| 4 Jul 2026 |
hexa | hydra: no idea | 12:54:41 |
hexa | we kinda regressed the the silent timeout from 2 to 3 hours on linux, because we couldn't set it per builder any more | 12:55:07 |
K900 | Maybe we should just set meta.maxSilent | 12:55:39 |
K900 | It looks like the runner understands it | 12:55:48 |
K900 | (at a glance) | 12:55:50 |
Vladimír Čunát | The very long linking in thunderbird is also not optimal wrt. utilization of big-parallel machines, but these timeouts are a way worse issue. | 12:58:30 |
hexa | linker times went up for firefox, too | 12:59:21 |
K900 | It's doing fatLTO, right? | 12:59:28 |
hexa | no idea | 12:59:54 |
emily | do linking in a separate derivation | 13:00:05 |
emily | surely firefox is ThinLTO? fat feels overkill | 13:00:30 |
hexa | https://github.com/mozilla-firefox/firefox/commit/e413a865c3f97bd9bc494373c470a088509ac293 hrm | 13:02:07 |
hexa | maybe in 155 :D | 13:02:21 |
hexa | https://bugzilla.mozilla.org/show_bug.cgi?id=2045567 | 13:02:49 |
hexa | 154* | 13:03:17 |
hexa | so we'll have a full ESR cycle without it | 13:03:25 |
hexa | Redacted or Malformed Event | 13:03:34 |
Vladimír Čunát | Maybe we could pick it. | 13:04:54 |
Vladimír Čunát | By glancing at the commit it doesn't really look relevant to firefox version. | 13:05:14 |
Vladimír Čunát | * Maybe we could pick the patch. | 13:06:00 |
Vladimír Čunát |
60% improvement on build time on opt-nightly
| 13:07:39 |
emily | FatLTO is very painful, I think for memory usage as well. so it makes sense to me | 13:09:24 |
hexa | we could most certainly try picking it | 13:09:24 |
emily | n.b. "since latest clang version bump" though | 13:09:34 |
Vladimír Čunát | But looking at the bug, it might not be easy. They backed it out a couple times. | 13:09:35 |
emily | if that was 21 → 22, we're still pending that bump ourselves | 13:09:41 |
emily | "Starting with clang 20 we're fine with the defaults" oh, nm. | 13:09:56 |
emily |
Revert "Bug 2045567 - Move back to ThinLTO for Linux builds and use default configuration r=glandium" for causing win shippable timeoutss
🤔
| 13:10:27 |
emily | not only did it cause "win shippable timeoutss", but also "perfetest crashes" | 13:10:52 |
Ihar Hrachyshka | re silenceTimeout being ignored, do we reuse daemon connections for multiple builds? | 13:10:56 |