| 20 May 2026 |
 Mic92 | In reply to @dramforever:matrix.org
the annoying thing is you have to be root to write to binfmt_misc, because the thing is hard coded to be owned by 0:0 in its userns I think we already have this for container enabled builds... But it is a bit annoying if we would have to rely on it | 14:48:03 |
 dramforever | as in, if you want to modify derivations | 14:48:11 |
| dramforever | you can just do the cursed dance to get yourself into a user namespace with a binfmt_misc you manage | 14:48:54 |
| dramforever | ... actually let me check something brb | 14:51:46 |
| Mic92 | @dramforever:matrix.org: I put it on my review list but let me know if I don't get back in a week | 14:55:47 |
| Mic92 | It looks mostly fine I did just want to read the double userns stuff again on a bigger screen | 14:56:46 |
| dramforever | Mic92: so i don't know how to do it with https://man7.org/linux/man-pages/man1/unshare.1.html, might try later, but what i did in #15539 shows that you can definitely give yourself binfmt_misc without privileges | 15:06:04 |
| dramforever | * Mic92: so i don't know how to do it with https://man7.org/linux/man-pages/man1/unshare.1.html, might try later, but what i did in #15539 shows that you can definitely give yourself a binfmt_misc without privileges | 15:06:11 |
| dramforever | as said in the pr description #15539 works even on rootless nix | 15:06:29 |
| dramforever | so if you're willing to do surgery on the bad perl stuff, i think you can get away with not touching nix at all | 15:07:08 |
| dramforever | * so if you're willing to do derivation surgery on the bad perl stuff, i think you can get away with not touching nix at all | 15:07:17 |
| dramforever | i'm thinking about the implications of a more literal "Allow write access to binfmt_misc when using Linux kernel >= 6.7", as in the title of that lxc issue, but currently i'm inclined to think we don't want that | 15:08:48 |
| dramforever | you might also want to know that there's no way to add binfmt_misc interpreters for a userns, only fully replace, so theoretically you can run into problems with an inner binfmt_misc on an binfmt_misc-emulated-cross build but at that point you're stretching everything too hard | 15:14:56 |
| dramforever | * you might also want to know that there's no way to add binfmt_misc interpreters for a userns, only fully replace, so theoretically you can run into problems with an inner binfmt_misc on an binfmt_misc-emulated-cross build but at that point you're stretching everything too hard (does qemu user even work for this? maybe there are less strech scenarios. idk.) | 15:15:21 |
| dramforever | okay i read the actual man page and you can do the double userns dance with
unshare --map-root-user --pid --fork --load-interp ":py:E::py::$(which python3):" unshare --map-user "$(id -u)" --map-group "$(id -g)" bash | 15:38:02 |
| dramforever | * okay i read the actual man page and you can do the double userns dance with something like
unshare --map-root-user --pid --fork --load-interp ":py:E::py::$(which python3):" unshare --map-user "$(id -u)" --map-group "$(id -g)" bash | 15:38:13 |
| dramforever | make of that what you will | 15:41:57 |
| 21 May 2026 |
|  rajudev joined the room. | 00:00:26 |
|  inifynastic joined the room. | 03:48:01 |
|  goatedxchef joined the room. | 19:38:00 |
| 22 May 2026 |
|  DJ2BD joined the room. | 13:02:40 |
|  Benedek joined the room. | 14:48:13 |
|  Jon Hermansen changed their display name from jonhermansen to Jon Hermansen. | 19:18:39 |
| 23 May 2026 |
 eveeifyeve | Has anyone considered a pr to flake registries to make them just an array of objects that contain owner repo and type? | 00:48:52 |
| eveeifyeve | * Has anyone considered a pr to flake registries to make them just an array of objects that contain owner, repo and type instead of what we have currently? | 00:49:13 |
| eveeifyeve | * Has anyone considered a pr to flake registries to make them just an array of objects that contain owner, repo and type instead of what we have currently, which IMO seems to deduplicate a lot of inputs? | 00:49:46 |
| eveeifyeve | * Has anyone considered a pr to flake registries to make them just an array of objects that contain owner, repo and type instead of what we have currently? Which IMO seems to deduplicate a lot of inputs that doesn't be needed? | 00:50:06 |
| eveeifyeve | * Has anyone considered a pr to flake registries to make them just an array of objects that contain owner, repo and type instead of what we have currently? Which IMO seems to deduplicate a lot of inputs that doesn't be needed. | 00:50:08 |
| eveeifyeve | * Has anyone considered a pr to flake registries to make them just an array of objects that contain owner, repo and type instead of what we have currently? Which IMO seems to deduplicate a lot of inputs that doesn't be needed like the to and from. | 00:51:27 |
| eveeifyeve | Like if there is no two and from, it should automatically assume that it is the to and from. | 00:53:47 |
