| 10 Sep 2024 |
 colemickens | the next step would be vmTools, but also :/ I'm starting to doubt ? | 05:07:59 |
| colemickens | maybe this is vmTools use of buildPackages, and maybe I can just set customQemu and get around it | 05:14:34 |
| colemickens | well no that doesn't make sense, build and host is aarch64-linux. what the heck then. | 05:14:52 |
| colemickens | matthewcroughan: https://github.com/NixOS/nixpkgs/pull/325619 doesn't this kinda not work since ${qemu} is still used for making the empty disk image, etc? | 05:40:54 |
| colemickens | I'm so confused. If I build qemu_kvm or buildPackages.qemu_kvm it works, but if I build diskoImageScript it acts like it's crosscompiling from x86_64-linux. I've checked nixpkgs host/buildPlatform. :/ | 05:43:10 |
| colemickens | xargs cp :/ | 06:25:21 |
| colemickens | runs out of mem | 06:25:25 |
| colemickens | https://github.com/nix-community/disko/blob/e55f9a8678adc02024a4877c2a403e3f6daf24fe/lib/make-disk-image.nix#L200
💀 | 06:31:03 |
| colemickens | minus a couple small things, disko was great. got a gpt rk3588 uboot+roofs emmc image image, all automated with disko. | 07:47:18 |
| colemickens | truly much more pleasant that sdImage | 07:47:39 |
 matthewcroughan | In reply to @colemickens:matrix.org
xargs cp :/ OMG you found the reason for the out of mem? | 11:36:09 |
| matthewcroughan | Can we fix it | 11:36:11 |
| matthewcroughan | lassulus: | 11:36:26 |
| matthewcroughan | https://github.com/nix-community/disko/blob/e55f9a8678adc02024a4877c2a403e3f6daf24fe/disko-install#L238 | 11:36:45 |
| matthewcroughan | xargs cp is running out of mem apparently according to cole | 11:36:56 |
 lassulus | hm? I thought the xargs cp was used instead of the nix copy since it uses less memory? | 11:37:01 |
| matthewcroughan | * xargs cp is running out of mem apparently in disko according to cole | 11:37:01 |
| matthewcroughan | maybe both do and we still haven't found the reason, but it's interesting that cole identified it almost immediately | 11:37:25 |
| lassulus | yeah copying derivations is usually the step where we do OOM | 11:38:51 |
|  @shinri:zince.net left the room. | 12:37:47 |
| colemickens | no no | 14:57:14 |
| colemickens | so the issue is that there's a module option for memSize, that is ignored in the impure/script route. | 14:57:29 |
| colemickens | I can try to remember to send a small patch for that. | 14:57:38 |
| colemickens | The postVM example is also ... wrong (you can't glob like that in a script), and ALSO a bit scary, because the script just clobbers $out/disk0.raw, no matter what is there, and worse, if the globbing worked, would just hit every "*raw" in whatever $out happens to be. | 14:58:30 |
| colemickens | But also, yeah, it would be cool if it didn't use up a fair bit of ram just running 'cp' :) | 14:59:03 |
| colemickens | * The postVM example is also ... wrong (you can't glob like that in a script), and ALSO a bit scary, because the script just clobbers $out/disk0.raw, no matter what is there, and worse, if the globbing worked, would just hit every "*raw" in whatever $out happens to be. (In "impure" $out is just where you run the image script from!) | 14:59:43 |
| matthewcroughan | In reply to @colemickens:matrix.org
But also, yeah, it would be cool if it didn't use up a fair bit of ram just running 'cp' :) the image builder occasionally ooms still to this day I think | 15:23:07 |
| matthewcroughan | it's not very reproducible, just happens sometimes | 15:23:13 |
| matthewcroughan | maybe we could replace xargs and cp with rust based things and see if it still happens | 15:24:16 |
| matthewcroughan | There is rargs and fcp | 15:24:46 |
