Enum `_missing_` function changes `__contains__` behaviour

[artur-sl]

Bug report

Bug description:

I observed different behavior for Enum __contains__ in 3.13.2 and 3.13.3 if there is _missing_ method implemented:

from enum import Enum

class Color(Enum):
    RED = 1
    
    @classmethod
    def _missing_(cls, value):
        return cls.RED

# 3.13.2
>>> "blue" in Color
False

# 3.13.3
>>> "blue" in Color
True

IMO _missing_ should only influence the __call__ method:

# both 3.13.2 and 3.13.3
>>> Color("blue")
<Color.RED: 1>

CPython versions tested on:

3.13

Operating systems tested on:

macOS

Linked PRs

• gh-132684: [Enum] only call _missing_ in __contains__ for Flags #132790
• [3.13] gh-132684: [Enum] only call _missing_ in __contains__ for Flags (GH-132790) #132896

[noamcohen97]

I am fairly certain this is a side effect of #131053, but I'm not sure what the correct behavior should be.
@picnixz @ethanfurman, what are your thoughts?

[picnixz]

Arf... this is something I didn't think of. This can be pretty annoying and we could also need to revert it in 3.12 :< I'll mark it as a release-blocker for now to draw attention.

[picnixz]

Actually, Ethan said on the previous issue:

In other words, if Weekday(x) would return a valid member, then x in Weekday should return True.

Now, here Color("blue") returns Color.RED. So... if we were to follow the same logic, "blue" in Color would still be correct. I'm not entirely sure whether we need to fix this or not but I know for sure that sometimes we do checks of the form x in ENUM and depending on how the ENUM is being done, we may or may not have broken some code out there.

[s3rius]

I guess it makes total sense, because usually you only check that the value is present in the enum to create its member.

But if you really want to mimic the behavior of 3.12 then you can use this workaround:

>>> 'blue' in {color.name for color in Color}
False

[noamcohen97]

Perhaps we should simply document the change in behavior?

[picnixz]

If possible, yes, but it's also sad that the behavior of some constructions have been broken in a stable release.

[ethanfurman]

The change was for Flags -- i'll look at narrowing the code to reflect that.