configure.build_arch

[Toby Peterson]

On Tue, Sep 15, 2009 at 17:25, Jack Howarth <howarth at bromo.med.uc.edu> wrote:
> On Tue, Sep 15, 2009 at 05:18:32PM -0700, Toby Peterson wrote:
>> On Tue, Sep 15, 2009 at 17:05, Mike Alexander <mta at umich.edu> wrote:
>> > --On September 15, 2009 4:01:43 PM -0700 Toby Peterson <toby at macports.org>
>> > wrote:
>> >
>> >> For the few ports that actually care, only --build makes any sense. We
>> >> don't support cross-compiling, and very few configure scripts get it
>> >> right anyway.
>> >
>> > I think Jack's point is that in 10.6 on a 32 bit kernel you effectively are
>> > cross-compiling whether you want to or not.  You're building 64 bit binaries
>> > on a system that claims to be a 32 bit system.  That seems like
>> > cross-compiling to me.  Am I misunderstanding something?
>>
>> As I've noted numerous times now, config.guess still reports i386 on
>> K64, so this perceived problem exists regardless of the kernel
>> architecture.
>>
>> - Toby
>
>   As I already admitted, I was in error about uname -p reporting x86_64
> on the 64-bit kernel. However that is completely orthogonal to the issue
> of "configure.build_arch x86_64" passing -m64 on the CFLAGS while neglecting
> to clue in configure that "uname -p" is wrong by passing "--target=x86_64-apple-darwin10"
> to it. You still have the situation of configure believing the host is behaving
> as a 32-bit system while actually the compiler is generating 64-bit code.
> You are missing the forest for the trees here.

To continue the somewhat appropriate metaphor, you're missing the
trees for the forest. The vast forest of ports couldn't care less what
config.guess says, because we're either building with multiple arch
flags (where we have to work around the inadequacies of autotools), or
simply because the detected architecture has no actual effect on the
build.

Fix the lonely trees where the detected architecture affects things,
but stop trying to burn down the entire forest just to keep a few
sickly trees alive.

- Toby