Cannot create a Function with firedrake-adjoint when using a MixedFunctionSpace

[Christian Jacobs]

I'm trying to use firedrake-adjoint with my shallow water model. However, I receive the following error message when trying to instantiate a Function object using a MixedFunctionSpace:

"christian@elevate ~ $ python test_fd_adjoint.py
Info : Increasing process stack size (8192 kB < 16 MB)
Traceback (most recent call last):
  File "test_fd_adjoint.py", line 9, in <module>
    mixed_function = Function(W)
  File "/data/dolfin-adjoint-dolfin-adjoint-7a6628f15dbc/dolfin_adjoint/function.py", line 134, in __init__
    assign.register_assign(self, args[0])
  File "/data/dolfin-adjoint-dolfin-adjoint-7a6628f15dbc/dolfin_adjoint/assign.py", line 11, in register_assign
    assert op is not None
AssertionError"

The following code reproduces the error:

"from firedrake import *
from firedrake_adjoint import *

mesh = UnitSquareMesh(10, 10)
fs_1 = FunctionSpace(mesh, "CG", 2)
fs_2 = FunctionSpace(mesh, "CG", 1)
W = MixedFunctionSpace([fs_1, fs_2])

mixed_function = Function(W)"

Firedrake version: 0.9.0
DOLFIN version: 1.0.0
DOLFIN-adjoint version: 0.6
libadjoint version: 0.9.3

[Simon Funke (simon-funke)]

MixedFunctionSpaces have not been implemented in firedrake-adjoint yet.

I will have a look at it once I have a working firedrake installation again.

[Christian Jacobs (ctjacobs)]

I'm still getting this error with dolfin-adjoint 1.4. Has any progress been made in a development branch?

[Patrick Farrell (pefarrell)]

Hi Christian,

Sorry, I didn't see this because we've moved the issue tracker to bitbucket.

Should be fixed in d-a default (on bitbucket) now. The problem was that in dolfin, MixedFunctionSpace is a subclass of FunctionSpace, whereas in firedrake it isn't. So our isinstance(something, FunctionSpace) failed and it went down the wrong branch.

Patrick

[Christian Jacobs (ctjacobs)]

Many thanks for fixing this, Patrick.