bpo-39101: Fixes BaseException hang in IsolatedAsyncioTestCase.

[lisroach]

If a BaseException is raised in IsolatedAsyncioTestCase the test will hang forever. This should fix that, added a unit test that would hang forever without the fix.

https://bugs.python.org/issue39101

[1st1]

Please add except (SystemExit, KeyboardInterrupt): raise before handling the BaseException.

[lisroach]

@1st1 this will still leave SystemExit and KeyboardInterrupt hanging if they are raised, I think they should be handled along with the BaseException.

[1st1]

lgtm but please take a look at the suggested change.

[danielnelson]

Should the except where we reraise asyncio.CancelledError be removed too? I ran into a hang with a test like:

import asyncio
import unittest

class TestHangsForever(unittest.IsolatedAsyncioTestCase):
    async def test_hangs_forever(self):
        task = asyncio.create_task(asyncio.sleep(1))
        task.cancel()
        await task

if __name__ == "__main__":
    import unittest
    unittest.main()

[lisroach]

@danielnelson you are right, moving asyncio.CancelledError to be handled the same way as BaseException fixes the hang.

@1st1 or @asvetlov, I am confused why KeyboardInterrupt and SystemExit are not handled along with BaseException. I read a little of python/asyncio#341 that seems to indicate there are some fundamental differences in how they are handled. I also see issue40152 that seems to reference the hanging problem.
A test like:

async def test_sys_exit(self):
    raise SystemExit()

Just hangs forever. Should I file a new bug for this?

[1st1]

@1st1 or @asvetlov, I am confused why KeyboardInterrupt and SystemExit are not handled along with BaseException. I read a little of python/asyncio#341 that seems to indicate there are some fundamental differences in how they are handled. I also see issue40152 that seems to reference the hanging problem.

They are specially handled because asyncio has no way of correctly recovering from them. They can happen at any time (including when asyncio runs its own private implementation code) wrecking the internal state. So if asyncio sees them it just propagates them immediately with the intent to crash early.

This is something we'd like to fix eventually, but it will require some substantial internal changes. For now I suggest to treat them just like they're treated in asyncio (propagate them immediately).

[1st1]

The fix looks correct to me.

[asvetlov]

LGTM!

[danielnelson]

In addition to the bug listed in the original comment, I believe this PR also fixes https://bugs.python.org/issue39706

[miss-islington]

Thanks @lisroach for the PR 🌮🎉.. I'm working now to backport this PR to: 3.8, 3.9.
🐍🍒⛏🤖

[bedevere-bot]

GH-22988 is a backport of this pull request to the 3.9 branch.

[bedevere-bot]

GH-22989 is a backport of this pull request to the 3.8 branch.