| 23 May 2024 |
 hexa | I thin not, in that I had to fiddle with the ttl in nix | 18:25:25 |
| hexa | because nix is not kind about failed substitution | 18:25:34 |
| hexa | narinfo-cache-positive-ttl = 3600;
| 18:26:09 |
| hexa | with gc.automatic enabled and no further tuning | 18:26:28 |
 cole-h | Then a follow up: is there something harmonia does better than attic? | 18:26:55 |
| hexa | the default is like a month | 18:26:59 |
| hexa | serves directly from the /nix/store, no central binary store required | 18:27:17 |
| hexa | attic with a small binary cache and aggressive gcing would work as well | 18:28:39 |
 ris_ | ahem https://github.com/risicle/nix-heuristic-gc | 20:25:16 |
| ris_ | i actually originally built that with the intention to use it on a binary cache | 20:27:20 |
| 24 May 2024 |
 trofi | I keep finding huge amount of accidental $foo.tests eval failures. An example fix is https://github.com/NixOS/nixpkgs/pull/314109. Looking at https://github.com/NixOS/nixpkgs/runs/24606477687 ofborg shown it as green while it said explicitly it fails to evaluate: The following builds were skipped because they don't evaluate on x86_64-linux: githooks.passthru.tests. Shouldn't it be at least gray in the UI? I vaguely remember it doing that in the past. | 06:08:11 |
 7c6f434c | Tests failing to evaluate is usually having no tests specified. A majority of packages failing to evaluate is something not being in the platforms. | 06:13:50 |
| 7c6f434c | Distinguishing the cases is some non-trivial work, not just switching a single constant | 06:15:26 |
| trofi | Sure, but it's so sad to see thing like:
$ nix build --no-link -f. nixVersions.git.tests
...
error: attribute 'nix_2_23' missing
Did you mean one of nix_2_13, nix_2_20, nix_2_21, nix_2_22 or nix_2_3?
| 06:15:26 |
| trofi | At least "no tests specified" feels like a simple case of changing the evaluator to discover tests with $foo.tests or [] instead of $foo.tests. | 06:16:34 |
| 7c6f434c | I think right evaluation failures are handled the same for tests and for packages | 06:17:12 |
 Lily Foster | there was a check-meta check to make sure tests was an attrset of derivations | 11:13:50 |
| Lily Foster | but | 11:13:52 |
| Lily Foster | when tests were moved from meta.tests to passthru.tests before that PR was merged | 11:14:14 |
| Lily Foster | the check-meta check was not updated | 11:14:20 |
| Lily Foster | so it's not useful right now | 11:14:27 |
| Lily Foster | but checking it there-ish would be ideal | 11:14:49 |
| Lily Foster | (and then abort-ish eval failures (e.g. syntax errors) vs throw-ish eval failures would also be differentiated, as intended by most eval checks) | 11:15:52 |
| Lily Foster | * (and then abort-ish eval failures (e.g. syntax errors) vs throw-ish eval failures (e.g. not available on platform) would also be differentiated, as intended by most eval checks) | 11:16:18 |
| 27 May 2024 |
 Vladimír Čunát | Any idea if the eval label mechanism is broken? It got completely wrong in this case: https://github.com/NixOS/nixpkgs/pull/314255#issuecomment-2132781439 | 07:06:12 |
| Vladimír Čunát | (also in the original PR from which this backport was created) | 07:07:28 |
| Vladimír Čunát | Or do the linux rebuild labels not take aarch64-linux into account? | 07:07:58 |
| hexa | vcunat: https://github.com/NixOS/ofborg/blob/1d74f577a10ca1adbc933bdfb966d380aa343eb1/ofborg/src/tagger.rs#L327 🙂 | 12:47:59 |
| hexa | aarch64 is not represented in there | 12:48:53 |
| hexa | * aarch64 is not represented in there, not for linux, not for darwin | 12:49:06 |
