Failure to select quadrature rule for empty forms

[Martin Sandve Alnæs]

Was - Empty integrals are not detected early enough

This dolfin program from @Gabrielbalaban results in an internal error in ffc that is the result of having 0 integrals in the form, which should probably be detected at an earlier stage.

from dolfin import *

mesh = UnitCubeMesh(1,1,1)
M = VectorFunctionSpace(mesh,"CG", 2)*FunctionSpace(mesh, "DG", 0)
U = Function(M)
vq = TestFunction(M)

#This one produces a cryptic error message.
u,p = U.split()

#This one works fine.
#u,p = split(U)

F = grad(u) + Identity(3)
psi = tr(F) - 3 + p*(det(F) - 1)

R = derivative(psi*dx, U, vq)
assemble(R)

[Martin Sandve Alnæs]

I've fixed the actual bug here, the choice of quadrature rule now defaults to canonical when the form is empty. Changing the last line in the example here to

r = assemble(R)
print r
print norm(r)

will work as it should, giving a zero vector instead of the cryptic error message.

Note that 'as it should' is not 'as Gabriel expected', because U.split() and split(U) are fundamentally different operations, but that is a whole other issue already in the bug tracker somewhere.

[Martin Sandve Alnæs]

Removing milestone: 1.5 (automated comment)