Incorrect union narrowing in expression with assignments

[elprans]

bug.py:

from __future__ import annotations
from typing import *


def foo() -> Optional[str]:
    return None


if ((f := foo()) is None or f.lower()):
    pass

$ mypy bug.py
bug.py:9: error: Item "None" of "Optional[str]" has no attribute "lower"
Found 1 error in 1 file (checked 1 source file)

[1st1]

@JukkaL @ilevkivskyi Jukka, Ivan, sorry for summoning you explicitly, but do you guys think you could take a look at this? This really blocks us on a few things... :(

[emmatyping]

@1st1 I realize that this could be painful, but for the provided example, a simple refactor to

f = foo()
if (f is None or f.lower()):
    pass

works as expected. If you have another example where it would be difficult or impossible to work around, we can bump the priority.

Also I believe @Michael0x2a has been working on type narrowing the most recently.

[elprans]

@ethanhs For some background, we have a lot of code that basically does this:

if is_foo(obj):
   obj.method()

Where is_foo is some non-trivial check on the obj type. Naturally, mypy would not understand constructs like this, since, unlike Typescript, it does not support user-defined type assertions. So, one way to handle this is to change is_foo to return Optional[Foo] instead of bool, which would be almost painless using the walrus operator:

if foo := is_foo(obj):
    foo.method()

The workaround you suggest does not work as easily when the conditional is part of some long chain of elif blocks.

[1st1]

works as expected. If you have another example where it would be difficult or impossible to work around, we can bump the priority.

Yeah, the workaround is obvious in simple cases. Take a look at @elprans' explanation of how we want to use walrus/mypy in our EdgeDB codebase. Would be amazing if you could bump the priority as having this fixed can simplify our code quite a bit.

[ilevkivskyi]

Just to double-check, did you try it on master? There is a chance #8258 fixed this.

In any case, unfortunately I will not have time to work on this.

[elprans]

@ilevkivskyi #8258 does seem to solve the simple case in #8128 (comment), but not the original issue reported here.

[elprans]

This appears to have been fixed in #8458 (at least the original issue).

[emmatyping]

Confirmed as fixed.