Update CI environment lockfiles

[github-actions]

Lockfiles updated to the latest resolvable environment.

[rcomer]

Oh dear, we may have to blame the new cf-units devs for these failures 😬. I don’t understand how though, as those changes were supposed to reinstate old behaviour. And why is py38 different?

[rcomer]

Aha! #3687 explains why previous behaviour != historical behaviour.

[rcomer]

py38 is still getting a cftime._cftime.real_datetime instance from num2date. I've no idea why that should be different from py36 and py37.

[jamesp]

Hmm, why has cython become a dependency? It wasn't in the previous build.

[trexfeathers]

Hmm, why has cython become a dependency? It wasn't in the previous build.

It was introduced for cf-units in v2.1.4 (conda-forge/cf_units-feedstock#32), so AFAICT it should have already propagated up to Iris before now?

[jamesp]

I would have thought that would only be a build dependency, not required to run cf-units

[trexfeathers]

Ah, yes, thanks. Didn't spot that fine distinction. Does that suggest one of the other changed dependencies has brought cython with it?

[jamesp]

Very odd, I'm not sure. I have a script somewhere that can read a lock file and tell you want depends on what. I'll dust it off and see if I can run it over this new version

[rcomer]

It seems the latest update has clobbered the commit from #4139 ☹

[trexfeathers]

@bjlittle informs me that cython is coming from python-stratify.

[bjlittle]

@jamesp python-stratify <0.2 will pull in cython as a runtime dependency. I fixed this as part of the recent 0.2.post0 release.

conda-lock has resolved against python-stratify 0.1.1 ... which will indeed be a source for pulling in cython.

If this is a problem, can you re-run to get the latest and greatest?

[jamesp]

Ok, since #4163 has now gone in this is only resolving py37 and py38 environments which is great. But now we're seeing inconsistencies between handling dayofwk between python 3.7 and 3.8... @trexfeathers any ideas why this might be?

[trexfeathers]

But now we're seeing inconsistencies between handling dayofwk between python 3.7 and 3.8... @trexfeathers any ideas why this might be?

It sounds related to cftime, but I don't know why there would be different behaviours between Python versions.

[jamesp]

Here's the diff between the two lock files

PACKAGE                        PY37          PY38
==================             ========      =========
cython                         0.29.23
importlib_metadata             4.4.0
python                         3.7.10        3.8.10
python-stratify                0.1.1         0.2.post0
python_abi                     3.7           3.8

Strange that the py37 is still using the old python-stratify

[trexfeathers]

But now we're seeing inconsistencies between handling dayofwk between python 3.7 and 3.8... @trexfeathers any ideas why this might be?

@jamesp see #4169.

[jamesp]

@trexfeathers @rcomer either of you want the honours of merging this?

[rcomer]

want the honours of merging this?

👀 🎉