LUSolver assumes repeated use of a single matrix

[James Maddison]

The dolfin-adjoint LUSolver assumes that a single matrix is used repeatedly. The following example:

#!/usr/bin/env python
from dolfin import *
from dolfin_adjoint import *
mesh = Rectangle(0.0, 0.0, 1.0, 1.0, 10, 10)
space = FunctionSpace(mesh, "CG", 1)
mass = assemble(inner(TestFunction(space), TrialFunction(space)) * dx)
rhs = assemble(inner(TestFunction(space), Constant(1.0)) * dx)
F = Function(space)
solver = LUSolver()
solver.solve(mass, F.vector(), rhs)
print assemble(F * dx)

leads to the error:

Original exception was:
Traceback (most recent call last):
  File "./dolfin-adjoint_bug_lusolver", line 9, in <module>
    solver = LUSolver()
  File "/usr/lib/python2.7/dist-packages/dolfin_adjoint/lusolver.py", line 57, in __init__
    self.operator = args[0].form
IndexError: tuple index out of range

while:

#!/usr/bin/env python
from dolfin import *
from dolfin_adjoint import *
mesh = Rectangle(0.0, 0.0, 1.0, 1.0, 10, 10)
space = FunctionSpace(mesh, "CG", 1)
mass = assemble(inner(TestFunction(space), TrialFunction(space)) * dx)
rhs = assemble(inner(TestFunction(space), Constant(1.0)) * dx)
F = Function(space)
solver = LUSolver(mass)
solver.solve(mass, F.vector(), rhs)
print assemble(F * dx)

leads to the error:

Original exception was:
Traceback (most recent call last):
  File "./dolfin-adjoint_bug_lusolver", line 10, in <module>
    solver.solve(mass, F.vector(), rhs)
  File "/usr/lib/python2.7/dist-packages/dolfin_adjoint/lusolver.py", line 84, in solve
    raise libadjoint.exceptions.LibadjointErrorInvalidInputs("The annotated LUSolver.solve must be called like solve(x, b).")
libadjoint.exceptions.LibadjointErrorInvalidInputs: The annotated LUSolver.solve must be called like solve(x, b).

Each of these is valid when used with dolfin.