| 6 May 2025 |
 Alex | In reply to @m1-s:matrix.org
Hello, does anyone have experience with building haskell on macOS? This project compiles fine on linux where nix uses gcc by default. it fails on darwin however as the default compiler there is clang.
> /nix/store/q7lf35kha4z40zkmhf3cyipaf7kkk6if-cctools-binutils-darwin-wrapper-1010.6/bin/ld: line 268: 5770 Segmentation fault: 11 /nix/store/qib1l5hjzaycqyw9slc2j127wma1ayix-cctools-binutils-darwin-1010.6/bin/ld "@$responseFile"
> clang-16: error: linker command failed with exit code 139 (use -v to see invocation)
> ghc-9.6.6: `cc' failed in phase `Linker'. (Exit code: 1)
The system has a linux builder vm with nix-darwin enabled but I dont think its related to this error.
You need to use a newer Nixpkgs, because this was fixed a few months ago.
I can't recall what the exact commit with the fix was, but one of the maintainers here likely knows. | 21:30:22 |
 alexfmpe | In reply to @alexfmpe:matrix.org
https://github.com/NixOS/nixpkgs/issues/149692 Fixed by https://github.com/NixOS/nixpkgs/pull/307880 | 22:24:53 |
| alexfmpe | I'd just bump to recent master and get 9.8.3 as default.
It took like 5s off my reloads on mac | 22:26:27 |
| alexfmpe | This here: https://gitlab.haskell.org/ghc/ghc/-/issues/23415 | 22:27:07 |
| 7 May 2025 |
|  Aerden joined the room. | 08:03:34 |
 sterni (he/him) | There is some kind of race condition in the install of cross compiled riscv64 ghc948 which is bizarre since I don't think we have the issue with the regular ghc | 09:30:23 |
| Alex | How frequently does it occur?
What does the error(?) look like? | 10:44:43 |
 maralorn | The Data.Text.show business is definitely aggravating. | 11:14:15 |
| maralorn | I have no clue how many ours of work and how many delays this will cost us … | 11:16:49 |
| maralorn | srsly, hackage needs a validator which plain forbids implict imports on packages without tight bounds. | 11:17:42 |
| maralorn | Although that wouldn’t help us much either … | 11:17:57 |
 Teo (he/him) | Are the broken packages on stackage? Or is it stuff in nixpkgs that's not on stackage? At some point I want to put some effort into improving our ecosystem testing code in head.hackage and making sure we test as much as possible before we cut a release of GHC | 11:32:06 |
| maralorn | Yes: https://hackage.haskell.org/package/http-link-header which was affected is on stackage apparently. | 11:36:20 |
| maralorn | Although they already fixed it in their newest release. | 11:36:47 |
| maralorn | Besides that I only encountered problems in more or less local code. | 11:37:08 |
| maralorn | So maybe its not going to be as bad as I am fearing. | 11:37:22 |
 chreekat | https://github.com/commercialhaskell/stackage/issues/7743 is the Stackage coordination point | 11:39:18 |
| chreekat | I count 48 packages dropped, and Jens said at least 18 of them are direct dependents of text | 11:41:54 |
| chreekat | Luckily it was "just" nightly, so snapshots continue to be released. Just with fewer packages | 11:42:34 |
| Teo (he/him) | Thanks! We have a draft PR to run head.hackage against stackage snapshosts somewhere. I want to tidy that up and merge that at some point. And then we could run that before a every release and have a better understanding of these things | 11:44:34 |
| Teo (he/him) | Btw I'll be at Zurihac this year and I'd be keen to have a chat about how we can improve GHC/head.hackage processes to make things like this less painful in the future | 11:49:33 |
| sterni (he/him) | i would guess more than 1/10; IIRC it is something like install: could not create file …: no such file or directory. Happens on Hydra, so you probably need a reasonable high -j (is marked big-parallel) and a busy machine | 11:52:58 |
| maralorn | I’ll be there, too. Let’s do that! | 12:33:02 |
| chreekat | I'm also in! | 12:53:01 |
 hellwolf | nice, I will be there too :) | 13:54:00 |
|  Brendon A Kay joined the room. | 15:00:46 |
|  Ethan Bodzioney joined the room. | 16:02:37 |
| maralorn | Does hydra kill jobs which need to much memory? | 21:15:38 |
| maralorn | Because the build I am currently trying is roughly at 100G RAM right now, so even if it succeeds, which seems unlikely, it doesn’t seem like a viable solution. 😄 | 21:16:51 |
| Alex | The kernel can if the overall system is low on memory (OoM killer). | 21:40:09 |
