sean yang wrote:
Is this aproblem for gcc 4.0.2 or gmp4.2.1?
Here I first list the numerical libraries.
objgcc>ls /usr/local/lib
-rw-r--r-- 1 root root 1180872 Jun 14 22:31 libmpfr.a
-rwxr-xr-x 1 root root 794 Jun 14 22:31 libmpfr.la
-rw-r--r-- 1 root root 503858 Jun 14 21:55 libgmp.a
-rwxr-xr-x 1 root root 783 Jun 14 21:55 libgmp.la
lrwxrwxrwx 1 root root 15 Jun 14 21:55 libgmp.so -> libgmp.so.3.4.1
lrwxrwxrwx 1 root root 15 Jun 14 21:55 libgmp.so.3 -> libgmp.so.3.4.1
-rwxr-xr-x 1 root root 241173 Jun 14 21:55 libgmp.so.3.4.1
Now, I list what I did for gcc4.0.2,
objgcc> ../gcc-4.0.2/configure --with-gmp=/usr/local
--with-mpfr=/usr/local --enable-languages=c,c++,f95 1>/dev/null
objgcc>make 1>/dev/null
...
/home/syang/software/objgcc/gcc/f951: symbol lookup error:
/home/syang/software/objgcc/gcc/f951: undefined symbol:
__gmp_get_memory_functions
make[2]: *** [selected_int_kind.lo] Error 1
Your --with-gmp and --with-mpfr options look strange to me. Did you find
reliable documentation somewhere showing they could be used this way?
For myself, I don't want any confusion about mpfr and gmp versions, so I
removed all old versions, both static and dynamic, before installing the
current ones.
I think gcc-4.0.2 is too old to have been tested with the current gmp.
I would infer that, if you want to use these versions, you shouldn't be
building fortran. configure --enable-languages='c c++' for example. You
should note the gfortran developers have disrecommended 4.0 strongly,
current released gcc is 4.1.1, and gfortran-4.2 is more reliable than
4.0, besides giving some advantages for the pain.
