master' into test-rustloop

20c04e0

Reverted unintended changelog changes

b4b2933

agronholm reviewed Jul 25, 2026

View reviewed changes

Comment thread tests/test_sockets.py Outdated

Update tests/test_sockets.py

3dc7d54

agronholm approved these changes Jul 25, 2026

View reviewed changes

agronholm merged commit caca0e0 into agronholm:master Jul 25, 2026
20 checks passed

Uh oh!

Conversation

tapetersen commented Jul 1, 2026 • edited by agronholm Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes

Checklist

Uh oh!

Vizonex left a comment

Choose a reason for hiding this comment

Uh oh!

tapetersen commented Jul 8, 2026

Uh oh!

Vizonex commented Jul 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

agronholm commented Jul 10, 2026

Uh oh!

tapetersen commented Jul 10, 2026

Uh oh!

agronholm commented Jul 10, 2026

Uh oh!

tapetersen commented Jul 10, 2026

Uh oh!

agronholm commented Jul 10, 2026

Uh oh!

tapetersen commented Jul 10, 2026

Uh oh!

tapetersen commented Jul 10, 2026

Uh oh!

agronholm commented Jul 10, 2026

Uh oh!

egorsmkv commented Jul 15, 2026

Uh oh!

tapetersen commented Jul 15, 2026

Uh oh!

agronholm commented Jul 15, 2026

Uh oh!

egorsmkv commented Jul 17, 2026

Uh oh!

egorsmkv commented Jul 18, 2026

Uh oh!

agronholm commented Jul 18, 2026

Uh oh!

egorsmkv commented Jul 18, 2026

Uh oh!

agronholm commented Jul 19, 2026

Uh oh!

egorsmkv commented Jul 19, 2026

Uh oh!

egorsmkv commented Jul 19, 2026

Uh oh!

tapetersen commented Jul 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

egorsmkv commented Jul 19, 2026

Uh oh!

agronholm commented Jul 19, 2026

Uh oh!

tapetersen commented Jul 19, 2026

Uh oh!

agronholm Jul 19, 2026

Choose a reason for hiding this comment

Uh oh!

tapetersen Jul 19, 2026

Choose a reason for hiding this comment

Uh oh!

tapetersen Jul 19, 2026

Choose a reason for hiding this comment

Uh oh!

tapetersen Jul 19, 2026

Choose a reason for hiding this comment

Uh oh!

agronholm commented Jul 20, 2026

Uh oh!

tapetersen commented Jul 21, 2026

Uh oh!

agronholm commented Jul 21, 2026

Uh oh!

egorsmkv commented Jul 21, 2026

Uh oh!

tapetersen commented Jul 1, 2026 •

edited by agronholm

Loading

Vizonex commented Jul 8, 2026 •

edited

Loading

tapetersen commented Jul 19, 2026 •

edited

Loading