The policy in libMesh has always been that NumericVector::close() isn't done at the end of the system assembly function, it's done at the beginning of the solver function.
Would anyone object to doing it the other way? We'd call close() in System::assemble() and related functions, so that user code wouldn't have to change at all to produce closed results. The advantages I see are More safety in new user codes: people calling an assemble function wouldn't have to worry about whether they could immediately use the outputs or not Conciseness: now that we've got various sensitivity methods using assembly results without calling solve(), there's a proliferation of new close() calls that could be removed. The disadvantages are Less safety in rare old user codes: if someone is modifying their matrix or rhs outside of the library assembly functions, they might be relying on the solver close() rather than calling that themselves. Potential new inefficiency in rare user codes: if the user wants to call his normal assembly function, but then assemble in additional terms before using the result, a close() in between would be unnecessary extra communications. --- Roy ------------------------------------------------------------------------------ _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
