| 5 Jul 2026 |
 emily | I mean we can just change the name | 21:20:05 |
 K900 | No like | 21:20:10 |
| K900 | Generally | 21:20:11 |
| emily | right | 21:20:20 |
 Scrumplex | I'm OOTL when it comes to cache.nixos.org, but does it deduplicate derivations? Otherwise I would be concerned that invalidating store paths might balloon the cache long term | 21:20:28 |
| K900 | The outpath won't chang | 21:20:43 |
| K900 | Redacted or Malformed Event | 21:20:44 |
| emily | I don't think a one-time payment of all the pnpm FODs should be a big deal | 21:20:44 |
| K900 | Also we have like | 21:20:50 |
| emily | it will | 21:20:48 |
| K900 | 100 pnpm FODs | 21:20:53 |
| K900 | Or something | 21:20:54 |
| emily | that's the only way you can invalidate a FOD | 21:20:56 |
| K900 | If we change the name yes | 21:20:59 |
| K900 | If we make the queue runner force rebuild them no | 21:21:04 |
| emily | okay, but we ought to change the name because downstream users might get their FODs broken by the SQLite update in case our sed isn't good enough | 21:21:25 |
| Scrumplex | In reply to @emilazy:matrix.org
I don't think a one-time payment of all the pnpm FODs should be a big deal Ah I thought you meant invalidating store paths going forward. One time should be fine :D | 21:21:29 |
| emily | i.e. it's not just Hydra even if we ignore the hypothetical nature of changing Hydra here | 21:22:03 |
| K900 | Valid | 21:22:36 |
| K900 | Fucking hell | 21:22:39 |
| emily | btw maybe check to see if there's some SQLite flag thing we can set to make it output the old way | 21:22:58 |
| emily | seems like the kind of thing they might have | 21:23:05 |
| Scrumplex | What if we bisect the exact sqlite commit that causes this miniscule change and revert it just for fetchPnpmDeps? | 21:36:19 |
| K900 | Disgusting but possible | 21:39:01 |
| emily | I prefer sed over that tbh | 21:40:18 |
| Scrumplex | Only to bridge the gap until we build a custom serializer/deserializer. We would just have to deprecate fetcherVersion 4 quite quickly though | 21:40:19 |
| K900 | Bisecting anyway | 21:44:45 |
| K900 | Maybe we'll find an option | 21:44:49 |
| K900 | My favorite part of sqlite is how it builds on one thread | 21:45:15 |
| K900 | NOTE TO SELF LOWER CASE IS BAD | 21:45:57 |
