--- a/DiffOps/src/laplace.jl	Wed Jun 26 21:19:00 2019 +0200
+++ b/DiffOps/src/laplace.jl	Wed Jun 26 21:22:36 2019 +0200
@@ -10,29 +10,28 @@
 end
 export NormalDerivative
 
+# TODO: This is obviouly strange. Is domain_size just discarded? Is there a way to avoid storing grid in BoundaryValue?
+# Can we give special treatment to TensorMappings that go to a higher dim?
+LazyTensors.range_size(e::NormalDerivative{T}, domain_size::NTuple{1,Integer}) where T = size(e.grid)
+LazyTensors.domain_size(e::NormalDerivative{T}, range_size::NTuple{2,Integer}) where T = (range_size[3-dim(e.bId)],)
+
 # Not correct abstraction level
 # TODO: Not type stable D:<
 function LazyTensors.apply(d::NormalDerivative, v::AbstractArray, I::NTuple{2,Int})
 	i = I[dim(d.bId)]
 	j = I[3-dim(d.bId)]
-	N_i = d.grid.size[dim(d.bId)]
-
-	r = getregion(i, closureSize(d.op), N_i)
+	N_i = size(d.grid)[dim(d.bId)]
 
-	if r != region(d.bId)
-		return 0
-	end
-
-	if r == Lower
+	if region(d.bId) == Lower
 		# Note, closures are indexed by offset. Fix this D:<
 		return d.grid.inverse_spacing[dim(d.bId)]*d.op.dClosure[i-1]*v[j]
-	elseif r == Upper
-		return d.grid.inverse_spacing[dim(d.bId)]*d.op.dClosure[N_i-j]*v[j]
+	elseif region(d.bId) == Upper
+		return -d.grid.inverse_spacing[dim(d.bId)]*d.op.dClosure[N_i-i]*v[j]
 	end
 end
 
 function LazyTensors.apply_transpose(d::NormalDerivative, v::AbstractArray, I::NTuple{1,Int})
-    u = selectdim(v,3-dim(d.bId),I)
+    u = selectdim(v,3-dim(d.bId),I[1])
     return apply_d(d.op, d.grid.inverse_spacing[dim(d.bId)], u, region(d.bId))
 end
 
@@ -52,7 +51,7 @@
 # TODO: This is obviouly strange. Is domain_size just discarded? Is there a way to avoid storing grid in BoundaryValue?
 # Can we give special treatment to TensorMappings that go to a higher dim?
 LazyTensors.range_size(e::BoundaryValue{T}, domain_size::NTuple{1,Integer}) where T = size(e.grid)
-LazyTensors.domain_size(e::BoundaryValue{T}, range_size::NTuple{2,Integer}) where T = (range_size[3-dim(e.bId)],);
+LazyTensors.domain_size(e::BoundaryValue{T}, range_size::NTuple{2,Integer}) where T = (range_size[3-dim(e.bId)],)
 
 function LazyTensors.apply(e::BoundaryValue, v::AbstractArray, I::NTuple{2,Int})
 	i = I[dim(e.bId)]

--- a/DiffOps/test/runtests.jl	Wed Jun 26 21:19:00 2019 +0200
+++ b/DiffOps/test/runtests.jl	Wed Jun 26 21:22:36 2019 +0200
@@ -5,8 +5,6 @@
 using RegionIndices
 using LazyTensors
 
-@test_broken false
-
 @testset "BoundaryValue" begin
     op = readOperator(sbp_operators_path()*"d2_4th.txt",sbp_operators_path()*"h_4th.txt")
     g = EquidistantGrid((4,5), (0.0, 0.0), (1.0,1.0))
@@ -56,9 +54,89 @@
     G_n = zeros(Float64, (4,5))
     G_n[:,5] = g_x
 
+    @test size(e_w*g_y) == (4,5)
+    @test size(e_e*g_y) == (4,5)
+    @test size(e_s*g_x) == (4,5)
+    @test size(e_n*g_x) == (4,5)
+
     @test collect(e_w*g_y) == G_w
     @test collect(e_e*g_y) == G_e
     @test collect(e_s*g_x) == G_s
     @test collect(e_n*g_x) == G_n
+end
 
+@testset "NormalDerivative" begin
+    op = readOperator(sbp_operators_path()*"d2_4th.txt",sbp_operators_path()*"h_4th.txt")
+    g = EquidistantGrid((5,6), (0.0, 0.0), (4.0,5.0))
+
+    d_w = NormalDerivative(op, g, CartesianBoundary{1,Lower}())
+    d_e = NormalDerivative(op, g, CartesianBoundary{1,Upper}())
+    d_s = NormalDerivative(op, g, CartesianBoundary{2,Lower}())
+    d_n = NormalDerivative(op, g, CartesianBoundary{2,Upper}())
+
+
+    v = evalOn(g, (x,y)-> x^2 + (y-1)^2 + x*y)
+    v∂x = evalOn(g, (x,y)-> 2*x + y)
+    v∂y = evalOn(g, (x,y)-> 2*(y-1) + x)
+
+    @test d_w  isa TensorMapping{T,2,1} where T
+    @test d_w' isa TensorMapping{T,1,2} where T
+
+    @test domain_size(d_w, (3,2)) == (2,)
+    @test domain_size(d_e, (3,2)) == (2,)
+    @test domain_size(d_s, (3,2)) == (3,)
+    @test domain_size(d_n, (3,2)) == (3,)
+
+    @test size(d_w'*v) == (6,)
+    @test size(d_e'*v) == (6,)
+    @test size(d_s'*v) == (5,)
+    @test size(d_n'*v) == (5,)
+
+    @test collect(d_w'*v) ≈ v∂x[1,:]
+    @test collect(d_e'*v) ≈ v∂x[5,:]
+    @test collect(d_s'*v) ≈ v∂y[:,1]
+    @test collect(d_n'*v) ≈ v∂y[:,6]
+
+
+    d_x_l = zeros(Float64, 5)
+    d_x_u = zeros(Float64, 5)
+    for i ∈ eachindex(d_x_l)
+        d_x_l[i] = op.dClosure[i-1]
+        d_x_u[i] = -op.dClosure[length(d_x_u)-i]
+    end
+
+    d_y_l = zeros(Float64, 6)
+    d_y_u = zeros(Float64, 6)
+    for i ∈ eachindex(d_y_l)
+        d_y_l[i] = op.dClosure[i-1]
+        d_y_u[i] = -op.dClosure[length(d_y_u)-i]
+    end
+
+    function ❓(x,y)
+        G = zeros(Float64, length(x), length(y))
+        for I ∈ CartesianIndices(G)
+            G[I] = x[I[1]]*y[I[2]]
+        end
+
+        return G
+    end
+
+    g_x = [1,2,3,4.0,5]
+    g_y = [5,4,3,2,1.0,11]
+
+    G_w = ❓(d_x_l, g_y)
+    G_e = ❓(d_x_u, g_y)
+    G_s = ❓(g_x, d_y_l)
+    G_n = ❓(g_x, d_y_u)
+
+
+    @test size(d_w*g_y) == (5,6)
+    @test size(d_e*g_y) == (5,6)
+    @test size(d_s*g_x) == (5,6)
+    @test size(d_n*g_x) == (5,6)
+
+    @test collect(d_w*g_y) ≈ G_w
+    @test collect(d_e*g_y) ≈ G_e
+    @test collect(d_s*g_x) ≈ G_s
+    @test collect(d_n*g_x) ≈ G_n
 end