Workflow and packaging improvements

[cdce8p]

Steps

• Add yourself to CONTRIBUTORS if you are a new contributor.
• Add a ChangeLog entry describing what your PR does.
• If it's a new feature or an important bug fix, add a What's New entry in doc/whatsnew/<current release.rst>.
• Write a good description on what the PR does.

Description

Replaces: #4111
It started with just some CI improvements, but got a bit bigger in the end.

Since almost all changes are workflow related it should be possible to later include the MR in the 2.7.1 release.

Changes

• CI: Improve usage of pinned astroid version
• CI: Add caching for Windows job (whole suite now takes 4:30 - 5:00, instead of 7:00min)
• Fixed some tests that occurred during the Windows CI run
• CI: Use specific versions for Github Actions
• Update requirements
  • Pin all test requirements to current versions
  • Rename requirements _test_pypy.txt to requirements_test_min.txt
  • Moved some dependencies from min to test
  • Updated coverage dependency to ==5.4.0, instead of <5.0
  • Use pytest-cov as wrapper for coverage
  • Pin docs requirements to current versions
• Add dependabot config. The bot will open MRs once dependency updates are available
• Update mypy pre-commit hook
• Update tox config to use requirements files
• Update metadata license to use the correct SPDX tag: GPL-2.0-or-later
• Update readme badges

Important!
I've configured dependabot to add the dependency label to each MR it opens. This label needs to be created for it to work properly.

Type of Changes

	Type
✓	🐛 Bug fix
✓	✨ New feature
	🔨 Refactoring
	📜 Docs

Related Issue

[coveralls]

Coverage increased (+0.06%) to 91.51% when pulling 2711cda on cdce8p:improve-ci into 8cbcf87 on PyCQA:master.

[cdce8p]

We should probably also update the Testing and Contributing docs at some point 🤔

[cdce8p]

@Pierre-Sassoulas I noticed an issue with pinning the astroid dependency. Pip now complains when if I want to install the latest astroid master commit in my dev environment. That's because 2.5.0 is incompatible with the version in the astroid master branch 2.5-dev (shouldn't that be 2.6-dev anyway?).

https://github.com/PyCQA/astroid/blob/5192907ff22fc0383206c0e5bbe3e1635d6c7132/astroid/__pkginfo__.py#L27

I see two solutions here:

1. Leave the astroid version at 2.5 until the next release
2. OR change pylint install_requires to something like astroid>= 2.5 after a release and then back to a specific version for a new one. Here the astroid version would need to be set to 2.6-dev.

Option 2 might be the better one IMO. For testing it should be enough to loosen the requirement in the master branch. Updating the release branches (2.7, ...) can happen once a new version of astroid is released. What is your opinion?

--
cc: @hippo91

[Pierre-Sassoulas]

I think we disabled the benchmark tests in Github Actions at some point, and we should launch them at least once, I'll create an issue.

[Pierre-Sassoulas]

#4119

[cdce8p]

I think we disabled the benchmark tests in Github Actions at some point, and we should launch them at least once, I'll create an issue.

I haven't added them since they did not run before either. Just pushed a commit to include them. I'll check what's causing the issue later. Interestingly tox -e benchmark runs without issues on my machine.

[hippo91]

@cdce8p @Pierre-Sassoulas may i change the astroid master version to 2.6-dev?
@cdce8p i just saw your PR in astroid.

[Pierre-Sassoulas]

I'll check what's causing the issue later. Interestingly tox -e benchmark runs without issues on my machine.

It's a problem with astroid 2.5.0 see #4119 (comment)

Astroid 2.4.2 is sufficient for pylint to pass all pylint tests in python 3.8, I guess 3.9 is another matter. I did not have time to bisect the bug in astroid yet.

[cdce8p]

I'll check what's causing the issue later. Interestingly tox -e benchmark runs without issues on my machine.

It's a problem with astroid 2.5.0 see #4119 (comment)

I though so, but tox was using astroid==2.5.0 and I also ran it just with pytest and astroid==2.6.0-dev and it worked fine. A bit strange since I would have though Linux and MacOS would behave similarly.

Anyway, I've removed the benchmark job for now. Will open a new one MR, once the issue is fixed.
It wouldn't make sense to hold this one just because of it, I think.

[cdce8p]

Rebase to catch up with current master (which added the benchmark CI job).

[Pierre-Sassoulas]

Looks good, I like the use of requirement file in tox in particular. I might have missed something as this is hard to review. I'll wait for other reviewers to check as well because continuous integration tests are really important.