Re: gmp 4.2 failures on the G5

[howarth at bromo dot msbb dot uc dot edu (Jack Howarth)]

Steve,
   I reported this to the GMP developers. I got the following comments back...

-------------------------------------------------------------------------------
howarth@bromo.msbb.uc.edu (Jack Howarth) writes:

      Has GMP 4.2 actually been tested on a G5?

A G5 is running GMP nightly tests.

  In attempting to build
  fink packages for gmp 4.2, we discovered that while GMP 4.2 can be buil=
t
  and passes its testsuite on a G4 there are many testsuite failures when
  building and testing on a G5.

Good, thus you're warned about that your compiler is buggy.

  Specifically the problem shows up when the
  configure option --enable-cxx is used. Without that option, the build p=
asses
  its make check on a G5.

You make it sound like a general meltdown happens with --enable-cxx,
but I assume just the C++ tests fails?

For me too, the C++ stuff fails when using Apple's compilers.  I
isolate perhaps one compiler bug per month; unfortunately GMP runs
into about 5 compiler bugs per month.  This number increases rapidly
with time.

I have not isolated this bug, and I do not intend to isolate more bugs
in Apple's compilers.

Please feel free to isolate this problem and report it to Apple.  If
you have a clean workaround for GMP, I'll consider putting it in.

--=20
Torbj=F6rn
----------------------------------------------------------------------------------

So it doesn't look like this issue will be addressed anytime soon.
                            Jack
ps In answer to his question, I see failures in the very first set of
GMP tests run. For example, t-bswap and t-count_zeros as well as others
fail in the initial check-TESTS. I don't have access to my G5 at the moment
to cite more of the failures.