Re: Possible c_ptr bug with gfortran 4.5.0 20090903 (experimental) [trunk revision 151392]

[Ian Rutt <I dot C dot Rutt at swansea dot ac dot uk>]

Hi Dennis,

Many thanks indeed for your quick response, and for kindly pointing out 
my error. That's certainly a bit of a gotcha for a C-Fortran interfacing 
novice like me, but one I shall remember in future! As you suggest, my 
code runs fine when I add the 'value' attribute - I agree that this is 
the preferable approach, as it keeps the C code idiomatic.

Thanks again,

Ian

Dennis Wassel wrote:
> Hi Ian,
>
> > I have had problems with passing pointers from C to Fortran and back again. This might be a bug, but of course I might be doing something silly.
>
> I skimmed your code and I think you are indeed, but I also think there
> is little shame in falling into this particular trap. Been there, done
> it myself!
>
> > Pointer location: 22040480
> > Value: 352
> > Pointer location (Fortran):             22040480
> > Pointer location: -47467344
> > Value: 22040480
>
> Clearly at some point, pointer and value get mixed up here.
> What the C caller does is to pass a copy of the pointer to the
> function, so the callee is supposed to receive the pointer by value --
> meaning the value of the pointer variable, not the value of the object
> it points to. Now re-read that sentence :-)
> The Fortran function OTOH expects to receive a c-pointer by
> reference*, i.e. the usual Fortran way of passing arguments (*by
> reference is not exactly true, but it behaves like that).
>
> You should be able to fix your code by either passing the address of
> the pointer variable (i.e. a pointer to the pointer) from C, or by
> adding the VALUE attribute to the c_ptr dummy argument in Fortran, so
> it expects a copy of the pointer variable, not a reference to it.
> Personally, I prefer the latter option.
>
> Cheers,
> Dennis