Test Python 3.9 and 3.12 on CI, test minimum dependencies

[krassowski]

• fixes /learn with wildcards not working in python 3.10 #1026

Related to multiple past issues where dependencies where misspecified.

[krassowski]

Good, it caught the reported issue:

Bad, the custom install script is reinstalling the pip packages invalidating the resolution in base action. I will remove it.

[krassowski]

Looking at the pip list output it does not actually install the minimum required versions in the minimum variant. I think this is likely either:

• because the resolution rules do not apply to dependencies (as in metaproject dependencies i.e. jupyter-ai-magics and jupyter-ai) or
• because in course of jlpm dev-install the resolutions get overwritten.

This PR could still be merged as-is just to ensure the Python versions are covered; I will iterate on the minimum version installation later either way.

[dlqqq]

@krassowski Thank you for opening this PR! This is a valuable contribution that will help prevent similar issues from arising in the future. I've left a few points of feedback for you to address below.

Looking at the pip list output it does not actually install the minimum required versions in the minimum variant.

I think to support dependency-type being used in GitHub job runs, we need to figure out how this is being passed to the shell command being run. For example, does this set an environment variable that scripts are supposed to use to determine how to install their dependencies? I can also investigate this myself later if you'd prefer my help on this.

[krassowski]

To mention it explicitly, the issue with minimum dependency_type is being solved with jupyterlab/maintainer-tools#250.

I think to support dependency-type being used in GitHub job runs, we need to figure out how this is being passed to the shell command being run. For example, does this set an environment variable that scripts are supposed to use to determine how to install their dependencies?

dependency_type is an input to base-setup action (I personally would prefer if it was passed explicitly but I don't feel like trying to fix something that ain't broke). When it is set to minimum three things happen:

• pip constraints are discovered from the extension source (how is an implementation detail) and written into a constraints.txt file
• the contsraints are installed using pip -c option
• the path to the constraints file is exported in PIP_CONSTRAINT environment variable (which is recognised by pip in any subsequent operations).

[krassowski]

@dlqqq does jupyterlab/maintainer-tools#250 look good to you?

[dlqqq]

@krassowski This PR looks great, thank you for contributing this! I've also reviewed jupyterlab/maintainer-tools#250.