|
| 1 | +--- |
| 2 | +tags: [meeting-notes] |
| 3 | +title: '2025-10-29' |
| 4 | +--- |
| 5 | +# conda-forge core meeting 2025-10-29 |
| 6 | + |
| 7 | +Add new agenda items under the `Your __new__() agenda items` heading |
| 8 | + |
| 9 | +- [Zoom link](https://zoom.us/j/9138593505?pwd=SWh3dE1IK05LV01Qa0FJZ1ZpMzJLZz09) |
| 10 | +- [What time is the meeting in my time zone](https://dateful.com/convert/utc?t=5pm) |
| 11 | +- [Previous meetings](https://conda-forge.org/community/minutes/) |
| 12 | + |
| 13 | +## Attendees |
| 14 | + |
| 15 | +| Name | Initials | GitHub ID | Affiliation | |
| 16 | +| ----------------------- | -------- | --------------- | --------------------------- | |
| 17 | +| Cheng H. Lee | CHL | chenghlee | Anaconda/cf | |
| 18 | +| Jaime Rodríguez-Guerra | JRG | jaimergp | Quansight/cf | |
| 19 | +| Mark Allen | MHA | markhallen | GitHub/Dependabot | |
| 20 | +| Sylvain Corlay | SC | QuantStack | | |
| 21 | +| Rob Aiken | RA | robaiken | Github/Dependabot | |
| 22 | +| Daniel Ching | DJC | carterbox | NVIDIA/cf | |
| 23 | +| | | | | |
| 24 | +| | | | | |
| 25 | +| | | | | |
| 26 | + |
| 27 | +X people total |
| 28 | + |
| 29 | +### Standing items |
| 30 | + |
| 31 | +- [ ] |
| 32 | + |
| 33 | +### From previous meeting(s) |
| 34 | + |
| 35 | +- [ ] |
| 36 | + |
| 37 | +### Active votes |
| 38 | + |
| 39 | +- [ ] |
| 40 | + |
| 41 | +### Your __new__() agenda items |
| 42 | + |
| 43 | +- [x] CHL/MHA/RA: GitHub/Dependabot team |
| 44 | + - (MHA) Have a plan to version updates using dependabot, independent of vulnerability feed |
| 45 | + - Queries the conda API for package versions |
| 46 | + - How to gather & provide CVE/vulnerability data for conda-forge packages? |
| 47 | + - (RA) Get information from GH Advisory database; do have support for Python security advisories |
| 48 | + - (RA) Unsure of how to add new ecosystem to advisory database |
| 49 | + - (MHA) Dependabot running within GHA runner; not feasible because of large download size |
| 50 | + - Could we consider tapping into the PyPI data feed and find matches in conda-forge? |
| 51 | + - (JRG) Add upstream PURLs into recipes; current name mapping is heuristic and subject to error |
| 52 | + - (JRG) complexities: not all versions available; multi-output packages; package renames (need to annotate which versions we switched) |
| 53 | + - (SC) Been looking into integrating conda-forge into repology. |
| 54 | + - XREF: https://conda-forge.org/community/minutes/2025-06-11/ |
| 55 | + - (JRG) Need to be careful about burdening volunteer maintainers |
| 56 | + - (CHL) Will invite the GitHub/Dependabot team to Zulip; create GitHub issue |
| 57 | +- [X] JRG: `zlib` -> `zlib-ng` migration: https://github.com/conda-forge/zlib-ng-feedstock/issues/10 |
| 58 | + - CPython 3.14 upstream ships zlibg-ng for Windows, with compatiblity mode; Pillow, various Linux distros switched to zlib-ng |
| 59 | + - Currently not building compat mode on c-f because it would create conflicts with existing `zlib` |
| 60 | + - (DJC) Continue to support non-compat mode and ask maintainers to explicitly enable zlib-ng |
| 61 | + - Could make compat-mode a `zlib` variant, using `blas` as a reference model |
| 62 | + - (CHL) Does zlib-ng support dynamic dispatch for vector instructions? If not, could break on older systems. |
| 63 | +- [X] DJC: Tegra support (demanded in robotics) |
| 64 | + - CTK 12.9 packages for Tegra sm87,sm101 devices are now live |
| 65 | + - Third-party packages may start building for Tegra |
| 66 | + - arm-variant not required for CUDA 13 (newer devices are SBSA), but we're not ready yet. |
| 67 | + - Once CUDA 12 is dropped, arm-variant can be retired. (No other packages are known to use `arm-variant`.) |
| 68 | +- [x] DJC: nvidia-virtual-packages |
| 69 | + - A conda virtual package plugin which detects the minimum CUDA architecture available on the system |
| 70 | + - Source: https://github.com/NVIDIA/nvidia-virtual-packages |
| 71 | + - RFC: https://github.com/conda-forge/conda-forge.github.io/issues/2623 |
| 72 | + - Motivation: Deep learning packages often have minimum supported CUDA archs which don't align with the CTK |
| 73 | + - https://github.com/conda-forge/cudnn-feedstock/issues/124 |
| 74 | + - https://github.com/conda-forge/flash-attn-feedstock/blob/b6e3742a7343268a33a285c593753fd49b46d268/recipe/meta.yaml#L23 |
| 75 | + - Motivation: Would be possible to break large binaries into smaller variants along CUDA arch |
| 76 | + - CHL: Apply for conda incubator |
| 77 | + - CHL: CUDA virtual packages should all live in the same place; though we can decide later exactly where. |
| 78 | + - JRG: There is a draft CEP about standard names for virtual packages |
| 79 | + - How to address bootstrap problem |
| 80 | + - conda-forge and Anaconda could just make `conda` depend on this/these plugins |
| 81 | + - pixi doesn't have a plug-in system, but could integrate virtual packages directly into pixi |
| 82 | +- [x] CHL: continued support for Windows 10? |
| 83 | + - [Regular security support](https://endoflife.date/windows) ended on 14-Oct-2025 |
| 84 | + - Took a quick look for `main` and `conda-forge` download data; as of 15-Oct, 25%-ish of downloads from `conda ... Windows/*` user agents are still on Window 10. Roughly matches what [Firefox reports](https://data.firefox.com/dashboard/hardware#operating-system-metric-overview-1) |
| 85 | + - Will open an issue on conda-forge.github.io to further discuss |
| 86 | +- [X] WV: Huge refactor of the `cache` output in rattler-build. More versatile, experiments with the staging output idea. |
| 87 | + |
| 88 | +### Pushed to next meeting |
| 89 | + |
| 90 | +- [ ] |
| 91 | + |
| 92 | +### CFEPs |
| 93 | + |
| 94 | +- [ ] |
0 commit comments