Thanks Roy... I'll give the init() one a bit more thought... but if I can't find a better place to cut it off.... then I'll just commit it where it is.
Derek On Apr 4, 2009, at 6:33 AM, Roy Stogner wrote: > > On Fri, 3 Apr 2009, Derek Gaston wrote: > >> Essentially it's just a short-circuiting before the >> user_initialization() call. > > Why not after user_initialization()? If a system with no variables is > being initialized, the user's doing something weird, and we should > allow the user to be weird as flexibly as possible. ;-) > >> The problem is that project_vector doesn't work properly with zero >> variables in the system... this fixes it. > > How does short-circuiting System::init and System::reinit fix that, > though? We're calling project_vector from EquationSystems::reinit. > > I'd also rather fix project_vector for the no-variables case, but if > your workaround is sufficient it sounds much much easier. > >> I've also made the same change to System::reinit()... just short >> circuiting the whole function.... there really is no reason to do any >> reiniting if there are no variables in the system.... > > For that matter, it looks like we've got some "no reason *ever* for > the reinit call to still exist" situations lying around! > > EigenSystem::reinit just calls Parent::reinit; that would have been > achieved anyway by just not overriding the virtual function again. > ExplicitSystem and TransientSystem are the same way now that rhs, > old_local_solution, etc are registered with add_vector rather than > initialized separatedly. NewmarkSystem::reinit() is a libmesh_error() > for some reason > >> With these two changes I can now properly debug runs containing a >> System without variables... > > And they're just "if (!nvars) return;" cases that definitely shouldn't > affect anyone else's usage. > > I'd say go ahead and commit them, but thanks for the warning. > --- > Roy ------------------------------------------------------------------------------ _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
