task completion handling in asyncio.run_until_complete()

[yoctopuce]

First a bit of context: as part of my implementation of a MicoPython code debugger, I have implemented an asyncio task inspector. In order to do so, I have instrumented functions task_make_new and task_attr in modasyncio.c to track creation and completion using a C callback hook.

I have noticed today that whenever a task is run via run_until_complete, the task object was not properly marked as done, possibly causing strange side effects.

Looking at asyncio/core.py, I see that indeed, when the so-called main task is done, the function exits before handling the task state:

    # This task is done, check if it's the main task and then loop should stop
    if t is main_task:
        cur_task = None
        if isinstance(er, StopIteration):
            return er.value
        raise er
    if t.state:
        # ...

I was able to fix my problem by setting t.state to False just after clearing cur_task, but I was wondering if the proper fix might involve a bit more than that. In particular, I wonder if t.state could possibly be a task queue, that should be pushed to the global _task_queue for proper continuation later on. In the most common case where run_until_complete is invoked as a unique entry point, this would not be necessary, but there are other cases where run_until_complete could be used in more subtle ways.

What do you think ?

[peterhinch]

It's not clear to me whether there are circumstances where this would affect normal code. If you can produce a simple repro demonstrating a bug, I suggest raising an issue.

[yoctopuce]

Yes, I have a simple code snipplet that shows that this incorrect handling may indeed cause adverse side effects, and is not in line with CPython implementation. I will make a pull request with the suggested fix.

[IHOXOHI]

If I understand correctly, maybe you could use one more 'main_task' to manipulate each task like you want...

[yoctopuce]

Not sure to understand this suggestion, but anyway I know how to work around the issue, or to fix it for myself, that is not the question.
My question was whether if this should be fixed in the official micropython distribution for everyone's benefit.