Regression in Async Libraries discovery

[mautini]

Things to check first

• I have searched the existing issues and didn't find my bug already reported there

• I have checked that my bug is still present in the latest release

AnyIO version

4.12.0

Python version

3.10.17

What happened?

Hello,

Before version 4.12.0, AnyIO used sniffio to detect available async libraries in the Python environment. Sniffio raised an AsyncLibraryNotFoundError when no async library was available, which was very convenient in some cases (for example, when running asynchronously code in production but synchronously in tests, with asynchronous libs in the classpath).

Starting with version 4.12, sniffio was removed and async library detection is now handled by custom code (PR #1021). However, this introduces two issues:

• The new NoCurrentAsyncBackend exception, which is raised when no event loop is present, lives in the protected _core module and therefore should not be accessed from outside the library.
• The newly added get_available_backends function only checks which backends are available using import; it does not verify whether an event loop is actually running.

I would recommend either exposing this exception so it is accessible to end users, or updating the backend availability check to reflect real availability (i.e., whether it is being called from within an active event loop).

How can we reproduce the bug?

anyio==4.12.0
asyncio==4.0.0

import anyio

if __name__ == '__main__':
    print(anyio.get_available_backends()) # ('asyncio',)
    anyio.to_thread.current_default_thread_limiter() # Trigger anyio._core._eventloop.NoCurrentAsyncBackend (protected member)

[agronholm]

I suppose I should make that exception public, as it does bubble to callers of certain functions when there is no running event loop.

As for get_available_backends(), that function is indeed only meant to look for importable back-ends, not check whether something is running. That was never the point. It was primarily intended to check which backends can be used to run the test suite with. Why is this a problem?

[agronholm]

Hmm...there's already a NoEventLoopError in the public API. Perhaps I should repurpose that to also cover the cases where NoCurrentAsyncBackend is currently raised.

[mautini]

As for get_available_backends(), that function is indeed only meant to look for importable back-ends, not check whether something is running. That was never the point. It was primarily intended to check which backends can be used to run the test suite with. Why is this a problem?

This is not an issue if a public exception is available to handle this case. I proposed two possible solutions, but it makes sense to keep get_available_backends as it is right now.