search

scikit-hep / scikit-hep-tutorials

Ecosystem tutorials, demos, examples
https://scikit-hep.org/scikit-hep-tutorials
BSD 3-Clause "New" or "Revised" License
8 stars 2 forks source link

Use scikit-hep metapackage in environment.yml #9

Closed eduardo-rodrigues closed 3 years ago

eduardo-rodrigues commented 3 years ago

Closes https://github.com/scikit-hep/scikit-hep-tutorials/issues/6.

henryiii commented 3 years ago

No, these things should always come from Conda if possible. The conda version of awkward and boost-histogram are custom compiled for the platform you are on (and are the only way to get these things on Apple Silicon right now, I believe). You should use pip as little as possible, ideally not at all.

henryiii commented 3 years ago

I think if you manually include the needed conda packages, and then include scikit-hep from pip, then it can tell you have the correct versions of the other packages and will not install the PyPI versions (unless the versions are too tight and mismatched).

(A scikit-hep-feedstock on conda forge might not be a ridiculous idea...)

eduardo-rodrigues commented 3 years ago

Thanks for your comments, @henryiii.

No, these things should always come from Conda if possible. The conda version of awkward and boost-histogram are custom compiled for the platform you are on (and are the only way to get these things on Apple Silicon right now, I believe). You should use pip as little as possible, ideally not at all.

No discussion there, certainly for "proper work". But do we really care about the custom compilation / greatest performance tuning to run a tutorial that is to be displayed in a book ;-). I'm just trying to simplify the settings here, but see below ..

I think if you manually include the needed conda packages, and then include scikit-hep from pip, then it can tell you have the correct versions of the other packages and will not install the PyPI versions (unless the versions are too tight and mismatched).

I can certainly try that.

(A scikit-hep-feedstock on conda forge might not be a ridiculous idea...)

Indeed I think it is a good idea in order to fully profit from the metapackage idea. We will have to get several of our org packages in conda, though, and that's going to take a bit of time. I've just created https://github.com/scikit-hep/scikit-hep/issues/131 to follow-up on this.

eduardo-rodrigues commented 3 years ago

Getting back to this - for reference the scikit-hep metapackage is available in Conda for the most recent versions. We can proceed with this MR.

eduardo-rodrigues commented 3 years ago

OK @henryiii, updated this MR to get pretty much everything from Conda. Ready for review :-).