Pyodide compatibility?

[hoodmane]

Would you be willing to accept patches to make this package work with Pyodide? Before version 1.2.0, it worked with Pyodide but from version 1.2.0 it fails with

RuntimeError: Runner.run() cannot be called from a running event loop

I'm not sure if this is better fixed in Pyodide or pytest-asyncio, probably it's better to fix in Pyodide. I think the cause of the issue is that Pyodide's WebLoop event loop doesn't handle lifecycle events. For instance, stop() is a no op. This is because our event loop schedules events on the JavaScript event loop and we have no control over the JavaScript event loop. But perhaps we can fake lifecycle events well enough to satisfy pytest-asyncio's needs.

Anyways if there is any interest it would be nice to add Pyodide to pytest-asyncio's CI once we get this working.

[seifertm]

Thanks! I wasn't aware that pytest-asyncio has issues on Pyodide. I certainly support efforts that make pytest-asyncio compatible, but I believe we should discuss how this can happen.

Currently, pytest-asyncio assumes that it has full control over the asyncio event loop. It uses asyncio.Runner internally, which takes care of event loop creation and cleanup. The transition from low-level asyncio functions to asyncio.Runner has been the goal since the deprecation of asyncio.set_event_loop in python/cpython#128218 (which was un-deprecated in v3.14)

Based on your report, I understand that Pyodide has its own event loop. This breaks pytest-asyncio's assumption of event loop control. Glib apparently has a similar issue (#1233).

I feel this all boils down to the "loop-in-loop" problem: Only one event loop can be running per thread at any particular time.

Essentially, we would have to find a solution for pytest-asyncio to piggy back on the Pyodide event loop or the other way around. In #235, Andrew proposed to integrate aioloop-proxy into pytest-asyncio a while ago. Since the loop-in-loop issue is a frequent source of issues for pytest-asyncio (#374 #383 #658 #543 #359 #186), it's maybe worth revisiting the loop proxy idea.

We should also look into how the Jupyter project tackles this, as they should a similar problem, as far as I know.

Feel free to share your opinions or additional ideas!