Do we want to change this to 3.9 instead?

Broadly, the unit & system test extras are not handled well in this (and likely all our noxfiles). As I will show below, we do fully test this library during unit tests, but I think it could be cleaned up (determine which extras need to be done with which Python versions, why, provide comments to clarify the why, and ensure that the logic across the setup.py, owlbot.py, noxfile.py are all in sync, etc.). I recommend that the clean up be a separate Issue as out of scope for the focus of this PR.

For unit tests, our noxfile currently overrides the effect of the variable
UNIT_TEST_EXTRAS_BY_PYTHON. UNIT_TEST_EXTRAS_BY_PYTHON is used in the function install_unittest_dependencies().

In the unit session there is another snippet that also selects extras to be installed. As can be seen below, for any Python version from 3.11-3.13 we test against alembic. But we see that for any Python version outside of 3.11-3.13 we test against all extras, which by definition in setup.py includes alembic.

if install_extras and session.python in ["3.11", "3.12", "3.13"]:
    install_target = ".[geography,alembic,tests,bqstorage]"
elif install_extras:
    install_target = ".[all]"
else:
    install_target = "."
session.install("-e", install_target, "-c", constraints_path)

The truthfulness of this can be see in the Kokoro CI/CD results which show:

• the install command: [all] versus [geography,alembic,tests,bqstorage] based on the Python version
• the actual install results from pip freeze (which includes alembic in every case)
• the successful run for the alembic tests (in every case)
• NOTE: I only show two versions, but I hand-checked every version.

3.9 (shortened for space reasons):

nox > Running session unit-3.9(protobuf_implementation='cpp')
...
nox > python -m pip install -e '.[all]' -c /tmpfs/src/github/python-bigquery-sqlalchemy/testing/constraints-3.9.txt
...
nox > python -m pip freeze
alembic==1.16.4
asyncmock==0.4.2
...
tests/unit/test__types.py ...........                                    [  9%]
tests/unit/test_alembic.py ..                                            [ 10%]
tests/unit/test_catalog_functions.py ................................... [ 21%]

3.13 (shortened for space reasons):

nox > Running session unit-3.13(protobuf_implementation='upb')
...
nox > python -m pip install -e '.[geography,alembic,tests,bqstorage]' -c /tmpfs/src/github/python-bigquery-sqlalchemy/testing/constraints-3.13.txt
nox > python -m pip freeze
alembic==1.16.4
asyncmock==0.4.2
...
tests/unit/test_alembic.py ..                                            [ 10%]
tests/unit/test_catalog_functions.py ................................... [ 21%]

I wonder if we still need coverage for alembic?

The compliance session selects extras slightly differently than the unit session.
I added alembic to this list of extras to account for that difference. DONE.

I assert that as part of the cleanup task discussed in another comment, all our noxfiles need to be checked to see when/why we choose to do some tests and not others. I suspect there was a historical reason that may OR may not apply.

It feels like we need to change this into 3.9 too?

I added 3.9 with alembic as a system test.

NOTE: there appear to be issues with how owlbot and noxfile, etc are interacting, much like described above for the unit tests. That should be looked into as part of a separate cleanup task.

Confirming - tests would fail if there were consequences to this +2 major version jump, right?

Tests will fail.
For context: the current version of pyarrow is 20.0
Why version 5.0?:

• Version 5.0 came out in 2021.
• Version 3.0 and version 4.0 do NOT run on Python 3.9.
• Version 5.0 is the first and oldest version that ran on Python 3.9.