Re: "Universal Binaries"

[David Fang]

> It seems that Apple's OSX has introduced something known as a
> "Universal Binary" in which the compiler conveniently compiles for
> both PowerPC and Intel in one whack.  Libraries and executables then
> automatically "work" for both architectures.
>
> Does anyone else see that this defeats the purpose of autoconf and
> will likely result in broken programs and libraries since the results
> of some autoconf tests are only applicable to the CPU type on which
> configure executes?

Hi,
        For discussion reference, the recipe(s) for manually configuring a
universal binary can be found here:

http://developer.apple.com/technotes/tn2005/tn2137.html

        There you can see a few ways of achieving the same result.  To get
around the problem of autoconf tests being applicable to one CPU, one
might be better off doing two single-arch builds (VPATH), and merging the
results with lipo.
        I haven't tried this myself, but some others have reported
success.  I'm not sure why this 'defeats' autoconf.  It does introduce a
challenge to possibly add some support macros to facilitate this.  For
example, what would be the ${exec_prefix}?  Arbitrarily choose one arch
and provide symlinks from the other?  libtool might also need some
revisions to automatically detect when one is building universal objects,
perhaps additional tagging?
        I, for one, would be very interested in how to cleanly accommodate
universal libs/bins using the GNU autotools.  I'd be grateful of some
documentation (manual section) came out of this thread.

David Fang