[major] Updates to cached-miniforge

[liamhuber]

As of five days ago, are getting CI issues. The mamba example on setup-miniconda stopped working too. I can't find or figure out a good direct attack, but mamba is deprecated anyhow so let's see whether it's possible to simply get around the problem by modernizing our setup-miniconda usage.

• Using mamba is deprecated, don't bother
• Localize env names containing paths with a ./ -- partially fixes CI failure debugging pyiron_workflow#634
• Remove the backwards compatible env name input (since removing the mamba flags breaks the interface anyhow)
• Sneaky unrelated: patch the sed invocation for osx

It's been just about a year since v3 dropped, so instead of deprecating the use_mamba flag, let's just get rid of it, clean up the other deprecated flag, and make a major bump to v4 here.

Whatever made the prepended ./ in the env names suddenly necessary broke other things too, so I'm hoping it gets rolled back/fixed. However, I don't see that including it here hurts anything, so let's keep it around either way.

[liamhuber]

Verified as working in pyiron/pyiron_workflow#636

[liamhuber]

It turns out that fixing this also seems to have resolved what was going wrong with build-docs, as seen here: pyiron/pyiron_workflow#637

[liamhuber]

internal references need to be re-targeted, but that is the last step pre-merge, because we need the branch-specific targeting to test behaviour.

[niklassiemer]

As far as I can tell this LGTM!

I have one question to the workflow to change the @commit/tag/branch, though:
Are the @use_miniconda rewritten with the merge? Or do we leave the already merged branch to pin the version until we release a new version and then use @4.0.0?

[liamhuber]

As far as I can tell this LGTM!

🚀

I have one question to the workflow to change the @commit/tag/branch, though: Are the @use_miniconda rewritten with the merge? Or do we leave the already merged branch to pin the version until we release a new version and then use @4.0.0?

Ah, I wish. It is not automated at all right now. Before merging I'll switch the targets all back to main; When we make a new release branch we set the target there and then make a release. To make a new release, we create and then freeze a new branch.

I discovered a potential weakness though and discuss it here. I think I'll try it now...