| 25 Mar 2023 |
 matthewcroughan - nix.how | In reply to @7c6f434c:nitro.chat
Is it a good idea to do rebuilds locally, though? You mean nixos-rebuild on the phone? | 12:45:09 |
| matthewcroughan - nix.how | Is it a good idea to nixos-rebuild on a raspberry pi? It's kind of up to the user, for sure | 12:45:27 |
| matthewcroughan - nix.how | I don't feel good when I do it, but sometimes I do! | 12:45:45 |
| matthewcroughan - nix.how | But yeah, I'd be deploying to the phone with nix copy and activating on the phone. | 12:46:02 |
| matthewcroughan - nix.how | nix build the toplevel with --store ssh-ng://user@buildserver works | 12:46:35 |
| matthewcroughan - nix.how | Or hosting the toplevel closure on a webserver which is pulled with nix copy by a systemd timer and activated, that's a way of doing automated updates. | 12:47:11 |
| matthewcroughan - nix.how | that systemd timer could even pop up a UI object that says "Do you wan to update" and wait for a dbus response or something | 12:47:44 |
| matthewcroughan - nix.how | * that systemd timer could even pop up a UI object that says "Do you want to update" and wait for a dbus response or something | 12:47:48 |
 7c6f434c | In reply to @matthewcroughan:defenestrate.it
Is it a good idea to nixos-rebuild on a raspberry pi? It's kind of up to the user, for sure I think RPi is easier to cool, though… | 12:47:51 |
| matthewcroughan - nix.how | I mean, one of the benefits for me of running Mobile NixOS was going to be the CPU control | 12:48:09 |
| matthewcroughan - nix.how | running at low MHz always, to get the best possible battery | 12:48:33 |
| matthewcroughan - nix.how | I don't need much CPU! | 12:48:35 |
| 7c6f434c | Hm OK, if you have a ton of RAM and just do the rebuilds at slow rates in background and you can have the cache just for the big packages… | 12:49:20 |
| matthewcroughan - nix.how | yeah you could even have the rebuild switch happen in a systemd service with very very very low resources in a slice | 12:49:58 |
| 7c6f434c | In reply to @matthewcroughan:defenestrate.it
running at low MHz always, to get the best possible battery Pretty sure that CPUs are designed to make this is not the best possible battery, but nothing has enough load control to make the design come true | 12:50:06 |
| matthewcroughan - nix.how | 1cpu 512M memory | 12:50:08 |
| 7c6f434c | Uhm, let me count the ways 512MiB is not enough for nixos-rebuild | 12:50:30 |
| matthewcroughan - nix.how | nix evaluation right? | 12:50:37 |
| matthewcroughan - nix.how | It would be enough if we could share the eval cache, something I know is not secure at the moment | 12:50:49 |
| matthewcroughan - nix.how | And that is a design proposal, to be able to share eval cache | 12:51:13 |
| 7c6f434c | Not sure how well NixOS module system would work for sharing eval caches between non-identical systems
| 12:51:22 |
| matthewcroughan - nix.how | Yeah me neither, I'm just thinking there are no limitations and what the design would look like without limitations | 12:51:41 |
| 7c6f434c | I mean, I don't like NixOS modules for any of the other reasons either… | 12:51:59 |
| matthewcroughan - nix.how | to me, the perfect setup is nix running everywhere for every purpose, to keep everything isomorphic | 12:52:00 |
| 7c6f434c | I kind of want enough RAM to run Nix evaluation on almost everything, that's true… | 12:52:52 |
| 7c6f434c | But for many purposes I am mostly fine with Nix-on-whatever | 12:53:15 |
| 7c6f434c | (Assuming a somewhat clueless but not too malicious and already platform-adapted carrier system) | 12:53:53 |
| matthewcroughan - nix.how | In reply to @7c6f434c:nitro.chat
I kind of want enough RAM to run Nix evaluation on almost everything, that's true… No I mean sharing eval cache would prevent that from being a problem, optionally | 12:56:19 |
| matthewcroughan - nix.how | there are ways we can keep it isomorphic just by messing around with caches and stuff, and fix these problems just like that | 12:56:39 |
| 7c6f434c | If you give up on «actually can eval anywhere», I won't call Nix the ideal setup. Maybe looking at what Guix did, noticing it is still not the best direction, and designing a new Nix-like-tolerable-on-low-RAM without giving up on data-crunching-completeness… | 12:58:39 |
