ppc buildserver problem

[Dan Williams]

On Tue, 2005-08-16 at 16:58 +0200, Joost van der Sluis wrote:
> > > the ppc buildsystem fails in compiling fpc. The pre-compiled binary
> > > gives a segfault. I asked someone on IRC to try to build the fpc-rpm on
> > > a old ppc, and there it works. (He used Debian Unstable, though)
> > > 
> > > Can someone give me some more information about what kind of powerpc the
> > > buildsystem has?
> > 
> > Not a PowerPC at all :)  It's got a couple of these:
> > 
> > cpu             : POWER5 (gr)
> > clock           : 1655.992000MHz
> 
> > If you've got altivec or something like that compiled in, you're hosed.
> > What's using a precompiled binary anyway?  That seems wrong, for the
> > exact reasons this job failed.
> 
> fpc is a compiler. It has to be build (bootstrapped) by itself, or an
> older version of itself. So the first time it needs a 'seed' to get it
> into the buildsystem. That's why there is a binary.
> 
> But according to some other fpc-developers fpc doesn't run on a Power5
> machine. Not only for the lack of Altivec (VMX) but also for some other
> things. (you also can't compile a version which does)
> 
> And the changes that we'll be able to fix that are very slim, since
> those machines aren't cheap...
> 
> > In any case, this is an interesting topic, since the build machines,
> > while blazingly fast, don't support Altivec, which most Fedora PPC users
> > have.  I'm not sure how to even differentiate between those jobs needing
> > Altivec (so they might go to a capable machine) and those jobs not
> > requiring it (so they could go to the POWER5).
> 
> Well, fpc _does_ run on powerpc, but not on power. How should I handle
> this?

Not sure.  There is _no_ chance that the upstream devs could fix what's
wrong on POWER, or at least work around it?  Does it actually require
Altivec, or is it the stuff like cache line instructions that are
missing?

Dan