GH-44314: [Packaging][Python] Use macOS 12 as deployment target to have macOS 12 pyarrow wheels

[raulcd]

Rationale for this change

We did just bump MACOSX_DEPLOYMENT_TARGET from 10 to 12 for pyarrow wheels on 18.0.0, we probably should wait a little before dropping.

What changes are included in this PR?

Revert moving macOS deployment target to 12

Are these changes tested?

Will trigger wheel jobs on archery

Are there any user-facing changes?

Yes, wheels will be available for macOS 12.

• GitHub Issue: [Python][Packaging] Revert macOS deployment target to 12 for pyarrow #44314

[github-actions]

⚠️ GitHub issue #44314 has been automatically assigned in GitHub to PR creator.

[raulcd]

@github-actions crossbow submit wheel-macos-monterey-*-amd64

[jorisvandenbossche]

Thanks @raulcd!

[raulcd]

@github-actions crossbow submit wheel-macos-monterey-*-amd64

[github-actions]

Revision: f5f407a

Submitted crossbow builds: ursacomputing/crossbow @ actions-bc85ef5cec

Task	Status
wheel-macos-monterey-cp310-cp310-amd64
wheel-macos-monterey-cp311-cp311-amd64
wheel-macos-monterey-cp312-cp312-amd64
wheel-macos-monterey-cp313-cp313-amd64
wheel-macos-monterey-cp313-cp313t-amd64
wheel-macos-monterey-cp39-cp39-amd64

[pitrou]

Is there someone with a macOS 12 machine that can test (one of) the generated wheels?

[kou]

+1

[raulcd]

Is there someone with a macOS 12 machine that can test (one of) the generated wheels?

Maybe @assignUser can spin up an AWS macOS 12 and just run the tests with one of those wheels? I don't have free access to an AWS account at the moment

[raulcd]

We should merge this to 18.0.0, would be nice if we can get a test @amoeba @assignUser maybe?

[jorisvandenbossche]

Ideally we can indeed test this, but FWIW I think many projects build with an older deployment target than what they are testing with (given the limitations of free CI).

We of course have the history of the macos-11 target actually failing in practice .. (#41696), because we were also not testing with it. But given it was working recently on CI, I assume running pyarrow on macos-12 is still fine.

[amoeba]

I don't have an amd64 macOS 12 machine laying around but I can probably find one. It sounds like it's not critical for this issue but I'll give it a try and report back.

[assignUser]

I was able to install the wheel on one of the deprecated macos-12 runners:

bash-3.2$ pip install wheel/pyarrow-18.0.0.dev426-cp312-cp312-macosx_12_0_x86_64.whl
Processing ./wheel/pyarrow-18.0.0.dev426-cp312-cp312-macosx_12_0_x86_64.whl
Collecting numpy>=1.16.6 (from pyarrow==18.0.0.dev426)
  Downloading numpy-2.1.2-cp312-cp312-macosx_10_13_x86_64.whl.metadata (60 kB)
Downloading numpy-2.1.2-cp312-cp312-macosx_10_13_x86_64.whl (20.8 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 20.8/20.8 MB 139.7 MB/s eta 0:00:00
Installing collected packages: numpy, pyarrow
Successfully installed numpy-2.1.2 pyarrow-18.0.0.dev426

[jorisvandenbossche]

FWIW installing is not enough to actually test it (installing just depends on the tag in the wheel name), you will have to at least import pyarrow in a python session (for macos-11 importing was sufficient to trigger a segfault (#41696), but if there is a problem in one of the submodules, even importing is not enough)

[assignUser]

🤦 of course ... sorry I'll do that soon

[assignUser]

I successfully tested reading and writing parquet locally and reading from s3 and imported and used compute. So it should be fine. I do think in general it's ok to keep the deployment target at 12 until we get user reports that it doesn't work.

We should probably note somewhere that eol macos support is best effort and can't be guaranteed due to the lack of testing infra. (support policy?)

[kou]

Thanks for testing this.
I merge this.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit deee9ac.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 10 possible false positives for unstable benchmarks that are known to sometimes produce them.