conda-forge core meeting 2024-11-13

Add new agenda items under the Your __new__() agenda items heading

• Zoom link
• What time is the meeting in my time zone
• Previous meetings

Attendees

Name	Initials	GitHub ID	Affiliation
Marco Esters	ME	marcoesters	Anaconda
Daniel J Ching	DJC	carterbox	NVIDIA/conda-forge
Klaus Zimmermann	KZ	zklaus	Quansight
Cheng H. Lee	CHL	chenghlee	Anaconda/conda-forge
Scott Hain	SMH	scotthain	Anaconda
Dasha Gurova	DG	dashagurova	Anaconda
John Kirkham	JK	jakirkham	NVIDIA/cf

X people total

Standing items

• [ ]

From previous meeting(s)

• [ ]

Active votes

• [ ]

Your __new__() agenda items

• (DJC) conda-forge default build containers should always have the latest glibc/sysroot package that we publish
  • https://github.com/conda-forge/conda-forge-pinning-feedstock/issues/6283#issuecomment-2453101928
  • defaulting to the latest os makes os_version irrelevant for most users because glibc backward compatability
  • glibc constraint still set by sysroot package at build time; this package can lag behind syroot in container
  • (HV) For clarity, I would formulate the topline as: "conda-forge should use the newest available image versions by default (in sync with max sysroot that we publish)"
  • (HV) Fully support this proposal; draft implementation here
  • (HV) Also propose to remove c_stdlib_version from CUDA zip -- with the policy of "always newest image", this is not necessary anymore (and actually harmful to common usecases)
  • Additional clarifications:
    • HV: System image mostly irrelevant to the build process, only relevant for runtime constraints that power the __glibc virtual package.
    • IF: Ok with proceeding, but should take care of making sure that the cuda-* repackaged stuff still works with the original GLIBC / Docker images. Override in those cases, because those repackaged builds do not use our sysroot, and we can't ensure otherwise that they do work with the lowest Docker image available.
    • HV: consequence would be using os_version: linux_*: alma8 in conda-forge.yml on feedstocks that do binary repackaing
  • Recap: Ok to go, but binary repackaging feedstocks should pin os_version as per above (to stay with whatever minimum version they claim to support) before bumping the default image.
• (HV) Propose to consolidate image names: https://github.com/conda-forge/docker-images/issues/293
  • IF/CHL: Jinja variables can't be used in conda_build_config.yaml
  • use distro-name in the tag (also for CUDA 11.8); despite the lack of templating over it
  • (Some conversations about dropping CUDA 11.8 so the corresponding Docker images are not needed. This will happen eventually, just not yet.)
• (JRG) conda-forge/miniforge considered "dangerous site" by Google.
  • See https://github.com/conda-forge/miniforge/issues/667
  • Proposed solution: Move content to conda-forge.org, where we have ownership for reviews and disputes.
  • Thoughts?
    • Consensus: Give it a try.
• (HV) How to deal with CUDA 12.x?

Pushed to next meeting

• (ME) Composite action to build installers (Miniconda, Miniforge, etc.)
• [ ]

CFEPs

• [ ]