False unsubscriptable-object positive with generic subprocess.Popen and python3.9

[scop]

Steps to reproduce

$ cat /tmp/t.py
from subprocess import Popen, PIPE

def foo(popen: Popen[str]) -> None:
    pass

foo(Popen(["ls"], stdout=PIPE))

Current behavior

$ pylint /tmp/t.py | grep subs
/tmp/t.py:3:15: E1136: Value 'Popen' is unsubscriptable (unsubscriptable-object)

Expected behavior

No unsubscriptable-object error.

pylint --version output

pylint 2.6.0
astroid 2.4.2
Python 3.8.0 (default, Dec 13 2020, 12:54:06) 
[GCC 7.5.0]

[dbaty]

tl;dr: on Python 3.8 and earlier, everything seems fine, I don't think there is a bug. With Python 3.9, though, this is a bug and a fix has been proposed in Astroid (a dependency of pylint): pylint-dev/astroid/pull/882.

Maintainers: I would mention "under Python 3.9" in the title and add the corresponding issue tag. :)

---

On Python 3.8 and earlier

Popen is indeed not subscriptable with in Python 3.8 and earlier. For example, running your code with Python 3.8 I get:

$ python3.8 test.py
Traceback (most recent call last):
  File "test.py", line 3, in <module>
    def foo(popen: Popen[str]) -> None:
TypeError: 'type' object is not subscriptable

So I think that pylint is right to report an error here up to Python 3.8.

On Python 3.9

However, under Python 3.9, Popen is subscriptable. Running your snippet with Python 3.9 does not raise any error.

So, with Python 3.9, pylint should not report the unsubscriptable-object warning. Yet it does, which is a bug.

To determine whether a class is subscriptable, pylint looks for a __class_getitem__ method. Here it cannot find one. This is because Astroid "overrides" the definition of subprocess.Popen (with what Astroid calls a "brain") and there the __class_getitem__ method is missing.

I proposed a fix in Astroid: pylint-dev/astroid/pull/882.

[scop]

Oh, interesting. The error was provoked by a fix to an issue flagged by running mypy with python 3.8 in python_version 3.8 mode. I think that makes it a mypy (well or typeshed) bug, too. Thanks!

[scop]

Ref python/typeshed#4948

[hippo91]

@scop thanks for the report. @dbaty thanks for your investigation and corresponding PR.