Type invalid due to import, only on alternating runs

Bug Report

I think this is a bug, as I get inconsistent results across multiple invocations.

Depending on how I define a Type Alias, it can or can not be used as a valid type in another file - on every other invocation of mypy.

from typing import Literal
from typing import Optional
from typing import Union

TYPE_FOO = Optional[str]
TYPE_BAR = Union[str, None]
    
def biz(a: TYPE_FOO, b: TYPE_BAR) -> Literal[True]:
    return True

However, if the type aliases are split into another file, TYPE_FOO (using Optional) is sometimes invalid in mypy and triggers "Variable "TYPE_FOO" is not valid as a type [valid-type]" – while the functionally equivalent TYPE_BAR (using Union) is always valid. The error from Optional seems to appear on every-other invocation of mypy. I can't seem to recreate the error reliably with the small test case alone, but dropping it into larger projects seems to trigger it on every-other run.

To Reproduce

custom_types.py

from typing import Optional
from typing import Union

TYPE_FOO = Optional[str]
TYPE_BAR = Union[str, None]

project.py

from typing import Literal
from .custom_types import TYPE_FOO
from .custom_types import TYPE_BAR

def biz(a: TYPE_FOO, b: TYPE_BAR) -> Literal[True]:
    return True

Expected Behavior

I expected no errors for either usage.
I expected consistent behavior from mypy across runs when it thinks there is an issue.

Actual Behavior

MyPy alternates an error for the imported type that uses Optional on every-other run:

Variable "TYPE_FOO" is not valid as a type [valid-type]

i.e. The first run has no errors, the second has the error, the third has no errors, the fourth has the error, etc

The type using Union never triggers an error.

Your Environment

• Mypy version used:
  1.10
• Mypy command-line flags:
  NA
• Mypy configuration options from mypy.ini (and other config files):
  check_untyped_defs = True
• Python version used:
  3.10

Inizia eseguendo ripetutamente mypy 1.10 sull’esempio composto da due file in custom_types.py e project.py con la configurazione Python 3.10 indicata. Confronta le esecuzioni in cui TYPE_FOO viene accettato con quelle che producono l’errore valid-type, quindi estendi la riproduzione a un progetto più grande come descritto. Il lavoro è completato quando gli alias basati su Optional importati si comportano in modo coerente e il falso errore non compare più.