| 21 Oct 2021 |
 sterni | nothing in nixpkgs needs GHC 8.8.4 as far as I am aware, but it's bit hard to tell whether there are a lot of downstream users | 10:28:17 |
| sterni | as long as we have GHC 8.8.4 we sadly need to keep GHC 8.6.5 (binary) around since that is required for bootstrapping | 10:28:38 |
 Alyssa Ross | What about 8.6? | 10:28:40 |
| Alyssa Ross | oh, right | 10:28:44 |
| sterni | reverse bootstrapping with 8.10.2 is possible since we do that for aarch64 | 10:29:40 |
| Alyssa Ross | oh cool | 10:30:10 |
| sterni | but it is not supported and reverse bootstrapping with 8.10.7 blows up in weird ways, so I'm a bit scared | 10:30:25 |
| sterni | but if 8.6.5 is the only blocker for removing an LLVM version that is an option, I'd say | 10:30:44 |
| Alyssa Ross | It looks like there are a couple of other users, so I'm not sure yet | 10:31:26 |
| Alyssa Ross | but it might be possible | 10:31:31 |
| Alyssa Ross | I think we'd want to keep llvmPackages_6 around because poetry2nix can use and stuff, but it would be nice if we could have no in-tree users and stop hydra building it, maybe | 10:32:00 |
| Alyssa Ross | I guess then it'd just bitrot | 10:32:08 |
| Alyssa Ross | toonn: what does your Hydra job actually build? Is it every package? | 10:32:21 |
 toonn | Yeah, stdenv rebuild because of LLVM bump. Building everything to check the impact. | 10:32:48 |
| Alyssa Ross | I see | 10:33:14 |
| Alyssa Ross | so the only way for you to not have to build every LLVM version would be to remove some LLVM versions entirely? | 10:33:29 |
| Alyssa Ross | that's unfortunate | 10:33:32 |
| Alyssa Ross | I think we probably don't want to do that | 10:33:53 |
| Alyssa Ross | I thought you meant all those things were in the bootstrap path, or something | 10:34:12 |
| toonn | No. It's just that sphinx got broken because of the Rust dependency and fixing that entailed two LLVM builds regularly. | 10:35:19 |
| Alyssa Ross | Right, before anything else can build | 10:35:46 |
| sterni | I guess one option would be to remove llvmPackages_6 and all its reverse deps from your jobset | 10:36:58 |
| toonn | No no, things are getting mixed up. | 10:38:21 |
| toonn | The issue was sphinx broke. And that meant building stdenv LLVM *and* LLVM for Rust. | 10:38:52 |
| toonn | That's not a solvable issue without aligning one or the other to the other version : ) | 10:39:13 |
| Alyssa Ross | When you said "Just checked my Hydra job. For x86_64-darwin we end up building LLVM 5, 6, 7, 8, 9, 10, 11, 12 and 13 >.<", was that unexpected? | 10:39:26 |
| toonn | The many LLVM versions just make complete evaluations to test stdenv harder but they're not as big an issue. If you're building every package anyway one more LLVM doesn't seem as much of an issue. | 10:40:06 |
| toonn | I didn't expect that many versions tbh. | 10:40:39 |
| 22 Oct 2021 |
| toonn | #126411 the x86_84-darwin LLVM bump is finally ready for review. | 15:56:18 |
 ris_ | hrm staging-next for darwin is stuck on a git failure https://hydra.nixos.org/build/156106839/nixlog/2 :S | 17:48:29 |
