integrands skeleton term bug
There are two problems that cause this model to diverge.
a += jump(grad(u[0]), n) * v[0] * dS
Here v[0] should not be allowed since it may not be a DG term. Instead the code runs but this term is silently set to zero, which isn't super helpful. This can be fixed by replacing v[0] with avg(v[0]):
a += jump(grad(u[0]), n) * avg(v[0]) * dS
An error should be raised for the first expression (or there should be some automatic substitution maybe?).
Secondly there is:
a4 += 1/hF * inner(jump(u - exact), jump(v)) * ds
This does not work as jump terms are not defined on the boundary. The following works instead:
a += 1/hF * inner(u - exact, v) * ds
Again this should be done automatically or an error should be raised.