| 8 Jul 2026 |
 hexa | which probably average ~4k/hr, so 96k/d, right now | 14:58:03 |
| hexa | Redacted or Malformed Event | 14:58:08 |
 r-burns | Specifically most interested in the time to build the entire package set, through and beyond stdenv out to all leaf pkgs, but just up to stdenv, or e.g. some of the more minimal jobsets, would also be good to know. I can do a local stdenv build, but I'd still be a bit curious how that compares to the current builders' hardware. | 14:58:07 |
| hexa | i'd say 3-4 days is realistic, if we don't run into regressions | 14:59:15 |
| hexa | I don't currently have a good picture of a full nixos test rebuild cycle on nixos/unstable with the new queue-runner | 15:00:06 |
| hexa | but as staging-next just merged we're going to see a reintegration of jobs and nixos tests will get rebuilt | 15:00:23 |
| r-burns | And I suppose the calculus there changes a bit near the end of the cycle, if a lot of those 5k straggler queued jobs are mega builds (like chromium or pytorch, perhaps) that are yet-to-complete in part because of how long they take to build, and are similarly expected to have slower-than-typical step completion rates? | 15:01:07 |
| hexa | correct | 15:01:25 |
| hexa | • Updated input 'nixpkgs':
'github:NixOS/nixpkgs/28d4faad49f8095ec21ebf0454aa8e9c7d7d0af9' (2026-07-05)
→ 'https://hydra.nixos.org/build/336706679/download/2/nixos-26.11pre1031291.1e94a0307f54.tar.zst?narHash=sha256-nh%2BVJ8A5Fiet8%2BXpWqpoN%2B0714dFpyDac1FIAPcRc0Q%3D' (1980-01-01)
| 15:02:54 |
| hexa | https://hydra.nixos.org/jobset/nixos/unstable | 15:54:50 |
| hexa | checking matching jobs for nixos/unstable is bloody fast | 15:55:02 |
| hexa | what took us a day or two before happens in minutes now | 15:55:22 |
 vcunat | That was because nixos.* came before nixpkgs.*, because lexicographically I guess. | 15:56:56 |
| vcunat | Well, mainly because we have a better queue-runner design now. | 15:57:50 |
| hexa | looks like we'll have to build 11.6k jobs on nixos/unstable | 15:58:48 |
| hexa | 3.8k of which are nixos.tests | 15:59:12 |
| vcunat | unstable-small advanced past the staging-next merge. | 16:11:41 |
 leona | but it seems like 11.6k - 7k failures of staging-next? as they are not yet marked as failed in nixos:unstable | 16:23:00 |
| hexa | plausible | 16:23:21 |
| hexa | so ~83% of rebuilds are nixos tests | 16:23:56 |
| hexa | Redacted or Malformed Event | 16:24:12 |
| hexa | Grimmauld (any/all) python313.pkgs.pgmpy still ignores core count limits and I see gomp barrier function call in perf top from 20% test progress | 22:57:57 |
| hexa | and I can't find OMP_NUM_THREADS in env in checkPhase | 23:01:47 |
 Grimmauld (any/all) | Okay will look | 23:09:54 |
| Grimmauld (any/all) | Current master? | 23:10:18 |
| hexa | yep | 23:12:23 |
| Grimmauld (any/all) | Do we have documentation for makePythonHook ? | 23:22:07 |
| hexa | probably not | 23:22:21 |
| Grimmauld (any/all) | diff --git a/pkgs/development/python-modules/pgmpy/default.nix b/pkgs/development/python-modules/pgmpy/default.nix
index cd30e613f033..d6f9b30b4640 100644
--- a/pkgs/development/python-modules/pgmpy/default.nix
+++ b/pkgs/development/python-modules/pgmpy/default.nix
@@ -94,6 +94,12 @@ buildPythonPackage (finalAttrs: {
writableTmpDirAsHomeHook
];
+ installCheckPhase = ''
+ runHook preInstallCheck
+ env
+ exit 1
+ '';
+
pythonImportsCheck = [ "pgmpy" ];
meta = {
This correctly lists OMP_NUM_THREADS=1. Catching it during testing in /proc is slightly more annoying, but will check that too. If that ends up not having the env var, something weird is going on in pytest check | 23:24:05 |
| hexa | hm, I see it now | 23:28:46 |
