Per conversation in a meeting today (will link once notes are up), we are now finding in our error messages disentangling work that we're rediscovering needs/features that are in PubGrub. Thus, we are discussing whether we should add a PubGrub implementation to pip (or ResolveLib).

Personally I agree PubGrub is the best technical choice for Python, and also the best choice for pip in theory. The problem here however is that there is no Python implementation suitable for pip’s usage. As mentioned above, it is difficult to integrate Mixology into pip due to data structure incompatibilities; on the other hand, both of our backtracking choices are designed to fit well into pip.

I’ve been thinking about implementing a PubGrub resolver (maybe based on Mixology, maybe not) that expose a similar interface to ResolveLib and Zazo, that can be better used by projects without forcing them into using dedicated data structures. But I don’t believe it is the best use of my time in the near term given my responsibilities (including bringing a resolver into pip within a deadline).

Given that proper resolvers are functionally interchangeable, my choice now is to pursue a more reachable goal. [insert Perfect is the Enemy of Good cliche here] But I also intend to help line up goals and efforts for a future alternative implementation, and work on that after pip has something to show for.

Originally posted by @uranusjr in #7406 (comment)