On Fri, Feb 24, 2006 at 10:02:52AM +0100, Gilles Dauphin wrote:
> > > because 'generic' solaris MACHINE_ARCH is i386 and don't take care of ABI.
> > > Second, x86_64 Solaris MACHINE_ARCH does not exist in pkgsrc.
> > > maybe it is also true for other system with bi-arch (32/64) ABI.
> > 
> > Solaris platform files should fix up the MACHINE_ARCH in that case.
> > A lot of packages will be broken by this because everyone else does not
> > use i386 for the 64bit ABI.
> 
> If it does not use i386 for 64 bit ABI, it fall into the elif part
> and use the correct x86 assembler (yasm). Is it wrong? 

That's not the problem. I object this because it means we have to
special case *every* i386/x86_64 decision in pkgsrc for the sake of
Solaris, when it could as well just provide a sensible MACHINE_ARCH.

I believe that you will see this kind of problems all over the place, as
soon as assembly is concerned, when most programs already support either
amd64 or x86_64 correctly.

Beside that, the patch is incorrect in that assumes that ABI is defined
at all, which is not the case on almost any platform pkgsrc supports.

Joerg