| 21 Jan 2026 |
 Ihar Hrachyshka | I think nix will always first use remote builders; localhost is least preferred | 14:57:05 |
 Austin Horstman | the problem comes that you set how many builds you'll run at once and if one of them is bringing the whole system down, they all become less effective. i have multiple builders, as well. this is more about not using all available cpu power that brings system to a halt | 14:57:25 |
| Ihar Hrachyshka | there was a bug for that... | 14:57:29 |
| Austin Horstman | i gave up on fighting this. i wanted to prefer local lol | 14:57:40 |
| Ihar Hrachyshka | I want same priority for localhost vs remote. I just want cpus go brrr. | 14:58:09 |
| Austin Horstman | i tried getting creative with setting my local as a self ssh to fit it into priority | 14:58:23 |
| Ihar Hrachyshka | https://github.com/NixOS/nix/issues/2457 I think | 14:58:38 |
 eveeifyeve | However I see a couple issues with that pr is, one it doesn't consider the new nix-commands and two it would need rebasing because it's super outdated. | 14:58:47 |
| Ihar Hrachyshka | self reference is prone to deadlock because of store locks | 14:59:00 |
| Austin Horstman | oh yeah it was a horrible experience lol | 14:59:26 |
| eveeifyeve | * However I see a couple issues with that pr is, one it doesn't consider the new nix-commands (eg. nix build) and two it would need rebasing because it's super outdated. | 14:59:27 |
| Ihar Hrachyshka | I have lots of builders (3 macs, 4+ linux machines, plus community builders) and what I learned is that you have to be very careful about making sure you don't have cycles in your build graph. otherwise machineA asks to build on machineB and machineB asks to build on machineA ... and you are stuck. | 15:00:52 |
| Austin Horstman | i didn't even consider that... i've had issues with builds seemingly stuck and not pinned it down... maybe i shouldn't let everyone access everyone the same.. | 15:02:16 |
| Ihar Hrachyshka | oh for sure you shouldn't. | 15:02:41 |
| eveeifyeve | Any other prs that are related to these below?
- https://github.com/NixOS/nix/pull/11143 (load limit)
- https://github.com/NixOS/nix/issues/2457 (speedFactor)
- And stuff about parallel eval implementation movement
| 15:05:24 |
 Randy Eckenrode | Once Nix is building stuff, eval is done. | 15:08:07 |
 Alyssa Ross | https://github.com/ninja-build/ninja/commit/a5da6f98c72d161f766f44fd7cacdcabaaa045ab | 15:22:18 |
| Randy Eckenrode | I had no idea it finally got merged. | 15:23:07 |
 WeetHet | How does thin lto compare to full? | 15:23:26 |
| WeetHet | IIRC it wasn’t much worse when I tested rust | 15:23:44 |
| WeetHet | * | 15:23:58 |
| Randy Eckenrode | Thin LTO was still too slow with Swift. | 15:24:19 |
| WeetHet | Damn, I thought ThinLTO was supposed to be near no lto speed | 15:27:47 |
| WeetHet | But I was mainly thinking about firefox | 15:28:18 |
| WeetHet | They have full LTO enabled for some reason | 15:28:49 |
| Randy Eckenrode | I think it makes different trade-offs, which makes it amenable to parallelization. | 15:28:56 |
| Randy Eckenrode | * | 15:29:05 |
| Randy Eckenrode | * | 15:29:10 |
| Randy Eckenrode | But I guess there can be pathological cases. | 15:29:24 |
| Randy Eckenrode | Oh, only client support? The PR for server support is https://github.com/ninja-build/ninja/pull/2634. | 15:54:14 |
