issue_comments: 1034340215

Thanks @rabernat (and others), I really appreciate your encouragement!

I think I start to see the light at the end of the tunnel here :). I have now refactored all the main parts of Xarray's internals that are using indexes and I managed to bring the failing tests down to a few. Now I'm going to try clearing all CI failures before working on the remaining TODO items. Although this is still a draft PR, it is now mostly ready for review I think. I'll mark it as ready when all CI is green.

Some of the re-implementations might still be improved, like align and concat. It's hard to know how best to do so before experimenting with custom indexes, though. We can still further iterate on this later...

IMO the best way to help will be to heavily test this branch (or the main branch once it is merged) against some real-world cases, as this PR will certainly introduce regressions that are not caught by the tests. A few release candidates might help in gathering feedback from the broader Xarray community. We'll also need to keep an eye on the asv benchmarks to ensure that there's no major regression in performance (I haven't checked yet).

This PR also intentionally introduces some minor changes in behavior (see my previous comments). I think those changes are welcome as they fix things that we can rather consider as bugs or limitations. That said, it is important to carefully check that they won't break Xarray users' (good or bad) habits.

And of course, once it is ready I'm eager to help anyone experiment with this feature (custom indexes)! I have a few ideas on convenient ways to implement custom selection / alignment behavior while mostly reusing the current Xarray indexing capabilities based on pandas indexes.