Sam Bull <m6f...@sambull.org> added the comment:
> If the task is already pending a cancellation targeted at a cancel scope, the > task itself cannot be cancelled anymore since calling cancel() again on the > task is a no-op. This would be solved by updating the cancel message on the > second call. > I think Andrew missed one case: in his second diagram, what if the explicit > cancel() happened *after* the timeout (but still in the same iteration)? > That's the case that makes me want to delete those two lines from > Task.cancel() (see my earlier message). To expand on this point, I've been looking at solving the race conditions in async-timeout. To see how such a race condition can end up with a task never exiting, take a look at this example: https://github.com/aio-libs/async-timeout/issues/229#issuecomment-908502523 In the condition Guido describes, the user's cancellation is suppressed and the code runs forever. I also wrote tests that seem to reliably reproduce the race condition (the 2nd one still seems unfixable with the current solutions, the 1st was fixed with the nonce/sentinel trick): https://github.com/aio-libs/async-timeout/commit/ab04eb53dcf49388b6e6eacf0a50bafe19c5c74b#diff-60a009a48129ae41018d588c32a6d94c54d1d2948cbc3b831fc27a9c8fdbac68L364-L421 You can see the flow of execution from the call_order assert at the end. I think most of the solutions proposed here will still not solve this race condition. I initially proposed a solution at: https://bugs.python.org/issue45098 In short, I think that every time we call .cancel(), we need to raise another CancelledError. So, in this race condition you would get 2 CancelledErrors (via an ExceptionGroup). Then our code can catch the error with our nonce/sentinel and handle that, but also reraise any other errors which are unrelated. ---------- nosy: +dreamsorcerer _______________________________________ Python tracker <rep...@bugs.python.org> <https://bugs.python.org/issue46771> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
