Make C prototypes

[MakisH]

With this PR, make prototype will generate a precice-prototype.h file. This includes warnings for non-interoperable types, as, for example:

void precicef_set_vertex_ (char *meshname, float *position /* WARNING: non-interoperable KIND */ , float *vertexid /* WARNING: non-interoperable KIND */ , int meshnamelength);

Not sure where the float comes from, I don't see it in any of the involved files:

• precice.f90
• precice/precice/extras/bindings/fortran/
• precice/precice/extras/bindings/c/

Applies https://fortran-lang.discourse.group/t/compilers-supporting-generation-of-c-function-prototypes/2688 suggested by @ivan-pi.

Not sure at the moment what would be the best way to use the generated prototype to detect inconsistencies. I need more time to investigate, which I currently do not have.

When complete, it should close #15.

[ivan-pi]

There is an interface mismatch:

    subroutine precicef_set_vertex(meshName, position, vertexID, meshNameLength) &
      &  bind(c, name='precicef_set_vertex_')

      use, intrinsic :: iso_c_binding
      character(kind=c_char), dimension(*) :: meshName
      real(kind=c_double) :: coordinates(3)             !<--- this should be named position
      integer(kind=c_int) :: id                         !<--- this should be named vertexID
      integer(kind=c_int), value :: meshNameLength
    end subroutine precicef_set_vertex

Because of Fortran language rules (...) the implicit none at module level, does not apply to interface blocks. Hence, the compiler assumed the variable position is of type real, and same for vertexID, per the default implicit typing rules.

With gfortran you can use -fimplicit-none to detect these errors: https://godbolt.org/z/5brssxEz4

[ivan-pi]

To detect inconsistencies, you would need to compile preciceFortran.cpp, but replace the true header precice/preciceFortran.hpp with the gfortran-generated one. ( Technically, this header is not needed it all; it shouldn't be used by any C or C++ users. )

Afterward, if you compile with -Wbuiltin-declaration-mismatch it should flag any inconsistencies, between the Fortran bind(c) interfaces, and the extern "C" procedures.

Here's an example of what happens: https://godbolt.org/z/EMWPExjjn. It appears like you need to have extern "C" on both the declaration and definition of the procedure for this to work.

This means that either the CI in this repository should check-out the precice trunk, or the precice trunk checks out this repository, and does the consistency check.

[MakisH]

There is an interface mismatch:

Hm... I did not think of checking the names of the variables... 😅

Because of Fortran language rules (...) the implicit none at module level, does not apply to interface blocks.

That's very useful information, thank you!

With gfortran you can use -fimplicit-none to detect these errors: https://godbolt.org/z/5brssxEz4

Very good suggestion, I added that to our Makefiles, thanks!

It appears like you need to have extern "C" on both the declaration and definition of the procedure for this to work.

I will then update the bindings in preCICE itself as well, but then this will need to wait.

(context: we are preparing releases of several components this week, I will continue with anything that needs changes in preCICE after that)

[ivan-pi]

Discussion summary:

• preciceFortran.hpp (can be removed to prevent accidental misuse)
• current Fortran bindings (with compiler-specific name mangling) should remain
  for backward compatibility
• the Fortran module can be rebased on top of the C bindings
  • gfortran header <-> preciceC.hpp
• new object-oriented bindings (F2008/F2018) (needs more work and testing)

[MakisH]

We could also use -fc-prototypes-external -fsyntax-only for checking prototypes not declared with bind(c).

However, this only works for external functions (thus, not here, because we are in a module).