In a function call, if a variable accessed in an iteratable doesn't have an attribute, mypy picks the first overload available, creating spurious further list-item errors.

Bug Report

If a variable accessed in an iteratable doesn't have an attribute ([attr-defined] error), and that iterable is then passed to a function, mypy seems to pick the first overload available to continue typechecking against that function, creating spurious further list-item errors, even when more information is available to deduce the correct type. The code below should make more clear what I mean.

To Reproduce
mypy-play gist

from typing import Any, Iterable, overload

d = 42 #example value without attributes

@overload
def foo(x: Iterable[str]) -> None: ...
@overload
def foo(x: Iterable[int]) -> None: ...

def foo(x: Any) -> None:
  return None

# Pass the checker, correctly
foo(["ok", "ok"])
foo([1, 1])

# Fail the checker, correctly, with [list-item] errors (although they all assume the str override is the authoritative one)
foo(["ok", 1])
foo([1, "ok"])
foo([None, "ok"])

# Fail the checker, but the int ones generate spurious list-item errors.
foo(["ok", d.x])
foo([2, d.x])
foo([d.x, "ok"])
foo([d.x, 2])

# This doesn't generate an [attr-defined] error, but still produces a spurious list-item error!
foo([d.x, 2]) #type: ignore[attr-defined]

Expected Behavior

If a function is declared to take an Iterable, with multiple types of Iterable specified by overload decorators, then an [attr-defined] error inside the iterable the function is called upon should not prevent mypy from deducing the correct type of that iterable from its other contents. The correct type of the iterable should still be deduced.

Actual Behavior

However, what actually happens is that if an [attr-defined] error occurs inside of the list I am passing to the function, mypy just choses the first overload. In this case, Iterable[str]. Therefore, spurious [list-item] errors are created when the other member of the list is an int. The [list-item] errors in the below output for lines 24, 26, and 29 are spurious:

test.py:18: error: List item 1 has incompatible type "int"; expected "str"  [list-item]
test.py:19: error: List item 0 has incompatible type "int"; expected "str"  [list-item]
test.py:20: error: List item 0 has incompatible type "None"; expected "str"  [list-item]
test.py:23: error: "int" has no attribute "x"  [attr-defined]
test.py:24: error: List item 0 has incompatible type "int"; expected "str"  [list-item]
test.py:24: error: "int" has no attribute "x"  [attr-defined]
test.py:25: error: "int" has no attribute "x"  [attr-defined]
test.py:26: error: "int" has no attribute "x"  [attr-defined]
test.py:26: error: List item 1 has incompatible type "int"; expected "str"  [list-item]
test.py:29: error: List item 1 has incompatible type "int"; expected "str"  [list-item]
test.py:29: note: Error code "list-item" not covered by "type: ignore" comment
Found 10 errors in 1 file (checked 1 source file)

Here is a --pretty listing of the relevant errors:

test.py:24: error: List item 0 has incompatible type "int"; expected "str"  [list-item]
    foo([2, d.x])
         ^
test.py:24: error: "int" has no attribute "x"  [attr-defined]
    foo([2, d.x])
            ^~~
test.py:26: error: "int" has no attribute "x"  [attr-defined]
    foo([d.x, 2])
         ^~~
test.py:26: error: List item 1 has incompatible type "int"; expected "str"  [list-item]
    foo([d.x, 2])
              ^
test.py:29: error: List item 1 has incompatible type "int"; expected "str"  [list-item]
    foo([d.x, 2]) #type: ignore[attr-defined]
test.py:29: note: Error code "list-item" not covered by "type: ignore" comment

29 is especially annoying, because the [attr-defined] ignore does not suppress all warnings that indirectly result from the [attr-defined] error!

You can also switch the order of the overloads in the code to see that mypy now complains that its arguments are str when it needs ints, instead of vice-versa.

Your Environment

• Mypy version used: 1.9.0
• Mypy command-line flags: none
• Mypy configuration options from mypy.ini (and other config files): n/a
• Python version used: Python 3.11.0 and 3.12

Commencez par reproduire le problème à partir de l’exemple inclus ou du gist mypy-play lié avec mypy 1.9.0, en incluant l’ordre des overloads et le cas d’attr-defined ignoré. Suivez la sélection des overloads et l’inférence des éléments de l’itérable, puis ajoutez une couverture de régression montrant que les éléments de liste connus déterminent le bon overload sans générer d’erreurs list-item superflues, tandis que l’erreur attr-defined reste présente.