conda-forge core meeting 2023-11-15
Add new agenda items under the Your __new__() agenda items heading
Attendees
| Name | Initials | GitHub ID | Affiliation |
|---|---|---|---|
| Marcel Bargull | MB | mbargull | Bioconda/cf |
| Bianca Henderson | BH | beeankha | Anaconda |
| Mark Anderson | MAA | markan | Anaconda |
| Marcelo Trevisani | MDT | marcelotrevisani | conda-forge |
| Isuru Fernando | IF | isuruf | Quansight |
| Wolf Vollprecht | WV | wolfv | prefix.dev |
| Dave Clements | DPC | tnabtaf | Anaconda |
| Jaime Rodríguez-Guerra | JRG | jaimergp | Quansight/cf |
| Matthew R Becker | MRB | beckermr | cf |
| John Kirkham | JK | jakirkham | NVIDIA/cf |
| 14 people total |
Standing items
- [ ]
From previous meeting(s)
- (HV) archspec-packages, next steps (feel free to discuss in my absence)
- we have microarch-level package now 🎉
- are we ready/willing to build packages for different arches?
- do we want to set minimal guidelines to avoid CI explosion from feedstocks indiscriminately wanting to build v2,v3,v4 because "it's obviously faster"?
- need to go through and check whether runtime dispatching is available
- how to detect macroarchitecture (x86_64 for example)? This was in
__arch, but is now not there. how should this be included?- Change existing string to include with micro architecture?
- New virtual package?
- Discussion continues in issue
- (JK) m2 recipes
- Isuru needs time.
- (IF) CDT build type thing for m2 (tools).
- (IF) m2w64 packages would be regular feedstocks
- (JK) Windows ARM
- (IF) Call last week with Finn (from Microsoft)
- (IF) ARM-64 windows CI set up.
- (IF) Not the whole things, but progress
- Use ARM64 image using an X86 installer, then use emulation
- (IF) Will also need m2 recipes (as Python needs these to build)
Active votes
- [ ]
Your __new__() agenda items
- (HV) / (WV) Discuss
{{ stdlib("c") }}vs.{{ compiler("c", stlib=...) }}, see here.- (WV)
- Still in favor of one Jinja function. Having 2 makes it messy
- Can try and fix it later if there is demand.
- (IF)
- This would add more technical debt to conda-build (?)
- (WV)
- conda-build is so much technical debt already.
- how much should we worry about it.
- (MB)
- agree with both
- (IF)
- One jinja function would be good, but there isn't a way to do that now.
- (WV)
- (JK) Travis CI update
- Hit issues a week ago with staged recipes because Travis gave us API issues
- Also long term issues with token resets from Travis.
- Had us resync the bots
- GitHub bots couldn' start CI...
- (MB) Did anyone from conda-forge ask for linux-arm?
- (JK) We haven't even discussed it.
- (IF) JRG addded a feature to admin-requests.
- We could stop registering all feedstocks when we add them.
- Could require developers to request them.
- 90% of developers don't really need this.
- (JK) Could maintainers ask for Travis CI support later?
- What's the prcess for that?
- https://github.com/conda-forge/admin-requests/blob/main/grant_access/travis/example.txt
- (JK) Windows CUDA 12
- Have done more testing with cupy - found small bugs which have been fixed.
- Ok to migrate? Yes
- Ok restarting existing migrator and adding Windows? Yes
- https://github.com/conda-forge/conda-forge-pinning-feedstock/pull/5121
- (JK) Results from conda-smithy 3.28.0
- https://github.com/conda-forge/conda-smithy/releases/tag/v3.28.0
- https://github.com/conda-forge/conda-smithy/releases/tag/v3.29.0
- How are things going with the new releases?
- libmamba solver is now the default
- Any issues
- (MRB)
- Seen a few issues
- Don't have the latest version of Boa
- (JRG)
- Seen reports where solver failed to write back because of key-errors
- Related to channels
- PR merged today. Hopefully released this week
- (IF)
- could specify miniforge version
- we use miniforge in all of our CI
- (JRG) Want to separate tooling issues from distribution issues
- (JRG) TL;DR Bumped into a couple issues. Working those out
- (HV) libboost 1.82 migration update & next steps
- almost 200 PRs merged
- long tail of packages that don't build (e.g. have open PRs for old boost migrations)
- ~70% are estimated to be done
- do one last pass for bot errors and unsolved feedstocks, then should be close
- (JK) Custom license discussion
- https://github.com/conda-forge/staged-recipes/pull/24449
- https://github.com/conda-forge/unicorn-binance-websocket-api-feedstock
- Claiming MIT when submitters actually are using a custom license
- How do we cope?
- We can't just eliminate custom licenses.
- (MB) in this specific case we can say you can't lie about licensing.
- They need to fix their metadata.
- "We are feeling ucomfortable with the licensing so not comfortable reviewing it."
Pushed to next meeting
- (JK) Miniforge 23.10
- (JK) NumPy 2.0
- (JK) CUDA Docker images
- (HV) what to do with CDTs for Alma 8
- make checklist with CDTs, for checking whether we can switch each to conda packages?
- (DJC) Policy for CUDA arch targets and pruning CUDA archs
- https://github.com/conda-forge/conda-forge.github.io/issues/1901
- Some packages are too big to build within the 6 hour CI limit while targeting many CUDA architectures
- examples include libmagma, libtorch
- The linked discussion is about which CUDA archs should be targeted when the upstream project does not have defaults and in what order to drop archs in order to complete builds within the 6 hours
- [ ]
CFEPs
- [ ]