chore(dev): Use Miniforge instead of Mambaforge

[CurtHagenlocher]

Use Miniforge instead of Mambaforge, as the latter has been deprecated.

Closes #2206

[kou]

+1

[CurtHagenlocher]

Not sure how to reconcile

Program /tmp/arrow-adbc-HEAD.ThdRU/miniforge/envs/conda-source/bin/glib-mkenums found: YES (/tmp/arrow-adbc-HEAD.ThdRU/miniforge/envs/conda-source/bin/glib-mkenums)

with

FileNotFoundError: [Errno 2] No such file or directory: '/tmp/arrow-adbc-HEAD.ThdRU/miniforge/envs/conda-source/bin/glib-mkenums'

[kou]

Ah, this is a known problem.
See also: apache/arrow#44270

Could you pin glib to 2.80.5 as a workaround?

[CurtHagenlocher]

That did indeed solve that problem. There are three more failures (so far). Two of them seem to be because R isn't installed, and I can't tell what's supposed to be doing that.

[paleolimbot]

Two of them seem to be because R isn't installed

I just looked for that error briefly and couldn't find it...which job is it on?

and I can't tell what's supposed to be doing that.

Here, I think:

https://github.com/apache/arrow-adbc/blob/d4163d94ce50d91b649fb50fe33d58e772bf0c54/ci/conda_env_r.txt

...but not sure why this change would cause that to stop working.

[CurtHagenlocher]

In "Verification (Nightly) / Verify Source (Conda)/macos-13 (pull_request)", for instance, at nearly the end of the log:
./arrow-adbc/dev/release/verify-release-candidate.sh: line 586: R: command not found

EDIT: Does "r-testthat" include both R and the "testthat" framework?

[paleolimbot]

It does! There seem to be a very long list of conflicts when installing the R dependencies, which I think is something about krb5, icu, and/or openssl.

https://github.com/apache/arrow-adbc/actions/runs/11137066683/job/30949901931?pr=2211#step:5:30453

It seems like pinning gobject may have introduced this? (I don't see those conflicts in the run before that pin)

[CurtHagenlocher]

It seems like pinning gobject may have introduced this? (I don't see those conflicts in the run before that pin)

It's possible. I tried just the pinning change in #2213, but that didn't trigger the verification runs that are failing with the R error.

[paleolimbot]

Hmm...I also can't trigger strictly based on the pin (although I know our conda env has a lot of stuff).

# docker run --rm -it condaforge/miniforge3
conda create -n foofy
conda install glib=2.80.5 compilers libsqlite libpq pkg-config r-testthat

If you want to skip the R part and open a separate issue I can plug away at that before the next release...getting CI to pass for everybody else is a far more pressing concern (we run separate R checks using environments more likely to show up for a user in CI).

[CurtHagenlocher]

If you want to skip the R part and open a separate issue I can plug away at that before the next release

Raúl has opened a separate issue (#2217) as it turns out that the R part is already failing without this Miniforge change. As such, I'm going to check this in.