gmp 4.3.0 failures on Intel Darwin9

[Emmanuel Thomé]

On Thu, Apr 23, 2009 at 08:55:13AM -0400, Jack Howarth wrote:
> On Thu, Apr 23, 2009 at 07:32:18AM +0200, Torbjorn Granlund wrote:
> > Jack Howarth <howarth at bromo.med.uc.edu> writes:
> > 
> >      I am seeing four segfaults during the testsuite of gmp 4.3.0
> >   when built on Mac OS X 10.5.6 using the following patch (which 
> >   the fink gmp-4.2.4-1000 package currently uses)...
> >   
> >   --- orig/configure.orig 2008-04-01 08:22:25.000000000 +0200
> >   +++ patched/configure   2008-04-01 10:00:15.000000000 +0200
> >   @@ -3454,6 +3454,10 @@
> >        # duplicate anything.
> >        #
> >        gcc_cflags_optlist="cpu arch"
> >   +    case $host in
> >   +          pentium3-apple-darwin*)
> >   +            host_cpu="core2" ;;
> >   +    esac  
> >        case $host_cpu in
> >          i386*)
> >            gcc_cflags_cpu="-mtune=i386 -mcpu=i386 -m386"
> >   
> > That is a very strange patch.
> > 
> > If config.guess returns the wrong cpu type, config.guess ought to be
> > fixed.  Patching configure is never right, since it is aa generated
> > file.
> > 
> > If you have problems with an unpatched GMP release, please let us know.
> > The information we need is described here:
> > http://gmplib.org/manual/Reporting-Bugs.html
> > 
> > 
> > 
> > -- 
> > Torbjörn
> 
> Torbjörn,
>     I see the same problem without the patch. The stock Apple gcc 4.0.1
> compiler doesn't understand the -mtune=core2 option so it ends up using
> -mtune=k8. The Apple gcc-4.2.1 compiler does and configure ends up using
> -mtune=core2. However both builds yield the same set of testsuite failures
> (none of which exist in gmp 4.2.1). Do you have access to a Intel based
> Mac to test gmp 4.3.0 directly?
>                  Jack

Disclaimer: I know nothing about macs.

Yet I have acess to a machine which says this:
Macintosh ~ $ uname -a
Darwin Macintosh.home 9.6.0 Darwin Kernel Version 9.6.0: Mon Nov 24 17:37:00 PST 2008; root:xnu-1228.9.59~1/RELEASE_I386 i386

And the stock gcc identifies itself as ``gcc version 4.0.1 (Apple Inc.  build 5465)''. Don't ask me to which xcode version it corresponds, I just don't know.

The cpu being ``Intel(R) Core(TM)2 Duo CPU     T7300  @ 2.00GHz'' (as per
cpuid).

I've tested several things which pass:

- gmp-4.3.0's ./config.guess says core2-apple-darwin9.6.0
- compiling gmp-4.3.0 with no configure options and the stock compiler
  passes all tests.
- compiling a non-bastardized gcc-4.3.2 with gmp-4.3.0 and mpfr-2.4.1
  passes (untar them in the source tree, and make symlinks. gcc
  recognizes this).
- compiling gmp-4.3.0 with the freshly compiled gcc-4.3.2 succeeds and
  also passes all tests.

Some other stuff does not pass, namely configuring a dynamic library with
ABI=32, with either compiler. The linker complains with ``ld: in
mpn/.libs/add_n.o, in section __TEXT,__text reloc 1: bad pc-rel vanilla
relocation length'', which urges me to check with the normal binutils.


So I suggest that you investigate your setup in more detail, and provide
possibly more appropriate information. In particular, is your OS as
up-to-date as your xcode package ? (Does ``Mac OS X 10.5.6 imply that
uname returns 9.5.6 ? If so, then it's not quite the same setup as me.'')

To start with, your config.status returning pentium3-* is just weird.

E.