Re: sbuild on hurd-amd64...

Le 22/08/2025 à 22:24, Samuel Thibault a écrit :
>> Thank you. I've managed to make it work. It was not "easy" to do without
>> root privileges. I've documented it at:
>>
>> https://salsa.debian.org/glondu/hurd-notes/-/blob/master/rootless-subhurd.md
> 
> tar -C mnt: you shouldn't need it, you should be able to use the -d
> option of mke2fs. That'd probably avoid the chown as well.

Right. Thank you so much! I was not aware of this option...

Before I initially wrote this, I also tried with genext2fs (as 
mmdebstrap does), but could not make it work in my workflow. "mke2fs -d" 
does work here.

> Concerning the second stage inside the subhurd, does mmdebstrap not set
> up everything properly? Normally it should be able to, since xattr are
> now used by default for translators.

It does use xattr to set up translators inside the tarball, but they 
were not preserved by (non-root) "tar -x".

With "mke2fs -d", I do no longer need to set up manually the translator 
on /servers/socket/1... (nor the chown)

Concerning the "dpkg --install" step, some maintainer scripts fail when 
run under fakechroot/fakeroot. It makes more sense to me to run them 
inside the subhurd anyway.

> I don't think halt-hurd would stop the subhurd, use poweroff-hurd

One is a symlink to the other, isn't it?

Le 22/08/2025 à 22:35, Samuel Thibault a écrit :
> I'm surprised that you see it not seemingly unmount the ext2 filesystem.
> Do you get
> 
> startup: notifying ext2fs.static pseudo-root of halt...done

Well... I do see this message. But e2fsck does find irregularities in 
the filesystem, and the filesystem randomly switches to read-only in 
subsequent boots (even after replacing --readonly by --writable in the 
boot script, of after a "mount -o remount,rw /").

If I do "mount -o remount,ro /" explicitly, it complains about being 
busy: maybe there is something to dig here? Maybe this is related to not 
using a proper init (I use init=/bin/bash).

> Stéphane Glondu, le ven. 22 août 2025 16:18:10 +0200, a ecrit:
>> Le 18/08/2025 à 08:06, Stéphane Glondu a écrit :
>>> I confirm that I could run a regular debootstrap, with the "brown-tape
>>> modification" of /usr/share/debootstrap/functions.
>>
>> I've documented my findings at:
>>
>>    https://salsa.debian.org/glondu/hurd-notes/-/blob/master/debootstrap.md
> 
> It'd probably be useful to propose a patch for
> 
> https://salsa.debian.org/webmaster-team/webwml/-/blob/master/english/ports/hurd/hurd-devel-debian.wml

Indeed. I will continue to investigate to understand things better. One 
of my goals is to have a subhurd sbuild backend with an experience as 
smooth as the unshare backend on Linux.


Cheers,