| 25 Jan 2023 |
 @rnhmjoj:maxwell.ydns.eu | * wow, that's a lot of haskell libraries. I had forgotten that GHC object files are not reproducible. are there any recent developments about this? | 09:59:05 |
 atemu12 | Perhaps you could ask nh2, they're into GHC stuff | 10:00:03 |
 raboof | I didn't notice much activity in https://github.com/NixOS/nixpkgs/issues/151347 / https://gitlab.haskell.org/ghc/ghc/-/issues/12935 | 10:06:54 |
| raboof | though it seems it doesn't necessarily impact all haskell builds? there's quite some green ones in https://reproducible.archlinux.org/ | 10:09:53 |
| raboof | quick experiment: haskellPackages.void seems reproducible for us as well | 10:10:58 |
| @rnhmjoj:maxwell.ydns.eu | could it be related to parallel builds/optimisation level? | 10:11:21 |
| raboof | yeah single threaded builds seem more reliable but also quite a price to pay (https://github.com/NixOS/nixpkgs/issues/151347#issuecomment-997817696, https://gitlab.haskell.org/ghc/ghc/-/issues/12935#note_373865) | 10:13:26 |
| raboof | perhaps an option for some small/fast modules that are currently unreproducible, but it might be good to put some energy in looking into the upstream ghc problem(s?) as well | 10:21:14 |
| atemu12 | Perhaps we could make a haskellPackagesReproducible set where parallel builds are off and use that in the ISO | 10:45:22 |
| @rnhmjoj:maxwell.ydns.eu | I think the problem is that the most commonly used libraries are not particularly small (pandoc, lens, aeson, etc). ~100 modules is typical | 10:45:43 |
 baloo | In reply to @raboof:matrix.org
yeah single threaded builds seem more reliable but also quite a price to pay (https://github.com/NixOS/nixpkgs/issues/151347#issuecomment-997817696, https://gitlab.haskell.org/ghc/ghc/-/issues/12935#note_373865) could that be that the objects needs to be fed in a deterministic order to the linker? It looks like the linker just gets them in random order and slaps them together. | 17:31:02 |
