| 9 Dec 2025 |
 Taeer Bar-Yam | No. null becomes empty string in my own example that i cooked up, not using inputDerivation. In that case it fails to build.
inputDerivation somehow gets rid of allowedReferences, despite setting it to null, and it succeeds to build
| 17:09:54 |
| Taeer Bar-Yam | I think so | 17:10:12 |
 John Ericson | OK so I was fixing the wrong branch :) | 17:10:35 |
| Taeer Bar-Yam | honestly both branches seem broken | 17:11:19 |
| John Ericson | code branch | 17:11:23 |
| John Ericson | oh? | 17:11:25 |
| Taeer Bar-Yam | even if nixpkgs is somehow getting away with it, it's still true that just setting allowedReferences = null; in a derivation that references something will fail | 17:12:06 |
| Taeer Bar-Yam | at least it does when I try it | 17:12:25 |
| Taeer Bar-Yam | I don't know what I'm doing differently than nixpkgs 😠| 17:12:33 |
| John Ericson | Sergei Zimmerman (xokdvium): https://github.com/NixOS/nix/pull/14754 I made this one with just the cleanup, and a warning for when exportReferencesGraph is not an object as we expect | 17:45:03 |
| John Ericson | and ^ Taeer Bar-Yam | 17:45:18 |
 Sergei Zimmerman (xokdvium) | In reply to @Ericson2314:matrix.org
Sergei Zimmerman (xokdvium): https://github.com/NixOS/nix/pull/14754 I made this one with just the cleanup, and a warning for when exportReferencesGraph is not an object as we expect Have you been able to repro with 2.4 / other versions too? | 17:46:23 |
| Taeer Bar-Yam | I'm super braindead right now, but I'll try to take a look this week | 17:46:36 |
| John Ericson | that PR is suppose to have 0 behavioral changes outside of diagnostics | 17:46:45 |
| John Ericson | not supposed to help or hurt 25.11 | 17:46:58 |
| Sergei Zimmerman (xokdvium) | In reply to @Ericson2314:matrix.org
that PR is suppose to have 0 behavioral changes outside of diagnostics For the next release cycle then since it’s a refactor | 17:47:22 |
| John Ericson | that's fine | 17:47:35 |
| John Ericson | (unless we want 2.33 to have 25.11) | 17:47:50 |
| John Ericson | * (unless we want 2.33 to have 25.11, maybe) | 17:47:59 |
| Sergei Zimmerman (xokdvium) | In reply to @Ericson2314:matrix.org
that PR is suppose to have 0 behavioral changes outside of diagnostics So the old PR should be closed then since it’s always behaved this way? I can double-check once more when I’m at the keyboard | 17:54:06 |
| John Ericson | I kinda do want null as a new feature rather than bug fix, but yes that takes the urgency away | 18:00:21 |
| John Ericson | marked it as draft | 18:01:10 |
| Taeer Bar-Yam | I will make a PR to nixpkgs inshallah this week | 18:02:16 |
| Taeer Bar-Yam | and then hopefully they'll backport it to 25.11? we'll see | 18:04:56 |
| John Ericson | Taeer Bar-Yam: yeah sounds good | 18:17:44 |
| John Ericson | It would be still nice to to know why non-structured attrs was working | 18:17:56 |
| John Ericson | but I just looked at 2.3 | 18:18:00 |
| John Ericson | and having it be an empty string in non-structured-attrs mode also seems wrong | 18:18:19 |
| John Ericson | in addition to null in structured attrs mode | 18:18:27 |
| John Ericson | https://github.com/NixOS/nix/pull/14755/files also here is a non-refactor version of just the warning and a test that I think would be nice to skip in before the release | 18:19:02 |
