Segfaults on powerpc platform

[Panu Matilainen]

On Tue, 17 Jul 2007, Tim Verhoeven wrote:

> Another update on getting this going. I was able to finaly build the
> latest devel version (0.5.15lorg3.92) on RHEL4. Thanks for the fixes
> to be able to compile without sqlite.
>
> Now, this still being PPC means that most userland stuff is build als
> ppc, not as ppc64. This includes apt itself because certains libs it
> requires are not available in ppc64, only as ppc.
>
> This has a a result that the multilib detection does not work it seems
> as I get lot of duplicate package warning. So I guess the multilib
> detection routine needs a update. (Check uname for example).
> Anyway, to fix this I add :
>   Architecture "ppc64";
> to the apt.conf file. And now it runs in multilib mode :-)

Ah, of course... PPC has everything backwards :) The real issue is that 
apt's "base architecture" is determined build, not runtime. So when it's 
built as 32bit PPC, it thinks it'll be running on a non-multilib system. 
Doh. It's a longstanding todo-item to move the basearch determination to 
runtime instead (although like you noticed, it can be overridden in the 
configuration so at least there's a workaround).

> Then running apt-get -s update runs fine but returns this error :
>
> W: There are multiple versions of "kernel.64bit" in your system.
>
> This package won't be cleanly updated, unless you leave
> only one version. To leave multiple versions installed,
> you may remove that warning by setting the following
> option in your configuration file:
>
> RPM::Allow-Duplicated { "^kernel.64bit$"; };
>
> I have this in my apt.conf :
>
>        Allow-Duplicated {
>                "^kernel$";
>                "^kernel-";
>                "^gpg-pubkey$";
>        };
>
> So I don't understand why it would complain about duplicate kernels.
>
> Any ideas ?

Yeah... allow-duplicated and multilib-autorenaming don't mix very well. 
And you'd get wonderfully sick names like kernel.64bit#2.6.21-1.3228.fc7 
if it did. If you're lucky, adding "^kernel.64bit$" to allow-duplicated 
might even work these days but I haven't tested at all.

These kind of things kinda make it obvious why yum won the depsolver wars 
despite it's other shortcomings (such as being hideously slow) :)

The only real fix is to teach apt about multiple versions and archs being 
simultaneously installable, which unfortunately requires ripping apart the 
basic design of libapt-pkg from the ground up :-/

 	- Panu -