Make MathOptInterface.jl a weak dependency

Project.toml

@@ -8,15 +8,21 @@ FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
 ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 LineSearches = "d3d80556-e9d4-5f37-9878-2ab0fcc64255"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
-MathOptInterface = "b8f27783-ece8-5eb3-8dc8-9495eed66fee"
 NLSolversBase = "d41bc354-129a-5804-8e4c-c37616107c6c"
 NaNMath = "77ba4419-2d1f-58cd-9bb1-8ffee604a2e3"
+PackageExtensionCompat = "65ce6f38-6b18-4e1d-a461-8949797d7930"
 Parameters = "d96e819e-fc66-5662-9728-84c9c7592b0a"
 PositiveFactorizations = "85a6dd25-e78a-55b7-8502-1745935b8125"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 
+[weakdeps]
+MathOptInterface = "b8f27783-ece8-5eb3-8dc8-9495eed66fee"
+
+[extensions]
+OptimMOIExt = "MathOptInterface"
+
 [compat]
 Compat = "3.2.0, 3.3.0, 3.4.0, 3.5.0, 3.6.0, 4"
 FillArrays = "0.6.2, 0.7, 0.8, 0.9, 0.10, 0.11, 0.12, 0.13, 1"
@@ -27,6 +33,7 @@ MathOptInterface = "1.17"
 NLSolversBase = "7.8.0"
 NaNMath = "0.3.2, 1"
 OptimTestProblems = "2.0.3"
+PackageExtensionCompat = "1"
 Parameters = "0.10, 0.11, 0.12"
 PositiveFactorizations = "0.2.2"
 Printf = "<0.0.1, 1.6"
@@ -39,6 +46,7 @@ julia = "1.6"
 [extras]
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 LineSearches = "d3d80556-e9d4-5f37-9878-2ab0fcc64255"
+MathOptInterface = "b8f27783-ece8-5eb3-8dc8-9495eed66fee"
 Measurements = "eff96d63-e80a-5855-80a2-b1b0885c5ab7"
 NLSolversBase = "d41bc354-129a-5804-8e4c-c37616107c6c"
 OptimTestProblems = "cec144fc-5a64-5bc6-99fb-dde8f63e154c"
@@ -49,4 +57,4 @@ StableRNGs = "860ef19b-820b-49d6-a774-d7a799459cd3"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["Test", "Distributions", "Measurements", "OptimTestProblems", "Random", "RecursiveArrayTools", "StableRNGs", "LineSearches", "NLSolversBase", "PositiveFactorizations"]
+test = ["Test", "Distributions", "MathOptInterface", "Measurements", "OptimTestProblems", "Random", "RecursiveArrayTools", "StableRNGs", "LineSearches", "NLSolversBase", "PositiveFactorizations"]

src/MOI_wrapper.jl → ext/OptimMOIExt.jl

@@ -1,5 +1,24 @@
+module OptimMOIExt
+
+using Optim
+using LinearAlgebra
 import MathOptInterface as MOI
 
+function __init__()
+    @static if VERSION >= v"1.9"
+        @eval Optim begin
+            OptimMOIExt = Base.get_extension(@__MODULE__, :OptimMOIExt)
+            const Optimizer = OptimMOIExt.Optimizer
+        end
+        # setglobal!(Optim, :Optimizer, Optimizer)
+    else
+        @eval Optim begin
+            using .OptimMOIExt
+            const Optimizer = OptimMOIExt.Optimizer
+        end
+    end
+end
+
 mutable struct Optimizer{T} <: MOI.AbstractOptimizer
     # Problem data.
     variables::MOI.Utilities.VariablesContainer{T}
@@ -8,12 +27,12 @@ mutable struct Optimizer{T} <: MOI.AbstractOptimizer
     sense::MOI.OptimizationSense
 
     # Parameters.
-    method::Union{AbstractOptimizer,Nothing}
+    method::Union{Optim.AbstractOptimizer,Nothing}
     silent::Bool
     options::Dict{Symbol,Any}
 
     # Solution attributes.
-    results::Union{Nothing,MultivariateOptimizationResults}
+    results::Union{Nothing,Optim.MultivariateOptimizationResults}
 end
 
 function Optimizer{T}() where {T}
@@ -37,7 +56,7 @@ function MOI.supports(::Optimizer, ::Union{MOI.ObjectiveSense,MOI.ObjectiveFunct
 end
 MOI.supports(::Optimizer, ::MOI.Silent) = true
 function MOI.supports(::Optimizer, p::MOI.RawOptimizerAttribute)
-    return p.name == "method" || hasfield(Options, Symbol(p.name))
+    return p.name == "method" || hasfield(Optim.Options, Symbol(p.name))
 end
 
 function MOI.supports(::Optimizer, ::MOI.VariablePrimalStart, ::Type{MOI.VariableIndex})
@@ -103,7 +122,7 @@ function MOI.get(model::Optimizer, ::MOI.TimeLimitSec)
     return get(model.options, Symbol(TIME_LIMIT), nothing)
 end
 
-MOI.Utilities.map_indices(::Function, opt::AbstractOptimizer) = opt
+MOI.Utilities.map_indices(::Function, opt::Optim.AbstractOptimizer) = opt
 
 function MOI.set(model::Optimizer, p::MOI.RawOptimizerAttribute, value)
     if p.name == "method"
@@ -151,7 +170,11 @@ function MOI.is_valid(model::Optimizer, index::Union{MOI.VariableIndex,MOI.Const
     return MOI.is_valid(model.variables, index)
 end
 
-function MOI.add_constraint(model::Optimizer{T}, vi::MOI.VariableIndex, set::BOUNDS{T}) where {T}
+function MOI.add_constraint(
+    model::Optimizer{T},
+    vi::MOI.VariableIndex,
+    set::BOUNDS{T},
+) where {T}
     return MOI.add_constraint(model.variables, vi, set)
 end
 
@@ -187,17 +210,17 @@ function MOI.set(
     return
 end
 
-function requested_features(::ZerothOrderOptimizer, has_constraints)
+function requested_features(::Optim.ZerothOrderOptimizer, has_constraints)
     return Symbol[]
 end
-function requested_features(::FirstOrderOptimizer, has_constraints)
+function requested_features(::Optim.FirstOrderOptimizer, has_constraints)
     features = [:Grad]
     if has_constraints
         push!(features, :Jac)
     end
     return features
 end
-function requested_features(::Union{IPNewton,SecondOrderOptimizer}, has_constraints)
+function requested_features(::Union{IPNewton,Optim.SecondOrderOptimizer}, has_constraints)
     features = [:Grad, :Hess]
     if has_constraints
         push!(features, :Jac)
@@ -255,7 +278,12 @@ function MOI.optimize!(model::Optimizer{T}) where {T}
     method = model.method
     nl_constrained = !isempty(nlp_data.constraint_bounds)
     features = MOI.features_available(evaluator)
-    has_bounds = any(vi -> isfinite(model.variables.lower[vi.value]) || isfinite(model.variables.upper[vi.value]), vars)
+    has_bounds = any(
+        vi ->
+            isfinite(model.variables.lower[vi.value]) ||
+                isfinite(model.variables.upper[vi.value]),
+        vars,
+    )
     if method === nothing
         if nl_constrained
             method = IPNewton()
@@ -264,12 +292,12 @@ function MOI.optimize!(model::Optimizer{T}) where {T}
             # are variable bounds, `Newton` is not supported. On the other hand,
             # `fallback_method(f, g!)` returns `LBFGS` which is supported if `has_bounds`.
             if :Hess in features && !has_bounds
-                method = fallback_method(f, g!, h!)
+                method = Optim.fallback_method(f, g!, h!)
             else
-                method = fallback_method(f, g!)
+                method = Optim.fallback_method(f, g!)
             end
         else
-            method = fallback_method(f)
+            method = Optim.fallback_method(f)
         end
     end
     used_features = requested_features(method, nl_constrained)
@@ -283,22 +311,35 @@ function MOI.optimize!(model::Optimizer{T}) where {T}
     initial_x = starting_value.(model, eachindex(model.starting_values))
     options = copy(model.options)
     if !nl_constrained && has_bounds && !(method isa IPNewton)
-        options = Options(; options...)
-        model.results = optimize(f, g!, model.variables.lower, model.variables.upper, initial_x, Fminbox(method), options; inplace = true)
+        options = Optim.Options(; options...)
+        model.results = optimize(
+            f,
+            g!,
+            model.variables.lower,
+            model.variables.upper,
+            initial_x,
+            Fminbox(method),
+            options;
+            inplace = true,
+        )
     else
-        d = promote_objtype(method, initial_x, :finite, true, f, g!, h!)
-        add_default_opts!(options, method)
-        options = Options(; options...)
+        d = Optim.promote_objtype(method, initial_x, :finite, true, f, g!, h!)
+        Optim.add_default_opts!(options, method)
+        options = Optim.Options(; options...)
         if nl_constrained || has_bounds
             if nl_constrained
                 lc = [b.lower for b in nlp_data.constraint_bounds]
                 uc = [b.upper for b in nlp_data.constraint_bounds]
                 c!(c, x) = MOI.eval_constraint(evaluator, c, x)
                 if !(:Jac in features)
-                    error("Nonlinear constraints should be differentiable to be used with Optim.")
+                    error(
+                        "Nonlinear constraints should be differentiable to be used with Optim.",
+                    )
                 end
                 if !(:Hess in features)
-                    error("Nonlinear constraints should be twice differentiable to be used with Optim.")
+                    error(
+                        "Nonlinear constraints should be twice differentiable to be used with Optim.",
+                    )
                 end
                 jacobian_structure = MOI.jacobian_structure(evaluator)
                 J_nzval = zeros(T, length(jacobian_structure))
@@ -315,13 +356,20 @@ function MOI.optimize!(model::Optimizer{T}) where {T}
                     return H
                 end
                 c = TwiceDifferentiableConstraints(
-                    c!, jacobian!, con_hessian!,
-                    model.variables.lower, model.variables.upper, lc, uc,
+                    c!,
+                    jacobian!,
+                    con_hessian!,
+                    model.variables.lower,
+                    model.variables.upper,
+                    lc,
+                    uc,
                 )
             else
                 @assert has_bounds
                 c = TwiceDifferentiableConstraints(
-                    model.variables.lower, model.variables.upper)
+                    model.variables.lower,
+                    model.variables.upper,
+                )
             end
             model.results = optimize(d, c, initial_x, method, options)
         else
@@ -334,7 +382,7 @@ end
 function MOI.get(model::Optimizer, ::MOI.TerminationStatus)
     if model.results === nothing
         return MOI.OPTIMIZE_NOT_CALLED
-    elseif converged(model.results)
+    elseif Optim.converged(model.results)
         return MOI.LOCALLY_SOLVED
     else
         return MOI.OTHER_ERROR
@@ -354,7 +402,7 @@ function MOI.get(model::Optimizer, attr::MOI.PrimalStatus)
     if !(1 <= attr.result_index <= MOI.get(model, MOI.ResultCount()))
         return MOI.NO_SOLUTION
     end
-    if converged(model.results)
+    if Optim.converged(model.results)
         return MOI.FEASIBLE_POINT
     else
         return MOI.UNKNOWN_RESULT_STATUS
@@ -374,7 +422,7 @@ end
 function MOI.get(model::Optimizer, attr::MOI.VariablePrimal, vi::MOI.VariableIndex)
     MOI.check_result_index_bounds(model, attr)
     MOI.throw_if_not_valid(model, vi)
-    return minimizer(model.results)[vi.value]
+    return Optim.minimizer(model.results)[vi.value]
 end
 
 function MOI.get(
@@ -384,5 +432,6 @@ function MOI.get(
 ) where {T}
     MOI.check_result_index_bounds(model, attr)
     MOI.throw_if_not_valid(model, ci)
-    return minimizer(model.results)[ci.value]
+    return Optim.minimizer(model.results)[ci.value]
 end
+end # module

src/Optim.jl

@@ -34,6 +34,7 @@ using Printf                 # For printing, maybe look into other options
 using FillArrays             # For handling scalar bounds in Fminbox
 
 #using Compat                 # for compatibility across multiple julia versions
+using PackageExtensionCompat # For retrocompatibility on package extensions
 
 # for extensions of functions defined in Base.
 import Base: length, push!, show, getindex, setindex!, maximum, minimum
@@ -220,7 +221,8 @@ include("multivariate/solvers/constrained/ipnewton/utilities/trace.jl")
 # Maximization convenience wrapper
 include("maximize.jl")
 
-# MathOptInterface wrapper
-include("MOI_wrapper.jl")
+function __init__()
+    @require_extensions
+end
 
 end