FWIW, I prefer writing a method to return early, and not have large else blocks, but this PR focuses on fixing the bug without refactoring or writing code that conflicts with the existing style.

I've updated the tests of both Specifier.filter and SpecifierSet.filter so they match each other, with the exception of the empty string specifier which SpecifierSet accepts but Specifier doesn't.

This is ready for review / merge. FYI the relevant line in the spec is:

accept remotely available pre-releases for version specifiers where there is no final or post release that satisfies the version specifier

Specifier.filter already conforms, but SpecifierSet.filter does not, which means dependency resolution tools like pip that rely on packaging do not correctly follow the spec here.

I'm pinging just to check this is on someone's queue @brettcannon @pradyunsg @uranusjr as you all approved my last pre-release related PR #794, no rush to review or respond.

I believe that this is a blocker for writing a correct CDCL or CDNL (such as pubgrub) Python packaging resolver that is spec-compliant. This PR is also motivating my work on pypa/pip#13221.

I will be at the Packaging Summit at PyCon US 2025 in a couple of weeks if any questions would better be resolved in real life if any of you are there.

This PR is superseded by #897, if accepted.

To reduce confusion I'm going to close this in favor of #897

If #897 is rejected I'll reopen or recreate this PR.