Facet normal with uflacs does not give the same results as ffc generated code

[Martin Sandve Alnæs]

I thought this was working, needs investigation.

[Martin Sandve Alnæs]

This test program gives the same results for uflacs and quadrature in the eye-norm:

from dolfin import *
import sys
if sys.argv[1] == "u":
    parameters["form_compiler"]["representation"] = "uflacs"

mesh = UnitIntervalMesh(3)
#mesh = UnitSquareMesh(10, 10)
#mesh = UnitCubeMesh(10, 10, 10)
#mesh = Mesh("dolfin_fine.xml.gz")

n = FacetNormal(mesh)
d = len(n)

V = VectorFunctionSpace(mesh, "CG", 1)

u = TrialFunction(V)
v = TestFunction(V)
N = Function(V)
alpha = Constant(1e-6)
solve(alpha*inner(grad(u), grad(v))*dx + inner(u, v)*ds == dot(n, v)*ds, N)

plot(N, title='N')
interactive()

[Martin Sandve Alnæs]

The bug is in the known bug in the ffc regression test for interior facet integrals: the same cell geometry is used from both sides which is not well defined input to an interior facet integral.