| 26 Jun 2026 |
 emily | oh right | 16:42:05 |
| emily | iirc Nix and Lix never actually set it to 0 in practice in the build sandbox these days? | 16:42:13 |
| emily | since it gets autodetected | 16:42:18 |
 hexa | https://github.com/OpenMathLib/OpenBLAS/pull/5808 🤔 | 16:42:28 |
| hexa | Redacted or Malformed Event | 16:42:31 |
| hexa | oh, possible | 16:42:40 |
| emily | but the things that divide it need to take into account that rounding down to 0 | 16:43:14 |
| hexa | I hope we're killing all of that | 16:43:34 |
 Grimmauld (any/all) | https://github.com/NixOS/nixpkgs/pull/535725
have fun | 16:44:19 |
| emily | we should probably default to $NIX_BUILD_CORES? | 16:44:52 |
| hexa | hell no | 16:45:01 |
| hexa | OMP threads are on top of NIX_BUILD_CORES already | 16:45:15 |
| Grimmauld (any/all) | you'd be multiplying^^ | 16:45:31 |
| emily | oh | 16:45:27 |
| emily | because Python spawns one test per thread, or…? | 16:45:32 |
| Grimmauld (any/all) | fixing to 1 is much much more reasonable | 16:45:40 |
| Grimmauld (any/all) | yes | 16:45:44 |
| Grimmauld (any/all) | pytest x omp = pain | 16:45:58 |
| emily | great way to exercise a concurrency framework | 16:46:09 |
| hexa | so shit squared | 16:46:58 |
| hexa | and builders sad | 16:47:13 |
| Grimmauld (any/all) | Don't forget the nondeterminisms from doing stuff in parallel | 16:47:16 |
| Grimmauld (any/all) | imo unless a single build takes literally ages, letting hydra scheduling and NIXBUILDCORES do the magic is just objectively better than trying to be smart here | 16:47:54 |
| Grimmauld (any/all) | imo unless a single build takes literally ages, letting hydra scheduling and NIX_BUILD_CORES do the magic is just objectively better than trying to be smart here | 16:48:04 |
| emily | sure yeah I just assumed we weren't stacking concurrency | 16:49:28 |
| Grimmauld (any/all) | We may want to propagate the hook on torch too. Torch gets it indirectly from numpy through some other package currently, so it does have it, just less obviously so. | 16:51:41 |
| emily | hexa: did we want to get Hydra building stdenv in advance of the cycle or is there not capacity for that yet anyway? | 17:15:00 |
| hexa | aarch64-darwin is still behind :/ | 17:15:23 |
| hexa | even building 9k old staging-next jobs | 17:15:34 |
| emily | cancel 'em? | 17:15:45 |
