CUDA 12 Migration includes Unsupported Windows

[carterbox]

conda-forge/libmagma-feedstock#4

The CUDA 12 Migration bot doesn't work for Windows. It fails on the conda-forge build setup step with the following error message:

2023-05-31T20:19:59.2050124Z ##[section]Starting: conda-forge build setup
2023-05-31T20:19:59.2216625Z ==============================================================================
2023-05-31T20:19:59.2217082Z Task         : Command line
2023-05-31T20:19:59.2217230Z Description  : Run a command line script using Bash on Linux and macOS and cmd.exe on Windows
2023-05-31T20:19:59.2217374Z Version      : 2.212.0
2023-05-31T20:19:59.2217493Z Author       : Microsoft Corporation
2023-05-31T20:19:59.2217671Z Help         : https://docs.microsoft.com/azure/devops/pipelines/tasks/utility/command-line
2023-05-31T20:19:59.2217818Z ==============================================================================
2023-05-31T20:20:00.3475222Z Generating script.
2023-05-31T20:20:00.3880761Z ========================== Starting Command Output ===========================
2023-05-31T20:20:00.4170133Z ##[command]"C:\Windows\system32\cmd.exe" /D /E:ON /V:OFF /S /C "CALL "D:\a\_temp\08a183b4-2c66-4613-a3be-6c3bfa7cc469.cmd""
2023-05-31T20:20:12.0113688Z Could Not Find C:\Windows\System32\msmpi.dll
2023-05-31T20:20:12.0120362Z Could Not Find C:\Windows\System32\msmpires.dll
2023-05-31T20:20:12.1608678Z CUDA '12.0' is not supported
2023-05-31T20:20:12.1613229Z Could not install CUDA
2023-05-31T20:20:12.2546635Z ##[error]Cmd.exe exited with code '1'.
2023-05-31T20:20:12.2948615Z ##[section]Finishing: conda-forge build setup

IMHO, the migration should be halted until all the Windows bits are in place (otherwise, packages will need to be migrated twice), and/or the bot should automatically add a skip statement like:

skip: True  # [win and cuda_compiler_version == "12.0"]

[beckermr]

Cc @jakirkham @h-vetinari

[jakirkham]

Thank for the ping! 🙏

Yep will take a look

[jakirkham]

Submitted PR ( conda-forge/conda-forge-pinning-feedstock#4526 ), which should do the trick

Tried re-rendering PR ( conda-forge/libmagma-feedstock#4 ) with those changes locally and it removed the Windows CUDA 12 variant (and related bits)

[jakirkham]

After merging that PR, tried re-rendering the libmagma feedstock and it appears to be doing the right thing now ( conda-forge/libmagma-feedstock#4 (comment) )