API with and without null-terminated strings

[MakisH]

Closes #1, implementing (to the extent I understand) the suggestions of @ivan-pi.

I also tried to make the length an optional argument, to maintain compatibility with user code. I am also trying to avoid touching the preCICE C/Fortran bindings for now, as we just had a breaking release.

Currently, building the (unmodified) solverdummy.f90 leads to the following error:

fortran-module/examples/solverdummy [nullterminated]$ make
gfortran -g solverdummy.f90 -o solverdummy -I../.. -L/home/gc/repos/precice/precice/build -lprecice
/usr/bin/ld: /tmp/ccQ2PSRq.o: in function `MAIN__':
/home/gc/repos/precice/fortran-module/examples/solverdummy/solverdummy.f90:38: undefined reference to `__precice_MOD_precicef_create'

Any clue what I am doing wrong?

[MakisH]

Together with @ivan-pi, we took now a different approach:

Instead of differentiating between null- and non-null-terminated strings in the C interface, we just adopt the de facto Fortran calling convention, as already done in the core library.

We dropped the bind(c) attribute. The way we now write it, automatically appends a trailing _ in each function name in the binary.

We also removed the dimension arguments present in the bind(c) version, and replaced them with len=* arguments. This enables the hidden length argument added by the compiler.

Important here would be that this would need special handling on Windows Fortran compilers, which adopt a different convention. We would need to specify FortranCInterface in the core library CMakeLists.txt, to automatically detect the right convention (add an underscore or not). This would assume the presence of a Fortran compiler.

We modified the precicef_create and improved the style of the example, I will adapt the rest of the files and ask for a review.