add test case: regression w/ mypy mis-interpreting generator comprehension

[zzzeek]

this seems to be a regression which I am getting with mypy master @ 0df8cf5, havent bisected to find the exact commit:

from typing import Dict
from typing import Generic
from typing import List
from typing import Tuple
from typing import TypeVar
from typing import Union


class Thing:
 pass


_THING = TypeVar("_THING", bound="Thing")


class ThingCollection(Generic[_THING]):
 _collection: List[Tuple[str, _THING]]

 _index: Dict[Union[None, str, int], _THING]

 def __init__(self, t: _THING):
 self._collection = [("x", t)]
 self._index = {}

 def do_thing(self) -> None:
 self._index.update(
 (idx, c) for idx, (k, c) in enumerate(self._collection)
 )

 def do_thing_workaround(self) -> None:
 self._index.update(
 (idx, c) for idx, c in enumerate(c for (k, c) in self._collection)
 )

output:

$ mypy test3.py 
test3.py:27: error: Generator has incompatible item type "Tuple[int, Tuple[str, _THING]]"; expected "Tuple[Union[None, str, int], _THING]" [misc]
Found 1 error in 1 file (checked 1 source file)

mypy seems to be mis-interpreting the first generator. works in mypy 0.942 and pyright

[hauntsaninja]

Thanks! This bisects to the recent typeshed sync: #12321

Ran mypy_primer -p test.py --bisect --new origin/master --old v0.942 --debug

[hauntsaninja]

Looks like the following diff "fixes" it:

diff --git a/mypy/typeshed/stdlib/builtins.pyi b/mypy/typeshed/stdlib/builtins.pyi
index 9c5dfcfef..870ed693d 100644
--- a/mypy/typeshed/stdlib/builtins.pyi
+++ b/mypy/typeshed/stdlib/builtins.pyi
@@ -995,7 +995,7 @@ class frozenset(AbstractSet[_T_co], Generic[_T_co]):
 
 class enumerate(Iterator[tuple[int, _T]], Generic[_T]):
 def __init__(self, iterable: Iterable[_T], start: int = ...) -> None: ...
- def __iter__(self: Self) -> Self: ...
+ def __iter__(self) -> Iterator[tuple[int, _T]]: ...
 def __next__(self) -> tuple[int, _T]: ...
 if sys.version_info >= (3, 9):
 def __class_getitem__(cls, __item: Any) -> GenericAlias: ...

[hauntsaninja]

Looks like I'm treading ground already covered in #12588

[AlexWaygood]

This was fixed by #12590, so I'm closing this (cc. @JukkaL)

[JelleZijlstra]

Could we add a test case specifically for @zzzeek's example? #12590 is pretty remote from its effect, and it would be good to get new regression tests for each bug we find.