Rename weave-feedstock

[mstimberg]

Comment:

A while ago, the weave package was split-off from scipy.weave in maintenance-only mode (in particular, it was Python2-only with no ongoing development). Recently, the package has again seen a bit of development, in particular it has now experimental support for Python 3. More importantly, it has been renamed to scipy-weave, to release the name weave on pypi. For more details, see https://mail.python.org/archives/list/scipy-dev@python.org/message/AGFIP5TMTIYDZAF6B5WAFYCXJ66LAHDW/

The conda-forge recipe for weave is still referring to weave on pypi which is now an unrelated package. I guess the best approach would be to rename the weave package to scipy-weave on conda-forge as well? I don't know what the correct procedure is in this case – I can open a new PR on staged recipes for scipy-weave, but I don't know how to delete/deprecate an existing package. Otherwise I could of course update the package recipe so that it refers to scipy-weave on pypi, but I feel that would only lead to more confusion in the future.

Thanks for your help!

[EDIT: forgot a few links]
Current weave feedstock: https://github.com/conda-forge/weave-feedstock
Scipy-weave on Pypi (formerly known as weave): https://pypi.org/project/scipy-weave/
PR that renamed weave to scipy-weave: scipy/weave#18
New, unrelated weave package on Pypi: https://pypi.org/project/weave/

[jaimergp]

Hey @mstimberg, thanks for the question!

There are two things to note here:

• Renaming feedstocks is not supported, so we'll have to archive the existing one after obtaining a new one through staged-recipes.
• I think it would be tricky to change what weave means. Many problems come to mind like conflicting versioning, existing environments with "the other weave" all of a sudden getting something different, etc. The safest option would be to publish the unrelated weave with a different name in conda-forge.

Alternatively, you can keep the weave-feedstock and change your recipe into a multi-output recipe where you publish scipy-weave and weave simply aliases to it. I think that's the less confusing story here. The Weight & Biases team can choose to publish their weave with a different name if required (e.g. wandb-weave or something).

[ocefpaf]

Alternatively, you can keep the weave-feedstock and change your recipe into a multi-output recipe where you publish scipy-weave and weave simply aliases to it.

If we want to keep the namespace weave for scipy-weave, this is the best solution. If we don't care about that we must archive weave-feedstock to prevent it from becoming the "new weave," which is a different package that clobbers this namespace.

Namespaces in conda-forge are first come first served. So, the name weave , belongs to "scipy-weave" in my view, to archive it, or keep it with multiple outputs along side scipy-weave. It is your choice @conda-forge/weave team.

[mstimberg]

Thank you for your comments/suggestions @jaimergp and @ocefpaf. Just to be clear, I don't have a stake in this and I am not even involved with the development of scipy-weave in the first place – I only had a personal use case for it and needed it to be on conda-forge, so I created the recipe and maintained it. I am happy with any solution really, and I wouldn't feel particularly sorry for Weight & Biases if they cannot claim the weave name on conda-forge. I am just CC'ing @rgommers here, since he's maintaining scipy-weave (and was involved in the negotiations around releasing the pypi name).

Just to give a quick summary, it seems there are two solutions:

1. Archive weave-feedstock and go through the usual route for a new scipy-weave-feedstock
2. Turn weave-feedstock into a multi-output recipe, creating both scipy-weave and weave packages.

I'm happy with either solution, really, the first would be a bit easier for me, but slightly less convenient for existing users of weave. Although I am not sure whether this is really much of an issue: the conda-forge weave package is still Python2-only and the new scipy-weave package is obviously now Python3-only.

[rgommers]

Hey all. For context: the SciPy team decided to hand over the weave name on PyPI to Weights & Biases (and W&B supported SciPy development with a significant donation, so it was a win-win situation). During that rename to scipy-weave I took care of incorporating open PRs and making some maintenance updates for the package, most importantly Python 3.x support. That was a one-off effort, the package is effectively stagnant since there are much better alternatives now (Cython, Pythran, Numba - take your pick).

I'd be happy with whatever solution is easiest here, and also with repurposing the existing weave name for W&B if they would like to have it at some point. The download count for the last weave release is very small, this won't affect many users and we've had no complaints about the rename on PyPI.

[mstimberg]

Thanks @rgommers ! Ok, from my side then the preferred solution would be to close down the current weave-feedstock (i.e., archive it), and I will go through the usual staged recipes process for scipy-weave. If Weights&Biases will show up some day to get the "new weave" onto conda-forge, I'll let them figure that out with conda-forge admins. Thanks everyone for your help!

PS: I don't think there's anything I can do myself to close down/archive weave-feedstock?

[jaimergp]

PS: I don't think there's anything I can do myself to close down/archive weave-feedstock?

You need to open a PR in admin-requests (here) to request the archival. Create an issue in the feedstock too (maybe pointing to this discussion) to have a reference on why it was archived. Thanks!

[mstimberg]

After having had a closer look at scipy/weave, I finally decided to not go ahead with submitting a new conda-forge package for scipy-weave. As @rgommers explained earlier, the package is not developed any further, and it will not be installable on up-to-date systems very soon. As it is, it is not installable with Python≥3.11, and its test suite does not work with numpy ≥ 1.25. Again, as @rgommers said, nowadays there are better alternatives, so it is unlikely that someone will invest the time for a major update. If someone does, by all means get in touch with me if you want me to maintain a new conda-forge package 😉

I have opened a new issue to archive the feedstock (admin-requests/issues/776), and have also opened an issue on the weave-feedstock to explain the situation to anyone stopping by (conda-forge/weave-feedstock#21).

[jakirkham]

Here's the PR to archive the weave-feedstock: #778