Fix docs

Author: odow

Diff for: docs/src/submodules/Nonlinear/overview.md

@@ -302,35 +302,17 @@ However, we cannot call gradient terms such as
 [`eval_objective_gradient`](@ref) because [`Nonlinear.ExprGraphOnly`](@ref) does
 not have the capability to differentiate a nonlinear expression.
 
-Instead of passing [`Nonlinear.Evaluator`](@ref) directly to solvers,
-solvers query the [`NLPBlock`](@ref) attribute, which returns an
-[`NLPBlockData`](@ref). This object wraps an [`Nonlinear.Evaluator`](@ref) and
-includes other information such as constraint bounds and whether the evaluator
-has a nonlinear objective. Create and set [`NLPBlockData`](@ref) as follows:
+If, instead, we pass [`Nonlinear.SparseReverseMode`](@ref), then we get access
+to `:Grad`, the gradient of the objective function, `:Jac`, the jacobian matrix
+of the constraints, `:JacVec`, the ability to compute Jacobian-vector products,
+and `:ExprGraph`.
 ```jldoctest nonlinear_developer
-julia> block = MOI.NLPBlockData(evaluator);
-
-julia> model = MOI.Utilities.UniversalFallback(MOI.Utilities.Model{Float64}());
-
-julia> MOI.set(model, MOI.NLPBlock(), block);
-```
-!!! warning
-    Only call [`NLPBlockData`](@ref) once you have finished modifying the
-    problem in `model`.
-
-If, instead, we set [`Nonlinear.SparseReverseMode`](@ref), then we get access to
-`:Grad`, the gradient of the objective function, `:Jac`, the jacobian matrix of
-the constraints, `:JacVec`, the ability to compute Jacobian-vector products, and
-`:ExprGraph`.
-```jldoctest nonlinear_developer
-julia> Nonlinear.set_differentiation_backend(
-           data,
+julia> evaluator = Nonlinear.Evaluator(
+           model,
            Nonlinear.SparseReverseMode(),
            [x],
        )
-
-julia> data
-NonlinearData with available features:
+Nonlinear.Evaluator with available features:
   * :Grad
   * :Jac
   * :JacVec
@@ -339,7 +321,7 @@ NonlinearData with available features:
 
 However, before calling anything, we need to call [`initialize`](@ref):
 ```jldoctest nonlinear_developer
-julia> MOI.initialize(data, [:Grad, :Jac, :JacVec, :ExprGraph])
+julia> MOI.initialize(evaluator, [:Grad, :Jac, :JacVec, :ExprGraph])
 ```
 
 Now we can call methods like [`eval_objective`](@ref):
@@ -348,7 +330,7 @@ julia> x = [1.0]
 1-element Vector{Float64}:
  1.0
 
-julia> MOI.eval_objective(data, x)
+julia> MOI.eval_objective(evaluator, x)
 7.268073418273571
 ```
 and [`eval_objective_gradient`](@ref):
@@ -357,13 +339,29 @@ julia> grad = [NaN]
 1-element Vector{Float64}:
  NaN
 
-julia> MOI.eval_objective_gradient(data, grad, x)
+julia> MOI.eval_objective_gradient(evaluator, grad, x)
 
 julia> grad
 1-element Vector{Float64}:
  1.909297426825682
 ```
 
+Instead of passing [`Nonlinear.Evaluator`](@ref) directly to solvers,
+solvers query the [`NLPBlock`](@ref) attribute, which returns an
+[`NLPBlockData`](@ref). This object wraps an [`Nonlinear.Evaluator`](@ref) and
+includes other information such as constraint bounds and whether the evaluator
+has a nonlinear objective. Create and set [`NLPBlockData`](@ref) as follows:
+```jldoctest nonlinear_developer
+julia> block = MOI.NLPBlockData(evaluator);
+
+julia> model = MOI.Utilities.UniversalFallback(MOI.Utilities.Model{Float64}());
+
+julia> MOI.set(model, MOI.NLPBlock(), block);
+```
+!!! warning
+    Only call [`NLPBlockData`](@ref) once you have finished modifying the
+    problem in `model`.
+
 ## Expression-graph representation
 
 [`Nonlinear.Model`](@ref) stores nonlinear expressions in