Drop support for Python 3.7

[cdce8p]

Description

Python 3.7 will reach its EOL in two months (2023-06-27). I think it would make sense to drop support with the next release 3.0. That will allow us to cleanup even more old code.

I'm not proposing to remove the ability to lint 3.7 compatible code. If users want that, they can still set py-version=3.7.

Refs #6306

[codecov]

Codecov Report

Merging #8609 (f2556fd) into main (3d036b7) will decrease coverage by 0.11%.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8609      +/-   ##
==========================================
- Coverage   95.92%   95.81%   -0.11%     
==========================================
  Files         174      174              
  Lines       18415    18317      -98     
==========================================
- Hits        17664    17551     -113     
- Misses        751      766      +15     

Impacted Files	Coverage Δ
pylint/checkers/utils.py	95.65% <ø> (-0.41%)	⬇️
pylint/typing.py	100.00% <ø> (ø)
pylint/checkers/__init__.py	93.10% <100.00%> (-0.84%)	⬇️
pylint/checkers/base/basic_checker.py	97.86% <100.00%> (-0.03%)	⬇️
pylint/checkers/base/docstring_checker.py	97.67% <100.00%> (-0.08%)	⬇️
pylint/checkers/classes/class_checker.py	93.61% <100.00%> (-0.02%)	⬇️
pylint/checkers/format.py	96.46% <100.00%> (-0.05%)	⬇️
pylint/checkers/imports.py	94.31% <100.00%> (-0.03%)	⬇️
pylint/checkers/logging.py	94.73% <100.00%> (-0.14%)	⬇️
pylint/checkers/raw_metrics.py	100.00% <100.00%> (ø)
... and 18 more

... and 4 files with indirect coverage changes

[DanielNoord]

I would defer this choice to pylint-dev/astroid#2137 (comment), as I think this is where we will make the first/final decision about this.

[Pierre-Sassoulas]

I would expect some change in the functional tests too (py 3.7 exclusive tests that now need a py-version to run).

[Pierre-Sassoulas]

I haven't considered that. This is (also) going to be a nice cleanup !

[cdce8p]

I would expect some change in the functional tests too (py 3.7 exclusive tests that now need a py-version to run).

This is just a first step. Besides the test cases, there are other areas which could be improved. Had to start somewhere.

Assignment expressions

Yeah, we could start using them with 3.8. Although I'm not sure we should enable the code-style check for it. Sometimes using := is an opinionated decision which the check just can't do.

[Pierre-Sassoulas]

Regarding the functional tests we could have a regression if it never run in the CI, right ? (Added #6306 (comment) so we do not forget later)

[cdce8p]

Regarding the functional tests we could have a regression if it never run in the CI, right ? (Added #6306 (comment) so we do not forget later)

py-version isn't a full replacement. For example for bad-reversed-sequence we check the argument locals to see if it contains the __reversed__ protocol method. This will always be true when pylint is run with >= 3.8 so it can't detect an issue for 3.7. That's why I would usually recommend to run pylint with the lowest supported Python version.

However, as this is the only small regression I still think dropping 3.7 now is worth it. If someone really depended on it, he can continue to use pylint 2.17.x.

[Pierre-Sassoulas]

I think the primer fail is expected in this PR. I'm a little worried that all MR will need to be rebased on main after we merge but I don't think we have a way to avoid that.

[cdce8p]

I think the primer fail is expected in this PR. I'm a little worried that all MR will need to be rebased on main after we merge but I don't think we have a way to avoid that.

Yeah, that's just the nature of these cleanups.
However, I would rebase on a need bases only, not proactive. It's likely some of the breaking changes in astroid will cause additional merge conflicts down the line.