Incorrect error in factory function on a dataclass

In Python 3.7.0rc1 and mypy 0.620+dev-5dc7454b1acb47dcbb3361c849e6113efcdbaaf2, if I save the following code to dc_test.py

from typing import Type, TypeVar
from dataclasses import dataclass

T = TypeVar('T', bound='A')

@dataclass
class A:
	@classmethod
	def make(cls: Type[T]) -> T:
		return cls()

I get the following mypy output

$ mypy dc_test.py 
dc_test.py:10: error: Incompatible return value type (got "A", expected "T")

If I comment out the @dataclass line, I get no error. I would expect to get no error with or without the @dataclass line.

Confirmed. All I can think of is that something's wrong with how class decorators are implemented in mypy.

The type first argument of class methods is patched by plugins for attrs and dataclasses (because __init__ is auto-generated). It looks like this patching is not done carefully (it needs to interact well with bind_self).

Workaround until this issue is fixed

Casting the result to type T silences mypy's false positive error.

-from typing import Type, TypeVar
+from typing import Type, TypeVar, cast
 from dataclasses import dataclass

 T = TypeVar('T', bound='A')

 @dataclass
 class A:
 	@classmethod
 	def make(cls: Type[T]) -> T:
-		return cls()
+		return cast(T, cls())

Make classmethod's first argument be Type[...]

Make classmethod's first argument be Type[...] (#5646)

Navigation Menu

Repository navigation

Incorrect error in factory function on a dataclass #5263

Workaround until this issue is fixed

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Incorrect error in factory function on a dataclass #5263

Description

Activity

gvanrossum commented on Jun 22, 2018

ilevkivskyi commented on Jun 22, 2018

wkschwartz commented on Jul 23, 2018

Workaround until this issue is fixed

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions