Parsing of Walrus Operator

[rvanlaar]

Mypy doesn't seem to handle variable type changes that happen with a walrus operator.

What happens here is: the walrus operator makes it a boolean:
while i:= update(i) > 0.

This is what needs to happen to check for i > 0 and i still being an int.
while (i:= update(i)) > 0 .

The following python code doesn't give any errors with mypy 0.750:

def walrus1(i: int) -> int:
    """Actually returns bool"""
    while i:= i > 0:
       return i 
    return i

def walrus2(i: int) -> int:
    """Works correctly."""
    while (i:= i) > 0:
        return i
    return i

assert walrus1(5) == True
assert walrus1(0) == False
assert walrus2(5) == 5
assert walrus2(0) == 0

[rvanlaar]

Where in the mypy code can I loop If I were to try to make a pull request?

[JelleZijlstra]

I don't know exactly, but #6899 is the PR where I added (limited) support for the walrus operator, so the bug may be in the code added there.

[rvanlaar]

It seems that bool is a covariant of int.

In the following example: x and is alright, y and z aren't.

x: int
x = True

y: bool
y = 5

z: str
z = True

[JelleZijlstra]

Right, bool is a subclass of int, so mypy sees that the function still returns a kind of int.

[rvanlaar]

Which wasn't what I expected. Even with years of python experience.

Could the table on https://mypy.readthedocs.io/en/latest/builtin_types.html be updated?
That's were I looked for an explanation.

I would expect that the definition of bool there would be a subtype of int.

[JukkaL]

@rvanlaar Sure, updating the table is good idea. Would like to create a PR (or a separate issue to track it)?

[rvanlaar]

Closing this issue, continue with issue #8069 to document it.