| 7 May 2025 |
 connor (burnt/out) (UTC-8) | God that’s so cool | 13:05:35 |
| connor (burnt/out) (UTC-8) | SomeoneSerge (UTC+U[-12,12]): I've got another one for you https://github.com/NixOS/nixpkgs/pull/404973 | 16:41:25 |
| connor (burnt/out) (UTC-8) | As an added benefit, one could float out imports so that only happens once per nixpkgs instantiation | 16:42:23 |
 SomeoneSerge (back on matrix) | If you're not using aderall idk what you're using xD | 16:57:02 |
| SomeoneSerge (back on matrix) | So fast | 16:57:06 |
| connor (burnt/out) (UTC-8) | Well, that one's sorta broken at the moment because cuda-library-samples doesn't mark derivations as broken if they rely on something unavailable in the current version of the package set | 17:12:37 |
| connor (burnt/out) (UTC-8) | I mean yeah for attention so I can survive my commute; the CUDA stuff is mostly because I think about it every waking moment. (Although I also usually have nightmares about it, too...) | 17:51:11 |
| connor (burnt/out) (UTC-8) | Okay, should be good to go assuming CI passes | 18:24:37 |
| connor (burnt/out) (UTC-8) | To recap, if anyone has spare cycles to review:
- https://github.com/NixOS/nixpkgs/pull/385960 -- very large, lots of bash, but lots of unit tests
- https://github.com/NixOS/nixpkgs/pull/404686 -- very small, no rebuilds
- https://github.com/NixOS/nixpkgs/pull/404973 -- ~900 line diff, no rebuilds outside of CUDA packages, largely just re-organizing to minimize the diff when landing setup hook changes
The setup hook changes (working on a PR) will need all three merged.
| 18:28:26 |
| SomeoneSerge (back on matrix) | Let's start by merging packagesFromDirectoryRecursive | 18:39:44 |
| SomeoneSerge (back on matrix) | Hmm can't we somehow adopt ci/eval and .github/workflows/eval.yml to evaluate, say, release-cuda.nix for PRs labeled "cuda" | 18:47:10 |
| SomeoneSerge (back on matrix) | (with builds and substitutions disabled) | 18:47:28 |
| SomeoneSerge (back on matrix) | connor (he/him) (UTC-7): feel like scratching an itch... git mv pkgs/development/cuda{-modules,} | 19:02:48 |
| connor (burnt/out) (UTC-8) | I kept cuda-modules just because it seems everything else in development follows that naming scheme, but no real reason to keep it that way, I suppose | 19:38:14 |
| SomeoneSerge (back on matrix) |
everything else in development follows that naming scheme
No idea why, but whatever
| 19:39:25 |
| SomeoneSerge (back on matrix) | connor (he/him) (UTC-7): your shims are basically fixups for manifests aren't they? | 20:03:39 |
| connor (burnt/out) (UTC-8) | Yep! | 20:04:17 |
| SomeoneSerge (back on matrix) | Now I want (aside from the stuff already listed in TODO): snapshots of nixpkgs attribute trees AND snapshots of callable arguments' functionArgs | 20:42:24 |
| connor (burnt/out) (UTC-8) | There's a builtin for functionArgs which is probably a little more in-depth than the __functionArgs I used: https://github.com/NixOS/nix/blob/1e822bd4149a8bce1da81ee2ad9404986b07914c/src/libexpr/primops.cc#L3135 | 20:47:12 |
| SomeoneSerge (back on matrix) | There's lib.functionArgs too | 20:48:11 |
|  kaya 𖤐 changed their profile picture. | 21:21:02 |
|  Tristan Ross joined the room. | 21:46:04 |
| Tristan Ross | I have joined here because I may do CUDA stuff with my Ampere Altra Max M128-26 and RTX 5070. | 21:46:43 |
 hacker1024 | Ooh is that SBSA then? Good luck lol | 21:47:47 |
| Tristan Ross | In reply to @hacker1024:matrix.org
Ooh is that SBSA then? Good luck lol Yep, SBSA and ACPI | 21:49:07 |
| SomeoneSerge (back on matrix) | Isn't this going to be like literally the first time anybody actually uses sbsa version of cuda in Nixpkgs? | 21:49:09 |
| Tristan Ross | In reply to @ss:someonex.net
Isn't this going to be like literally the first time anybody actually uses sbsa version of cuda in Nixpkgs? Maybe | 21:49:28 |
| Tristan Ross | I built a bunch of CUDA stuff that @[connor (he/him) (UTC-7)] has on Saturday | 21:50:07 |
| Tristan Ross | There's a lot of things broken lol. | 21:50:17 |
| connor (burnt/out) (UTC-8) | FWIW, I was able to build and train an MLP mixer model this weekend on a GH200 through Lambda Labs using my repo, so proper support is at least in progress lol | 22:04:12 |
