Some questions about using @overload decorator correctly

[szw0407]

Below is an example code:

from typing import overload

class A:
    data: list[int]  # A
    def __init__(self, *args: int) -> None:
        self.data = list(args)

class B:
    @overload
    def __init__(self, *args: int) -> None: ...
    @overload
    def __init__(self, *args: A) -> None: ...

    def __init__(self, *args: int | A) -> None:
        self.data = []
        if all(isinstance(arg, A) for arg in args):
            for arg in args:
                self.data.extend(arg.data) # typeerror here, saying arg is Unknown type, but arg has been checked as A in the if statement, thus arg.data should be list[int] according to the line marked A
        elif all(isinstance(arg, int) for arg in args):
            self.data.extend(args)  # typeerror here, saying args is Unknown type, but it has been checked as list[int] in the if statement
        else:
            raise TypeError

May I give another example code:

from typing import overload, Literal
    
class Obj:
    data: list[int]
    @overload
    def __init__(self, *args: int, all_int: Literal[True]) -> None: ...
    @overload
    def __init__(self, *args: str, all_str: Literal[True]) -> None: ...

    def __init__(self, *args:int | str, **kwargs: bool) -> None:
        if kwargs.keys() == {'all_int'}:
            self.data = list(args) # TypeError: list[int | str] is incompatible with list[int]
        elif kwargs.keys() == {'all_str'}:
            self.data = list(map(int, args))
        else:
            raise TypeError("Invalid arguments")

I have declared the Type of args when all_int is in kwargs.keys() the args must be all integers, but when all_str appears, it will be all strings. I can use all(isinstance(args, int)) to check the types but I guess that was not necessary if Pylance could read the overload function declearations. Moreover, even if I did so, all function is not supported yet.

It might be asked why I am using kwargs instead of writing all the names of variables in the function body. Well, I think using kwargs makes my code more readable and clear.

Perhaps I misunderstood what overload could do. And I found #4526 . It had been marked as completed, and I am not sure whether some exceptions still exist. Thanks for your help.

[szw0407]

The previous version of this question is fixed when I switch to the pre-release version of Pylance, but perhaps there are still some issues.

The new example may be due to usage of the all function.

[erictraut]

The all function is not a supported type guard. For a complete list of supported type guard patterns, please refer to this documentation. For complex type guards like all, you would need to write your own "user-defined type guard" using the TypeGuard capability introduced in PEP 647.

[erictraut]

Also, it's important to understand that @overload signatures do not affect the type checking of the implementation of the overload. The @overload signatures are used only when evaluating calls of the function, not the implementation of the function. For more information about overloads, refer to this section of PEP 484.