| 10 Mar 2026 |
 Lun | Might prefer LLVM_TOOL_LLVM_DRIVER_BUILD, iirc there was some cursed issue with multiple differing llvms in the same process with shared libs when you got further out to things like torch. | 19:16:54 |
| Lun | that only saves about 500MiB :c | 19:28:40 |
 Sergei Zimmerman (xokdvium) | for x86-linux the diff is 1.8G -> 200M | 19:47:17 |
| Lun | * that (LLVM_TOOL_LLVM_DRIVER_BUILD) only saves about 500MiB :c | 19:49:06 |
| Sergei Zimmerman (xokdvium) | Ah yeah, but having a static and a shared library in the same process is even more cursed. | 19:51:16 |
| Lun | If you put something up give me a ping so I can test some leaf packages with highest chance of hitting the cursed | 19:53:29 |
 hexa | The answer is always: reluctantly | 19:55:31 |
| hexa | And most of the time for the graphical ISO | 19:55:52 |
| hexa | We've not established precedent for AI/ML packages and that's a good thing | 19:56:21 |
 Gaétan Lepage | Got it! Just to be sure I get it right, when some output is too big, the thing is just... not cached, right? | 22:22:56 |
| hexa | yep, it means it does not get pushed to s3 | 23:10:37 |
| 11 Mar 2026 |
 yorik.sar | Please retrigger https://hydra.nixos.org/build/323649912 - configure failed with segfault again. It looks like we'll never have both node 20 and 24 built in one eval... | 13:41:29 |
| yorik.sar | (the same package on x86_64-darwin went much further, but died with No space left on device) | 13:42:04 |
| hexa | rebooting the two macs with jobs stuck at sending inputs | 13:46:14 |
|  Theuni changed their display name from Theuni to Christian Theune. | 14:12:14 |
| 12 Mar 2026 |
| Theuni changed their display name from Christian Theune to Theuni. | 07:18:17 |
| 13 Mar 2026 |
| yorik.sar | Please retrigger https://hydra.nixos.org/build/323621817 - it was failing because of the dependencies that got rebuilt successfully earlier. | 12:37:00 |
| yorik.sar | Thanks a lot! Finally we have all dependencies and github-runner itself in the cache. | 12:59:50 |
|  @codayon:matrix.org joined the room. | 15:21:18 |
| @codayon:matrix.org | I have a Gigabit internet but the download speed I'm getting while updating are not great and very inconsistent. Here is the results of this diagnostics script I got from cache.nixos.org. | 15:32:32 |
| @codayon:matrix.org | * I have a Gigabit internet but the download speed I'm getting while updating are not great and very inconsistent. Here is the results of this diagnostics script I got from here. | 15:42:57 |
| hexa | ends early | 23:12:50 |
| hexa | with no mtr result | 23:12:57 |
| hexa | what is peculiar is that your ipv4 latency is much higher than your ipv6 latency | 23:13:19 |
| 14 Mar 2026 |
| @codayon:matrix.org | Is there anything I can do from my end? | 06:32:09 |
|  amadaluzia -> 4d2.org changed their display name from amadaluzia to amadaluzia[uorg]. | 18:48:48 |
|  amadaluzia joined the room. | 19:29:29 |
| amadaluzia -> 4d2.org changed their display name from amadaluzia[uorg] to amadaluzia -> 4d2.org. | 21:22:28 |
 tomberek | Scaleway is interested in sponsoring in the form of providing hardware and compute. What would the infra team be most interested in from them? They mentioned providing builders from idle availability zones and Apple Silicon. Also interest in RISC-V and a cache for it. Don't limit to just that though, what would have the greatest impact? | 21:37:15 |
| hexa | I guess something of each would be interesting. We do have an Apple bottleneck, but with x86_64-darwin loosing support after 26.05 we shouldn't optimize too much on that. | 22:11:29 |
