Fortran-C++ interoperability: Passing array via void pointer

Question

I roughly have the following situation. I have a C++ function which is called from Fortran code and takes a function pointer and a void pointer as arguments like this

int STDCALL FORTRAN_NAME(CPPFunction, CPPFUNCTION)(
    int (*userFunction)(const int *object,
                        const void *userFunctionUserData),
    const void *userData)
{
  // ...
    int index;
    int result;
  // ...
    result = userFunction(&index, userData);
  // ...
}

This is called from Fortran like this

! ...
DOUBLE PRECISION, ALLOCATABLE :: data(:,:)
INTEGER :: n, result

! ...

ALLOCATE(data(3,n)); data = 0.0

! ... fill data with something

result = CPPFUNCTION(FORTRANFUNCTION, data)
! ...

The Fortran function which I want to pass via the function pointer looks like

INTEGER FUNCTION FORTRANFUNCTION(idx, data)
IMPLICIT NONE

INTEGER, INTENT(IN) :: idx
DOUBLE PRECISION, INTENT(IN) :: data(*)
INTEGER :: i, offset

offset = 3 * (idx - 1)
WRITE(*,*) 'data(offset + 1) = ', data(offset + 1)
WRITE(*,*) 'data(offset + 2) = ', data(offset + 2)
WRITE(*,*) 'data(offset + 3) = ', data(offset + 3)

END FUNCTION FORTRANFUNCTION

If I start the whole thing CPPFunction seems to be correctly called, it calls FORTRANFUNCTION and I get an exception code

c0000005 ACCESS_VIOLATION

exactly in the line in FORTRANFUNCTION where the first access to the array data is done.

WRITE(*,*) 'data(offset + 1) = ', data(offset + 1)

Can somebody tell me where my mistake is? I should also mention that the C++-function is not mine. It is not impossible to change it but it would affect lots of other code. The Fortan part is completely under my control and I can do what I want with it.

Thank you.

Gabionade

Shouldn't the first parameter be just int object? – Debidebilitate

I changed the tag to fortran to get you more attention, but, for the record, the OP originally requested Fortran 90. – Sorption

@Anton Savin I thought that since Fortran passes its variables by reference, I MUST give addresses in the calling C++-program, but this is the first time I use C++ and Fortran together, so I am not sure. – Gabionade

@Gabionade Yes, you should pass addresses unless the FORTRAN_NAME macro does something very strange. – Sorption

The calling convention for the top level C++ function appears to be set to stdcall, presumably to suit the calling convention of your Fortran compiler (which compiler?). I don't see a similar calling convention qualification for the calling convention of the function pointer argument in the C++ code. – Fetal

I have a very limited experience mixing C++ and fortran (I'm on the C++ side of a mixed project), and looking at my examples, I see we also use extern "C" { ... } and __stdcall to wrap all functions and global vars exposed to fortran code. I think the arguments you receive are ok, I don't know about the return value, but may be the error is related to calling conventions. – Magna

@Fetal The compiler is IFORT 11.1. – Gabionade

I'm building for x64 architecture, so the calling convention should be x64 calling convention, right? – Gabionade

Do you have to stick to F90? If not, I'd suggest to give the ISO_C_Binding from F2003 a try. – Streamway

I don't know any reason to set stdcall in X64 development, although it may make no difference (as it would in 32-bit mode). ifort has an option to change from default to stdcall. iso_c_binding obviously is preferable and covered by all maintained compilers, but I don't know that compatibility with void * is assured. Is void * really appropriate? – Bailsman