| 12 Sep 2025 |
 bglgwyng | * Do you mean that it's possible to override 'cabal2nix' value with caba-insstall, which is used to build the haskell package? | 08:25:54 |
 maralorn | Well, no. If your project actually needs cabal-install as a Haskell dependency (not just as a build tool), then you will have to figure out how to massage the override in a way that cabal-install works. | 08:27:27 |
| maralorn | If however you only need cabal-install as a tool dependency then you should really just use the cached pkgs.cabal-install without any overrides. | 08:28:14 |
| bglgwyng | I need cabal-install as dependency | 08:28:39 |
| maralorn | If you have to do that then you need to understand why configure is pulling two different version of hackage-security in. | 08:29:11 |
| bglgwyng | Yes I agree | 08:29:54 |
| maralorn | There is one other possibility. 1. Only override hackage-security on your project not on the whole dependency closure. 2. Enable "allowInconsistentDependencies". 3. Hope that the types from the two different hackage-security versions never meet during compilation.
However last time I tried something like this I failed. So it’s probably not a good idea. | 08:31:25 |
| bglgwyng | Hmm I'm trying haskell.nix now. Actually, I should build my package to bootstrap things, so that how to build it not important. | 08:33:10 |
| bglgwyng | Wait for minutes please. | 08:33:17 |
| bglgwyng | * Hmm I'm trying haskell.nix now. Actually, I should build my package to bootstrap things, so that how to build it is not important. | 08:33:34 |
| maralorn | Best explanation I can think of is that there is a dependency chain cabal-install -> X -> hackage-security, where X is a core package distributed with ghc. Then X would not be affected by the overrideScope and thus use the wrong hackage-security.
But this is close to where I normally give up … | 08:34:07 |
 oak 🏳️🌈♥️ | The maintainer scripts maintainers/scripts/haskell/ seem broken, I get this error with multiple scripts: error: access to absolute path '/home/myusername/.local' is forbidden in restricted mode | 13:08:51 |
| maralorn | Overriding core adjacent packages is just a pain and actually the one time where I switched to haskell.nix … | 08:34:42 |
| oak 🏳️🌈♥️ | ./maintainers/scripts/haskell/regenerate-hackage-packages.sh --fast goes through, but without --fast I get the above error | 13:11:37 |
|  @aidalgol:tchncs.de set a profile picture. | 09:21:40 |
| oak 🏳️🌈♥️ | Same seems to happen with maintainers/scripts/haskell/regenerate-transitive-broken-packages.sh | 13:11:56 |
 sterni (he/him) | okay I fell down one or two rabbit holes | 16:03:07 |
| oak 🏳️🌈♥️ | Actually I think it happens while regenerate-hackage-packages.sh calls regenerate-transitive-broken-packages.sh | 13:14:27 |
| oak 🏳️🌈♥️ | https://github.com/NixOS/nixpkgs/issues/400784 I see there's this issue which was closed by sterni last week, I was able to reproduce it :D I have nix (Nix) 2.28.4 | 14:08:19 |
| oak 🏳️🌈♥️ | * https://github.com/NixOS/nixpkgs/issues/400784 I see there's this issue which was opened by sterni which was closed last week, I was able to reproduce it :D I have nix (Nix) 2.28.4. I'm trying to update to 2.28.5 which seems to be the latest on stable | 14:11:55 |
| oak 🏳️🌈♥️ | Seems to happen also with Nix version 2.30.2 and 2.30.3, I first saw this on my macOS system and then I switched to NixOS to see it also happening there | 14:24:34 |
 jean-paul. | Anyone recognize this?
network> tests/Network/SocketSpec.hs:111:9:
network> 1) Network.Socket.getAddrInfo works for IPv6 address
network> uncaught exception: IOException of type NoSuchThing
network> Network.Socket.getAddrInfo (called with preferred socket type/protocol: AddrInfo {addrFlags = [AI_NUMERICHOST,AI_ADDRCONFIG], addrFamily = AF_UNSPEC, addrSocketType = NoSocketType, addrProtocol = 0, addrAddress = 0.0.0.0:0, addrCanonName = Nothing}, host name: "2001:0db8:85a3:0000:0000:8a2e:0370:7334", service name: Nothing): does not exist (Address family for hostname not supported)
network> To rerun use: --match "/Network.Socket/getAddrInfo/works for IPv6 address/" --seed 102858746
network> tests/Network/Socket/ByteStringSpec.hs:268:9:
network> 2) Network.Socket.ByteString.recvMsg receives control messages for IPv6
network> uncaught exception: IOException of type UnsupportedOperation
network> Network.Socket.bind: unsupported operation (Cannot assign requested address)
network> To rerun use: --match "/Network.Socket.ByteString/recvMsg/receives control messages for IPv6/" --seed 102858746
From 092c565d333be1e17b4779ac22104338941d913f (~5 days ago)
| 20:37:13 |
| sterni (he/him) | darwin? | 21:04:11 |
| jean-paul. | nope, Linux - but CircleCI | 21:20:02 |
 Alex | In reply to @jean-paul.:matrix.org
nope, Linux - but CircleCI Just guessing, but is it because the kernel has IPv6 disabled via sysctl or perhaps even via kconfig?
Nix builds are sensitive to the host kernel. | 22:17:23 |
| 13 Sep 2025 |
| oak 🏳️🌈♥️ changed their profile picture. | 09:46:10 |
 Teo (he/him) | I know the nixpkgs build of GHC edits the settings file, so I wanted to give y'all a heads up that the ghc-toolchain line of work is replacing the settings file, so it's likely that you'll need to make some changes here. See: https://gitlab.haskell.org/ghc/ghc/-/merge_requests/14500 | 13:48:05 |
| Teo (he/him) | * I know the nixpkgs build of GHC edits the settings file, so I wanted to give y'all a heads up that the ghc-toolchain line of work is replacing the settings file, so it's likely that you'll need to make some changes here. See: https://gitlab.haskell.org/ghc/ghc/-/merge_requests/14554 | 13:48:56 |
| jean-paul. | Seems likely I guess. I don't know how to learn about CircleCI's kernel configuration.
Annoyingly, zlib test suite also fails. Seems like that should be less sensitive to the kernel but 🤷 | 14:45:54 |
| jean-paul. | I assume more folks than me want to use Nix on CircleCI | 14:46:10 |
