I think it's simpler to stay close to numpy while we are still getting to know this new release procedure.
I can see a few advantages to having SOURCE_REF_TO_BUILD in the yaml rather than a manual input field:
- harder to mess up, it's set once in a release branch rather than having to type it manually a few times
- audatibility. Right now nothing in the log is telling you what the input field was. If it's in the repo and an environment variable it's probably in the CI log.
I could not find more details by looking in numpy and numpy-release. The context around not using it in scikit-learn is in scikit-learn/scikit-learn#34320 (comment).
Also once it is set in a release branch you don't have to think too much about it, rather than typing a few times during the release process.
I think we could also set SOURCE_BUILD_REF to refs/pull/<pr_number>/merge as we did if we want to test on a scikit-learn PR, which did for 1.9.1 since this is part of our release process.
Note that numpy has an issue about this numpy/numpy-release#13, not sure if I am missing something or they haven't realized they can set it to a PR merge ref.
I think it's simpler to stay close to numpy while we are still getting to know this new release procedure.
I can see a few advantages to having
SOURCE_REF_TO_BUILDin the yaml rather than a manual input field:I could not find more details by looking in numpy and numpy-release. The context around not using it in scikit-learn is in scikit-learn/scikit-learn#34320 (comment).
Also once it is set in a release branch you don't have to think too much about it, rather than typing a few times during the release process.
I think we could also set SOURCE_BUILD_REF to
refs/pull/<pr_number>/mergeas we did if we want to test on a scikit-learn PR, which did for 1.9.1 since this is part of our release process.Note that numpy has an issue about this numpy/numpy-release#13, not sure if I am missing something or they haven't realized they can set it to a PR merge ref.