Hi, Pascal Hambourg <pascal@plouf.fr.eu.org> writes: > On 16/05/2025 at 16:49, Cyril Brulebois wrote: >> >> with a tentative netinst for amd64, I deployed a laptop, full >> disk, switching the / FS from the default ext4 to btrfs, and proceeding >> as always for the rest. >> >> Then I broke the boot by removing the initramfs. >> >> Then ran d-i in (graphical) rescue mode, picked the right partition (I >> had no idea which, but I could give it several tries), then executed a >> shell in the installed system, `update-initramfs -c -k all` which was >> required, and `update-grub` for peace of mind, probably not required. >> >> And the system boots again. >> >> I'm not going to try and compare that with earlier images, but that >> looks to be “not entirely broken”. If that confirms what people have >> been working on, yay. Yes :) Thanks for testing, since more eyes is better! > Indeed it does, thanks for the feedback. > > However it may not work with a system installed from Debian live with > Calamares which appears to set a different btrfs subvolume layout (see > #1104552). I guess repair is possible in a live session but probably not > as automated as rescue-mode (select root partition, mount proper > subvolume, mount /dev, /proc, /sys...) so would it be desirable to add > (trivial) support for this case in d-i RC2 ? I'm CCing Debian's Calamares maintainer, because I think this is an upstream Calamares bug. I hypothesise that the nature of the Calamares bug is that upstream assumes Ubuntu-style subvolume layout that we never intended to support in Debian. In case anyone missed the following reply to that bug: > Have you been able to track down those discussions where we decided on > @rootfs? One of the arguments against installing to '@' was that we're > letting Ubuntu claim that namespace, we're letting Fedora claim 'root' > (and rootfs), and we're staying out of their way. I was surprised to > learn that people use btrfs in this way, but it's not that much of a > stretch from using one VG to hold multiple distribution's LVs. > > Another topic in that (and subsequent) discussions is basically this: > > 1. Ubuntu never implemented subvolume creation, because they chose > ZFS. At some point their installer began to statically create @ and > @home. > 2. Fedora and SUSE implemented full support for any custom subvolume > topology. > 3. Due to Ubuntu's popularity, some developers exclusively support > Ubuntu's static nonconfigurable default as a kind of emergent > bug-for-bug API. > 4. This results in DFSG-free software having an Ubuntu-specific > implementation; DFSG-free software should also work on Debian, Fedora, > SUSE and everywhere else. It also result in utterly wasting the time > and effort the Fedora and SUSE developers took to implement an actual > solution rather than a stop-gap measure. > 5. Our users want our installer to have the same basic features as > Fedora and SUSE. > 6. Implementing this is a waste of time if our users will only use it > to make their Debian systems behave like Ubuntu so that > Ubuntu-specific software will work on their Debian systems. > 7. We need to support the effort for portable software. I think we need to fix the bug in Calamares and provide a trivial migration script to unbug affected Debian Live installations. Cheers, Nicholas
Attachment:
signature.asc
Description: PGP signature