ASYNC100 now treats start_soon() as a cancel point #327
Conversation
flake8_async/visitors/visitor91x.py
Outdated
| def visit_Call(self, node: cst.Call) -> None: | ||
| # [Nursery/TaskGroup].start_soon introduces a cancel point | ||
| if ( | ||
| isinstance(node.func, cst.Attribute) | ||
| and isinstance(node.func.value, cst.Name) | ||
| and node.func.attr.value == "start_soon" | ||
| and node.func.value.value in self.taskgroups | ||
| ): | ||
| self.checkpoint_cancel_point() |
There was a problem hiding this comment.
I don't think this is quite correct; .start_soon(...) is not itself a cancel point, but rather it means that Nursery.__aexit__ will be a cancel point.
(unless there is an earlier checkpoint in the nursery block, during which the task finishes - in which case we can rely on that checkpoint instead)
It's a pretty subtle distinction but we should be careful about that in both implementation and docs.
There was a problem hiding this comment.
Ah good catch, I thought it wouldn't make a difference to the implementation at first but turns out it does. Fixed~
|
Upon further consideration a rule for "redundant" cancel scopes would probably be quite tricky, and only be applicable if the potentially redundant scope has no timeout or shielding, and is not getting manually canceled, etc etc. So probably not worth considering |
unreverts #317 (partially reverted in #326) and fixes #325