conda-forge core meeting 2025-05-28
Add new agenda items under the Your __new__() agenda items heading
Attendees
| Name | Initials | GitHub ID | Affiliation |
|---|---|---|---|
| Daniel Ching | DJC | carterbox | NVIDIA/cf |
| Filipe Fernandes | FF | ocefpaf | conda-forge |
| John Kirkham | JK | jakirkham | NVIDIA/cf |
| Jaime Rodríguez-Guerra | JRG | jaimergp | Quansight/cf |
| Dasha Gurova | DG | dashagurova | Anaconda |
X people total
Standing items
- [ ]
From previous meeting(s)
- [ ]
Active votes
- [ ]
Your __new__() agenda items
- (HV) Drop CUDA 11.8: https://github.com/conda-forge/conda-forge-pinning-feedstock/issues/7404
- this is currently blocking the CUDA 12.(8,9) migration, which we need for supporting newer architectures like sm_120.
- Also is getting more painful to maintain (more complex to maintain than other CUDA versions)
- Seeing fewer user downloads
- (DJC) Dropping CUDA Toolkit (CTK) 11.8 support from conda-forge is fine by NVIDIA; we'll support CTK 11.8 in our products' feedstocks if we decide we need to extend support
- Overall consensus to do this, make announcement and execute in 1-2 weeks.
- (JRG) When doing CUDA Docker images get dropped?
- (JK) Support is based on associated driver version support
- (HV) Compiler upgrades: GCC 14 & Clang 19: https://github.com/conda-forge/conda-forge-pinning-feedstock/pull/7421
- Make announcement once remaining PRs are in and tested
- (HV) Move to VS2022: https://github.com/conda-forge/conda-forge.github.io/issues/2138
- (DJC) We should just try it in some CUDA feedstocks if we're concerned about compatability
- (DJC) CUDA 11.8 only works with a earlier VS2022 version. It is not compatible with the latest VS2022. For CUDA 12.4+, this is not an issue. Details in https://github.com/conda-forge/conda-forge.github.io/issues/2138#issuecomment-2913390542 and following.
- (HV) Given we are ok dropping CUDA 11.8, we should be ok.
- (IF) Azure/GitHub Actions still have VS2019, but it is in a different location.
- Plan is to move forward with VS2022.
- (HV) Make announcement once ready
- (DJC) We should just try it in some CUDA feedstocks if we're concerned about compatability
- (HV) Dealing with C++/Fortran modules (https://github.com/conda-forge/conda-forge.github.io/issues/2525) -- we probably need to implement
host_exports(https://github.com/conda/ceps/issues/77)- Not a theoretical concern; we're hitting this on some Fortran feedstocks: https://github.com/conda-forge/flang-activation-feedstock/issues/14
- Planning to follow up on issue
- (HV) Follow through on numpy 2 migration closure: https://github.com/conda-forge/conda-forge.github.io/pull/2522
- (JK) Bulk of libraries support NumPy 2
- (HV) Pytorch 2.7 migration blocked since ~one month due to -- is there a way to debug/fix this, or do we want to migrate without windows?
- (IF) Is the package on cf-staging that we can copy over?
- (HV) Need to check whether this could be an option
- (IF) Other option would be caching on CI and copying over
- (HV) Think this doesn't work with the GPU builds
- (JRG) Checking with Amit just to make sure
- (IF) Is the package on cf-staging that we can copy over?
- (DJC) Tegra CUDA Toolkit Migration announcement https://github.com/conda-forge/cuda-feedstock/issues/57
- (IF) Can we figure out a way of building for SBSA & Tegra in the same build?
- (DJC) Would be unsupported
- (IF) We do other custom things to make life easier on our maintainers. Think this would still be valuable
- (IF) This would also avoid the need for a migrator.
- (DJC) Would like to do some simple tests to see how packages behave
- (DJC) Would be unsupported
- (IF) Can we figure out a way of building for SBSA & Tegra in the same build?
- (JRG) Core activity check in https://github.com/conda-forge/governance/issues/1
- (JRG) CPython supported releases policy discussion at https://github.com/conda-forge/python-feedstock/issues/794 (prompted by 3.14 rcs coming soon)
- (JRG) Please comment here
- (IF) Is this the right place to discuss the matrix?
- (HV) Useful for discoverability
- (JK) Important to note we will keep the python package updated to continue to support the matrix.
Pushed to next meeting
- [ ]
CFEPs
- [ ]