CPython only makes these methods class method when a class created. If you set __class_getitem__ method after the creation it doesn't work unless you use classmethod decorator manually.

>>> class B:
...     pass
... 
>>> def y(*a, **k):
...     return a, k
... 
>>> B.__class_getitem__ = y 
>>> B[int]
((<class 'int'>,), {})
>>> B.__class_getitem__ = classmethod(y)
>>> B[int]
((<class '__main__.B'>, <class 'int'>), {})

Is this an issue though? Shouldn't methods be defined in the class definition, we don't expect setting a function as an attribute to a class to transform it automatically to a method.

Why would we special case __class_getitem__ and not __init_subclass__?

> CPython only makes these methods class method when a class created.
> If you set __class_getitem__ method after the creation it 
> doesn't work unless you use classmethod decorator manually.

Give the intended use case for __class_getitem__ and that it was replacement for metaclass code with the same effect, the current behavior doesn't seem unreasonable.  That said, I don't see any downside to allowing the method to attach after class creation.

Assigning to Ivan to make the call on this one.

__new__ is special function too.  It is converted as staticmethod when class creation.

Since manually converting to (static|class)method is allowed, I don't think
automatic conversion should be applied when attach after class creation.

$ python3
Python 3.7.2 (default, Feb 12 2019, 08:15:36)
[Clang 10.0.0 (clang-1000.11.45.5)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> class C:
...   def __new__(cls):
...     pass
...
>>> type(C.__dict__['__new__'])
<class 'staticmethod'>
>>> class D:
...   pass
...
>>> D.__new__ = lambda cls: cls
>>> type(D.__dict__['__new__'])
<class 'function'>

I don't like complicating the code and adding a performance penalty to support such uncommon case.

The documentation for __class_getitem__ precisely describes the current behavior: "when defined in the class body, this method is implicitly a class method". Nothing should be changed here.

The documentation for __init_subclass__ could be rewritten in similar words.

I totally agree with Serhiy

messages: + msg336026