Bug: Operations on empty FESpaces

[JordiManyer]

When evaluating the gradient of an FEFunction defined on an empty FESpace, en error occurs.

The error can be avoided by doing the change of domain first then doing the gradient, i.e the change of domain and the gradient do not commute. It has to do with how extend works for different kinds of arrays.

using Gridap
using Gridap.CellData, Gridap.FESpaces, Gridap.Arrays, Gridap.Geometry
using Gridap.ReferenceFEs, Gridap.TensorValues

model = CartesianDiscreteModel((0,1,0,1),(3,3))

Ω = Triangulation(model)
Ωe = Triangulation(model,Int[])

dΩ = Measure(Ω,2)
dΩe = Measure(Ωe,2)

reffe = ReferenceFE(lagrangian,Float64,1)
V = FESpace(Ωe,reffe)
uh = zero(V)

Π(u) = change_domain(u,Ω,DomainStyle(u))
x = get_cell_points(Ω)

uh(x) # OK
∇(uh)(x) # ERROR - Equivalent to Π(∇(uh))(x)
∇(Π(uh))(x) # OK