warp-lang-feedstock

[shi-eric]

Starting a thread here with @flferretti and @traversaro:

To what extent can we help with maintaining the https://github.com/conda-forge/warp-lang-feedstock?

Is there anything we can do as part of our release process to make the job of maintaining the Warp feedstock repository easier?

By the way, we're now on Apache 2.0.

[flferretti]

Thanks for starting the thread @shi-eric! There are a couple of ways we could streamline things together:

• Adding CI and Maintainership
  We'd be glad to give to anyone from your team maintainer access on the feedstock. That way you can push updates directly, review PRs, and adjust packaging logic as needed. Similarly, if you’d like to add CI steps (e.g., for upstream integration or testing variants) let us know what you need, we’re happy to help wire things up.

• Minimizing Patch Overhead
  Right now, we carry a few patches mostly to:

  • Avoid hardcoded GCC usage
  • Enable dynamic CUDA linking
  • Make the build a bit more configurable for packaging

  In general we try to keep upstream defaults untouched, and prefer introducing optional flags/env variables to adjust behavior only when needed for packaging. If you’re open to it, we can help upstream these tweaks as optional compile-time flags, keeping the defaults exactly as you have them, and letting us toggle behavior in the feedstock when needed.

• Release Coordination
  If there’s anything we can do during your release process, e.g. validate compatibility, we’d be happy to set that up. Just let us know what would be useful from your side.

Also, congrats on the switch to Apache 2.0, great news!

Let us know how you'd like to proceed, we're all ears

C.C. @traversaro

[flferretti]

By the way, I think there are some missing some info in the license

warp/LICENSE.md

Line 190 in 5495dc7

Copyright [yyyy] [name of copyright owner]

[shi-eric]

• Thanks for pointing out the license issue, I'll get that fixed! (Edit: After doing some searching, it seems that this may be intentional and we just rely on individual file headers for owner and year)
• For now I can help out as a maintainer on the feedstock, there may be more developers in the future who will be able to help.
• We're definitely interested in upstreaming your patches for the feedstock into Warp as non-default options.
• Not sure what we should do about release coordination yet, but I think the goal we would work towards is ensuring feature parity with the Warp we distribute via PyPi (e.g. MathDx support in tile operations if it's not there already) and release version parity.

[flferretti]

Thank you! I added you in the maintainers list in conda-forge/warp-lang-feedstock#53

In the next days I'll double check if MathDx support is enabled

[flferretti]

@shi-eric, sorry for the late reply. As discussed on conda-forge/warp-lang-feedstock#54, I can confirm that the CUDA12 recipes on Linux have MathDx support enabled