If conda-forge-pinning package has migrations installed, use those

[isuruf]

instead of the ones from the feedstock if the
timestamp field match and remove if the migration yaml has a
timestamp and there's no corresponding one in conda-forge-pinning
which indicates that the migration is over.

Checklist

• Added a news entry

[beckermr]

Do we need to consider the migration number too? If that gets changed it is a new migration IIUIC.

[scopatz]

Thanks for putting this in @isuruf. It isn't clear to me what would happen if not all of the dependencies have been rebuilt with the pin the migration. Would it fall back to using the original pin in the pinning file?

[isuruf]

It isn't clear to me what would happen if not all of the dependencies have been rebuilt with the pin the migration. Would it fall back to using the original pin in the pinning file?

I don't understand what you are asking here. Can you explain more?

In this PR what we are doing is using the migration file in the cfp package if there's a corresponding file in the feedstocks. For example if there's 2 migrations in the cfp package, python38 (A), boost172 (B) and there's 2 migrations in the feedstock, python38 (C) and boost171 (D), then conda-smithy will use (A) only because (B) is not in the feedstock, (C) is overriden by (A), and (D) is an old migration that has been merged to the pinning file.

[beckermr]

@isuruf given that none of the files are yet in the pinnings package, this pr will remove current migrations as of now. I guess we need to merge the change to the pinnings package first?

[beckermr]

Alright. This all seems fine to me. I’d personally like to see some tests of the various cases we have discussed to ensure that they all work and we don’t accidentally wreak havoc.

[beckermr]

The only case we are missing is an empty PREFIX migrations and old ones in the feedstock. Everything else is covered.

[isuruf]

Falling back to old behaviour where there's no $PREFIX/share/conda-forge/migrations is covered by other test cases.

[beckermr]

We probably also want a test where we keep one and delete one.

[beckermr]

Yep the old tests cover fall back. We don’t have tests of the deletion parts.

[isuruf]

Added one for deletion.

[beckermr]

LGTM!

[isuruf]

@conda-forge/bot, can one of you also have a look? This is the last piece missing in getting a PyPy migrator.

[CJ-Wright]

It would be great if some of the new functions could get some doc strings.

[isuruf]

@CJ-Wright, done.

[CJ-Wright]

It would be great if some of the new functions could get some doc strings.

[jakirkham]

They seem too (with the exception of tests). Were there some in particular you were looking at?