Operator update with LU solver (Python)

Question

I've got a concern about the LU solver.
Given two different operators A and B (i.e. build from different weak forms), the code below produces sol_a == sol_b if PETSc is the LA backend but returns different solutions (as it should) if the backend is set to uBLAS.

solver = LUSolver()
solver.parameters['reuse_factorization'] = True

solver.set_operator(A)
sol_a = Function(V)
solver.solve(sol_a.vector(), rhs)

solver.set_operator(B)
sol_b = Function(V)
solver.solve(sol_b.vector(), rhs)

Same behavior if the solution is computed in the form

solver.solve(A, sol_a.vector(), rhs)

But the solutions are different (i.e. correct) if the parameter key 'reuse_factorization' is set to False.
Is this a bug (in which case I'll post this on the mailing-list) or am I missing something? I can post the complete code if someone feels it is necessary.
Thanks for any help!

Comments

This is probably related to https://bitbucket.org/fenics-project/dolfin/issue/269. I've linked this question to the Issue tracker to get this question answered once Issue 269 is fixed.