MAINT: Use mamba in CIs

[larsoner]

One way to move toward being able to recommend mamba is by including it in environment.yml, so I added it here. I made our solvers use mamba by including it in the setup-miniconda action according to the instructions here. Note that they even say it's experimental, and I'm assuming that's because mamba is moreso than the action itself:

Example 6: Mamba
Experimental! Use mamba to handle conda installs in a faster way. mamba-version accepts a version string x.y (including "*"). It requires you specify conda-forge as part of the channels, ideally with the highest priority.

FWIW @drammock Azure was already using mamba env update on Windows, IIRC to speed up the solve time.

[larsoner]

Okay, just asking it to install mamba as the solver broke things / created conflicts. This does not bode well for people trying conda install -c conda-forge mamba in their base environments :(

I'm going to try it now with better proper channel priority (as suggested in the setup-miniconda docs) and hope that works. I guess that's more in line with our instructions anyway, which suggest --strict-channel-priority at least some of the time.

[larsoner]

... apparently this works on macOS but not Linux. I'm tempted to close for now and revisit in a few months hoping some conda, mamba, and/or setup-miniconda fixes land in the meantime

[drammock]

using mambaforge fixed compat-old and compat-minimal. Only notebook is failing now, with:

 Warning: 
  UnsatisfiableError: The following specifications were found
  to be incompatible with the existing python installation in your environment:
  
  Specifications:
  
    - defaults/linux-64::brotlipy==0.7.0=py310h7f8727e_1002[md5=a8dfa5aefcc49d5e78000b16d2b7714e] -> python[version='2.7.*|3.5.*|3.6.*|3.6.12|3.6.12|3.7.10|3.7.10|3.7.12|3.8.12|3.8.16|3.9.16|>=3.11,<3.12.0a0|>=3.9,<3.10.0a0|>=3.8,<3.9.0a0|>=3.7,<3.8.0a0|3.9.10|>=3.6,<3.7.0a0|3.7.9|3.6.9|3.6.9|3.6.9|>=2.7,<2.8.0a0|3.6.9|>=3.5,<3.6.0a0|3.4.*',build='0_73_pypy|4_73_pypy|0_73_pypy|1_73_pypy|0_73_pypy|5_73_pypy|5_73_pypy|3_73_pypy|2_73_pypy|1_73_pypy']
    - defaults/linux-64::conda-content-trust==0.1.3=py310h06a4308_0[md5=69f1c06e7c9ce52d4a9b5787c74cac20] -> python[version='2.7.*|3.5.*|3.6.*|>=2.7,<2.8.0a0|>=3.11,<3.12.0a0|>=3.8,<3.9.0a0|>=3.9,<3.10.0a0|>=3.7,<3.8.0a0|>=3.6,<3.7.0a0|>=3.5,<3.6.0a0|3.4.*']
    - defaults/linux-64::cryptography==38.0.1=py310h9ce1e76_0[md5=f9aba54a3e93124c31846c700f502f9d] -> python[version='3.6.12|3.6.12|3.7.10|3.7.10|3.7.12|3.8.12|3.8.16|3.9.16|>=3.11,<3.12.0a0|>=3.9,<3.10.0a0|>=3.8,<3.9.0a0|>=3.7,<3.8.0a0|3.9.10|>=3.6,<3.7.0a0|3.7.9|3.6.9|3.6.9|3.6.9|>=2.7,<2.8.0a0|3.6.9',build='0_73_pypy|4_73_pypy|0_73_pypy|1_73_pypy|0_73_pypy|5_73_pypy|5_73_pypy|3_73_pypy|2_73_pypy|1_73_pypy']
    - defaults/linux-64::ruamel.yaml==0.17.21=py310h5eee18b_0[md5=0a54e321078e043b72b6f68d10472b62] -> python[version='>=3.11,<3.12.0a0|>=3.9,<3.10.0a0|>=3.8,<3.9.0a0|>=3.7,<3.8.0a0']
  
  Your python: defaults/linux-64::python==3.10.8=h7a1cb2a_1[md5=c50aebbb632aea22b13bd654dfe80a5b]

[... a lot more omitted]

If I'm reading it correctly, those versions of those packages simply don't allow python 3.10 (though they do allow 3.11?) Should I try using 3.11 in the notebook image, or is that known to fail for other reasons?

[drammock]

All green! mambaforge FTW. @larsoner merge if happy.

[drammock]

actually, we should check the job timings. maybe I'm reading things wrong but it looks like the usage minutes went up for the last run of CIs on this PR, when compared to recent runs on #11473 (a randomly chosen PR for comparison). IDK how to query GH Actions timings efficiently though, maybe you do?

[larsoner]

Okay I restarted linux_conda, compat_minimal, and compat_old and will check against runs from #11479 which are happening concurrently

[larsoner]

Timings are 6min / 2min / 2min, respectively, on this PR, and 8min / 1min / 1min on #11479 (roughly). But macOS improves from 11 to 7 min on this PR, too, which is nice. Since it's at least as fast overall and tests mamba let's merge and continue!

[agramfort]

🙏 @larsoner