Possible regression with attrs.asdict related to type guarding?

[dougthor42]

Possibly related issues:

• Use of AttrsInstance protocol is incompatible with type checkers other than mypy #996
• attr.asdict -> attrs.asdict regression for sets of instances #1180

Related PRs:

• Use TypeGuard for has in Python 3.10 and above #997
• Add AttrsInstance protocol to forgotten attrs.asdict and attrs.astuple #1090

Summary

It looks like there's a regression with types guards and asdict from 21.4.0 to 23.1.0, and that they TypeGuard on attrs.has should be

def has(cls: type) -> TypeGuard[AttrsInstance]:

Or attrs.asdict should accept inst: Type[AttrsInstance]

Steps to Reproduce:

In 21.4.0, this worked:

# attrs-issue.py
from typing import Any
import attrs

@attrs.frozen
class Foo:
    a: int

obj: Any = Foo(0)

if attrs.has(obj):
    print(attrs.asdict(obj))

$ python attrs-issue.py 
{'a': 0}
$ mypy attrs-issue.py 
Success: no issues found in 1 source file

In 23.1.0, that same file fails mypy:

$ pip install attrs==23.1.0
Looking in indexes: https://pypi.org/simple
Collecting attrs==23.1.0
  Using cached attrs-23.1.0-py3-none-any.whl (61 kB)
Installing collected packages: attrs
  Attempting uninstall: attrs
    Found existing installation: attrs 21.4.0
    Uninstalling attrs-21.4.0:
      Successfully uninstalled attrs-21.4.0
Successfully installed attrs-23.1.0
$ python attrs-issue.py 
{'a': 0}
$ mypy attrs-issue.py 
attrs-issue.py:11: error: Argument 1 to "asdict" has incompatible type "type[AttrsInstance]"; expected "AttrsInstance"  [arg-type]
attrs-issue.py:11: note: ClassVar protocol member AttrsInstance.__attrs_attrs__ can never be matched by a class object
Found 1 error in 1 file (checked 1 source file)

Other Info

• Python version: 3.10.10
• Mypy version: 1.5.1 (compiled: yes)
• Operating system: Debian GNU/Linux rodete, 64-bit

[Tinche]

Looks like we have a regression but the type hints are correct. attrs.has is supposed to be called with a class, not an instance of a class, so your snippet should read if attrs.has(type(obj)):.

I think Mypy isn't clever enough to figure out that obj is an AttrsInstance if type(obj) is type[AttrsInstance]. Not sure what to do here.

[dougthor42]

Ah, of course it's RTFM 🤣.

Looks like we have a regression

I guess it's just a regression in the sense that "incorrect code that happened to previously work no longer does". IMO that's less-bad than alternatives.

For the record, mypy is happy in both attrs versions when using attrs.has(type(obj)).

(Also, I noticed that the example code in the OP was missing a print(), so the issue was edited.)

[Tinche]

attrs.has(type(obj)) didn't work for me, in the following block Mypy treated obj as Any. Which I guess is fine here, but not in the general case.

[hynek]

So uh, can this be closed?

[dougthor42]

I'm fine with it, but given @Tinche's comment:

Looks like we have a regression but ...

I'll let them have the final say.

[hynek]

Tin, close if OK, otherwise pls expound next actions

[Tinche]

I don't think we can do anything here, this seems like a Mypy limitation. We could start a discussion in the typing section of DPO. Closing this for now.