| 22 Apr 2026 |
 Randy Eckenrode | I will have a very brief time this morning to look at it and then lunch. This is my no-rebuild hack. How did it pass the eval check if it now fails to eval? | 11:06:39 |
 emily | it's only failing on x86_64-darwin | 11:09:03 |
| Randy Eckenrode | Okay, but it evals for me on x86_64-darwin, and it evaled in the CI action. | 11:12:39 |
| Randy Eckenrode | That’s the weird part. | 11:12:43 |
 vcunat | Oh right, it's broken on master already, not staging-next
https://hydra.nixos.org/job/nixpkgs/unstable/release-checks/all | 11:13:06 |
| vcunat | * Oh right, it's broken on master already, not just staging-next
https://hydra.nixos.org/job/nixpkgs/unstable/release-checks/all | 11:13:22 |
| Randy Eckenrode | x86_64-darwin evals for me on staging-next. | 11:14:08 |
| emily | perhaps it's a Nix version thing(???) | 11:14:20 |
| Randy Eckenrode | I’m using Lix, but shouldn’t CI be using whatever the default Nix is? | 11:15:53 |
| Randy Eckenrode | Are release checks always run from Linux? | 11:16:23 |
| vcunat | Yes. | 11:18:18 |
| vcunat | It does look like an impure problem. | 11:18:40 |
| Randy Eckenrode | The release check also succeeds locally for x86_64-darwin. I’ll try it from Linux next. | 11:20:48 |
| Randy Eckenrode | It also evals on Linux …. | 11:26:57 |
 Sergei Zimmerman (xokdvium) | k, looking | 11:36:13 |
| Sergei Zimmerman (xokdvium) | Hm reproduces locally on 2.28/2.31/2.33/2.34 | 11:39:13 |
| Randy Eckenrode | https://github.com/NixOS/nixpkgs/pull/512356 tries to fix it based on the Hydra log. I’m not able to reproduce locally, so if that doesn’t work, we may need to revert the PR splitting out the source releases. | 11:39:24 |
 whispers [& it/fae] | nom-build pkgs/top-level/release.nix -A release-checks fails on latest master (be011d5e) for me on nix (Lix, like Nix) 2.96.0-dev-23b2797321e6 fwiw | 11:40:04 |
| whispers [& it/fae] | nom-build pkgs/top-level/release.nix -A release-checks fails on latest master (be011d5e) for me on nix (Lix, like Nix) 2.96.0-dev-23b2797321e6/x86_64-linux fwiw | 11:40:41 |
| Sergei Zimmerman (xokdvium) | Also on lix 2.93 but not with lix git | 11:40:38 |
| whispers [& it/fae] | nom-build pkgs/top-level/release.nix -A release-checks fails on latest master (be011d5e) for me on nix (Lix, like Nix) 2.96.0-dev-23b2797321e6 & x86_64-linux fwiw | 11:40:56 |
| Sergei Zimmerman (xokdvium) | So I assume this might be some fallout from lix changes to pointer equality or some other regression on the lix side? | 11:41:43 |
| Randy Eckenrode | It should pass the nixpkgs-vet check now …. | 11:42:07 |
| Sergei Zimmerman (xokdvium) | It's using cppnix inside the build | 11:43:06 |
| Randy Eckenrode | I was running the checks individually. I’m trying them now using nix-build pkgs/top-level/release.nix -A release-checks . I’m on Lix 2.94.1. | 11:43:29 |
| Randy Eckenrode | It doesn’t run on Darwin. It fails with error: creating directory "/homeless-shelter/.local/state/nix": Read-only file system. Trying on Linux now. | 11:46:24 |
| Sergei Zimmerman (xokdvium) | Ok so narrowed it down to 2.93 -> 2.94 now. If anybody has the capacity to bisect lix be my guest. It might be the intentional pointer equality changes between those versions, but uncertain | 11:49:07 |
 raitobezarius | What is the command you are running? | 11:49:25 |
| raitobezarius | Wait, the eval error is a missing attribute on some bootstrap component or an infinite recursion? | 11:52:09 |
| Sergei Zimmerman (xokdvium) | Same one as in the buiild: nix-env -f /nix/store/phggf546c3ig9x349z4rf0qr4ckdci62-source --show-trace --argstr system x86_64-darwin --arg config '{ allowAliases = false; allowDeprecatedx86_64Darwin = true; }' -qa --drv-path --system-filter '*' | 11:52:23 |
