False Positive: `no-member` for `ParamSpec` under Python 3.12 about pylint HOT 8 CLOSED

would you mind updating the snippet to include imports and assignments of P and T?

Sure! Well, you already figured it out but for the sake of completeness I edited my post.

Thanks!

from pylint.

@DanielNoord it's actually not as bad as the collections issue.

I looked into the reason that I stubbed out the typing module. It turns out it was really just to pass about a dozen or so cases in astroid that depend on being able to infer the bases of a class using subscript syntax. It doesn't yet work with the new PEP 695 syntax. We should fix it in a minor release when someone volunteers. In the meantime, I think my stopgap is good for a patch release.

I'll open an astroid issue with my finding.

from pylint.

Hi @lord-haffi, would you mind updating the snippet to include imports and assignments of P and T? Thanks!

from pylint.

Thanks for the report!

This diff to astroid should fix it:

diff --git a/astroid/brain/brain_typing.py b/astroid/brain/brain_typing.py
index ea496440..20276b6c 100644
--- a/astroid/brain/brain_typing.py
+++ b/astroid/brain/brain_typing.py
@@ -437,7 +437,13 @@ def _typing_transform():
     class Generic:
         @classmethod
         def __class_getitem__(cls, item):  return cls
-    class ParamSpec: ...
+    class ParamSpec:
+        @property
+        def args(self):
+            return ParamSpecArgs(self)
+        @property
+        def kwargs(self):
+            return ParamSpecKwargs(self)
     class ParamSpecArgs: ...
     class ParamSpecKwargs: ...
     class TypeAlias: ...

I took a quick look at the typing package in 3.12 and couldn't find any declaration of ParamSpec

This is exactly pylint's issue. In 3.12 CPython decided to C-accelerate the typing module, and that hosed a lot of pylint's ability to inspect typing classes. I stubbed out the bare minimum in pylint-dev/astroid@f2120db, but there's a lot missing.

from pylint.

@jacobtylerwalls Is this the same issue as with collections? I thought we fixed that by making our import resolved accept the frozen modules?

from pylint.

I don't know what happened with collections. I'd love to know of a better solution for this, and this sounds like it.

from pylint.

See pylint-dev/astroid#1512 and related PRs. We might be able to come up with a similar fix here.

from pylint.

Opened pylint-dev/astroid#2370 for the real fix. Will prepare a patch with the band-aid.

from pylint.