itertools.product called on a Flag enum is inferred as object rather than class name

[jatkinson1000]

Bug Report

When defining a class based on Flag from the enum package I would expect items of the class to be of the type of the class when I iterate over them using itertools functions.
However, instead mypy raises an error that they are instead of type object.

To Reproduce

See playground link here: https://mypy-play.net/?mypy=latest&python=3.12&gist=5dc24e94f25872e5c5d8ac032c4d8309

Expand for code in playground to reproduce.

from enum import Flag, auto
import itertools


class Test_Flag_1(Flag):
    TF1a = auto()
    TF1b = auto()
    TF1c = auto()

class Test_Flag_2(Flag):
    TF2a = auto()
    TF2b = auto()
    TF2c = auto()

def test_func_1(flagitem: Test_Flag_1) -> None:
    print(flagitem.name)

def test_func_2(flagitem: Test_Flag_2) -> None:
    print(flagitem.name)


if __name__ == "__main__":
    for flagitem1, flagitem2 in itertools.product(
        Test_Flag_1, Test_Flag_2
    ):
        test_func_1(flagitem1)
        test_func_2(flagitem2)

Expected Behavior

I would expect this to pass fine through mypy.

Actual Behavior

Instead items of the Test_Flag_1 or Test_Flag_2 subclasses of Flag when iterated over using itertools.product are picked up by mypy as type object instead of type Test_Flag_1 or Test_Flag_2.

(venv) [user@machine ~]$ mypy mypy_minimal.py
mypy_minimal.py:26: error: Argument 1 to "test_func_1" has incompatible type "object"; expected "Test_Flag_1"  [arg-type]
mypy_minimal.py:27: error: Argument 1 to "test_func_2" has incompatible type "object"; expected "Test_Flag_2"  [arg-type]
Found 2 errors in 1 file (checked 1 source file)

Your Environment

• Mypy version used: 1.15.0 (compiled: yes)
• Mypy command-line flags: None
• Mypy configuration options from mypy.ini (and other config files): None
• Python version used: 3.13.1 (also reproduced with 3.12 in the playground link)

Full pip list:

Package           Version
----------------- -------
mypy              1.15.0
mypy-extensions   1.0.0
typing-extensions 4.12.2

[jatkinson1000]

Strangely, when I run this on a subset of the Flag subclass using the | operator rather than the entire subclass mypy passes fine:

if __name__ == "__main__":
    for flagitem1, flagitem2 in itertools.product(
        Test_Flag_1.TF1a | Test_Flag_1.TF1b, Test_Flag_2.TF2a | Test_Flag_2.TF2b
    ):
        test_func_1(flagitem1)
        test_func_2(flagitem2)

[jatkinson1000]

Similarly, expanding the function of itertools.product into two nested loops passes fine, so it seems to be the way in which mypy interacts with the itertools.product over a Flag enum:

if __name__ == "__main__":
    for flagitem1 in Test_Flag_1:
        for flagitem2 in Test_Flag_2:
            test_func_1(flagitem1)
            test_func_2(flagitem2)

[ilevkivskyi]

I can't reproduce this on the master branch, and we usually close things as soon as they are fixed on master. (You can also try it in the playground like this https://mypy-play.net/?mypy=master&python=3.12&gist=5dc24e94f25872e5c5d8ac032c4d8309)

[jatkinson1000]

Thanks, can confirm that gist passes fine using master branch.

Any idea when the fix will make it into a release, or advice on how one can use the master branch in a project in anticipation?

Also, do you know which commit/PR resolved this - I would be interested in understanding better how mypy works. Don't worry if not though.

[ilevkivskyi]

To use dev build you can do something like this https://mypy.readthedocs.io/en/stable/common_issues.html#using-a-development-mypy-build (although this is not recommended for production code). Next release should be around mid-March.

Also, do you know which commit/PR resolved this - I would be interested in understanding better how mypy works. Don't worry if not though.

I didn't bisect, but most likely it is my recent PR #18587 where I stopped using both class and instance for protocol inference against class object if an attribute is present on both class and the metclass (like __iter__ in this case). Only metaclass one is used now as it should.