Use micromamba 1.5.10 where conda is needed

[kmuehlbauer]

Updated behaviour

Since micromamba 2.0 installing conda seems to be broken in CI. In #9732 conda was removed from the install where not absolutely needed and the micromamba version was pinned to 2.0.2-0 due to broken 2.0.3-0.

The broken version was yanked and this PR reverts to the default micromamba version (now 2.0.2-0). In CI runs where conda is absolutely needed the micromamba-version is set to 1.5.10-0 (last version of v1-release line).

• When changes for mamba 2.0 airspeed-velocity/asv#1439 is resolved, we can remove the pin in the benchmark CI runs.
• By resolving Use micromamba 1.5.10 where conda is needed #9737 (comment) the pin in the min-deps-check CI run will be removed.

Former ramblings:

conda.api and libmambapy are not API compatible

In https://github.com/pydata/xarray/blob/main/ci/min_deps_check.py we use conda.api to gather package versions from ["conda-forge", "defaults"] channel (shouldn't we remove defaults, because of the legal stuff?).

The replacement for micromamba v2 is libmambapy which unfortunately isn't API compatible. It should have all the capabilities to do the same, but I wasn`t successful ( 🤷‍♂️ ) in gathering the correct incantations from the docs:

• https://mamba.readthedocs.io/en/latest/usage/specs.html
• https://mamba.readthedocs.io/en/latest/usage/solver.html

If figured out the correct way, this can eventually be replaced.

Tested micromamba/libmamba 2.0.2-0 also for:

• ✅ upstream
• ❌ benchmark, asv doesn't work, need to step back here, too
• ✅ hypothesis
• ❌ pyright, currently disabled

[kmuehlbauer]

@keewis I've missed this yesterday, so checked it today.

asv is also not yet ready for micromamba 2 , so I've reverted min-deps-check and benchmark CI runs to use last 1.5.10-0 version of mamba/micromamba. All other CI runs look good.

WOuld be nice, if you could have another look here. Thanks!

[keewis]

looks good, just a couple of comments:

• I don't have knowledge about the benchmarks, so I can't comment on that
• according to the issue you opened we might be able to remove most of the micromamba version pins again?
• regarding libmambapy, I think it is time to rewrite the min-versions script anyways, so I think pinning for now is fine

shouldn't we remove defaults, because of the legal stuff?

we're firmly open source, so that shouldn't affect us (I think?). However, we don't really use the defaults channel anywhere else, so my guess is that this was to get more accurate python release dates? All that would change, though, is that we would get to advance the bump date by a few days – which we usually miss by a couple of weeks anyways, so doesn't really matter either way. And finally, see my note above about rewriting the script.

[keewis]

I think we can use py-rattler's version of query instead, which will get rid of the conda dependency, is much faster, and we can avoid the undocumented libmambapy API.

I'll open a PR for the rewrite of the script once it's close to being ready.

[kmuehlbauer]

@keewis I've followed your suggestion to use default settings where applicable, as the problematic version was removed.

Let's see how this works.

[kmuehlbauer]

Nice, array-api-strict 2.1.2 (released 5 hours ago) breaks the tests.