The issue is that a deboostrap for hurd as done bu pbuilder sets up
individual translators on /dev/* and /servers/* just before
"shadowing" them by setting up "firmlinks" on /dev and /servers. This
results in a filesystem with active translators that prevents it being
clean up, which for instance breaks the pbuilder create step.

There is an attempt in pbuilder to fix this, but it now fails as some
"new" (wrt the pbuilder fix) translator has been introduced.

The attached patch for debootstrap aims to resolve the issue
completely by avoiding setting up the translators that get shadowed.
After implementing this patch, the pbuilder fix may or should be
removed.

Ralph

Hello,

Thanks for looking at this!

Ralph Ronnquist, le ven. 01 mai 2026 18:11:18 +1000, a ecrit:

Mmm, in my memory this is needed in the d-i case, which needs to setup
the translators in the installed system before firmlinking the installer
directories, otherwise at reboot you do not have any translator set up.

So we need to know whether the chroot will always be used as a mere
chroot, or as a standalone system.

Why is that needed exactly?

(We don't really want a dummy directory in socket/, at least)

Which new translator was introduced? We could generalize what is done
there.

Samuel

Samuel Thibault, le mer. 06 mai 2026 01:32:29 +0200, a ecrit:

Ah, acpi?

I have done so in
https://salsa.debian.org/pbuilder-team/pbuilder/-/merge_requests/43

to at least fix the workaround while fixing debootstrap in the pbuilder
case without breaking other use cases.

Samuel