Re: Configuration bug in GMP 4.1.2

[Kevin Ryde]

"Krzysztof Kozminski" <address@hidden> writes:
>
> checking compiler gcc -g -O2 -m64 -mptr64 -Wa,-xarch=v9 -mcpu=v9 ... no

configure first tries to use ABI=64, and rejects gcc because it
doesn't support that.  I think gcc 3.2 is the first to do so.

> checking compiler cc -fast -fns=no -fsimple=1 -xarch=v9 ... yes

Still trying to use ABI=64, configure falls back to cc.

> checking what prefix to use for a local label... configure: WARNING:
> "/usr/local/gnu/bin/nm -B" failure

Hmm.  I guess binutils nm isn't 64-bit capable either.  Unfortunately
this is a case of needing a coherent set of tools.  We expect "nm" to
be able to process what "cc" produces.

Ah, but maybe we need to beef up AC_PROG_NM (from libtool) so it's run
on an object produced with our flags.  I've got a horrible suspicion
it's only run on /dev/null, which means we won't look further for a
better one in $PATH :-(.

> configure: error: cannot determine how to define a 32-bit word

Hmm.  Well, this is a consequence of not having a working nm, but I
suspect the gmp sparc64 code doesn't actually care about a 32-bit word
directive, so it's a rather spurious failure.


In any case, if you want to use gcc you can ask for it explicitly,

        ./configure CC=gcc

in this case configure won't look at cc, and will fallback to ABI=32
if that's all gcc can do.