test: extend regression test on self-referential extras

[Trenza1ore]

What does this PR do?

As discussed in #14160, the current regression test introduced by #14157 was too naive, even old versions of pip (v20.3 - v21.1.3) were able to pass.

This PR added 9 more test cases to cover expected behaviour and prevent regression (local test results for v20.2 to latest & noxfile used are provided in #14160).

Closes #14160

Should pass with 20.3+

1. nested — nested self-referential extras should work.
2. external — a self-referential extra can mix pkg[a] with an unrelated third-party dep in the same extra.
3. partial — after pkg[a] is already installed, installing pkg[all] with the same version adds only the missing deps.
4. unknown — a self-reference with unknown extra (pkg[missing]) must raise a warning as usual.
5. upgrade-dep — a self-referential extra whose dependency version requirements change across versions (from requiring dep==1 to requiring dep==2) must bump dep to version 2.
6. circular — mutually dependent extras (extra a depends on extra b and vice versa) must resolve normally (pip install pkg[a] installs dependencies from both extras).

Should pass with 21.2+

7 & 8. upgrade-different-extras (two scenarios: had-no-extra or had-different-extras) — with an older pkg already installed (with no extras defined or with different extras defined and installed), upgrading to a new version with an all extra should: 1) pull new deps and 2) downgrade an existing dep.

Should pass with 24.1+

9. normalize — PEP 685 normalization should allow "all": ["pkg[x-y]"] to install extra defined as x_y in pyproject.toml.

Corresponding test cases in tests/functional/test_install_extras.py:

Column	Test
nested	test_install_self_referential_extras_nested
external	test_install_self_referential_extras_with_external_dep
partial	test_install_self_referential_extras_after_partial_install
unknown	test_install_self_referential_extras_unknown_nested
normalize	test_install_self_referential_extras_name_normalization
upgrade-dep	test_install_self_referential_extras_upgrade_changes_dep_version
circular	test_install_self_referential_extras_circular
upgrade-different-extras	test_install_self_referential_extras_upgrade_different_extras

PR Checklist:

• I agree to follow the PSF Code of Conduct.
• I have read and have followed the CONTRIBUTING.md file.
• I have added a news file fragment (or this PR does not need one).
• I have read and followed the AI_POLICY.md file, and if any AI tools were used, I have disclosed it below.

Disclosure:

Assisted-by: Cursor
It was used to write the nox file for testing the full matrix, and I have discussed the design of these test cases with it before implementation.

[pfmoore]

Please clarify your description of the tests here. In your own words, I don't want to have a discussion with an LLM.

What do you mean by tests being "Highly Unlikely to Fail" or "Could Potentially Fail". Tests are intended to never fail - any test that "might" fail is flaky, and we don't want flaky tests. If what you mean is that the tests check more or less likely scenarios, that's a very different matter (and I'd question why you thing one scenario is more likely than another).

Sorry for nitpicking here, but I have no patience for dealing with LLM verbosity, and I'd like a human explanation of what you're hoping to achieve here. As I said on the issue, brevity is a virtue.

[Trenza1ore]

Please clarify your description of the tests here. In your own words, I don't want to have a discussion with an LLM.

What do you mean by tests being "Highly Unlikely to Fail" or "Could Potentially Fail". Tests are intended to never fail - any test that "might" fail is flaky, and we don't want flaky tests. If what you mean is that the tests check more or less likely scenarios, that's a very different matter (and I'd question why you thing one scenario is more likely than another).

Sorry for nitpicking here, but I have no patience for dealing with LLM verbosity, and I'd like a human explanation of what you're hoping to achieve here. As I said on the issue, brevity is a virtue.

Sorry about the confusing titles, I meant that case 1-7 described above should pass between 20.3 and 21.1.3; and 8-9 should only pass with 21.2+, I subdivided the first 7 cases into 2 groups based of how I felt like, but that may not make sense.

The test case with PEP 685 normalization would start passing with 24.1+ so I have reordered to have it as the 9th case in PR description.

I've also updated the PR description and merged 7-8 into one test function parameterized with two scenarios:

1. old package version with no extra definition installed along with dep_a==2
2. old package version with extra a requiring dep_a==2 installed

Then a new package version with an "all": ["pkg[a]", "pkg[b]"] extra with updated extra a requiring dep_a==1 is installed for both scenarios.

@pfmoore I've updated the PR description with how I'd describe each case, again, really sorry for the initial descriptions with too much LLM-ness inside.