PPC/PPC64 builders not working?

[David Woodhouse]

On Tue, 2008-06-17 at 11:11 -0400, Jesse Keating wrote:
> On Tue, 2008-06-17 at 16:00 +0100, David Woodhouse wrote:
> > 
> > Just taking a moment to add 'strace -f' to the doxygen invocation in the
> > libburn specfile shows interesting behaviour...
> > 
> > [pid  1268] execve("/usr/bin/dot", ["dot"..., "libburn_8h__incl.dot"..., "-Tpng"..., "-o"..., "libburn_8h__incl.png"...], [/* 31 vars */]
> >  ....
> > [pid  1268] --- SIGILL (Illegal instruction) @ 0 (0) ---
> > [pid  1268] rt_sigreturn(0x4)           = 0
> > [pid  1268] --- SIGILL (Illegal instruction) @ 0 (0) ---
> > [pid  1268] rt_sigreturn(0x4)           = 0
> > [pid  1268] --- SIGILL (Illegal instruction) @ 0 (0) ---
> > 
> > Did anyone even bother to file that bug yet?
> > 
> > If not, how do you expect it to get fixed?
> > 
> > I'm used to having to explain the basic mechanisms of "first you tell us
> > about the bug, and only _then_ can we fix it" to the Great Unwashed on
> > #fedora, but I kind of expect better from you lot :)
> 
> To be fair, it likely requires a ppc host to be able to investigate the
> problem to this level, and that's just not something everybody has
> sitting on their desk.

I can't reproduce it on any of my machines -- the above is from adding
'strace -f' to the specfile, and trying a scratch build.

I suspect it's either 64KiB pages, or something trying to use Altivec
when it shouldn't. Probably the latter, given the nature of the problem.

Will try to reproduce on the POWER5, although I think one of the routers
between me and it is dead so I can't reach it at the moment.

-- 
dwmw2