[possibly-used-before-assignment] Avoid FP for typing.NoReturn

[jacobtylerwalls]

Type of Changes

	Type
✓	🐛 Bug fix

Description

Closes #9674 for the general case, but not necessarily for pytest.skip due to the following:

from astroid import extract_node
n = extract_node("""
... import pytest
... pytest.skip('foo')""")
>>> n.func.inferred()
[Uninferable]

[jacobtylerwalls]

We'll be dropping support for 3.8 in the next minor release, so this inelegance doesn't bother me ATM.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.85%. Comparing base (3f1f7b8) to head (6481da5).
Report is 5 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #9714   +/-   ##
=======================================
  Coverage   95.84%   95.85%           
=======================================
  Files         174      174           
  Lines       18865    18870    +5     
=======================================
+ Hits        18082    18087    +5     
  Misses        783      783           

Files	Coverage Δ
pylint/checkers/utils.py	96.16% <100.00%> (+0.01%)	⬆️

[github-actions]

🤖 Effect of this PR on checked open source code: 🤖

Effect on home-assistant:
The following messages are no longer emitted:

1. possibly-used-before-assignment:
   Possibly using variable 'max_bind_vars' before assignment
   https://github.com/home-assistant/core/blob/522a1e9d56ce8571b730c9ddf2108d58969b5044/homeassistant/components/recorder/util.py#L601

Effect on black:
The following messages are now emitted:

1. unreachable:
   Unreachable code
   https://github.com/psf/black/blob/b677a643c5322b6af2dbca1f8f7a4f8c98e732aa/src/blib2to3/pgen2/pgen.py#L346

Effect on pytest:
The following messages are no longer emitted:

1. possibly-used-before-assignment:
   Possibly using variable 'reprentry' before assignment
   https://github.com/pytest-dev/pytest/blob/f85289ba872a26eb41836cdaefe97a93b311bf29/src/_pytest/reports.py#L573

This comment was generated for commit 6481da5

[Pierre-Sassoulas]

I'm on the fence about this one. On the one hand, it's going to prevent false positives. And we have false positives for code that is too dynamic for astroid to understand. But on the other hand, one of the value proposition of pylint is not trusting the typing. If we start trusting the typing then we're going to raise the same class of error that type checker already raise, and they do that better than us and faster than us. Also, if we don't do it generically we're going to have some message that trust the typing and some that don't and how are we drawing the line when someone ask for typing use making the point that "possibly-user-before-assignment" does it for "NoReturn" already ? Then again I don't see massive improvements happening in astroid's inference any time soon.

[jacobtylerwalls]

I hear that. I think NoReturn is very special -- it's basically assert_never, which is basically a more expressive pylint disable.

Allowing someone to use a stdlib feature that works well with type checkers instead of a pylint disable is a pretty isolated case compared to always trusting typing.

[Pierre-Sassoulas]

I like the idea if seeing some type hint as implicit disables. A passing thought : Taking other tool's noqas into account is somewhat similar and would also make pylint more useful.

[Pierre-Sassoulas]

I like the fact that the genericity if the code permits to also makes 'unreachable' better. It's done properly because it's easy to see that kind of helper reuse as a dirty source of unintented side effects.

[Pierre-Sassoulas]

@DanielNoord do you want to chime in :) ?

[DanielNoord]

I like it! However, I would like to document that this is a special case and not something we do for other checks. Then we have something we can point to when more requests start coming in.

[jacobtylerwalls]

Other checks or other type annotation concepts? I've made change in the general util, I think we probably should simplify the code for other checks by using the common utils more, so I don't know there's much more to document.