On Wed, Oct 30, 2013 at 10:49 AM, Jed Brown <[email protected]> wrote:
> Matthew Knepley <[email protected]> writes: > > So the rationale for this is when we change backend types? > > Yes, when we changed implementations, we shouldn't be contaminated by > whatever came before, but if the user does > > PCSetType(pc,PCASM); > PetscObjectComposeFunction(pc,"TheirFunction",TheirFunction); > PCSetType(pc,PCGAMG); > > Their function should not be removed. Thus the thought of clearing > based on a prefix. > > > Is it worth anything more than a convention? If so, what about just > having > > a count in PetscObject that gets checked in the teardown process. > > 1. How do we check the convention? It's very error-prone at present. > > 2. See above for why a simple count is not good enough. > You can mask it so that it only increments inside Create(). Matt -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener
