Only zip c_stdlib_version with compilers for CUDA by mbargull · Pull Request #5672 · conda-forge/conda-forge-pinning-feedstock (original) (raw)
I'd like to see it tested (like #5607; instructions therein) against some relevant feedstocks, e.g. https://github.com/conda-forge/ucx-split-feedstock.
I did this now (both with and without conda-forge/conda-smithy#1889), and ucx-split, cupy, abseil-cpp, numpy all worked fine.
It's important to note that we cannot remove (from #5607)
| c_stdlib_version: # [linux and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] |
|---|
| - 2.17 # [linux and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] |
as the CUDA 12 builds still need that for the zip to work (though the effect is somewhat surprising - not an error, but the CUDA 12 builds simply getting dropped); I experienced that because
conda-smithy rerender --no-check-uptodate -e conda_forge_pinning.yaml
ends up not pulling in the newest migrator versions, and so this happened for ucx-split & cupy.