How to narrow a field in a dataclass in __init__?

[mmerlo]

I am having trouble narrowing a dataclass field. See the example below. The last line gives the error. How can this be typed correctly? Adding assert self.info.data is not None before return self.info.data fixes this issue, but I was expecting the check in __init__ to be enough.

from dataclasses import dataclass

@dataclass
class Info:
    data: str | None = None

class Foo:
    def __init__(self, info: Info):
        if info.data is None:
            raise ValueError("Data cannot be None")
        self.info = info

    def get_data(self) -> str:
        return self.info.data  #pylance error on this line

Pylance Error

Type "str | None" is not assignable to return type "str"
  Type "str | None" is not assignable to type "str"
    "None" is not assignable to "str"Pylance[reportReturnType](https://github.com/microsoft/pyright/blob/main/docs/configuration.md#reportReturnType)

[erictraut]

In your example, the type of info is Info. There's no narrower type than Info, so narrowing can't be performed.

Pyright will narrow more complex expressions like self.info.data locally within an execution scope (i.e. a function or method body), but type narrowing of expressions does not extend outside that scope because there's no guarantee that the narrowing still applies.

Consider the following:

f = Foo(Info(data="not None"))

f.info.data = None  # This is legal

f.get_data() # Crash

If you want to narrow self.info.data within the get_data method, you'd need to add some logic to that method. For example, an assert statement will document your assumptions for both human readers and a type checker.

    def get_data(self) -> str:
        assert self.info.data is not None
        return self.info.data

For more details about type narrowing, refer to this documentation.

[mmerlo]

Thank you for the explanation. It helped me understand where I was going wrong.