Description of problem: Not sure that it is fixable. gfortran is still immature and lacking g77 features. So there is still a need to compile with g77. Now i am trying to link statically with lapack, using g77 as a linker, but it fails. Here is what I get: [dumas@nor75-15-82-67-190-22 examples]$ cat call_sgemm.f program main call sgemm end [dumas@nor75-15-82-67-190-22 examples]$ g77 -static -o call_sgemm -g -O2 call_sgemm.f -lblas /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/../../../libblas.a(xerbla.o): In function `xerbla_': undefined reference to `_gfortran_st_write' /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/../../../libblas.a(xerbla.o): In function `xerbla_': undefined reference to `_gfortran_transfer_character' /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/../../../libblas.a(xerbla.o): In function `xerbla_': undefined reference to `_gfortran_transfer_integer' /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/../../../libblas.a(xerbla.o): In function `xerbla_': undefined reference to `_gfortran_st_write_done' /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/../../../libblas.a(xerbla.o): In function `xerbla_': undefined reference to `_gfortran_stop_numeric' collect2: ld returned 1 exit status It works when called with -lgfortran: g77 -static -o call_sgemm -g -O2 call_sgemm.f -lblas -L /usr/lib/gcc/i386-redhat-linux/4.1.0/ -lgfortran Version-Release number of selected component (if applicable): How reproducible: always Steps to Reproduce: 1. link statically with g77 without an expplicit call to -lgfortran 2. 3. Actual results: link failure. Expected results: No link error. Additional info: This is one of many issues with g77/gfortran incompatibilities issues. The ABI is different, too, but it doesn't affect lapack because it only affect functions.
i'm not sure how this is resolvable. g77 is dead branch code, instead of trying to use it, I recommend that you file bugs against gfortran where it does not provide needed feature functionality.
There is allready a meta bug with g77 regressions. There aren't that much move on the intrinsic front... http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19292 but the gfortran team isn't very big. Given that fact their solution is to keep g77 in distributions. But if it isn't usable... For example you can have a look on what kind of 'discussion' this leads to: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13082 I have reported the lack of intrinsics that prevent cernlib to build on fedora in the fedora bugzilla, but I was (rightfully) redirected to upstream.
Erm, why is the Component for this bug set to aalib? That can't be right.
It should be on lapack or blas. I'll try to fix it.