Adding constants to parameters in the objective

[bnarum]

Hi

I found an issue when trying out this package. It seems an error occurs when trying to add a constant to a parameter in the objective.

Here is an example:

using JuMP, ParametricOptInterface, Gurobi
const POI = ParametricOptInterface
optimizer = () -> POI.Optimizer(Gurobi.Optimizer())

mdl = Model(optimizer)
@variable(mdl, p in POI.Parameter(1.0))
@variable(mdl, 0 <= x <= 1)
@objective(mdl, Max, (p + 0.5) * x)
optimize!(mdl)
@show objective_value(mdl)
@show value(objective_function(mdl))

objective_value(mdl) = 0.5

value(objective_function(mdl)) = 1.5

The optimal solution should be 1.5, but it becomes 0.5. It is as if the parameter goes to zero when added to a constant. This seems to only happen in the objective, not for constraints.

Using JuMP v1.3.0, ParametricOptInterface v0.4.3 and Gurobi v0.11.3

[odow]

I can reproduce:

julia> model = Model(() -> POI.Optimizer(GLPK.Optimizer()))
A JuMP Model
Feasibility problem with:
Variables: 0
Model mode: AUTOMATIC
CachingOptimizer state: EMPTY_OPTIMIZER
Solver name: Parametric Optimizer with GLPK attached

julia> @variable(model, p in POI.Parameter(1.0))
p

julia> @variable(model, 0 <= x <= 1)
x

julia> @objective(model, Max, (p + 0.5) * x)
p*x + 0.5 x

julia> optimize!(model)

julia> @test value(x) ≈ 1.0
Test Passed

julia> @test objective_value(model) ≈ 1.5
Test Failed at REPL[20]:1
  Expression: objective_value(model) ≈ 1.5
   Evaluated: 0.5 ≈ 1.5
ERROR: There was an error during testing

julia> @test value(objective_function(model)) ≈ 1.5
Test Passed

[odow]

The issue is that the objective passed to GLPK is just completely wrong:

julia> print(unsafe_backend(model).optimizer)
Maximize ScalarAffineFunction{Float64}:
 0.0 + 0.5 x

Subject to:

VariableIndex-in-GreaterThan{Float64}
 x >= 0.0

VariableIndex-in-LessThan{Float64}
 x <= 1.0