Type narrowing when one of two fields may be None

[Peter9192]

The following example:

def a_or_b(a: int | None = None, b: int | None = None):
    assert a or b
    if not b:
        return a + 5

complains about: Operator "+" not supported for "None" Pylance (reportOptionalOperand). However, from the context it is clear that a cannot be None. Is there a way to get successful type narrowing in this case?

FYI, my real use case is a bit more complex, using pydantic validators to ensure the input is valid. Something like this, as discussed here.

from pydantic import BaseModel, model_validator

class AorB(BaseModel):
    a: int | None
    b: int | None

    @model_validator(mode='after')
    def check_a_or_b(self):
        if not self.a and not self.b:
            raise ValueError('Either a or b (or both) is required')
        return self
    
    def make_pylance_unhappy(self):
        if self.a is None:
            return self.b + 5

[erictraut]

The behavior of pyright (the type checker upon which pylance is built) is as designed here, so this isn't a bug.

Type narrowing is applied to individual symbols (in this case, a and b). Types are not tracked conditionally between symbols (e.g. "the type of b is int if a is None but int | None if a is not None"). The assert statement in this case is not able to narrow the type of either a or b in this case. You'll see the same error if you run other static type checkers (e.g. mypy) on this code.

If you want to learn more about type narrowing, refer to this documentation.

You can change your simple example as follows:

def a_or_b(a: int | None = None, b: int | None = None):
    if not b:
        assert a
        return a + 5

In your "real" example, you can fix it by doing the following:

    def make_pylance_unhappy(self):
        if self.a is None:
            assert self.b is not None
            return self.b + 5

[Peter9192]

Thanks for the swift response, I suspected as much. Just noting that I clicked "question" when creating this issue, so I never intended to report it as a bug.

[debonte]

Just noting that I clicked "question" when creating this issue, so I never intended to report it as a bug.

@rchiodo, this makes me wonder if we should have a "Question" option when creating new issues or perhaps just steer questions towards the discussions area. Or maybe we need a "question" label.

[rchiodo]

@rchiodo, this makes me wonder if we should have a "Question" option when creating new issues or perhaps just steer questions towards the discussions area. Or maybe we need a "question" label.

There's actually a question option already:

I wonder if we can just redirect that to the discussions tab?

[debonte]

Sorry, I didn't phrase that well -- what I meant was, maybe we should remove it.