StrEnum.__getitem__ raises TypeError

[injust]

Bug report

Bug description:

from enum import Enum, StrEnum, auto

class Foo(Enum):
    one = auto()
    two = auto()
    
class Bar(StrEnum):
    one = auto()
    two = auto()
    
names = ["one", "two"]

print([Foo[name] for name in names])
print(list(map(Foo.__getitem__, names)))

print([Bar[name] for name in names])
print(list(map(Bar.__getitem__, names)))  # TypeError: expected 1 argument, got 0

CPython versions tested on:

3.12, 3.13

Operating systems tested on:

macOS

[tom-pytel]

str class slots are overriding EnumType mapping for __getitem__ and others in StrEnum for certain uses.

>>> Foo.__getitem__
<bound method EnumType.__getitem__ of <enum 'Foo'>>

>>> Bar.__getitem__
<slot wrapper '__getitem__' of 'str' objects>

They work implicitly but not explicitly:

>>> Bar['one']
<Bar.one: 'one'>

>>> Bar.__getitem__('one')
Traceback (most recent call last):
  File "<python-input-19>", line 1, in <module>
    Bar.__getitem__('one')
    ~~~~~~~~~~~~~~~^^^^^^^
TypeError: expected 1 argument, got 0

>>> Bar.__getitem__(Bar, 'one')
Traceback (most recent call last):
  File "<python-input-20>", line 1, in <module>
    Bar.__getitem__(Bar, 'one')
    ~~~~~~~~~~~~~~~^^^^^^^^^^^^
TypeError: descriptor '__getitem__' requires a 'str' object but received a 'EnumType'

Because Bar['one'] calls EnumType.__getitem__() and Bar.__getitem__('one') calls str.__getitem__(). This also applies to other magics like __contains__ or __len__:

>>> Bar.__contains__
<slot wrapper '__contains__' of 'str' objects>

>>> Bar.__len__
<slot wrapper '__len__' of 'str' objects>

>>> len(Bar)
2

>>> Bar.__len__()
Traceback (most recent call last):
  File "<python-input-22>", line 1, in <module>
    Bar.__len__()
    ~~~~~~~~~~~^^
TypeError: descriptor '__len__' of 'str' object needs an argument

>>> Bar.__len__(Bar)
Traceback (most recent call last):
  File "<python-input-23>", line 1, in <module>
    Bar.__len__(Bar)
    ~~~~~~~~~~~^^^^^
TypeError: descriptor '__len__' requires a 'str' object but received a 'EnumType'

>>> Bar.__len__('abc')
3

Quick hack fix, add the following to StrEnum to re-override methods overridden by str (if this is the desired behavior):

    # re-override methods overridden by str, see gh-128659
    __contains__ = classmethod(EnumType.__contains__)
    __getitem__ = classmethod(EnumType.__getitem__)
    __iter__ = classmethod(EnumType.__iter__)
    __len__ = classmethod(EnumType.__len__)

Before:

>>> Foo.__getitem__
<bound method EnumType.__getitem__ of <enum 'Foo'>>
>>> Bar.__getitem__
<slot wrapper '__getitem__' of 'str' objects>

After:

>>> Foo.__getitem__
<bound method EnumType.__getitem__ of <enum 'Foo'>>
>>> Bar.__getitem__
<bound method EnumType.__getitem__ of <enum 'Bar'>>
>>> Bar.__getitem__('one')
<Bar.one: 'one'>

[serhiy-storchaka]

Then what Bar.one[0] will return?

[tom-pytel]

Then what Bar.one[0] will return?

It will not act as a str with respect to those magics as you point out but rather as an EnumType (see qualifier above "if this is the desired behavior"). So maybe not the desired behavior and the original code should just be invalid?

[ethanfurman]

I'll need to get some tests written for the expected behavior of those dunders.

[serhiy-storchaka]

We need different, well defined behavior for Bar['one'] and Bar.one[0], len(Bar) and len(Bar.one), etc. An enum class should behave as other enum classes, and an enum instance should behave as other instance of a parent class.

As for the dunder methods, we have precedents. For example, int.__or__ is a method that corresponds to an instance operation, while str.__or__ is a method that corresponds to a type operation.

>>> int.__or__
<slot wrapper '__or__' of 'int' objects>
>>> int.__or__(1, 2)
3
>>> str.__or__
<method-wrapper '__or__' of type object at 0x563a4b54f960>
>>> str.__or__(int)
str | int

So I think we should not change anything in StrEnum, at least not before we change the larger image.