gh-99553: fix bug where an ExceptionGroup subclass can wrap a BaseException #99572
Conversation
iritkatriel
added
type-bug
|
Thanks @iritkatriel for the PR 🌮🎉.. I'm working now to backport this PR to: 3.11. |
…aseException (pythonGH-99572) (cherry picked from commit c8c6113) Co-authored-by: Irit Katriel <1055913+iritkatriel@users.noreply.github.com>
|
GH-99580 is a backport of this pull request to the 3.11 branch. |
| Like :exc:`ExceptionGroup`, any subclass of :exc:`BaseExceptionGroup` which | ||
| is also a subclass of :exc:`Exception` can only wrap instances of | ||
| :exc:`Exception`. |
There was a problem hiding this comment.
As written, consistency suggests that class MyEG(BaseExceptionGroup, ValueError): should be unable to wrap anything but ValueErrors!
That may in fact be the intention, and it wouldn't break the trio.MultiError-as-BaseExceptionGroup deprecation pathway, but it's not the implementation and the tests don't distinguish these cases.
Otherwise, I think we'd want to document the behavior which is that "any subclass of ExceptionGroup can only wrap instances of Exception, like ExceptionGroup itself.", and therefore class MyEG(BaseExceptionGroup, ValueError): would be able to contain BaseExceptions 🤔
There was a problem hiding this comment.
As written, this suggests that class MyEG(BaseExceptionGroup, ValueError): should be unable to wrap anything but ValueErrors!
Does it? In the doc the Exception becomes a hyperlink.
There was a problem hiding this comment.
But feel free to make a PR with better wording.
There was a problem hiding this comment.
I expect some users to read that as "if a subclass of BaseExceptionGroup is also a subclass of X, it can only wrap instances of X"; I'll open a PR to rephrase as "if it's also a subclass of Exception, it can't wrap BaseExceptions".
But then we're still treating Exception as particularly special for subclasses, which feels a bit strange.
There was a problem hiding this comment.
Zac, I'm confused why you think the sentence as written could be interpreted that way. It says Exception twice. In this context that is very clearly a reference to the specific built-in class named Exception (it even links there), not a "free variable" that must refer to the same class in both places.
Regardless, even if we clear that up, it feels like it would be expensive to inspect the class hierarchy to find other subclasses of BaseException and try to constrain the arguments to inherit from those. And that's definitely not what the PEP intended. We want the common case (where you're using one of the two built-in EG classes) to be foolproof, and other cases possible.
There was a problem hiding this comment.
Apologies for the confusion, that's on me 😞
I do think the behavior in this PR is good, and I'll open a PR with some additional tests and maybe docs to distinguish it from the case that I was wondering about. Thanks again to both of you for everything.
Fixes #99553.
ExceptionGroupcan wrapBaseExceptions #99553