configure.build_arch

[Toby Peterson]

On Wed, Sep 16, 2009 at 06:41, Jack Howarth <howarth at bromo.med.uc.edu> wrote:
> On Wed, Sep 16, 2009 at 09:00:46AM +0200, Anders F Björklund wrote:
>> Jack Howarth wrote:
>>
>>>    Reading through the commits on the configure.build_arch feature
>>> in port, I am rather surprised that invoking that doesn't also
>>> set the triplets for configure as well as passing the -m32 or
>>> -m64 compiler flags. Wouldn't you at least want to set
>>> --target= to the appropriate value so that configure understood
>>> what architecture it was going to build for? I suppose such a
>>> change might be an option to patching config.guess for those
>>> Portfiles that use configure.build_arch. This would make sure
>>> that configure was given --target=x86_64-apple-darwin9/10
>>> when -m64 (or -arch x86_64?) was passed through the invoking
>>> configure.build_arch.
>>
>> We set the triplets back when we were (ab)using +universal
>> to do cross-compilation, with the three universal_ flags:
>>
>> http://lists.macosforge.org/pipermail/macports-dev/2008-February/
>> 004358.html
>>
>> It never worked very well, so in MacPorts 1.8.0 it was all
>> ripped out in favor of the much simpler build_arch solution...
>>
>> --anders
>
> Anders,
>   My main concern is that the autoconf folks fully understand
> all of the details of darwin10 so that whatever change created
> upstream is as well thought out as possible.
>   I still think that anytime port (on at least darwin10) is
> passing -m64 or -arch x86_64 to configure, via build_arch, that
> we should at least pass --target=x86_64. Currently to configure,
> darwin10 acts effectively like a cross-compiler in that the detected triplet
> for --host is i386-apple-darwin10 but the actual code generation
> is x86_64-apple-darwin10. Simply making sure that the --target
> is set by build_arch, so that configure properly understands that actual
> code generation being used, should help a lot.
>           Jack
> ps We are starting to get reponses on autoconf mailing list so hopefully
> they cant arrive at a consensus on whether we should pass --target on
> darwin10 with the existing config.guess. I would note that Ben
> Elliston, the config.guess maintainer, is already on board with
> the concept of decoupling the output of config.guess from the
> kernel and instead on the code generation of the c compiler
> as defined by CC.

We will not be passing --target (or --build, for that matter), because
we're not cross compiling. When we do begin supporting cross
compiling, which is not in the foreseeable future, we can re-evaluate
this.

Even then, I don't see any reason to, because we've already spent
years working around configure results. Unless they completely
rearchitect, this isn't going to change.

- Toby