ENH make sure packages in current repo data w/ features have versions without features

[beckermr]

This PR makes sure that packages in the current repo data which have features also have version without features. This change makes sure that feature minimization works.

cc @isuruf @jjhelmus

[beckermr]

So the osx test that failed passes locally for me and the windows failures appear unrelated but I don't quite follow them.

@marcelotrevisani do you know what is going on here?

[beckermr]

Ah it appears all of these tests are failing on master too. So I think this one is ready to go!

[beckermr]

Bump here! I fixed one bug in the pr and it appears to be working now in my tests of the new sysroot stuff for conda forge.

[jjhelmus]

I'm not certain excluding packages with features entirely from current_repodata.json is a good idea. Excluding packages from current_repodata could result in packages without features having un-met dependencies because the dependency has a feature.

I think a better solution is to treat features similar to "pins". That is, if package with features is included in current_repodata the newest version package without features should also be include, provided that one exists.

[beckermr]

I'm not certain excluding packages with features entirely from current_repodata.json is a good idea. Excluding packages from current_repodata could result in packages without features having un-met dependencies because the dependency has a feature.

So, in this case, the solver will fall back to the full repodata, but that is a performance hit. Is the performance the concern here?

[beckermr]

I think a better solution is to treat features similar to "pins". That is, if package with features is included in current_repodata the newest version package without features should also be include, provided that one exists.

This should work.

[jjhelmus]

So, in this case, the solver will fall back to the full repodata, but that is a performance hit. Is the performance the concern here?

I'm not worried about performance, more my worry is that if all of the packages for a particular name have a feature then that entire package is excluded current_repodata which may cause the solver to prefer the solution from current_repodata over something that is more logical.

As an example: Currently conda create -n test blas=*=openblas numpy will give a solution using current_repodata. With this proposed change the blas-1.0-openblas package would not be included in current_repodata and conda should fall back to the full repodata.

My concern is that there will be some other solution that conda does find in current_repodata for these edge cases that will make no sense. That said, I cannot provide a non-contrived concrete example. Expanding, current_repodata to always include a non-feature package seems like a lower risk than removing all feature packages.

[beckermr]

Ahhhhh thank you! That makes sense to me!

[beckermr]

ok @jjhelmus this is ready for another look!

[beckermr]

Bump here @jjhelmus

[beckermr]

Bump here

[jjhelmus]

LGTM thanks again for putting this together @beckermr

[github-actions]

Hi there, thank you for your contribution!

This pull request has been automatically locked because it has not had recent activity after being closed.

Please open a new issue or pull request if needed.

Thanks!