The way the GenericVector interface is implemented, a size argument to the constructor will partition the vector on every process in parallel. We need to be able to allocate locally sized vectors. Should I simply use a non-parallel aware linear algebra backend for this, like uBLAS? This will introduce a mix of different linear algebra backends, and I'm not sure what problems lurk i the shadows if I go down that road :)
An alternative would be to add functionality (default argument?) to specify scope when calling resize and friends. -- Ola Skavhaug _______________________________________________ DOLFIN-dev mailing list DOLFIN-dev@fenics.org http://www.fenics.org/mailman/listinfo/dolfin-dev