RLS: 2.0.2

[datapythonista]

This is the tracker for the 2.0.2 release. Probably happening mid-May.

cc @pandas-dev/pandas-core

[phofl]

Thoughts on releasing next week? Would this work for you @datapythonista? We have a bunch of fixes, so a release would b worth it

[datapythonista]

Yes, I was thinking about that. I'm at a conference this week, but I can release early next week.

[datapythonista]

There were some unexpected things and I haven't been able to release yet as you've seen. I can release Friday, or if people prefer to avoid "deployments" on Friday, I can release Monday too. I'll do it Friday of no objections.

[phofl]

I'd be in favour of releasing Friday, it's only a patch release, so should be ok

[datapythonista]

I started the release few hours ago, but the docs build of the 2.0.x branch is failing. Not sure exactly why, but I think I may have a fix, will keep this thread posted.

[datapythonista]

CI is green for 2.0.x, and everything is ready. I'll tag now.

[datapythonista]

pandas 2.0.2 tagged, CI for wheels and docs running, PR for conda-forge opened: conda-forge/pandas-feedstock#164

I'll continue tomorrow morning (Middle East time) with publishing the wheels, merging the conda-forge PR, and the rest of the release stuff.

[datapythonista]

There is something odd with the tests. All tests work just fine until the release, but the Linux 32 test failed to start with the empty commit we use for the release, and for the tag builds. This is the error:

ImportError while loading conftest '/__w/pandas/pandas/pandas/conftest.py'.
pandas/__init__.py:22: in <module>
    from pandas.compat import is_numpy_dev as _is_numpy_dev  # pyright: ignore # noqa:F401
pandas/compat/__init__.py:25: in <module>
    from pandas.compat.numpy import (
pandas/compat/numpy/__init__.py:4: in <module>
    from pandas.util.version import Version
pandas/util/__init__.py:2: in <module>
    from pandas.util._decorators import (  # noqa:F401
pandas/util/_decorators.py:14: in <module>
    from pandas._libs.properties import cache_readonly
pandas/_libs/__init__.py:13: in <module>
    from pandas._libs.interval import Interval
pandas/_libs/interval.pyx:1: in init pandas._libs.interval
    import numbers
pandas/_libs/hashtable.pyx:1: in init pandas._libs.hashtable
    cimport cython
pandas/_libs/missing.pyx:1: in init pandas._libs.missing
    from decimal import Decimal
pandas/_libs/tslibs/__init__.py:66: in <module>
    from pandas._libs.tslibs.period import (
E   ImportError: /__w/pandas/pandas/pandas/_libs/tslibs/period.cpython-38-i386-linux-gnu.so: undefined symbol: pandas_datetime_to_datetimestruct

Source: https://github.com/pandas-dev/pandas/actions/runs/5106394697/jobs/9178573560

All wheels have been generated correctly (we don't have wheels for Linux 32 anyway), so moving forward with the release, but it'd be good to understand what's going on.

[WillAyd]

I don't think we backported #51525 so if you are trying to build in parallel on that branch you might run into that issue (not at computer to double check - but error message matches what that resolved)

[datapythonista]

I don't think we backported #51525 so if you are trying to build in parallel on that branch you might run into that issue (not at computer to double check - but error message matches what that resolved)

Ah,I see. I guess we can try to backport now, and see if that fixes the problem. What I'm not sure is why we have tests independent from the wheels. If we're not going to release something, I'm not sure what value adds to test it. Personally I think we should have one set of jobs to build the wheels, and then run the tests on those. And when tagging, run the same exact jobs, but publishing the wheels if everything is fine.

[datapythonista]

Still some conda-forge ARM builds running, but other than that the release is complete. Closing.

[datapythonista]

All ARM builds in conda-forge finished correctly.