✨ Add support for solver at optimize with bridges

REQUIRE

@@ -1,5 +1,5 @@
 julia 0.6
-MathOptInterface 0.6 0.7
+MathOptInterface 0.6.3 0.7
 ForwardDiff 0.5 0.11
 Calculus
 DataStructures

src/JuMP.jl

@@ -134,8 +134,8 @@ end
 # Model
 
 # Model has three modes:
-# 1) Automatic: moi_backend field holds a LazyBridgeOptimizer{CachingOptimizer} in Automatic mode.
-# 2) Manual: moi_backend field holds a LazyBridgeOptimizer{CachingOptimizer} in Manual mode.
+# 1) Automatic: moi_backend field holds a CachingOptimizer in Automatic mode.
+# 2) Manual: moi_backend field holds a CachingOptimizer in Manual mode.
 # 3) Direct: moi_backend field holds an AbstractOptimizer. No extra copy of the model is stored. The moi_backend must support add_constraint etc.
 # Methods to interact with the CachingOptimizer are defined in solverinterface.jl.
 @enum ModelMode Automatic Manual Direct
@@ -157,7 +157,7 @@ mutable struct Model <: AbstractModel
     variable_to_fix::Dict{MOIVAR, MOIFIX}
     variable_to_integrality::Dict{MOIVAR, MOIINT}
     variable_to_zero_one::Dict{MOIVAR, MOIBIN}
-    # In Manual and Automatic modes, LazyBridgeOptimizer{CachingOptimizer}.
+    # In Manual and Automatic modes, CachingOptimizer.
     # In Direct mode, will hold an AbstractOptimizer.
     moi_backend::MOI.AbstractOptimizer
     # Hook into a solve call...function of the form f(m::Model; kwargs...),
@@ -184,11 +184,10 @@ a cache. The mode of the `CachingOptimizer` storing this cache is
 `caching_mode`. The optimizer can be set later in the [`JuMP.optimize!`](@ref)
 call. If `bridge_constraints` is true, constraints that are not supported by the
 optimizer are automatically bridged to equivalent supported constraints when
-an appropriate is defined in the `MathOptInterface.Bridges` module or is
-defined in another module and is explicitely added.
+an appropriate transformation is defined in the `MathOptInterface.Bridges`
+module or is defined in another module and is explicitely added.
 """
 function Model(; caching_mode::MOIU.CachingOptimizerMode=MOIU.Automatic,
-                 bridge_constraints::Bool=true,
                  solver=nothing)
     if solver !== nothing
         error("The solver= keyword is no longer available in JuMP 0.19 and " *
@@ -198,13 +197,7 @@ function Model(; caching_mode::MOIU.CachingOptimizerMode=MOIU.Automatic,
     universal_fallback = MOIU.UniversalFallback(JuMPMOIModel{Float64}())
     caching_opt = MOIU.CachingOptimizer(universal_fallback,
                                         caching_mode)
-    if bridge_constraints
-        backend = MOI.Bridges.fullbridgeoptimizer(caching_opt,
-                                                  Float64)
-    else
-        backend = caching_opt
-    end
-    return direct_model(backend)
+    return direct_model(caching_opt)
 end
 
 """
@@ -224,10 +217,11 @@ The following creates a model using the optimizer
 model = JuMP.Model(with_optimizer(IpoptOptimizer, print_level=0))
 ```
 """
-function Model(optimizer_factory::OptimizerFactory; kwargs...)
+function Model(optimizer_factory::OptimizerFactory;
+               bridge_constraints::Bool=true, kwargs...)
     model = Model(; kwargs...)
-    optimizer = optimizer_factory()
-    MOIU.resetoptimizer!(model, optimizer)
+    set_optimizer(model, optimizer_factory,
+                  bridge_constraints=bridge_constraints)
     return model
 end
 
@@ -269,22 +263,6 @@ if VERSION >= v"0.7-"
 end
 
 
-# In Automatic and Manual mode, `backend(model)` is either directly the
-# `CachingOptimizer` if `bridge_constraints=false` was passed in the constructor
-# or it is a `LazyBridgeOptimizer` and the `CachingOptimizer` is stored in the
-# `model` field
-function caching_optimizer(model::Model)
-    if backend(model) isa MOIU.CachingOptimizer
-        return backend(model)
-    elseif (backend(model) isa
-            MOI.Bridges.LazyBridgeOptimizer{<:MOIU.CachingOptimizer})
-        return backend(model).model
-    else
-        error("The function `caching_optimizer` cannot be called on a model " *
-              "in `Direct` mode.")
-    end
-end
-
 """
     backend(model::Model)
 
@@ -294,8 +272,7 @@ and whether there are any bridges in the model.
 
 If JuMP is in direct mode (i.e., the model was created using [`JuMP.direct_model`](@ref)),
 the backend with be the optimizer passed to `direct_model`. If JuMP is in manual
-or automatic mode, the backend will either be a `MOI.Utilities.CachingOptimizer`
-or a `MOI.Bridges.LazyBridgeOptimizer`.
+or automatic mode, the backend is a `MOI.Utilities.CachingOptimizer`.
 
 This function should only be used by advanced users looking to access low-level
 MathOptInterface or solver-specific functionality.
@@ -308,16 +285,34 @@ backend(model::Model) = model.moi_backend
 Return mode (Direct, Automatic, Manual) of model.
 """
 function mode(model::Model)
-    if !(backend(model) isa MOI.Bridges.LazyBridgeOptimizer{<:MOIU.CachingOptimizer} ||
-         backend(model) isa MOIU.CachingOptimizer)
+    if !(backend(model) isa MOIU.CachingOptimizer)
         return Direct
-    elseif caching_optimizer(model).mode == MOIU.Automatic
+    elseif backend(model).mode == MOIU.Automatic
         return Automatic
     else
         return Manual
     end
 end
 
+"""
+    bridge_constraints(model::Model)
+
+Return a `Bool` indicating whether the model `model` is in manual or automatic
+mode, the optimizer is set and unsupported constraints are automatically bridged
+to equivalent supported constraints when an appropriate transformation is
+available.
+"""
+function bridge_constraints(model::Model)
+    caching_optimizer = backend(model)
+    if caching_optimizer isa MOIU.CachingOptimizer
+        return caching_optimizer.optimizer isa MOI.Bridges.LazyBridgeOptimizer
+    else
+        # Direct mode
+        return false
+    end
+end
+
+
 """
     num_variables(model::Model)
 
@@ -439,17 +434,17 @@ function optimizer_index(v::VariableRef)
     if mode(model) == Direct
         return index(v)
     else
-        @assert caching_optimizer(model).state == MOIU.AttachedOptimizer
-        return caching_optimizer(model).model_to_optimizer_map[index(v)]
+        @assert backend(model).state == MOIU.AttachedOptimizer
+        return backend(model).model_to_optimizer_map[index(v)]
     end
 end
 
 function optimizer_index(cr::ConstraintRef{Model})
     if mode(cr.model) == Direct
         return index(cr)
     else
-        @assert caching_optimizer(cr.model).state == MOIU.AttachedOptimizer
-        return caching_optimizer(cr.model).model_to_optimizer_map[index(cr)]
+        @assert backend(cr.model).state == MOIU.AttachedOptimizer
+        return backend(cr.model).model_to_optimizer_map[index(cr)]
     end
 end
 

src/constraints.jl

@@ -218,23 +218,25 @@ function constraint_object(ref::ConstraintRef{Model, MOICON{FuncType, SetType}})
 end
 
 """
-    add_constraint(m::Model, c::AbstractConstraint, name::String="")
+    add_constraint(model::Model, c::AbstractConstraint, name::String="")
 
-Add a constraint `c` to `Model m` and sets its name.
+Add a constraint `c` to `Model model` and sets its name.
 """
-function add_constraint(m::Model, c::AbstractConstraint, name::String="")
+function add_constraint(model::Model, c::AbstractConstraint, name::String="")
     f = moi_function(c)
     s = moi_set(c)
-    if !MOI.supports_constraint(backend(m), typeof(f), typeof(s))
-        if backend(m) isa MOI.Bridges.LazyBridgeOptimizer
+    if !MOI.supports_constraint(backend(model), typeof(f), typeof(s))
+        if mode(model) == Direct
+            bridge_message = "."
+        elseif bridge_constraints(model)
             bridge_message = " and there are no bridges that can reformulate it into supported constraints."
         else
             bridge_message = ", try using `bridge_constraints=true` in the `JuMP.Model` constructor if you believe the constraint can be reformulated to constraints supported by the solver."
         end
         error("Constraints of type $(typeof(f))-in-$(typeof(s)) are not supported by the solver" * bridge_message)
     end
-    cindex = MOI.add_constraint(backend(m), f, s)
-    cref = ConstraintRef(m, cindex, shape(c))
+    cindex = MOI.add_constraint(backend(model), f, s)
+    cref = ConstraintRef(model, cindex, shape(c))
     if !isempty(name)
         set_name(cref, name)
     end

src/copy.jl

@@ -97,12 +97,8 @@ function copy_model(model::Model)
               " instead of the `direct_model` constructor to be able to copy",
               " the constructed model.")
     end
-    caching_mode = caching_optimizer(model).mode
-    # TODO add bridges added to the bridge optimizer that are not part of the
-    #      fullbridgeoptimizer
-    bridge_constraints = backend(model) isa MOI.Bridges.LazyBridgeOptimizer{<:MOIU.CachingOptimizer}
-    new_model = Model(caching_mode = caching_mode,
-                      bridge_constraints = bridge_constraints)
+    caching_mode = backend(model).mode
+    new_model = Model(caching_mode = caching_mode)
 
     # Copy the MOI backend, note that variable and constraint indices may have
     # changed, the `index_map` gives the map between the indices of

src/objective.jl

@@ -115,12 +115,12 @@ However, it is not convertible to a variable.
 julia> JuMP.objective_function(model, JuMP.VariableRef)
 ERROR: InexactError: convert(MathOptInterface.SingleVariable, MathOptInterface.ScalarAffineFunction{Float64}(MathOptInterface.ScalarAffineTerm{Float64}[ScalarAffineTerm{Float64}(2.0, VariableIndex(1))], 1.0))
 Stacktrace:
- [1] convert at /home/blegat/.julia/dev/MathOptInterface/src/functions.jl:393 [inlined]
- [2] get(::JuMP.JuMPMOIModel{Float64}, ::MathOptInterface.ObjectiveFunction{MathOptInterface.SingleVariable}) at /home/blegat/.julia/dev/MathOptInterface/src/Utilities/model.jl:259
- [3] get at /home/blegat/.julia/dev/MathOptInterface/src/Utilities/universalfallback.jl:105 [inlined]
- [4] get at /home/blegat/.julia/dev/MathOptInterface/src/Utilities/cachingoptimizer.jl:436 [inlined]
+ [1] convert at /home/blegat/.julia/dev/MathOptInterface/src/functions.jl:398 [inlined]
+ [2] get(::JuMP.JuMPMOIModel{Float64}, ::MathOptInterface.ObjectiveFunction{MathOptInterface.SingleVariable}) at /home/blegat/.julia/dev/MathOptInterface/src/Utilities/model.jl:290
+ [3] get at /home/blegat/.julia/dev/MathOptInterface/src/Utilities/universalfallback.jl:114 [inlined]
+ [4] get at /home/blegat/.julia/dev/MathOptInterface/src/Utilities/cachingoptimizer.jl:439 [inlined]
  [5] get(::MathOptInterface.Bridges.LazyBridgeOptimizer{MathOptInterface.Utilities.CachingOptimizer{MathOptInterface.AbstractOptimizer,MathOptInterface.Utilities.UniversalFallback{JuMP.JuMPMOIModel{Float64}}},MathOptInterface.Bridges.AllBridgedConstraints{Float64}}, ::MathOptInterface.ObjectiveFunction{MathOptInterface.SingleVariable}) at /home/blegat/.julia/dev/MathOptInterface/src/Bridges/bridgeoptimizer.jl:172
- [6] objective_function(::Model, ::Type{VariableRef}) at /home/blegat/.julia/dev/JuMP/src/objective.jl:121
+ [6] objective_function(::Model, ::Type{VariableRef}) at /home/blegat/.julia/dev/JuMP/src/objective.jl:129
  [7] top-level scope at none:0
 ```
 """

src/optimizer_interface.jl

@@ -3,28 +3,57 @@
 #  License, v. 2.0. If a copy of the MPL was not distributed with this
 #  file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
+function error_if_direct_mode(model::Model, func::Symbol)
+    if mode(model) == Direct
+        error("The `$func` function is not supported in Direct mode.")
+    end
+end
+
 # These methods directly map to CachingOptimizer methods.
 # They cannot be called in Direct mode.
-function MOIU.resetoptimizer!(model::Model, optimizer::MOI.AbstractOptimizer)
-    @assert mode(model) != Direct
-    MOIU.resetoptimizer!(caching_optimizer(model), optimizer)
+function MOIU.resetoptimizer!(model::Model, optimizer::MOI.AbstractOptimizer,
+                              bridge_constraints::Bool=true)
+    error_if_direct_mode(model, :resetoptimizer!)
+    MOIU.resetoptimizer!(backend(model), optimizer)
 end
 
 function MOIU.resetoptimizer!(model::Model)
-    @assert mode(model) != Direct
-    MOIU.resetoptimizer!(caching_optimizer(model))
+    error_if_direct_mode(model, :resetoptimizer!)
+    MOIU.resetoptimizer!(backend(model))
 end
 
 function MOIU.dropoptimizer!(model::Model)
+    error_if_direct_mode(model, :dropoptimizer!)
     @assert mode(model) != Direct
-    MOIU.dropoptimizer!(caching_optimizer(model))
+    MOIU.dropoptimizer!(backend(model))
 end
 
 function MOIU.attachoptimizer!(model::Model)
+    error_if_direct_mode(model, :attachoptimizer!)
     @assert mode(model) != Direct
-    MOIU.attachoptimizer!(caching_optimizer(model))
+    MOIU.attachoptimizer!(backend(model))
 end
 
+function set_optimizer(model::Model, optimizer_factory::OptimizerFactory;
+                       bridge_constraints::Bool=true)
+    error_if_direct_mode(model, :set_optimizer)
+    optimizer = optimizer_factory()
+    if bridge_constraints
+        # The names are handled by the first caching optimizer.
+        # If default_copy_to without names is supported, no need for a second
+        # cache.
+        if !MOIU.supports_default_copy_to(optimizer, false)
+            if mode(model) == Manual
+                # TODO figure out what to do in manual mode with the two caches
+                error("Bridges in Manual mode with an optimizer not supporting `default_copy_to` is not supported yet")
+            end
+            universal_fallback = MOIU.UniversalFallback(JuMPMOIModel{Float64}())
+            optimizer = MOIU.CachingOptimizer(universal_fallback, optimizer)
+        end
+        optimizer = MOI.Bridges.fullbridgeoptimizer(optimizer, Float64)
+    end
+    MOIU.resetoptimizer!(model, optimizer)
+end
 
 """
     optimize!(model::Model,
@@ -55,6 +84,7 @@ JuMP.optimize!(model, with_optimizer(GLPK.Optimizer))
 """
 function optimize!(model::Model,
                    optimizer_factory::Union{Nothing, OptimizerFactory}=nothing;
+                   bridge_constraints::Bool=true,
                    ignore_optimize_hook=(model.optimize_hook === nothing))
     # The nlp_data is not kept in sync, so re-set it here.
     # TODO: Consider how to handle incremental solves.
@@ -67,11 +97,11 @@ function optimize!(model::Model,
         if mode(model) == Direct
             error("An optimizer factory cannot be provided at the `optimize` call in Direct mode.")
         end
-        if MOIU.state(caching_optimizer(model)) != MOIU.NoOptimizer
+        if MOIU.state(backend(model)) != MOIU.NoOptimizer
             error("An optimizer factory cannot both be provided in the `Model` constructor and at the `optimize` call.")
         end
-        optimizer = optimizer_factory()
-        MOIU.resetoptimizer!(model, optimizer)
+        set_optimizer(model, optimizer_factory,
+                      bridge_constraints=bridge_constraints)
         MOIU.attachoptimizer!(model)
     end
 

test/constraint.jl

@@ -353,13 +353,13 @@ function test_shadow_price(model_string, constraint_dual, constraint_shadow)
                                          JuMP.JuMPMOIModel{Float64}(),
                                          eval_objective_value=false,
                                          eval_variable_constraint_dual=false))
-    mock_optimizer = JuMP.caching_optimizer(model).optimizer
+    mock_optimizer = JuMP.backend(model).optimizer.model
     MOI.set(mock_optimizer, MOI.TerminationStatus(), MOI.Success)
     MOI.set(mock_optimizer, MOI.DualStatus(), MOI.FeasiblePoint)
     JuMP.optimize!(model)
 
     @testset "shadow price of $constraint_name" for constraint_name in keys(constraint_dual)
-        ci = MOI.get(JuMP.caching_optimizer(model), MOI.ConstraintIndex,
+        ci = MOI.get(JuMP.backend(model), MOI.ConstraintIndex,
                      constraint_name)
         constraint_ref = JuMP.ConstraintRef(model, ci, JuMP.ScalarShape())
         MOI.set(mock_optimizer, MOI.ConstraintDual(),

test/generate_and_solve.jl

@@ -37,13 +37,13 @@
 
         model = JuMP.JuMPMOIModel{Float64}()
         MOIU.loadfromstring!(model, modelstring)
-        MOIU.test_models_equal(JuMP.caching_optimizer(m).model_cache, model, ["x","y"], ["c", "xub", "ylb"])
+        MOIU.test_models_equal(JuMP.backend(m).model_cache, model, ["x","y"], ["c", "xub", "ylb"])
 
         JuMP.optimize!(m, with_optimizer(MOIU.MockOptimizer,
                                          JuMP.JuMPMOIModel{Float64}(),
                                          eval_objective_value=false))
 
-        mockoptimizer = JuMP.caching_optimizer(m).optimizer
+        mockoptimizer = JuMP.backend(m).optimizer.model
         MOI.set(mockoptimizer, MOI.TerminationStatus(), MOI.Success)
         MOI.set(mockoptimizer, MOI.ObjectiveValue(), -1.0)
         MOI.set(mockoptimizer, MOI.ResultCount(), 1)
@@ -138,11 +138,11 @@
 
         model = JuMP.JuMPMOIModel{Float64}()
         MOIU.loadfromstring!(model, modelstring)
-        MOIU.test_models_equal(JuMP.caching_optimizer(m).model_cache, model, ["x","y"], ["xfix", "xint", "ybin"])
+        MOIU.test_models_equal(JuMP.backend(m).model_cache, model, ["x","y"], ["xfix", "xint", "ybin"])
 
         MOIU.attachoptimizer!(m)
 
-        mockoptimizer = JuMP.caching_optimizer(m).optimizer
+        mockoptimizer = JuMP.backend(m).optimizer.model
         MOI.set(mockoptimizer, MOI.TerminationStatus(), MOI.Success)
         MOI.set(mockoptimizer, MOI.ObjectiveValue(), 1.0)
         MOI.set(mockoptimizer, MOI.ResultCount(), 1)
@@ -186,13 +186,13 @@
 
         model = JuMP.JuMPMOIModel{Float64}()
         MOIU.loadfromstring!(model, modelstring)
-        MOIU.test_models_equal(JuMP.caching_optimizer(m).model_cache, model, ["x","y"], ["c1", "c2", "c3"])
+        MOIU.test_models_equal(JuMP.backend(m).model_cache, model, ["x","y"], ["c1", "c2", "c3"])
 
         JuMP.optimize!(m, with_optimizer(MOIU.MockOptimizer,
                                          JuMP.JuMPMOIModel{Float64}(),
                                          eval_objective_value=false))
 
-        mockoptimizer = JuMP.caching_optimizer(m).optimizer
+        mockoptimizer = JuMP.backend(m).optimizer.model
         MOI.set(mockoptimizer, MOI.TerminationStatus(), MOI.Success)
         MOI.set(mockoptimizer, MOI.ObjectiveValue(), -1.0)
         MOI.set(mockoptimizer, MOI.ResultCount(), 1)
@@ -245,7 +245,7 @@
 
         model = JuMP.JuMPMOIModel{Float64}()
         MOIU.loadfromstring!(model, modelstring)
-        MOIU.test_models_equal(JuMP.caching_optimizer(m).model_cache, model, ["x","y","z"], ["varsoc", "affsoc", "rotsoc"])
+        MOIU.test_models_equal(JuMP.backend(m).model_cache, model, ["x","y","z"], ["varsoc", "affsoc", "rotsoc"])
 
         mockoptimizer = MOIU.MockOptimizer(JuMP.JuMPMOIModel{Float64}(),
                                            eval_objective_value=false,
@@ -308,7 +308,7 @@
 
         model = JuMP.JuMPMOIModel{Float64}()
         MOIU.loadfromstring!(model, modelstring)
-        MOIU.test_models_equal(JuMP.caching_optimizer(m).model_cache, model,
+        MOIU.test_models_equal(JuMP.backend(m).model_cache, model,
                                ["x11","x12","x22"],
                                ["var_psd", "sym_psd", "con_psd"])