Summation of preassembled matrices breaks dolfin-adjoint

[Simon Funke]

Consider the dolfin-consider/tests/supg example.
It preassembles the lhs of the system with:

  A = assemble(a)

and uses it later in the solve command:

  solver = LUSolver(A)

While this works, a simple variation breaks this test: Namely replacing the assembly part with:

  A1 = assemble(a)
  A = 2*A1

crashes with:
  libadjoint.exceptions.LibadjointErrorInvalidInputs: Your matrix A has to have the .form attribute: was it assembled after from dolfin_adjoint import *?

The problem is that the new A object does not carry the "form" attribute anymore.
The dirty way to fix this is:

  A1 = assemble(a)
  A = 2*A1
  A.form = 2*A1.form

A short term solution to this bug would be to moneky patch the Matrix objects in dolfin-adjoint so that it always carries the .form attribute with it.

Alternatively, we can submit a fix to dolfin I guess?