Tzu-ping Chung

The parsing part is likely already supported, but it’d be helpful to be about to _output_ to that form. [Currently I’m formatting the line myself](https://github.com/uranusjr/pfff/blob/master/pfff/build.py#L11).

Personally I prefer to not put tests in sdists since they’re dead weight to most people. What prevents the Git tags (and artifacts generated by GitHub against them) from being...

When debugging Molinilo fixtures I constantly add logging and breakpoints at three locations: * New candidate introduced (pinned). * Retracting a previously-pinned candidate due to backtracking. * A requirement exhausted...

Next on my todo after pulling in test cases. I also want to modernise the metadata a bit, maybe move to Flit.

CI added in #23. Release automation needs some clean up, but there’s no hurry since we’re not releasing any time soon (until at least I finish with tests).

I had the automation down using invoke and parver (for version bumping). I’ll use this script manually for a while to make sure it satisfies the need, and make it...

Sure, I probably won’t have much time working on this for the next week, and if I find some I’ll work on creating a better mock PyPI server first.

> is it necessary that the same candidate is returned in both calls, or is it enough that "equivalent" candidates are returned? The resolver does not compare candidates with each...

A full write-up on how the interface should be implemented, why each one is needed, and things the implementer should consider. This will be easier to write when we finish...

I don’t think this requires a resolvelib change. To make the resolver to not backtrack, you can simply return only one candidate from `find_matches()`.