Repository navigation
should mypy recognize pep-681 @dataclass_transform on a method as well as a function? #19824
Description
Just noticed that trying to make an object method provide @dataclass_transform functionality is not recognized by mypy as a pep-681 use of @dataclass_transform (it does not complain either, just silently ignores it):
from typing import dataclass_transform
class SomeClassThing:
@dataclass_transform()
def make_a_dataclass(self, cls: type) -> type:
return cls
sct = SomeClassThing()
@sct.make_a_dataclass
class MyClass:
x: int
y: str
m1 = MyClass(x=5, y="y")mypy:
test4.py:16: error: Unexpected keyword argument "x" for "MyClass" [call-arg]
test4.py:16: error: Unexpected keyword argument "y" for "MyClass" [call-arg]
Found 2 errors in 1 file (checked 1 source file)
the above would need to be reorganized like this:
from typing import Callable
from typing import dataclass_transform
class SomeClassThing:
pass
@dataclass_transform()
def make_a_dataclass(sct: SomeClassThing) -> Callable[[type], type]:
def decorate(cls: type) -> type:
return cls
return decorate
sct = SomeClassThing()
@make_a_dataclass(sct)
class MyClass:
x: int
y: str
m1 = MyClass(x=5, y="y")where mypy is happy:
Success: no issues found in 1 source file
SQLAlchemy has a pep-681 feature that mostly uses the superclass approach, however we also offer a decorator that looks like @mapper_registry.mapped_as_dataclass(), and I just noticed that while pyright has always accepted this, mypy does not. So I'm adding an alternative decorator of the above form @mapped_as_dataclass(mapper_registry). Took a look at pep-681 and it discusses @dataclass_transform applied to a function or a class, not a method. So I'm out of luck. but it seems ....like it should work? pyright does it....
Activity
sterliakov commentedon Sep 10, 2025
Funny enough, the typing spec does not mention methods either. I'm inclined to say that
@dataclass_transformshould be supported on methods (and it doesn't contradict PEP language, as methods are actually functions at the time when decorator is applied to then), but this is really somewhat grey area.tommie-lie commentedon Nov 9, 2025
I have another usecase for that: Decorators that are implemented as a class using
__call__to do the decoration needs@dataclass_transformon methods (and in this case methods that are not "visible" on the call site) as well:In this case, instances of DataclassLike are not recognized as dataclasses by mypy.