zip.__iter__ and reversed.__iter__ return self

[JelleZijlstra]

(On 3.9)

>>> class X(reversed): pass
... 
>>> x = X("abc")
>>> type(x)
<class '__main__.X'>
>>> type(iter(x))
<class '__main__.X'>
>>> class Y(zip):
...     pass
... 
>>> y = Y("abc", "def")
>>> type(y)
<class '__main__.Y'>
>>> type(iter(y))
<class '__main__.Y'>

[JelleZijlstra]

Hold on, this isn't true if you're reversing a list (thanks @AlexWaygood):

% ./python3.10 -c '''class X(reversed): pass           
print(type(iter(X([1, 2, 3]))))'''
<class 'list_reverseiterator'>

[AlexWaygood]

I think we should make the change to zip but leave reversed as it is and add a # noqa: Y034 comment. It will be the only Y034 noqa in typeshed.

[JelleZijlstra]

I looked into the C code and realized this PR is right after all. The problem isn't that iter(reversed([])) doesn't return a reversed object, it's that reversed([]) doesn't return a reversed object in the first place, but a list_reverseiterator, because list implements the __reversed__ special method. We could model that in typeshed by overloading __new__, but I'd like to defer that until we actually need it.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[AlexWaygood]

I looked into the C code and realized this PR is right after all. The problem isn't that iter(reversed([])) doesn't return a reversed object, it's that reversed([]) doesn't return a reversed object in the first place, but a list_reverseiterator, because list implements the __reversed__ special method. We could model that in typeshed by overloading __new__, but I'd like to defer that until we actually need it.

Huh. Bizarre :)