False Positive with subclassed Enums

[jonapich]

Adding custom methods to an Enum subclass and then implementing it: pylint can't see the new methods.

from enum import Enum, auto


class TestBase(Enum):
    """ Adds a special method to enums. """

    def hello_pylint(self) -> str:
        """ False positive. """
        return self.name


class TestEnum(TestBase):
    """ Tests the false positive for enums. """
    a = auto()
    b = auto()


test_enum = TestEnum.a
assert test_enum.hello_pylint() == test_enum.name

[no-member] Instance of 'auto' has no 'hello_pylint' member
[no-member] Instance of 'auto' has no 'name' member

Similar to #1932 and #2062

[PCManticore]

Thanks for the report!

[belm0]

I believe this problem happens regardless of subclassing.

What will it take for pylint/astroid to be aware of Enum's patching of class members?

[bhargavrpatel]

Just adding more examples on this as it may help close this issue.

import typing
from enum import Enum


class MyEnum(Enum):
    def to_pb2(self) -> int:
        return typing.cast(int, self.value)

    def __str__(self) -> str:
        return self.name.upper()  # pylint: disable=no-member


class SomeEnum(MyEnum):
    A = "A"


class AnotherEnum(Enum):
    B = "B"


SomeEnum.__members__.items()
AnotherEnum.__members__.items()

This will report an issue for SomeEnum.__members__

[belm0]

@nelfin do you have a PR forthcoming? nelfin/astroid@afd0c2b

It seems as astroid gets smarter about propagating inference, I hit this problem more and more in my codebase.

[nelfin]

@belm0, the fix you reference made it into the main astroid branch as pylint-dev/astroid@8181dc9 (see pylint-dev/astroid#941). It should be available in astroid 2.5.7 and later

[Pierre-Sassoulas]

Will be available in pylint 2.9.0 (see milestone for release planning), or you can have it with pylint 2.8.2 if you set the astroid version to 2.5.8 yourself.

[belm0]

@nelfin I'm actually still seeing this problem with subclassed enums on the latest astroid (2.5.8) and pylint pre-release. Has anyone confirmed otherwise?

[nelfin]

Ah, now I understand. The commits we've been talking about actually only address the issue in the comment by bhargavrpatel above (#2626 (comment)), which talks about Enum.__members__. I'd misread the issue when I was first looking for duplicates about false-positives on Enum.__members__. I don't know if any work has been done to address incorrect inference of attributes on enum.Enum subclasses.

[belm0]

If there's any guidance you can offer on what this problem might be or what parts of the implementation to poke at, I can try looking at it.

[nelfin]

@belm0, I think astroid/brain/brain_namedtuple_enum.py:infer_enum_class would probably be the most likely place to start