| 4 Aug 2025 |
 Marc Jakobi | Has anyone else noticed a significant increase in memory when compiling Haskell packages with ghc 9.8.4 compared to ghc 9.6.6? I haven't been able to pinpoint it yet and I can't reproduce it outside of nix builds. | 09:45:42 |
 ners | Does this happen on the same nixpkgs pin, e.g. nix build nixpkgs#haskell.packages.ghc984.rhine vs nix build nixpkgs#haskell.packages.ghc966.rhine? | 10:09:07 |
| Marc Jakobi | In reply to @ners:nixos.dev
Does this happen on the same nixpkgs pin, e.g. nix build nixpkgs#haskell.packages.ghc984.rhine vs nix build nixpkgs#haskell.packages.ghc966.rhine? It doesn't even happen for all packages. I'm trying to build the same package with ghc 9.6, but it's failing to build singletons-th (or singletons-th_3_5), which is one of the dependencies - which kind of adds to my suspicion that this may be a template-haskell regression. | 14:28:51 |
| Marc Jakobi | Yep, no matter how I try to override or jailbreak packages, I can't seem escape dependency hell if I use another ghc version. | 15:13:23 |
 teo (they/he) | Are you able to reproduce this outside of nix? Might be good to open a GHC ticket if you can confirm it's a GHC issue | 15:31:08 |
| teo (they/he) | Oh you already said you couldn't reproduce it outside of nix builds | 15:34:38 |
 sterni | teo (they/he): I mean what you could put on your GHC review pile is my never reviewed MR :p https://gitlab.haskell.org/ghc/ghc/-/merge_requests/13932 | 16:30:06 |
| teo (they/he) | Will do! Feel free to tag me in the future too. It's a real shame that that slipped through the cracks | 16:31:20 |
| teo (they/he) | I hadn't realized you were already trying to build a stage2 compiler for these configurations. I know that isn't supported in hadrian in general and that's being fixed in https://gitlab.haskell.org/ghc/ghc/-/merge_requests/11444. But maybe it would be supported in this case and something is just going wrong with this flag. I'll look into it and see if there's an easy fix | 16:33:24 |
| sterni | I was convinced it worked, but when I checked the log I saw that only ghc the library is built in _build/stage1/compiler just before it finishes. Checking ghc --info revealed that it's indeed just Stage1 :( | 16:48:18 |
| teo (they/he) | Ah annoying! | 16:56:45 |
 oak 🏳️🌈♥️ | Is it possible to ignore specific testcases when packaging a Haskell package, instead of disabling all the tests with dontCheck? With overrideCabal? | 19:03:10 |
| oak 🏳️🌈♥️ | Found a solution, modify testFlags through overrideCabal | 21:09:34 |
| 5 Aug 2025 |
| Marc Jakobi | I think I've narrowed the high memory consumption down to the use of inline-python's marshalling functions (fromPy, ...).
Will investigate further... | 06:59:22 |
| Marc Jakobi | It's not inline-python either. But I can work around it by disabling library profiling, which probably explains why I wasn't able to reproduce it outside of nix. | 09:48:07 |
|  Fargate joined the room. | 12:44:58 |
|  @fargate-official:matrix.org left the room. | 12:45:33 |
| Marc Jakobi | It's a combination of library profiling + template haskell in large modules (500 LOC). Please don't ask me why we have modules with ~500 LOC 🙈 | 13:39:15 |
 magic_rb | 500 loc is not that much? Id think | 13:39:53 |
| magic_rb | If you said 5000loc then sure | 13:40:00 |
| Marc Jakobi | I don't know how much the TH is adding to those lines | 13:44:38 |
| magic_rb | Ah right, lots of TH | 13:44:56 |
| teo (they/he) | You can use https://downloads.haskell.org/ghc/latest/docs/users_guide/debugging.html#ghc-flag-ddump-splices if you want to see the stuff TH outputs | 14:07:00 |
| sterni | you should also probably try to figure out whether GHC or the linker is the problem | 14:44:05 |
| 6 Aug 2025 |
|  Mikel Pérez changed their profile picture. | 02:53:36 |
| 7 Aug 2025 |
|  J. Ryan Stinnett joined the room. | 13:34:41 |
| 8 Aug 2025 |
|  0x4A6F joined the room. | 06:54:24 |
| sterni | teo (they/he): is there an issue alreday for the issue that finalStage = Stage2 is silently ignored? | 10:06:06 |
