Re: Update of math/R in pkgsrc

[coypu%sdf.org@localhost]

There are some issues with gfortran that do not exist with g95.
the lang/gcc* packages install libgcc_s.so and libstdc++.so, and add it
to RPATH.

Having mixed libstdc++ is non-ideal, at the very least we want any
resulting binaries as well as their dependencies to pick the newer
libstdc++ of the bunch.

Otherwise, we may have issues, some of which are hopefully visible at
compile time, but no guarantee of this - they may be runtime failures.

For fortran only, I suspect defaulting to the highest GFORTRAN_VERSION
or close will eliminate most of the issues, but there's a lot of 'maybe'
and 'we shouldn't be doing this'. If a non-fortran package depends on a
fortran one, we run into the same issue of an older libstdc++ being
used.

Additionally I had an issue with octave's configure which seemed to
invoke gcc rather than gfortran for mixing C and fortran, and my gcc
doesn't have fortran support, but disabling the configure test made it
work.

Everything is subpar, as is using an old fortran implementation which is
showing its age. I've proposed to always use the newest compiler
available, and tie gfortran with gcc, which covers almost everything,
but that will run us into other issues. recently gcc started defaulting
to C++11/C11, which e.g. for SunOS introduced a lot of build failures as
it disallows mixing _XOPEN_SOURCE=500 with C99 or newer, so it
introduced some build failures.

I'd like to switch to gfortran and GFORTRAN_VERSION=6. I think almost
all fortran packages are leaf ones or their dependencies are fortran
packages themselves, and the ones that aren't can be forced to continue
using g95.