| 28 Mar 2026 |
 Alex |
we autogenerate expressions
And as a result many of the broken ones are a doJailbreak away from building.
(Random idea: would automatically trying jailbreak = true on the failing packages as part of the Stackage bump workflow be worth doing?) | 19:16:06 |
 alexfmpe | If you just want to make them build, then yeah that's not too different than what my sweep script does (actually in the other direction to see if jailbreak/etc are no longer needed) | 20:47:38 |
| alexfmpe | In reply to @maralorn:maralorn.de
catsarecute: Unless the override is very nix specific we normally don’t accept patches which don’t have open PR upstream. So basically most of what would be done from our side is already been taken care of. But if we then merge as is it goes against this criteria | 20:48:05 |
| alexfmpe | But maybe it could be somewhat more automated than is | 20:48:53 |
 maralorn | This would in practice help a lot. Although error messages for build errors would get worse. And some people hate nixpkgs and will never touch it because we have doJailbreak. I remember hearing one story where someone lost data in production because we ignored a bound which didn’t prevent a compile error but a runtime error … | 21:51:47 |
 sterni | of course 9.12.4 released when I was busy. I've updated the RC PR now: https://github.com/NixOS/nixpkgs/pull/500108 | 21:53:05 |
| 29 Mar 2026 |
| Alex | Yeah, preventable runtime errors are the risk of automated jailbreaking. It might be best not to so that any jailbreaking people do must be a deliberate and conscious decision. | 01:53:39 |
 catsarecute | In reply to @maralorn:maralorn.de
That being said there are certainly a lot of things that could be improved in our setup. I am just not sure if we can manage to bring in GSoC form. Ah I see, that makes sense. I didn’t realize how much of the set was autogenerated from Hackage.
Out of curiosity then, what parts of the current Haskell packaging setup in nixpkgs do you feel are the roughest right now? I’d be interested in contributing around there. | 04:09:16 |
| catsarecute | Thanks everyone for the context and explanations, this was really helpful. I appreciate it a lot. | 04:11:25 |
| sterni | alexfmpe (or anyone else): do you have a minute to look this over? | 10:21:41 |
| alexfmpe | Lesseee | 10:22:45 |
| alexfmpe | LGTM, need some testing of builds? | 10:27:34 |
| sterni | i tested HEAD and 9.12.4, currently building cabal2nix and haskell-language-server and gonna merge after i think | 10:28:07 |
| sterni | on x86_64-linux | 10:28:10 |
| sterni | found a GHC 9.12.4 panic -.- | 12:03:20 |
| sterni | * found a GHC 9.12.4 panic -.- https://gitlab.haskell.org/ghc/ghc/-/issues/27121 | 12:40:58 |
| alexfmpe | Sigh | 12:45:44 |
| alexfmpe | https://gitlab.haskell.org/ghc/ghc/-/issues/27061#note_666912 | 12:49:06 |
| alexfmpe | This really needs to hapoen | 12:49:31 |
| alexfmpe | We even have a bunch of cross platforms with one-line level of support | 12:50:17 |
| alexfmpe | How do we dump all that on ghc hq as a regression test suite or whatever | 12:50:50 |
 chreekat | This is the kind of thing that needs a gsoc project | 12:54:30 |
| alexfmpe | Hehehe | 12:58:19 |
| alexfmpe | Somewhat related, I wonder if the ghc.nix shell could be massively reduced by being based in the shell for our hadrian derivation or something. | 13:02:15 |
| alexfmpe | Not sure if relevant here but when opening issues on ghc I tend to add a nixpkgs hash and the invocation just in case they can't reproduce it out of the box | 13:06:10 |
|  M̸̙̜̔̇Ǎ̴͎̙͔G̸̞̈N̸͔͍̝͗͋̾Ő̷͖̼͈̽̚L̷̻͚̓̔I̷̛͔̰̟̔Å̴̩̍ ̷̦̒̇͝M̷̱̠̺̉̎A̵̼̎͗͘Ỹ̸̬̲͂̕H̷̙̖͂Ē̷͉̦̌͒M̶͈̥̽̐ (you don't get my real name) changed their display name from M̸̙̜̔̇Ǎ̴͎̙͔G̸̞̈N̸͔͍̝͗͋̾Ő̷͖̼͈̽̚L̷̻͚̓̔I̷̛͔̰̟̔Å̴̩̍ ̷̦̒̇͝M̷̱̠̺̉̎A̵̼̎͗͘Ỹ̸̬̲͂̕H̷̙̖͂Ē̷͉̦̌͒M̶͈̥̽̐ to M̸̙̜̔̇Ǎ̴͎̙͔G̸̞̈N̸͔͍̝͗͋̾Ő̷͖̼͈̽̚L̷̻͚̓̔I̷̛͔̰̟̔Å̴̩̍ ̷̦̒̇͝M̷̱̠̺̉̎A̵̼̎͗͘Ỹ̸̬̲͂̕H̷̙̖͂Ē̷͉̦̌͒M̶͈̥̽̐ (you don't get my real name). | 13:22:43 |
| sterni | alexfmpe: well I did test it ahead of time, but I only tried a couple of packages that were affected by the subword division issue and cabal2nix iirc | 18:19:53 |
| 30 Mar 2026 |
|  Big man joined the room. | 04:41:17 |
 MangoIV | It can be. It just grew organically over many years. | 08:48:09 |
 Janus | I am wondering whether the plan is for nixpkgs to patch packages using memory to use ram instead? That would be interesting to see, as it could provide a relatively painless experience compared to all the builds currently failing (when mixing these) using the cabal solver... | 08:49:21 |
