superfluous-parens false positive for is not tests

[ryansdowning]

Bug description

I just got the lint C0325: Unnecessary parens after 'not' keyword (superfluous-parens), so I removed the parens it referred to, but thankfully a unit test caught that these are not the same, and the parens are needed (removing them changes the logic!).

I have been able to create a minimal example that I believe reproduces this

In [1]: (None is None) is not (None == 1)  # Original, pylint lints with C0325
Out[1]: True

In [1]: (None is None) is not None == 1  # Recommended by C0325
Out[1]: False

Configuration

No response

Command used

pylint path/to/pkg/ --max-line-length=120 --fail-under=9 --load-plugins pylint_django

Pylint output

************* Module pkg.module
path/to/pkg/module.py:63:0: C0325: Unnecessary parens after 'not' keyword (superfluous-parens)

Expected behavior

C0325 should not be output here.

Pylint version

pylint 2.11.1
astroid 2.8.6
Python 3.9.7 (default, Sep 16 2021, 13:09:58) 
[GCC 7.5.0]

OS / Environment

No response

Additional dependencies

No response

[jacobtylerwalls]

Thanks for the report. This is not a duplicate of #5803 (my diff for that doesn't fix this!)

There are numerous operators that aren't working. At a glance, it looks like the implementation of the check confuses is not with not, which have different precedences. That would explain why the operators in the same group as is not produce false positives:

x is not (x <= y)
True is not (True == False)
True is not (True is not False)

[timmartin]

If I understand correctly, the problem isn't so much that not and is not have different precedence (they are different, but no other operator is between them), but that the formatting checker assumes that a not necessarily indicates a unary operator when in fact it can be part of a binary operator expression in the form of is not.