F401+F811 false positive

[dalcinl]

trafficstars

Minimal reproducer

from foo import Foo
from typing import TypeAlias

class Bar:
    Foo: TypeAlias = Foo
    def bar(self) -> Foo: ...

Output

$ ruff stub.pyi 
stub.pyi:1:17: F401 `foo.Foo` imported but unused
stub.pyi:5:5: F811 Redefinition of unused `Foo` from line 1
Found 2 error(s).
1 potentially fixable with the --fix option.

Ruff version: 0.0.227

[ngnpope]

Confirmed with 0.0.236 with a .pyi file. Doesn't occur with a .py file unless prepending with from __future__ import annotations.

[charliermarsh]

Is this, maybe, the same as #1401?

[dalcinl]

I believe it is not exactly the same. My reproducer would be equivalent to:

from foo import Foo

class Bar:
    Foo = Foo
    def bar(self) -> Foo: ...

and ruff is happy with this code.

[dalcinl]

This is a small variation to my original reproducer:

from typing import TypeAlias
class Foo: ...

class Bar:
    Foo: TypeAlias = Foo
    def bar(self) -> Foo: ...

This time Ruff is happy with it.

1. My original reproducer triggers the warnings.
2. Similar code with a locally defined symbol does not.
3. Removing the TypeAlias annotation silences the warning.
4. Changing the annotation to Foo: Type = Foo also silences the warning. So far this looks like a subtle bug related to visibility and/or the TypeAlias annotation.

[charliermarsh]

Yeah this is really tricky.

[charliermarsh]

Obligatory link to Tweet: https://twitter.com/charliermarsh/status/1659233087786459139

[charliermarsh]

I believe we need to change the logic in this case to eagerly resolve if the binding is already defined, and otherwise, go through standard deferred resolution.