fix: allow even more complex bindings

One of the outstanding problems is a two model mutli-bind where

        B.K_1 -> A.K_1
        B.K_2 -> A.K_2
        A.K_2 -> A.K_1

in which case the bindings for `B` might come out as index `[1, 2]`, but
`A` only has 1 parameter in its parameter vector, leading to index out
of bounds.

src/fitting/binding.jl

@@ -17,18 +17,22 @@ function _construct_bound_mapping(bindings, parameter_count)
             parameter_mapping[b[1]][b[2]] = reference[2]
 
             # mark for removal: find the parameter index in the global array
-            N = sum(length(parameter_mapping[q]) for q = 1:(b[1]-1))
+            N = if b[1] > 1
+                sum(length(parameter_mapping[q]) for q = 1:b[1]-1)
+            else
+                0
+            end
             index = N + b[2]
             push!(remove, index)
 
             # need to now shuffle all the indices greater than this one down by 1
-            for k = (b[2]+1):length(parameter_mapping[b[1]])
+            for k = b[2]+1:length(parameter_mapping[b[1]])
                 if (parameter_mapping[b[1]][k] > parameter_number)
                     parameter_mapping[b[1]][k] -= 1
                 end
             end
             # and for subsequent models
-            for j = (b[1]+1):length(parameter_count)
+            for j = b[1]+1:length(parameter_count)
                 for k = 1:length(parameter_mapping[j])
                     if parameter_mapping[j][k] > parameter_number
                         parameter_mapping[j][k] -= 1

test/fitting/test-binding.jl

@@ -123,3 +123,22 @@ new_bindings = SpectralFitting.adjust_free_bindings(prob.model, prob.bindings)
 @test new_bindings == [[1 => 1, 2 => 3]]
 
 # TODO: free parameters should not be allowed to bind to frozen parameters
+
+
+# can we bind parameters within the same model
+prob = FittingProblem(model1 => dummy_data1)
+bind!(prob, 1 => :K_1, 1 => :a_1)
+_, mapping = SpectralFitting._build_parameter_mapping(prob.model, prob.bindings)
+@test mapping == ([1, 1, 2],)
+
+# can we bind parameters within the same model
+model1.a_2.frozen = false
+prob = FittingProblem(model1 => dummy_data1, model1 => dummy_data1)
+bind!(prob, 1 => :K_1, 2 => :K_1)
+bind!(prob, 1 => :a_1, 2 => :a_1)
+bind!(prob, 1 => :K_2, 2 => :K_2)
+bind!(prob, 1 => :a_2, 2 => :a_2)
+bind!(prob, 1 => :K_1, 1 => :a_1)
+details(prob)
+params, mapping = SpectralFitting._build_parameter_mapping(prob.model, prob.bindings)
+@test mapping == ([1, 1, 2],)