[petsc-dev] new Fortran function pointer stuff broken

[Jed Brown]

This looks like my fault, I'll fix it.


On Wed, Feb 6, 2013 at 10:02 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:

>
> For example
>
> static PetscErrorCode oursnesfunction(SNES snes,Vec x,Vec f,void *ctx)
> {
>
> PetscObjectUseFortranCallback(snes,_cb.function,(SNES*,Vec*,Vec*,void*,PetscErrorCode*),(&snes,&x,&f,_ctx,&ierr));
>   return 0;
> }
>
> with
>
> #define PetscObjectAllocateFortranPointers(obj,N) do {                  \
>     if (!((PetscObject)(obj))->fortran_func_pointers) {                 \
>       *ierr =
> PetscMalloc((N)*sizeof(void(*)(void)),&((PetscObject)(obj))->fortran_func_pointers);if
> (*ierr) return; \
>       *ierr =
> PetscMemzero(((PetscObject)(obj))->fortran_func_pointers,(N)*sizeof(void(*)(void)));if
> (*ierr) return; \
>       ((PetscObject)obj)->num_fortran_func_pointers = (N);              \
>     }                                                                   \
>   } while (0)
>
> /* Entire function body, _ctx is a "special" variable that can be passed
> along */
> #define PetscObjectUseFortranCallback_Private(obj,cid,types,args,cbclass)
> { \
>     PetscErrorCode ierr;                                                \
>     void (PETSC_STDCALL *func) types,*_ctx;                             \
>     PetscFunctionBegin;                                                 \
>     ierr =
> PetscObjectGetFortranCallback((PetscObject)(obj),(cbclass),(cid),(PetscVoidFunction*)&func,&_ctx);CHKERRQ(ierr);
> \
>     (*func)args;CHKERRQ(ierr);                                          \
>     PetscFunctionReturn(0);                                             \
>   }
> #define PetscObjectUseFortranCallback(obj,cid,types,args)
> PetscObjectUseFortranCallback_Private(obj,cid,types,args,PETSC_FORTRAN_CALLBACK_CLASS)
> #define PetscObjectUseFortranCallbackSubType(obj,cid,types,args)
> PetscObjectUseFortranCallback_Private(obj,cid,types,args,PETSC_FORTRAN_CALLBACK_SUBTYPE)
>
> But any function with PetscFunctionBegin/Return needs to have
>
> #undef __FUNCT__
> #define __FUNCT__
>
> so most of the fortran stubs are broken with error messages about wrong
> function names.
>
> Does anyone test before they push anymore :-)?
>
>   Barry
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130206/4170fc55/attachment.html>