| 2 Aug 2025 |
K900 | If we just use a different partition layout | 11:10:54 |
emily | tbf systemd upstream lightly recommends /efi for ESPs so we could just do that and it would fix it | 11:10:58 |
K900 | Or honestly it will take like five lines of code to fix this upstream too | 11:11:10 |
emily | and a separate XBOOTLDR is probably not a terrible idea given Windows | 11:11:16 |
emily | but you cannot have /boot as unencrypted FAT32 that is an ESP on UEFI and an XBOOTLDR on BIOS with unpatched Calamares | 11:11:57 |
emily | which is going to be the natural default setup once we switch BIOS to Limine | 11:12:10 |
emily | I wouldn't mind doing /efi though. but I bet it would annoy people | 11:12:46 |
aloisw | Since when do they recommend /efi? | 11:15:23 |
ElvishJerricco | systemd-gpt-auto-generator mounts the ESP on /efi unless /boot is an existing empty directory and isn't going to be used for XBOOTLDR | 11:16:52 |
ElvishJerricco | i.e. it only mounts the ESP at /boot if it really really looks like this person is used to it being there | 11:17:11 |
emily | yeah and the Lennart blog post advocating for the modern ESP/XBOOTLDR handling explicitly advocates for /efi for it | 11:17:48 |
emily | that's the standard if you have XBOOTLDR as /boot but the post and man page are pretty clear that the recommendation is not conditional on that | 11:18:26 |
emily | of course nobody does this in practice if you don't XBOOTLDR and probably some things still even do /boot/efi but it's the word of god on the matter | 11:19:06 |
emily | XBOOTLDR is frankly probably a good idea anyway. our kernels and initrds are fairly big and we make more of them than most distros thanks to our generation handling | 11:20:13 |
emily | and our handling when you get out of disk space on the boot partition is bad | 11:20:31 |
emily | and graphical installer users are much more likely to be dual booting with an existing Windows-created anaemically-sized ESP | 11:21:00 |
emily | FWIW https://gitlab.gnome.org/p3732/os-installer is a generic installer framework that looked like it might be nicer than Calamares when I took a brief look a while ago but I didn't go very in depth. there's a prototype NixOS thing for it linked but I wouldn't be surprised if it's wonky in the same ways as our Calamares one | 11:43:18 |
emily | there's also https://github.com/snowfallorg/icicle. vlinkz has written more graphical NixOS installers than anyone else on the planet apparently | 11:43:47 |
ElvishJerricco | oh also, it should also be noted why you want ESP at /efi and not /boot/efi. You can't have /boot as an automount if the ESP is at /boot/efi. It'll have to be always mounted instead of on demand. | 11:47:40 |
emily | it could theoretically automount both but yes (although frankly that whole automounting business feels a little overengineered to me anyway) | 11:49:51 |
emily | but that's only relevant when you have both | 11:49:58 |
emily | I'm just saying that the rec is technically /efi with only the ESP even though it'd weird and people would say it's weird :) | 11:50:36 |
emily | *it's | 11:51:03 |
aloisw | In reply to @elvishjerricco:matrix.org oh also, it should also be noted why you want ESP at /efi and not /boot/efi. You can't have /boot as an automount if the ESP is at /boot/efi. It'll have to be always mounted instead of on demand. To be clear I was confused about /boot vs /efi for the ESP. | 11:52:14 |
emily | well when you have ESP + XBOOTLDR it's /efi and /boot | 11:53:27 |
ElvishJerricco | In reply to @emilazy:matrix.org it could theoretically automount both but yes (although frankly that whole automounting business feels a little overengineered to me anyway) Well my point was that nested automounts don't work. The need for the child dir to exist forces the parent to be actually mounted | 11:54:02 |
emily | it's just that the behaviour and statements from systemd upstream imply that ESP on /boot when you have no XBOOTLDR is legacy compat behaviour | 11:54:04 |
aloisw | So the new recommendation is basically to have the ESP at the same place regardless of whether you have XBOOTLDR, understood. | 11:54:15 |
emily | it's actually as old as the initial recs around XBOOTLDR on UEFI | 11:54:33 |
emily | people just didn't read the blog post | 11:54:39 |