2025-12-10 - HackMD

--- tags: [meeting-notes] title: '2025-12-10' --- # conda-forge core meeting 2025-12-10 Add new agenda items under the `Your __new__() agenda items` heading - [Zoom link](https://zoom.us/j/9138593505?pwd=SWh3dE1IK05LV01Qa0FJZ1ZpMzJLZz09) - [What time is the meeting in my time zone](https://dateful.com/convert/utc?t=5pm) - [Previous meetings](https://conda-forge.org/community/minutes/) ## Attendees | Name | Initials | GitHub ID | Affiliation | | ----------------------- | -------- | --------------- | --------------------------- | | John Kirkham | JK | jakirkham | NVIDIA/cf | | Klaus Zimmermann | KZ | zklaus | Quansight | | Cheng H. Lee | CHL | chenghlee | Anaconda/cf | | Lilly Winfree | LW | lwinfree | Anaconda | | 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 - [X] (DJC) Tegra YAML migrator - https://github.com/conda-forge/conda-forge-pinning-feedstock/pull/7956 - Looking for reviews after a few weeks. Main criticism: using `key_override` instead of `key_add`, which replaces the whole zipped keys set, which requires changes in the global pinning. Tried using `key_add` but couldn't make it work. Reason: tegra only needed in 12.9, but using key_add will also add it to 13.0. Due to a bug in conda-smithy, even a skip statement will not work (skip cuda 13 + tegra), it will show up in the feedstock render. - JRG: Link to conda-smithy bug? - https://github.com/conda-forge/conda-smithy/issues/1617 - IF: The `key_add` migrator will do the job; it's just that the variant won't be skipped until runtime (an extra job will be spawned, but won't publish anything). Similar issue with tensorflow migrator. - DJC: I will open a new MR which uses `key_add` instead of `key_override` - [X] (JRG) Last meeting of the year, we will reconvene in 2026 under a unified conda + conda-forge meeting scheme! Calendar invites will be updated and blog posts published in conda-forge.org and conda.org. - [X] (LW & DG) Discussion about package data - (JRG) missing granular usage data - this helps decide things like deprecations & updates, and prioritize - Python version usage per package - Architecture per package - e.g. GLIBC usage tracking (equivalent for macOS and Windows) - e.g. Like these manylinux plots: https://mayeut.github.io/manylinux-timeline/ - (JRG) User agent - More info about where the download occurred (for example CI downloads compared to non-CI) - Can users say who they are? For example could conda-forge add this info to its downloads to more easily segment this? - https://github.com/conda/infrastructure/issues/1018 - https://github.com/anaconda/anaconda-package-data/issues/64 - Example strings at: https://github.com/anaconda/anaconda-package-data/issues/64#issuecomment-3516511486 - (LW) Can we separate visualizations from API? - Anaconda can deliver API - Not every visualization. Users would need to build these - (LW) Would like to make Anaconda's infrastructure more transparent - (JRG) Add back documentation for Anaconda.org - (LW) Also working on fixing some wonky functionality (public & private package?) - (LW) Where to ask for feedback= Zulip? - JRG: Zulip and meetings - CHL: Add to agenda and claim "special session" - JRG: To add PRs so meeting notes are published earlier in the week - [X] (LW) Short update on cloudflare infrastructure issues - There have been a few issues with Cloudflare - However these look like mysterious Cloudflare issues to outside users - Still trying to understand the Cloudflare issue - Please ping me when you see Cloudflare issues - (JRG) How frequent should status update alarms be? - Threshholds are ~40mins. Should these be more or less - (LW) How do you feel about that? - (JRG) Remember 1hr CDN syncs. Then this got faster ([2022 June minutes claimed 5-7 mins!](https://conda-forge.org/community/minutes/2022-06-29/#your-new-agenda-items)) and then degrade - (JK) Agree with Jaime - Would love to see quicker CDN sync - Also better APIs into visibility helps (thanks Lilly!) - Lastly would add post-mortem writeups would help when communicating with other stakeholders - (CHL) Would like to get rid of CDN syncs as a special case and make the system CDN-ready in general. - (JK) Would sharded repodata help with this? - CHL: Not sure. - (WV): Prefix.dev also uses Cloudflare (and was affected by the outages), but setup uses a different approach with no CDN sync delays. A 5+ min delay is unacceptable. - (JK) If this could be done, this would make it desirable for all channels and would take load off Anaconda.org. - (LW) We'll sync internally on my team to look into the infrastructure and look for improvements - [X] (HV) Use RHEL 10 infrastructure from Rocky instead of Alma (for riscv support) - JRG: https://github.com/conda-forge/linux-sysroot-feedstock/pull/93 - JRG: https://github.com/conda-forge/conda-forge.github.io/issues/1744 - JK: Does Rocky support RISCV and Alma doesn't? - Yes. https://rockylinux.org/news/rockylinux-support-for-riscv - IF: We probably don't need CDTs; would rather not build CDTs unless someone requests them. Death to CDTs. - https://github.com/conda-forge/cdt-builds/issues/89 - JK: Skeptical that CDTs aren't used - IF: In almost all cases, there are CDT alternatives. - IF: We had a discussion about keeping PAM as a CDT, but now no one uses it. - JK: What about mesa? - IF: Since CentOS7, we now have non-CDT alternatives for those packages - JRG: We have already a linter which asks maintainers to migrate away from CDTs - [X] KZ: Pending migrations with ghost "awaiting parents"? e.g. see pyside6 https://conda-forge.org/status/migration/?name=python314t - Looks like feedstock names and packages names are mixed at some point in the code base. Is that a bug? Can feedstock branches block a migration - IF: Yes, they can block. Was the PR opened for the branch? - KZ: Pyside2 still waiting for Python 3.14, python 3.14t not opened yet. Question is whether a feedstock should block or a package _name_ should block. - IF: May be ok to remove abi_migration_branches from pyisde2 and mark the migrations as done. There should be a syntax in the migration file to tell the bots to ignore a certain dependency (parent feedstock). Similar was done for matplotlib and qt for pypy. ### Pushed to next meeting - [ ] (HV) Update on exports, self-exports and conditional dependencies - [ ] (HV) CDTs versus security versus compatibility ### CFEPs - [ ]