I had some trouble with parallel calculations using the PETSc interface in the following way:
A = PETScMatrix()
b = PETScVector()
assemble(a, tensor=A)
assemble(L, tensor=b)
ksp = PETSC.KSP().create()
# some PETSc KSP setup...
ksp.setOperators(A)
# the solution function
w = Function(W)
ksp.solve(b.vec(), as_backend_type(w.vector().vec())
because I forgot to call update_ghost_values() after the solve. In that case, the 'interfaces' of the parallel partition of the problem are zero.
If, instead, I do
(x, _) = A.mat().getVecs()
ksp.solve(b.vec(), x)
w.vector()[:] = x
calling update_ghost_values() is apparently not necessary.
So my simple question is: When do I have to call update_ghost_values()?
