Static typing with functools.partial

[benjaminsaljooghi]

Is your feature request related to a problem? Please describe.
Calls to functools.partial returns an object of the partial class which cannot be used for static typing.

Describe the solution you'd like
Pyright analyzes the args and kwargs of calls to functools.partial to perform static typing with the resulting partial object.

from functools import partial

def add(a: int, b: int) -> int:
    return a + b

add3 = partial(add, 3)

add3("hello") # Pyright should raise an error

[erictraut]

Thanks for the suggestion.

Pyright is performing correct type checking based on the definition of the partial class in the typeshed stub:

class partial(Generic[_T]):
    func = ...  # type: Callable[..., _T]
    args = ...  # type: Tuple[Any, ...]
    keywords = ...  # type: Dict[str, Any]
    def __init__(self, func: Callable[..., _T], *args: Any, **kwargs: Any) -> None: ...
    def __call__(self, *args: Any, **kwargs: Any) -> _T: ...

Unfortunately the python type language isn't expressive enough to specify the full behavior of partial. The only way to provide more detailed type checking here would be to hard-code knowledge about the internals of the partial class within pyright. In general, that's not something I want to do. There have been a few exceptions for very popular classes like namedtuple. I don't think partial meets the bar though.

If you want static typing in this case, it's trivial to create your own custom class that does what partial does but doesn't rely on untyped args and kwargs.

[codethief]

@erictraut

The only way to provide more detailed type checking here would be to hard-code knowledge about the internals of the partial class within pyright.

It seems Mypy recently went that way and added a plugin for functools.partial, see python/mypy#1484 and python/mypy#16939 . In light of this, would it make sense to revisit the issue?

[erictraut]

@codethief, pyright does have extensive special-casing for functools.partial. I added this several years ago (after this issue was created) based on feedback from users that this was an important case to handle. In fact, pyright's handling of functools.partial is more complete than mypy's at this point. If you're experiencing problems or have questions, feel free to start a new discussion thread.

[bschnurr]

yup. see an error now.

[codethief]

Interesting, I was under the impression it only tracked/type-checked the function's return value. I'll have another look! Thanks so much!