Ok - perhaps we can simulate current behavior with: typedef char* VecType; typedef const char* ConstVecType;
and use VecType/ConstVecType as needed in code? Satish On Wed, 27 Oct 2010, Barry Smith wrote: > > Satish, > > We already had this fight and this change lost. Making this change means > that one can never build a VecType on the fly, so all VecType must be > declared at compile time which defeats the entire purpose of PETSc. > > I don't understand why "const" is scoped differently for a macro than for > the typedef." but it seems to me an annoying feature of C. > > Barry > > > On Oct 27, 2010, at 5:08 PM, Satish Balay wrote: > > > On Wed, 27 Oct 2010, Barry Smith wrote: > > > >> > >> On Oct 27, 2010, at 4:48 PM, Jed Brown wrote: > >> > >>> On Wed, Oct 27, 2010 at 17:45, Barry Smith <bsmith at mcs.anl.gov> wrote: > >>> hmm, is there any reason we can't/shouldn't use a typedef for our VecType > >>> (and its many friends)? > >>> > >>> I think the issue is because "const" is scoped differently for a macro > >>> than for the typedef. > >>> > >>> typedef char *Foo; > >>> > >>> now "const Foo" means "char *const" instead of "const char *". > >> > >> Damn, damn, damn, damn, > > > > -#define VecType char* > > +typedef const char* VecType; > > > > -EXTERN PetscErrorCode PETSCVEC_DLLEXPORT VecGetType(Vec, const VecType *); > > +EXTERN PetscErrorCode PETSCVEC_DLLEXPORT VecGetType(Vec, VecType *); > > > > The above type of change [i.e 'const VecType' to 'VecType'] gets > > src/vec/vec/interface sources compiled.. > > > > Satish > >
