| 26 Jul 2025 |
 emily | it's just the bind mount that scares me | 21:04:43 |
 raitobezarius | In reply to @emilazy:matrix.org
what if you… replace it with a Unix socket <insert meme of person unhappy> | 21:04:51 |
| emily | no I mean like | 21:04:56 |
| emily | what if something switches out the cert store for a Unix socket | 21:05:04 |
| emily | I dunno | 21:05:10 |
| raitobezarius | can it change inode type? | 21:05:20 |
| emily | well you could do it between builds even (maybe? if that's not checked for) | 21:05:31 |
| raitobezarius | the setup happens at each build | 21:05:43 |
| emily | not having to think about all of this, and making it faster for Darwin, is why I landed on the store path solution 😅 – esp. since on Darwin the path already canonicalizes to something in the store already, so zero copying needed | 21:05:46 |
| raitobezarius | so if it becomes a socket after build 1 | 21:05:49 |
| emily | (though you do have to canonicalize again, in that case) | 21:05:54 |
| raitobezarius | it gets rejected | 21:05:58 |
| raitobezarius | i paged pennae on this question | 21:06:26 |
| emily | fwiw, not sure about this | 21:06:26 |
| raitobezarius | i'm slowlycrashing | 21:06:36 |
| emily | anyway, ideally there'd be some Darwin benchmark numbers but … that's hard | 21:06:38 |
| emily | I don't oppose this path | 21:06:44 |
| emily | I like the store thing the most still, but I'm not the one implementing it | 21:06:53 |
| raitobezarius | given my time budget, i do not think the store solution will be implementable | 21:07:08 |
| emily | I worry about things constantly getting slightly slower in ways nobody notices until it all adds up | 21:07:10 |
| raitobezarius | we care about sandbox setup times | 21:07:18 |
| raitobezarius | but i also need help to make this better | 21:07:33 |
| raitobezarius | for example, I'm still seeking someone to package bencher.dev | 21:07:39 |
| raitobezarius | if i had that, i would setup continuous benchmark infra | 21:07:47 |
| raitobezarius | and we could pile up benchmark targets | 21:07:52 |
| emily | FWIW, everything feels mysteriously slow on Darwin in ways that i suspect are trivial and dumb but that nobody has figured out | 21:08:02 |
| emily | I forget if I mentioned Randy having compared build times inside a Nix build and out and the latter being way faster | 21:08:15 |
| raitobezarius | this is very possible | 21:08:21 |
| emily | unfortunately I may end up having to be that person | 21:08:32 |
| raitobezarius | i'm happy to make myself available and answer questions and bounce ideas and look beyond the shoulder | 21:08:46 |
