| 7 Mar 2026 |
 Yureka (she/her) | https://github.com/NixOS/nixpkgs/pull/450106 | 11:36:38 |
 K900 | So do we have to bisect into openblas now | 11:37:11 |
| Yureka (she/her) | or just revert and do that later? | 11:37:51 |
| K900 | Depends on how big the bisect is I guess | 11:38:27 |
| K900 | Do you want to try it? | 11:38:47 |
| Yureka (she/her) | big | 11:39:09 |
| Yureka (she/her) | 558 commits | 11:39:17 |
| K900 | And how long does it take to fail? | 11:39:28 |
| Yureka (she/her) | not very long, but looking at the commits between openblas .30 ... .31, it's likely that broke and was fixed more than once in between | 11:41:37 |
| K900 | Ugh | 11:42:13 |
| Yureka (she/her) | just the fact that this package is so irreproducible in the first place, and they added full new code paths for the neoverse platform in between | 11:42:52 |
| Yureka (she/her) | Redacted or Malformed Event | 11:42:55 |
| Yureka (she/her) | 
Download foo.png | 11:43:32 |
| Yureka (she/her) | it likely broke at a different point on the Ampere vs the M1 Pro | 11:43:41 |
| Yureka (she/her) | and the dynamicArch ? null suggests the people who wrote this package initially fully intended the non-reproducibility | 11:44:25 |
 dramforever | does openblas build different objects with different -march | 11:44:33 |
| dramforever | and then uh, does something | 11:44:41 |
| Yureka (she/her) | it builds different objects depending on what /proc/cpuinfo you have during the build | 11:44:52 |
| dramforever | if dynamicArch means runtime that's good | 11:44:58 |
| dramforever | uh oh | 11:45:04 |
| K900 | I think dynamicArch does mean runtime | 11:45:06 |
| Yureka (she/her) | hmm | 11:45:17 |
| K900 | But it ALSO tries to be smart somehow | 11:45:15 |
| Yureka (she/her) | so maybe both | 11:45:19 |
| Yureka (she/her) | why would my different build hosts fail with different errors then? | 11:45:39 |
| dramforever | but it is passing different -march to different objects | 11:45:39 |
| dramforever | so something is going on | 11:45:43 |
| K900 | That postPatch line is also hella sus | 11:46:00 |
| K900 | In the nixpkgs PR change | 11:46:06 |
| K900 | In reply to @yuka:yuka.dev
why would my different build hosts fail with different errors then? Could just be ordering | 11:46:10 |
