| 21 Sep 2025 |
emily | I'm tired but can try to look at regressions in a few hours. | 14:06:56 |
emily | not sure building my systems will be much use | 14:07:10 |
emily | since I already built Darwin configs earlier | 14:07:16 |
ghpzin | llvm_18 is not getting dropped any time soon, right ? | 14:08:32 |
K900 | I just want to get as much as possible built first | 14:08:40 |
K900 | And then start looking at the failures | 14:08:55 |
emily | I plan to drop it in 26.05. | 14:11:02 |
emily | why? | 14:11:08 |
ghpzin | Noticed it would need patches for gcc15 | 14:11:26 |
emily | I guess the question is whether we GCC 15 or drop LLVM 18 first 😅 | 14:11:57 |
emily | depends on whether we do it this release, which Fabián Heredia seemed to lean against | 14:12:04 |
K900 | Uhh | 14:12:29 |
emily | (I'm in favour of doing it personally but admittedly this cycle might be enough hell for several months) | 14:12:26 |
K900 | Postgres JIT is busted | 14:12:32 |
K900 | By LLVM 21 I think | 14:12:34 |
emily | cc Wolfgang Walther | 14:12:50 |
Wolfgang Walther | I remember reading something about that upstream, lemme check. | 14:12:56 |
Wolfgang Walther | https://www.postgresql.org/message-id/flat/d25e6e4a-d1b4-84d3-2f8a-6c45b975f53d%40applied-asynchrony.com
Haven't fully read through that, yet. There are some patches there. | 14:15:27 |
Wolfgang Walther | Might be better to not use LLVM 21 until the next minor releases here. | 14:16:26 |
K900 | Yeah I don't think we should risk it | 14:16:37 |
K900 | That thread looks very wacky | 14:16:55 |
emily | (fwiw though I am always fine backporting upstream patches like this) | 14:19:33 |
emily | (if they are not a huge pain at least, like a clean fetchpatch is totally fine by me) | 14:19:48 |
emily | let's pin LLVM 20 for Postgres. | 14:19:58 |
Wolfgang Walther | In https://github.com/NixOS/nixpkgs/pull/444981 | 14:33:19 |
emily | the lib.warnIf definitely shouldn't cause issues? (trauma from that Haskell LLVM warning) | 14:33:57 |
emily | maybe lib.versionAtLeast version "17.7"… just in case we skip a version (unless that's inconceivable) | 14:34:13 |
Wolfgang Walther | Only when we update to the next minor release. | 14:34:15 |
Wolfgang Walther | Nah, then I need to check it's less than major 18, too. | 14:34:24 |
Wolfgang Walther | We 100% won't skip a minor version of PostgreSQL. | 14:34:34 |