[tlbuild] Testing cxx-runtime-hack -- V 1.1

[Peter Breitenlohner]

On Mon, 26 Jan 2009, Thomas A. Schmitz wrote:

> Peter, one little remark to the little table you have compiled: linux-ppc is 
> in fact == powerpc-linux; I think Karl renamed it to that. And I remember 
> why, and that is something that we/you should think about when you ship 
> TL2009: my system identifies itself as ppc64. The processor is indeed a 
> ppc64, but as far as I know, the complete userland is 32-bit. There are only 
> a few linux distros that actually offer ppc-versions, but it's unlikely that 
> any of them provide 64-bit binaries; most of them will (like my SuSE 
> installation) just provide a 64-kernel. Which means that the install script 
> should recognize ppc64 and then suggest installing powerpc-linux-binaries. I 
> installed texlive 2008 on this box a few days ago, and even though I 
> specifically checked the box, no binaries were installed; I had to download 
> them manually (and they work fine).

Hi Thomas,

that is very similar to our situation with lots of x86_64 CPUs plus a few
i686.  All systems have a 32-bit kernel, a complete 32-bit userspace
(including gcc and binutils, capable to produce 32-bit or 64-bit code with
32-bit the default) plus a 64-bit kernel and just a few 64-bit libraries
(required by 64-bit Mathematica and Maple).

Thus the x86_64 CPUs can be booted with the 64-bit (default) or 32-bit
kernel.

If I want to compile a package for 32-bit I usually say
 	uname -m		=> x86_64
 	linux32
 		uname -m	=> i686
 		configure
 		make
 		install
 	exit
and I suppose the same should work for ppc64/ppc.

On the other hand, as long as there is no powerpc64-linux binary
distribution, the TL install script probably should automatically translate
linux-ppc64 => powerpc-linux.

Regards
Peter Breitenlohner <peb at mppmu.mpg.de>