Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

mypy is unable to assign Callable taking Concatenate to Union of that Callable and another using the same ParamSpec #15177

Closed
schuelermine opened this issue May 3, 2023 · 6 comments
Labels
bug mypy got something wrong topic-paramspec PEP 612, ParamSpec, Concatenate

Comments

@schuelermine
Copy link

schuelermine commented May 3, 2023

Bug Report

mypy is unable to assign Callable[Concatenate[B, C], D] to Union[Callable[Concatenate[B, C], D], Callable[C, D].

To Reproduce

from typing import TypeVar, ParamSpec, Concatenate, cast
from collections.abc import Callable

A = TypeVar("A")
B = TypeVar("B")
C = ParamSpec("C")
D = TypeVar("D")


def precall_guard_none(
    a: A, b: B | None, f: Callable[Concatenate[A, B, C], D]
) -> Callable[Concatenate[B, C], D] | Callable[C, D]:
    r: Callable[Concatenate[B, C], D] | Callable[C, D]
    if b is None:

        def g(b: B, *args: C.args, **kwargs: C.kwargs) -> D:
            return f(a, b, *args, **kwargs)

        r = g  # This should be permitted because the type of g is one of the variants of the type of r, a Union
    else:

        def h(*args: C.args, **kwargs: C.kwargs) -> D:
            return f(a, cast(B, b), *args, **kwargs)  # This cast is surprisingly necessary

        r = h

    return r

Expected Behavior

I expect no error.

Actual Behavior

mypy produces this error:

main.py:19: error: Incompatible types in assignment (expression has type "Callable[[B, **C], D]", variable has type "Union[Callable[[B, **C], D], Callable[C, D]]")  [assignment]

Your Environment

mypy playground

  • Mypy version used: 1.2.0
  • Mypy command-line flags: --strict
  • Python version used: 3.11
@schuelermine schuelermine added the bug mypy got something wrong label May 3, 2023
@hauntsaninja hauntsaninja added the topic-paramspec PEP 612, ParamSpec, Concatenate label May 3, 2023
@schuelermine schuelermine changed the title mypy is unable to unify Callable taking Concatenate with Union of that Callable and another using the same ParamSpec mypy is unable to assign Callable taking Concatenate to Union of that Callable and another using the same ParamSpec May 3, 2023
@nikochiko
Copy link

The problem seems to be with using Unions inside some other type.

I ran into the same issue, but with Dict.

... has incompatible type "Dict[str, str]"; expected "Dict[str, Union[str, bool]]"

@JelleZijlstra
Copy link
Member

@nikochiko that's an unrelated issue. Mypy is correct here; see https://mypy.readthedocs.io/en/stable/common_issues.html#invariance-vs-covariance. Let's keep this issue focused on the ParamSpec problem.

@nikochiko
Copy link

Ah, TIL. Thanks for the pointer.

@matperc
Copy link

matperc commented May 29, 2023

I may have the same issue but with Optional

from typing import Concatenate, ParamSpec, Callable, Optional


P = ParamSpec("P")

def func1(func: Callable[Concatenate[str, P], None], *args):
    pass

def func2(func: Optional[Callable[Concatenate[str, P], None]], *args):
    pass

def func3(a: str, b: str):
    pass

func1(func3, "hello")
func2(func3, "hello")
test.py:18: error: Argument 1 to "func2" has incompatible type "Callable[[str, str], Any]"; expected "Optional[Callable[[str, VarArg(<nothing>), KwArg(<nothing>)], None]]"  [arg-type]

func1 works, func2 not.

@matperc
Copy link

matperc commented May 29, 2023

I have done some research, those are my findings.

The issue seems to be originated here

mypy/mypy/constraints.py

Lines 256 to 288 in 7fe1fdd

# Now the potential subtype is known not to be a Union or a type
# variable that we are solving for. In that case, for a Union to
# be a supertype of the potential subtype, some item of the Union
# must be a supertype of it.
if direction == SUBTYPE_OF and isinstance(actual, UnionType):
# If some of items is not a complete type, disregard that.
items = simplify_away_incomplete_types(actual.items)
# We infer constraints eagerly -- try to find constraints for a type
# variable if possible. This seems to help with some real-world
# use cases.
return any_constraints(
[infer_constraints_if_possible(template, a_item, direction) for a_item in items],
eager=True,
)
if direction == SUPERTYPE_OF and isinstance(template, UnionType):
# When the template is a union, we are okay with leaving some
# type variables indeterminate. This helps with some special
# cases, though this isn't very principled.
result = any_constraints(
[
infer_constraints_if_possible(t_item, actual, direction)
for t_item in template.items
],
eager=False,
)
if result:
return result
elif has_recursive_types(template) and not has_recursive_types(actual):
return handle_recursive_union(template, actual, direction)
return []
# Remaining cases are handled by ConstraintBuilderVisitor.
return template.accept(ConstraintBuilderVisitor(actual, direction))

While for func1 it checks constraints with

return template.accept(ConstraintBuilderVisitor(actual, direction))

For func2 it enters here

mypy/mypy/constraints.py

Lines 270 to 285 in 7fe1fdd

if direction == SUPERTYPE_OF and isinstance(template, UnionType):
# When the template is a union, we are okay with leaving some
# type variables indeterminate. This helps with some special
# cases, though this isn't very principled.
result = any_constraints(
[
infer_constraints_if_possible(t_item, actual, direction)
for t_item in template.items
],
eager=False,
)
if result:
return result
elif has_recursive_types(template) and not has_recursive_types(actual):
return handle_recursive_union(template, actual, direction)
return []
then, in infer_constraints_if_possible, here

mypy/mypy/constraints.py

Lines 301 to 304 in 7fe1fdd

if direction == SUPERTYPE_OF and not mypy.subtypes.is_subtype(
actual, erase_typevars(template)
):
return None
because the two functions (func1 Callable[[str, str], Any] and Callable[[str, *Any, **Any], None]) are not one the subtypes of the other (I don't know if this is correct but it seems reasonable). In is_subtype the function is checked with
return left.accept(SubtypeVisitor(orig_right, subtype_context, proper_subtype))

To sum up, func1 is checked using ConstraintBuilderVisitor while func2 is checked with SubtypeVisitor

I have tried with this change and it works good, but it seems like an hack to me.

diff --git a/mypy/constraints.py b/mypy/constraints.py
index 33230871b..2de4ea1e9 100644
--- a/mypy/constraints.py
+++ b/mypy/constraints.py
@@ -298,8 +298,10 @@ def infer_constraints_if_possible(
     """
     if direction == SUBTYPE_OF and not mypy.subtypes.is_subtype(erase_typevars(template), actual):
         return None
-    if direction == SUPERTYPE_OF and not mypy.subtypes.is_subtype(
-        actual, erase_typevars(template)
+    if (
+        direction == SUPERTYPE_OF
+        and not isinstance(get_proper_type(template), CallableType)
+        and not mypy.subtypes.is_subtype(actual, erase_typevars(template))
     ):
         return None
     if (

I hope that this would be useful for someone

@ilevkivskyi
Copy link
Member

The original example works on master, likely fixed by #15837

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug mypy got something wrong topic-paramspec PEP 612, ParamSpec, Concatenate
Projects
None yet
Development

No branches or pull requests

6 participants