Wrong return type of `read_csv` if using IO-Objects about python-type-stubs HOT 5 CLOSED

I think the overloads overall need to be different besides just that one change; reading the implementation and docs, it's dependant on chunksize or iterator being set that causes it to return a TextFileReader.

from python-type-stubs.

I think the overloads overall need to be different besides just that one change; reading the implementation and docs, it's dependant on chunksize or iterator being set that causes it to return a TextFileReader.

I think, you are completely right. I further think, that a first measure should be, to take care of the ReturnValue, which would help others most. As we only have stubs here, I think changing them will not harm.

from python-type-stubs.

I spent way too much time on this today, and then discovered this issue. I'm not sure how python type checking with overloads can do this right. Maybe @jakebailey or @erictraut have some ideas (see example below):

The design of the API changes the return type to TextFileReader under any of the following conditions:

1. chunksize is specified and integer and not None
2. iterator is specified and set to True

So you have to find a way in the overloads to handle all the possible combinations of chunksize and iterator, both of which are keyword arguments, and either or both may not be specified. That is reflected in the following table

To help me figure this out, I wrote a simple test, where i1 is iterator and chunksize is cs, and a return type of int corresponds to DataFrame and a return type of str corresponds to TextFileReader :

@overload
def myfun(fake: str, i1: Literal[True], cs: Literal[None]) -> str:
    ...


@overload
def myfun(fake: str, i1: Literal[True], cs: int) -> str:
    ...


@overload
def myfun(fake: str, i1: Literal[True]) -> str:
    ...


@overload
def myfun(fake: str) -> int:
    ...


@overload
def myfun(fake: str, i1: Literal[False]) -> int:
    ...


@overload
def myfun(fake: str, i1: Literal[False], cs: Literal[None]) -> int:
    ...


@overload
def myfun(fake: str, i1: Literal[False], cs: int) -> str:
    ...


# @overload
# def myfun(fake: str, i1: bool = False, cs: None = None) -> int:
#     ...


# @overload
# def myfun(fake: str, i1: bool = ..., cs: int = ...) -> str:
#     ...


def myfun(fake: str, i1: bool = False, cs: Optional[int] = None) -> Union[str, int]:
    print(f"i1 is {i1} cs is {cs} result is ", end="")
    if i1:
        if cs is not None:
            return "TextFileReader"
        else:
            return "TextFileReader"
    else:
        if cs is not None:
            return "TextFileReader"
        else:
            return -1


res1: int = myfun("meh")
res2: int = myfun("meh", i1=False)
res3: int = myfun("meh", i1=False, cs=None)
res4: str = myfun("meh", i1=False, cs=23)
res5: str = myfun("meh", i1=True)
res6: str = myfun("meh", i1=True, cs=None)
res7: str = myfun("meh", i1=True, cs=23)
res8: int = myfun("meh", cs=None)
res9: str = myfun("meh", cs=23)

I could not figure out how to get the last 2 function calls (which correspond to the last 2 lines in the table) to have a proper overload, because in those cases, i1 (equivalently iterator) is not specified, but cs is specified. I've commented out some attempts but I'd get complaints about overlapping signatures.

If someone can show me how to get the above to work, then I could make the changes to read_csv() typing to correspond.

from python-type-stubs.

I think this can be achieved by the following:

@overload
def myfun(fake: str, i1: Literal[True], cs: int | None = ...) -> str:
    ...

@overload
def myfun(fake: str, i1: Literal[False], cs: int) -> str:
    ...

@overload
def myfun(fake: str, i1: Literal[False] = ..., cs: None = ...) -> int:
    ...

@overload
def myfun(fake: str, i1: bool = ..., cs: int = ...) -> str:
    ...

from python-type-stubs.

Thanks @erictraut . That makes pyright happy, but mypy complains. I'll open up an issue there.

from python-type-stubs.