Dependencies updates can break releases

[anissa111]

Currently, the latest version (2022.10.0) of geocat-comp breaks on import (See GH actions logs)

>>> import geocat.comp
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/anissaz/opt/miniconda3/envs/test_import/lib/python3.10/site-packages/geocat/comp/__init__.py", line 16, in <module>
    from .skewt_params import get_skewt_vars, showalter_index
  File "/Users/anissaz/opt/miniconda3/envs/test_import/lib/python3.10/site-packages/geocat/comp/skewt_params.py", line 5, in <module>
    import pint.quantity
ModuleNotFoundError: No module named 'pint.quantity'

PR #277 addresses this, but the published package is still broken.

We should do the following:

• Patch release of 2022.10.0 as soon as possible that includes the pint pinning
• Look in to how other packages handle pinning dependencies for releases so this doesn't happen again
• Consider making a nightly released package import test to catch potential issues as soon as possible

[dcherian]

Consider making a nightly released package import test to catch potential issues as soon as possible

Highly recommend xarray's infrastructure for this. keewis just created a new action with most of the scripting (https://github.com/xarray-contrib/issue-from-pytest-log) so it should be relatively easy now

[anissa111]

@dcherian thank you very much, I'll look in to that!

[erogluorhan]

Consider making a nightly released package import test to catch potential issues as soon as possible

Highly recommend xarray's infrastructure for this. keewis just created a new action with most of the scripting (https://github.com/xarray-contrib/issue-from-pytest-log) so it should be relatively easy now

Looks like we already have this infrastructure in our upstream workflow, don't we? However, we may want to modify our upstream environment file to install as many of those dependencies as we want to install from their GitHub repo.

@dcherian xarray's workflow seems to be conda-installing everything, then removing many of them, and pip-installing through GitHub repositories. What is the intention with conda-install and uninstall stage? Couldn't they directly be pip-installed?

[dcherian]

Perhaps to keep dependencies at the same versions as in the main env?

The real magic is the automated opening of an issue when the upstream-dev env fails.

e.g. pydata/xarray#7175

[erogluorhan]

Okay, yeah, that automated issue opening looks good!

[anissa111]

Close as completed with #300 and #301