Wheels

[henryiii]

• Move testing to GHA
• Add wheel builds
• Enable parallel compile (seems to break, can be reenabled later)
• Use PEP 517/PEP 518 builds
• Simplify setup.py quite a bit
  • Use setuptools built-in Cython support
• Add python_requires

[kratsg]

Hi Henry, this looks great. Two questions:

1) what happens with copyrights? (I see you have your copyrights around a few of these files, but I'm not sure what the appropriate policy here is, given scikit-hep/iris-hep/etc...) 2) do we really want to support python2?

[henryiii]

1. what happens with copyrights?

I think the general procedure is to put the author as the copyright holder and use a free license. It was discussed a few times and I think this was the conclusion. @HDembinski @eduardo-rodrigues

2. do we really want to support python2?

See https://scikit-hep.org/supported-python-versions. And all packages must have a release with python_requires before dropping Python 2 (if they ever supported Python 2, that is)!

[henryiii]

Passes!

[henryiii]

Standard reminder: needs to be regular-merged, as rebase breaks subtree (at least locally it does).

[kratsg]

Standard reminder: needs to be regular-merged, as rebase breaks subtree (at least locally it does).

Will squash-n-merge work?

[henryiii]

Not to my knowledge. The “merge” commit must be preserved for it to be updatable later via a subtree pull. That’s why rebase fails, it thinks the merge commit is empty and drops it, which causes all the files to move out of their prefix and into the base repository! You have to rebase with mergers preserved (don’t know what github does for PRs). I’m by no means an expert, I prefer sub modules, but azure doesn’t check out submodules before configuring.

[kratsg]

If I look at squash+merge on GH, this is what I see

* Squashed '.ci/azure-wheel-helpers/' content from commit e52961a

git-subtree-dir: .ci/azure-wheel-helpers
git-subtree-split: e52961a9467954d8393b8fcbc714d44c97ec2166

* Add build, parallel, simplify

I'm generally a fan of a single commit, rather than N+merge -- but if this is the issue with subtree, we can do a regular merge here.

[eduardo-rodrigues]

what happens with copyrights?

I think the general procedure is to put the author as the copyright holder and use a free license. It was discussed a few times and I think this was the conclusion. @HDembinski @eduardo-rodrigues

Seems often the most adequate think to do, e.g. as in scikit-learn. For sure the copyright cannot be DIANA/HEP or IRIS-HEP or similar, but rather a university, etc. Having the main author is best is many situations.

[henryiii]

I'm generally a fan of a single commit, rather than N+merge -- but if this is the issue with subtree, we can do a regular merge here.

I'd have to play with it to see what happens if it's squashed. I've not even done a subtree pull yet.

[henryiii]

I'm generally a fan of a single commit, rather than N+merge -- but if this is the issue with subtree, we can do a regular merge here.

Github won't rebase-and-merge, it detects that the files will change if it does. And squash and merge does lose the history:

~/g/s/numpythia $ git subtree pull --prefix .ci/azure-wheel-helpers git@github.com:scikit-hep/azure-wheel-helpers.git master --squash
From github.com:scikit-hep/azure-wheel-helpers
 * branch            master     -> FETCH_HEAD
Can't squash-merge: '.ci/azure-wheel-helpers' was never added.

~/g/s/numpythia $ git checkout henryiii-wheels 
Switched to branch 'henryiii-wheels'
Your branch is up to date with 'origin/henryiii-wheels'.

~/g/s/numpythia $ git subtree pull --prefix .ci/azure-wheel-helpers git@github.com:scikit-hep/azure-wheel-helpers.git master --squash
From github.com:scikit-hep/azure-wheel-helpers
 * branch            master     -> FETCH_HEAD
Subtree is already at commit 02274e749eaeb9903b185c015b0d2f6d633a2913.