| 7 Sep 2025 |
 Sergei Zimmerman (xokdvium) |
- Which lix is the CI running and 2. Nixpkgs input probably needs https://github.com/NixOS/nixpkgs/pull/434761
| 11:12:15 |
 aloisw |
- is for the infra folks to answer. 2. should probably be done on all branches then?
| 11:34:07 |
| aloisw | I guess a different Lix running on the x86_64-linux and aarch64-linux builders could explain the difference. | 11:34:42 |
| Sergei Zimmerman (xokdvium) | I imagine some runners are running newer Lix that has emily’s stack merged. That issue is exactly why we had to revert the initial patch to nix and only reapplied it only when that PR backport landed. | 11:40:35 |
| aloisw | Yeah I just don't have the insights into the infra to be sure this is indeed the case. | 11:41:47 |
| Sergei Zimmerman (xokdvium) | Everything points to that tbh. In cppnix we are dogfooding all PR GHA CI with the top of trunk as the host nix | 11:42:50 |
| Sergei Zimmerman (xokdvium) | And the exact issue immediately surfaced | 11:44:52 |
 raitobezarius | We don't do the same because the host Nix in the Buildkite agents is not dynamic | 11:59:56 |
| raitobezarius | A different Lix though is unexpected, everything should be upgraded at the same time | 12:01:00 |
| aloisw | Is the one on the aarch64-linux builders updated to toml11 4? | 12:01:04 |
| aloisw | I'm preparing a nixpkgs bump fwiw that should fix this. | 12:01:14 |
| raitobezarius | In reply to @aloisw:julia0815.de
Is the one on the aarch64-linux builders updated to toml11 4? I don't know OTOH | 12:02:08 |
| aloisw | Let's see whether https://gerrit.lix.systems/c/lix/+/4142 succeeds in CI, I will post backports to all supported releases if it does. | 12:04:07 |
 Simon Hauser | Redacted or Malformed Event | 12:30:40 |
 emily | should we decide what releases are supported first? 😅 | 14:05:35 |
| emily | my understanding is that it was going to be just 2.93 and 2.91-on-25.05? | 14:06:00 |
| emily | my stack forward-ports the new toml11, or do you mean it happens before that point in the stack? | 14:07:11 |
| aloisw | It happens before that point because the host nix evaluates something for the nixpkgs lib tests. | 14:07:43 |
| emily | yeah then it's just what @xokdvium:matrix.org says | 14:08:06 |
| aloisw | So all evals in CI will fail unless bumped past that nixpkgs PR. | 14:08:11 |
| emily | because lib tests were very silly | 14:08:15 |
| emily | the Nixpkgs PR adding that absurd test should never have been merged either :( | 14:09:04 |
| aloisw | Wait it was fully intentional to test the buggy integer saturation? | 14:10:14 |
| emily | yes and there was a blocking review from lib maintainer saying "uh. no?" | 14:11:36 |
| emily | and then it got cherry-picked into a GSoC PR without modification | 14:11:51 |
| emily | https://github.com/NixOS/nixpkgs/pull/433710#issuecomment-3189728351 | 14:12:47 |
| emily | feeding hashString into fromHexString was not only a test, it was one of two examples in the docs!! | 14:13:28 |
| emily | 🫠 | 14:13:42 |
| emily | it never worked on 2.3 either | 14:19:21 |
| emily | so nobody ever tried to eval lib tests with 2.3 on 24.11 or 25.05, at least as the host Nix | 14:19:49 |
