GH-95289: Always call uncancel() when parent cancellation is requested

[kumaraditya303]

• Issue: asyncio.TaskGroup corrupts uncancel() stack when the parent task replaces the cancelled error #95289

Co-Authored-By: Guido van Rossum [email protected]

[gvanrossum]

We're getting very close to the RC1 release date (this Friday!). I'd like to get this in. @graingert can you bless this as strictly better than what we had before?

[kumaraditya303]

We're getting very close to the RC1 release date (this Friday!). I'd like to get this in

I agree, I would like to get this in ASAP.

[graingert]

We're getting very close to the RC1 release date (this Friday!). I'd like to get this in. @graingert can you bless this as strictly better than what we had before?

This now follows the five (six? Cc @ambv) rules of uncancellation so I think therefore it's strictly better. However I can't say for sure. It looks to me like TaskGroup was relying on task.cancelling() >= 1 traveling up the stack to communicate with parent scopes without anticipating that it also travels down the rest of the function body.

[graingert]

In addition I think there's other more fundamental problems with uncancel, for example when a cancellation is delivered from a chained concurrent.fututes.Future or when a TaskGroup task waits on another TaskGroup task and cancellation is chained between them

[miss-islington]

Thanks @kumaraditya303 for the PR, and @gvanrossum for merging it 🌮🎉.. I'm working now to backport this PR to: 3.11.
🐍🍒⛏🤖

[bedevere-bot]

GH-95654 is a backport of this pull request to the 3.11 branch.