Skip to content

Method overloads ignored when dispatched via protocols #9464

Description

@BvB93

Bug Report

At the moment I'm dealing with a number of numpy functions which, effectively, serve as wrappers
around a set of matching ndarray methods. Taking a simplified version of argmax as example,
a custom SupportsArgmax Protocol is herein used for dispatching the internals of the argmax
function to the ndarray.argmax method.

As it turns out, this works perfectly fine when the ndarray.argmax mods does not contain any
overloads, but when it does then any overloads beyond the first are ignored.

from typing import Protocol, overload, TypeVar

# Step 1: Define the class
class ndarray:
    @overload
    def argmax(self, axis: None) -> int: ...
    @overload
    def argmax(self, axis: int) -> ndarray: ...


# Step 2: Define a Protocol for describing the classes' argmax method
T_contra = TypeVar("T_contra", contravariant=True)
T_co = TypeVar("T_co", covariant=True)

class SupportsArgmax(Protocol[T_contra, T_co]):
    def argmax(self, axis: T_contra) -> T_co: ...


# Step 3: Define a wrapper function around the classes' argmax method
T1 = TypeVar("T1")
T2 = TypeVar("T2")

def argmax(a: SupportsArgmax[T1, T2], axis: T1) -> T2: ...

Now, when an ndarray instance is passed to the argmax function I'd expect the
return type to be int if axis=None and ndarray if axis=int(...), just as is
described in the ndarray.argmax method:

array: ndarray

# note: Revealed type is 'builtins.int*'
reveal_type(argmax(array, axis=None))

# error: Cannot infer type argument 1 of "argmax"
# note: Revealed type is 'Any'
reveal_type(argmax(array, axis=0))

As can be seen above, the expected behavior is indeed observed when passing
through the first ndarray.argmax overload, but the presence of any subsequent
overloads seems to be ignored. Swapping the two overloads in the original
ndarray.argmax method only serves to reverse the issue.

Expected Behavior

Mypy successfully infers the return type, regardless of whether the protocol should
extract it from the methods' first overload or any subsequent ones.

Actual Behavior

Mypy fails for any method overload overload beyond the first.

Your Environment

  • Mypy version used: mypy 0.782
  • Mypy command-line flags: --show-error-codes (including or excluding --strict makes no difference here)
  • Mypy configuration options from mypy.ini (and other config files): n.a.
  • Python version used: Python 3.8.5
  • Operating system and version: macOS 10.15.6

Activity

  1. BvB93 commented on Sep 21, 2020

    @BvB93
    ContributorAuthor

    For further reference: Manually annotating a variable with above-mentioned SupportsArgmax protocol works without a hitch, it's only when used in annotating function parameters that the original methods' overloads are ignored:

    # Both assignments are accepted without issue
    a: SupportsArgmax[None, int] = ndarray() 
    b: SupportsArgmax[int, ndarray] = ndarray() 
  2. gvanrossum commented on Sep 27, 2020

    @gvanrossum
    Member

    Thanks for the highly detailed bug report. Are you interested in submitting a PR with a fix yourself? We're currently swamped with other things, alas.

  3. BvB93 commented on Sep 28, 2020

    @BvB93
    ContributorAuthor

    I'm willing to give it a shot, though I'm not very familiar with the internals of mypy (it might take a bit of time).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions