| 8 Oct 2025 |
 K900 | Anyone want a funni | 21:03:54 |
| K900 | https://github.com/NixOS/nixpkgs/actions/runs/18355504442/job/52286419787 | 21:03:55 |
| K900 | opentbb segfaults on aarch64 github runners | 21:04:03 |
| K900 | * onetbb segfaults on aarch64 github runners | 21:04:54 |
| K900 | OK it also reproduces on not github runners | 21:26:50 |
| K900 | So it sounds like it's real | 21:26:57 |
| K900 | There's one change to onetbb between master and next | 21:28:04 |
| K900 | https://github.com/nixOS/nixpkgs/commit/666a42cdf57e7d64d108b9ae96e013cebf3f6c02 | 21:28:15 |
| K900 | @jopejoe1 joooooooo | 21:28:18 |
| K900 | OK reverting that commit fixes it | 21:35:49 |
| K900 | So we have a thing | 21:35:50 |
 dramforever | just disable those tests, i think | 21:37:30 |
| dramforever | they're trying to access sysfs | 21:37:37 |
| K900 | One of them segfaults | 21:37:55 |
| K900 | I don't think we should disable that | 21:38:02 |
| K900 | I mean, we should maybe disable whatever is segfaulting, but we should not just disable the test and pretend it's fine | 21:38:17 |
| dramforever | wait what the hell | 21:38:46 |
| dramforever | yeah i think it's just accessing the front() of an empty std::vector
auto constraints = tbb::task_arena::constraints{}
.set_numa_id(tbb::info::numa_nodes().front())
.set_core_type(tbb::info::core_types().front())
.set_max_threads_per_core(1);
| 21:42:24 |
| dramforever | which understandably segfaults | 21:42:29 |
| dramforever | https://github.com/uxlfoundation/oneTBB/blob/v2022.2.0/test/tbb/test_arena_constraints.cpp#L102-L105 | 21:43:15 |
| K900 | Great | 21:43:38 |
| K900 | And those only fail on aarch64 because...? | 21:43:50 |
| K900 | Let me guess, Intel has a special code path for x86 | 21:43:57 |
| dramforever | what do you get when you run hwloc-ls from hwloc in the nix builder on x86_64-linux | 21:46:09 |
| dramforever | ... wait i just did that | 21:47:31 |
| dramforever | you get the same sysfs error | 21:47:36 |
| dramforever | what on earth | 21:49:04 |
| dramforever | okay hwloc has magic on x86 to read cache info and stuff from cpuid | 21:53:07 |
| dramforever | nothing like that on arm64 or loongarch | 21:53:30 |
| K900 | Jajajajjaajjajaajajajanakaka | 21:54:08 |
