Mypy incorrectly requires __next__() on the iterable of a for loop (regression from 0.942).

[md2468]

Bug Report

For loops in Python require only an iterable, for which an iterator is subsequently created. However, in the code below, Mypy incorrectly requires __next__ to be available on the iterable.

To Reproduce

from collections.abc import Iterable

xs: zip[tuple[object, ...]] | Iterable[object] = zip(*[[0], [0]])
for x in xs: pass

Actual Behavior

main.py:4: error: "Iterable[object]" has no attribute "__next__"  [attr-defined]

Your Environment

• Mypy version used: 1.8.0.
• Mypy command-line flags: None.
• Mypy configuration options from mypy.ini (and other config files): None.
• Python version used: 3.12.

This is a regression from version 0.942.
1.8.0 Playground
0.942 Playground

[graingert]

I can repeat this on 0.800, so I think it's just that zip changed definition in typeshed after 0.942

https://mypy-play.net/?mypy=0.800&python=3.12&gist=94e05458b907384b70af1be75c60cfeb

[JamesParrott]

I can recreate this in v1.8.0 too. But interestingly there's no error with either type hint alone. The both annotations together with the | is causing the bug.

[hauntsaninja]

Thanks for the issue!

It looks like mypy isn't handling the self type annotation on zip.__iter__ correctly. It loses track of the upper bound implied by the self type, so it gets the wrong return type when trying to resolve __iter__ in

mypy/mypy/checker.py

Line 4795 in 5350c0e

iterator = echk.check_method_call_by_name("__iter__", iterable, [], [], context)[0]

This looked familiar to me and it turns out I have a relevant patch in #13221. That patch would indeed fix this case, but for reasons beyond my understanding that patch makes mypy pretty unhappy in general.

Maybe @ilevkivskyi has time to take a look, since he made #16184 which solved the issue #13221 was trying to solve, and also is generally awesome at fixing things.