[petsc-dev] (no subject)

[Barry Smith]

On Feb 16, 2013, at 5:14 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> 
> On Sat, Feb 16, 2013 at 5:02 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
>     No one would "classify" PETSC_HAVE_HYPRE as being relevant to any particular thing. Rather if you are writing code anywhere in PETSc (Mat, PC, whatever) and wished to use hypre you would have
> 
> #include <petsc-private/petscconfhypre.h>
> 
> Ah, this is finer-grained than I had imagined, but may work. Reconfiguring with a new package would typically only require rebuilding a few files, which is nice.

   Plus there is already code to figure out the dependencies.

> OTOH, this would create a zillion one-line files.

   Maybe only a few dozen. 
>  
> 
> #if defined(PETSC_HAVE_HYPE)
> just like now
> 
> Of course this requires always creating includes for all packages but maybe no biggie?
> 
> I was concerned with how they would be classified. The majority of petscconf.h is machine/system stuff that changes rarely.

   There might be a petsconf.h petsc-private/petscconfimpl.h petsc-private/petscconfsys.h petsc-private/petscconfdraw.h petsc-privatre/petscconfPACKAGE.h