CI: correct azure-36-locale slow name + move pyarrow

[jorisvandenbossche]

Follow-up on #30039

I put it in the wrong file, because the file names were swapped...

Now, here, I just renamed both files to the other name (the diff apparently cannot the rename, therefore you see a lot of changes, but I can ensure I didn't change anything except for pyarrow).
But, I could also leave the files as is and switch them in the posix.yml file ? (use the envs for the different build)

I don't know if there was something specific in either environment that was specifically added for the slow tests?

cc @datapythonista

[jbrockmendel]

is the test failure in the py37_locale build one of the things this PR is intended to address?

[jorisvandenbossche]

I didn't change anything about the 37 build (only 36), so no ;)
I also didn't see that one yet, thanks for pointing out. It's from the PR I merged earlier today (#29961, and which I should have updated with master after the CI changes to catch this failing test, in hindsight)

[jorisvandenbossche]

I quickly "fixed" CI (-> #30069), will look into it tomorrow properly

[datapythonista]

The renaming makes sense. I saw that before, was part of #29685.

For me the way the tests are specified is still too complex to understand. So I don't really know what makes sense regarding the changes in versions. But if you know what you're doing, I'm happy with this, and hopefully we revisit all those dependency files for #29685 soon.

[jorisvandenbossche]

But if you know what you're doing, I'm happy with this

To be clear, I explicitly don't know what I am doing ;) Because there is no guideline or docs or rationale on what is the purpose of which build or what is supposed to be tested in which build.

But, basically I didn't change anything in this PR (except for the pyarrow stuff), I just fixed the naming to be logical. The main thing I was wondering is that maybe the names weren't wrong in the first places, but we were accidentally using the wrong file in the posix.yml, and maybe I should only have changed it there (which would actually change which packages are used in the slow test build).

[jorisvandenbossche]

For the actual issue I was trying to solve: I checked that in the Linux py36_locale build pyarrow 0.12 is installed, and the parquet tests are actually run (not listed in the skipped ones), so now it is working as intended.

[datapythonista]

I was referring to unpinning all the packages. I think the renaming is the best we can do so far, at least the names match what's being tested.

[jorisvandenbossche]

I was referring to unpinning all the packages.

To be clear, in case you are talking about this PR: I am trying to explain I didn't do any of that. The "big diff" is just because git/github couldn't handle the rename

[datapythonista]

ah, I see, didn't get that. All good then.

[jorisvandenbossche]

OK, merging then!