Invalid type for redefined variable inside nested function

[JukkaL]

The revealed type in this example is int, even though it should be str:

# mypy: allow-redefinition

def f() -> None:
    x = 0
    print(x)
    x = 'x'

    def g() -> None:
        reveal_type(x)  # int, but should be str

[rushabh-v]

Hey @JukkaL
I looked into it and found out that,
that is not because of the nested function. it is because probably mypy doesn't run the line giving an error. For example if you run below code with mypy,

x = 'x'
reveal_type(x)
x = 0
reveal_type(x)

it will give output:

rev.py:2: note: Revealed type is 'builtins.str'
rev.py:3: error: Incompatible types in assignment (expression has type "int", variable has type "str")
rev.py:4: note: Revealed type is 'builtins.str'
Found 1 error in 1 file (checked 1 source file)

So basically either " x = 0" line has not run or the type of x has not changed because it gave the incompatible type in assignment error.

[JukkaL]

When using --allow-redefinition the second reveal_type actually shows builtins.int. This issue is about --allow-redefinition (see the first line of the original code fragment).

[DevilXD]

I've just ran into an issue that's pretty similar to this one, even tho I do not use the allow-redefinition flag.

Here's the minimal code:

def func():
    x: Optional[str] = "xyz"
    if not x:
        return
    reveal_type(x)  # Revealed type is 'builtins.str'

    def func2():
        reveal_type(x)  # Revealed type is 'Union[builtins.str, None]'
        pass

It looks like the second function definition completely ignores the if not x: check, falling back to the originally defined type.
Using Python 3.8.1 and latest master branch, flags used are default VSCode ones + untyped defs:

--follow-imports=silent
--show-column-numbers
--check-untyped-defs
--ignore-missing-imports