| 9 Jun 2023 |
 infinisil | Redacted or Malformed Event | 18:50:36 |
 Robert Hensing (roberth) | I don't think we've considered pkgs/by-name yet? | 19:14:03 |
| infinisil | Robert Hensing (roberth): I just thought of pkgs/by-prefix a couple minutes ago :o | 19:14:40 |
| infinisil | Should definitely consider, reminds me of /dev/disk/by-* | 19:15:09 |
| Robert Hensing (roberth) | yeah, I don't think we'll have many views into the same "structure" like dev has, but I don't think that's an objection | 19:16:08 |
| infinisil | I guess it's also focusing on the structure of it, not the contents, which I don't think is great | 19:16:22 |
| Robert Hensing (roberth) | that's kind of the point though, because we can't agree on what the contents should be | 19:16:52 |
| Robert Hensing (roberth) | and we've already acknowledged that we may want to use it for more than packages | 19:17:09 |
| Robert Hensing (roberth) | and pkgs/by-name does suggest "packages", so that's actually covered | 19:17:35 |
| Robert Hensing (roberth) | like unit it lends itself to a transposition like by-name/<shard+name>/package. It's really quite similar | 19:18:54 |
| infinisil | True that's not bad. I'm not sure if we'd want to keep such a name in a future with more than just packages in these directories though. But that's not a strong argument | 19:21:23 |
| Robert Hensing (roberth) | not by-attr because we allow usage that doesn't strictly follow that; all-packages.nix may reuse the path-based name for attributes that have a different name, such as mypkg_unstable and whatnot | 19:22:21 |
| infinisil | Hmm, that's not great though, by-name has the same problem | 19:23:26 |
| Robert Hensing (roberth) | pkgs/software; just a catch-all category. Even plain data is software according to some definitions | 19:23:39 |
 K900 | Also, if I may throw in a nit, it would actually be nice to have it as _by-name or similar | 19:23:48 |
| Robert Hensing (roberth) | I think we can add that to the list of considered but rejected quite easily (software) | 19:24:00 |
| K900 | So it's sorted first | 19:24:00 |
| K900 | Also, I don't think mypkg-unstable breaks by-prefix? | 19:24:16 |
| K900 | It is the same prefix after all | 19:24:21 |
| K900 | Now unstable-mypkg would | 19:24:35 |
| K900 | But that's a different story | 19:24:39 |
| infinisil | If we have a single directory containing all the shards then I don't think it needs to be sorted at the beginning. Or at least I wouldn't want to add a _ just for that | 19:24:51 |
| Robert Hensing (roberth) | Right, I think in a few cases we might have soft forks with different names though | 19:24:59 |
| Robert Hensing (roberth) | Also the other categories will wither away eventually | 19:25:21 |
| Robert Hensing (roberth) | I mean pkgs/* subdirs, to the point that there's like five want to keep perhaps | 19:25:42 |
| Robert Hensing (roberth) | at that point sorting isn't super relevant | 19:25:48 |
| Robert Hensing (roberth) | wouldn't want to be stuck with an underscore then | 19:26:02 |
| Robert Hensing (roberth) | What I don't like about by-prefix is that it focuses on the mechanism too much. People will understand it as relating to the two letter shard directories, which is not a useful concept | 19:26:47 |
| Robert Hensing (roberth) | I want everyone to think of sharding as a useless but necessary syntax-level concern that really doesn't matter for anything they care about | 19:27:18 |
| Robert Hensing (roberth) | Also I want the names that we do use to be meaningful, if at all possible. by-name carries more relevant meaning than by-prefix, because prefixes are a syntax level concern that has no bearing on the actual meaning of what we're declaring in those files | 19:28:37 |
