| 29 Jun 2025 |
 hexa | it probably grew organically, yes | 04:15:19 |
| hexa | https://github.com/NixOS/nixpkgs/pull/420912 | 04:17:00 |
| hexa | https://github.com/NixOS/nixpkgs/issues/287308 | 04:18:55 |
| hexa | https://github.com/orgs/NixOS/projects/88 | 04:19:10 |
| hexa | oh god | 04:19:11 |
 emily | https://github.com/NixOS/nixpkgs/pull/287105 relevant PR by some guy | 04:22:24 |
 ElvishJerricco | I think we should not do nixos-generate-config changes yet | 04:23:36 |
| ElvishJerricco | just flipping useNetworkd = true should do the trick | 04:23:48 |
| emily | if we had done them years ago we'd not be quite this many stack layers deep into talk about how to deprecate the tangled interface :p | 04:24:20 |
| ElvishJerricco | Well | 04:24:54 |
| ElvishJerricco | whatever we generate with nixos-generate-config should work for those who decide to disable useNetworkd | 04:25:09 |
| emily | I don't understand why | 04:25:57 |
| emily | why should generated configs be written to accommodate a migration option? | 04:26:18 |
| ElvishJerricco | I still don't see it as a migration option | 04:26:36 |
| ElvishJerricco | I think the networking.* things are here to stay | 04:26:43 |
| ElvishJerricco | at least, many of them | 04:26:54 |
| emily | that isn't what I meant (but we're returning to the long conversation this started with) | 04:27:01 |
| emily | choosing scripted networking is a migration thing | 04:27:12 |
| emily | a new user who doesn't know how to adjust their networking config but wants to opt into scripted networking is not a relevant use case | 04:27:24 |
| emily | the only reason we don't remove scripted networking tomorrow is existing users, not new ones | 04:27:41 |
| ElvishJerricco | In an ideal world, nixos-generate-config generates as little as possible, and calls on abstractions to do the real work. That way we can improve what was generated without having to regenerate anything | 04:28:18 |
| ElvishJerricco | it's worse to make a bunch of specific networkd configs than it is to just enable the most abstract possible options | 04:28:35 |
| emily | but networkd config is their best attempt at abstract declarative configuration already | 04:29:03 |
| ElvishJerricco | I mean | 04:29:13 |
| ElvishJerricco | eh | 04:29:14 |
| hexa | the configs we generate are basic enough that we should just generate the real thing and not the abstraction | 04:29:28 |
| ElvishJerricco | useDHCP does not have a clean networkd implementation even though it is the sane default for a new thing | 04:29:28 |
| emily | by a project with much more comparative expertise at that task compared to whatever random interface we accumulated since 2006 | 04:29:33 |
| ElvishJerricco | I mean no, my whole point is that I don't want hardware-configuration.nix to contain any sort of ideas about how things should be implemented | 04:30:15 |
| ElvishJerricco | because then it gets stale and we can't change it | 04:30:21 |
