Control: severity -1 important
(adding the bug, and my work hat, to the CC)
Moritz Mühlenhoff writes ("Re: Xen jessie testing"):
> On Thu, May 11, 2017 at 11:04:21AM +0200, Axel Beckert wrote:
> > Moritz Mühlenhoff wrote:
> > > Ian's upload was built on i386, while the previous ones were built
> > > on amd64.
I built it on i386 because then it would produce packages I could at
least do some kind of test of it on my system at home. We could
rebuild it on amd64 and issue another update. However, I think if
something was going to break it would have done already.
> The severity seems exaggarated, though. If that's really relying
> on the build architecture, this would have been broken on i386
> all the time until the DSA. And while i386 is arguably not a good
> choice to run Xen, I still think this would have been noticed earlier
> if it were a severe issue.
The only place this seems to be used is to prepend it to
the LD_LIBRARY_PATH in force during execution of the hotplug scripts.
This is inherited from upstream, where it is needed (I think) so that
in non-packaged builds of Xen, on non-Debian systems, the Xen
libraries in /usr/local are found when trying to execute the Xen tools
inside the hotplug scripts.
Our dynamic linker is clever enough to ignore irrelevant files, so I
think this is largely harmless. Of course the noise in /etc ought to
be got rid of.
I think this should be fixed by not dropping this setting (and
probably the consequent LD_LIBRARY_PATH too). Ideally via some kind
of upstream knob but if not by a Debian patch.
I think such a change is buster material. For now, I suggest that I
continue to build security updates for jessie on i386 as I am able to
conveniently test that.
If I need to do another update to stretch (eg, a security update)
before its release, I will simply drop this line from this file.
Does that plan seem good ? Obviously, feel free to try to convince me
that this is RC for stretch. I wouldn't want to let this slide if
it's going to cause real trouble.
Thanks,
Ian.
