[Pkg-sysvinit-devel] Bug#784567: intent to NMU sysvinit really soon -- followup fix

[Andreas Henriksson]

Hello Petter Reinholdtsen.

On Sat, May 30, 2015 at 07:59:51AM +0200, Petter Reinholdtsen wrote:
> 
> I noticed something strange when I upgraded my sid chroot for the first
> time in a while.  I only used 'apt-get upgrade', and the util-linux and
> most sysvinit packages were held back:
> 
> The following packages have been kept back:
[...]
>   initscripts 
[...]
>   sysvinit-utils util-linux
[...]
> 
> But even then, I saw this unexpected message at the end of the upgrade:
> 
>   Running hooks in /etc/ca-certificates/update.d...
> 
>   /etc/ca-certificates/update.d/jks-keystore: 33:
>     /etc/ca-certificates/update.d/jks-keystore: mountpoint: not found
>   the keytool command requires a mounted proc fs (/proc).
>   E: /etc/ca-certificates/update.d/jks-keystore exited with code 1.
>   done.
> 
> I guess things will improve with a dist-upgrade, but this tell me the
> ordering of dependencies/breaks is not quite right.

Yes, this doesn't look good. Hard to guess exactly what the problem
might be without exact version numbers. Since initscripts was
one of the packages held back + mountpoint missing, you probably
already had upgraded to a version where initscripts no longer shipped
mountpoint.... mount should have provided it by then, but it was since
moved to util-linux which was held back and possibly mount was upgraded
to a version no longer shipping mountpoint... Thus you ended up
without mountpoint on your system.

Thats atleast the best explanation I can guess. This should thus only
affect those which are upgrading from the versions no longer available
in the archive. I think we don't have to worry too much about that,
even though it's ofcourse not very nice. Anyone fully upgrading
their system should get back to a working state.

For the longer term I think we should consider simplifying the dependencies
by turning them around. eg. Switch from Breaks to actual Depends in
sysvinit and from Depends to Breaks in util-linux.
See http://fatal.se/tmp/pkg-sysvinit/ for examples.

The main thing to consider here is the dropped initscripts dependency
in util-linux which I'm still pondering if it's correct.
The hwclock.sh init script in util-linux has an LSB dependency header
which needs initscripts installed. On the other hand, it's my understanding
that since (new installs of) Jessie you'll now need sysvinit-core installed
for a fully functional sysvinit - which depends on initscripts already.
(The problem with keeping util-linux depending on initscripts is
the circular dependency which apparently affects bootstrapping.)

Anyone who has opinions on how to best handle this situation is very
much welcome to add their insight. If anyone has plans on how the
sysvinit packaging is going to look when we comes closer to the
freeze time for stretch, that is something I'd also like to hear about.

Regards,
Andreas Henriksson