Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.35%. Comparing base (61cb138) to head (e326b9a).

@@            Coverage Diff             @@
##           master     #131      +/-   ##
==========================================
+ Coverage   90.32%   90.35%   +0.04%     
==========================================
  Files          15       15              
  Lines        1125     1129       +4     
==========================================
+ Hits         1016     1020       +4     
  Misses        109      109              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@sphuber, can you also add the fixes for #138? I think we can just pinning the version of aio-pika to ~=9.4.0?

Docs failing with

WARNING: failed to reach any of the inventories with the following issues:
intersphinx inventory 'https://aio-pika.readthedocs.io/en/latest/objects.inv' not fetchable due to <class 'requests.exceptions.HTTPError'>: 404 Client Error: Not Found for url: https://aio-pika.readthedocs.io/en/latest/objects.inv

@unkcpz applied the aio-pika pin but unfortunately there are still two tests failing that don't seem trivial to fix.

Right, I'll look into it.

EDIT: the doc build is fixed by updating the aio-pika doc link.

I think the failed test came from a070bfe
In _outcom_destroyed the coroutine might be called from the other thread therefore it requires the run_coroutine_threadsafe. Why you made this change in the first place @sphuber?

This eager_task_factory feature of py3.12 seems more like call_soon_threadsafe.
In the outcome destroy case it seems the correct async primitive to use, but I think it is better for another PR to check. @sphuber can you revert that commit?

Thanks for revert the change @sphuber. Now there are two tests failed and I'll try to look into it.

I didn't find anything in https://docs.python.org/3/whatsnew/3.12.html#asyncio that may cause the failed test after using python3.12. But my guess is that performance improvement was by tightly schedule the task into even loop and run without another round of event loop between schedule requeue and task_queue.get. The change in the #139 is explicitly relinquish the control before get task from queue to make sure all the requeue coroutines are scheduled.

Fantastic, thanks a lot for the help @unkcpz