gmp 4.2 fails tests on G5

[Jack Howarth]

Torbjorn,
    Actually if you had skimmed through the acinclude.m4 in gmp 4.2 you would
have noticed the following comments...

dnl  GMP_ASM_POWERPC_PIC_ALWAYS
dnl  --------------------------
dnl  Determine whether PIC is the default compiler output.
dnl
dnl  SVR4 style "foo at ha" addressing is interpreted as non-PIC, and anything
dnl  else is assumed to require PIC always (Darwin or AIX).  SVR4 is the
dnl  only non-PIC addressing syntax the asm files have at the moment anyway.
dnl
dnl  Libtool does this by taking "*-*-aix* | *-*-darwin* | *-*-rhapsody*" to
dnl  mean PIC always, but it seems more reliable to grep the compiler
dnl  output.
dnl
dnl The next paragraph is untrue for Tiger.  Was it ever true?  For tiger,
dnl "cc -fast" makes non-PIC the default (and the binaries do run).
dnl  On Darwin "cc -static" is non-PIC with syntax "ha16(_foo)", but that's
dnl  apparently only for use in the kernel, which we're not attempting to
dnl  target at the moment, so don't look for that.

Its clearly documented right there that on Tiger at least it unreliable
to assume PIC code for 'cc -fast'. This issue shouldn't be a surprise at
all!
                         Jack