assign giving wrong results assigning subfunctions of a mixed FE function

Question

In some circumstances when assign is used to assign the subfunctions of a function in a mixed function space, it gives a completely wrong answer. Here is a very simplified program that shows the problem. Is this a bug in assign, or have I misunderstood something?

from dolfin import *
# create a mixed function space X = X0 x X1
mesh = UnitSquareMesh(1, 1)
X0 = VectorFunctionSpace(mesh, 'BDM', 1)
X1 = FunctionSpace(mesh, 'DG', 0)
X = MixedFunctionSpace([X0, X1])
# create any nonzero functions in X0 and X1
x0 = interpolate(Constant((("1", "2"), ("3", "4"))), X0)
x1 = interpolate(Constant("5"), X1)
# combine into a function in X: x = (x0, x1)
## method 1: use projection, this gives the right answer
y0, y1 = TrialFunctions(X)
z0, z1 = TestFunctions(X)
a = (inner(y0, z0) + inner(y1, z1) )*dx
L = (inner(x0, z0) + inner(x1, z1)  )*dx
xproj = Function(X)
solve(a == L, xproj)
print  "Evaluating projected fn.  Should give 1, 2, 3, 4, 5."
print xproj(.1,.1)
## method 2: use assignment, this gives the wrong answer
xassign = Function(X)
assign(xassign.sub(0), x0)
assign(xassign.sub(1), x1)
print  "Evaluating assigned fn.  Should give 1, 2, 3, 4, 5."
print xassign(.1, .1)

The output is

Solving linear variational problem.
Evaluating projected fn.  Should give 1, 2, 3, 4, 5.
[ 1.  2.  3.  4.  5.]
Evaluating assigned fn.  Should give 1, 2, 3, 4, 5.
[ 1.4  2.  -1.5 -0.5  5. ]

As the example program shows, a way around this is to compute a projection. But it seems like it should not be necessary to solve a linear system just to assign known coefficients.

Answer 1

This should be fixed in the development version of DOLFIN.

Answer 2

This looks like a bug in the FunctionAssigner class. I'll report the bug on bitbucket.

Here is a temporary workaround in case you don't want to do the project:

dofs = X.sub(0).dofmap().collapse(mesh)[1].values()
xassign.vector()[dofs] = x0.vector()
assign(xassign.sub(1), x1)  # This one works