Do we really need to still protect and segregate the complex stuff from petscmath.h? You want to jump straight to MPI 3 but we can’t use complex? If it is just hypre then get hypre fixed.
Barry On Jul 28, 2014, at 12:53 PM, Jed Brown <[email protected]> wrote: > Barry Smith <[email protected]> writes: > >> Where petsccomplex.h simply defines PETSC_DESIRE_COMPLEX? > > I'd rather have it contain the code specific to complex (so no > order-dependence). When scalar-type=complex, petscmath.h would include > petsccomplex.h. > >> And what happens if petsccomplex.h is included after petscsys.h it will >> have no affect? Should it generate an error if not defined first? >> >> So petsccomplex.h in pseudo code would be >> >> #if defined(__PETSCMATH_H) >> #error must include petsccomplex.h BEFORE other petsc include files >> #endif >> #define PETSC_DESIRE_COMPLEX >> >> Seems reasonable to me. >> >> Barry
