Slow calculations for dual_objective_value()

[trulsf]

I have a primal infeasible model where I want to find the dual objective value (following the Benders' tutorial on generating feasibility cuts). While solving the problem in HiGHS (v1.9.3) takes less than a second, the calculation for dual_objective_value takes around 2 minutes. It seems that this time is spent in MOI.Utilities.get_fallback which is triggered by the existence of a dual ray for the solution (MOI_wrapper.jl, line 2046):

if model.solution.has_dual_ray
    return MOI.Utilities.get_fallback(model, attr, Float64)

Is this expected behavior? My problem is fairly large, but I did not expect such long running times. If needed, I can share the model.

A JuMP Model
├ solver: HiGHS
├ objective_sense: MIN_SENSE
│ └ objective_function_type: AffExpr
├ num_variables: 91171
├ num_constraints: 182159
│ ├ AffExpr in MOI.EqualTo{Float64}: 81350
│ ├ AffExpr in MOI.GreaterThan{Float64}: 1848
│ ├ AffExpr in MOI.LessThan{Float64}: 13524
│ ├ VariableRef in MOI.EqualTo{Float64}: 7473
│ ├ VariableRef in MOI.GreaterThan{Float64}: 77963
│ └ VariableRef in MOI.LessThan{Float64}: 1
└ Names registered in the model: none

Looking at the profiled results, almost all time is spent in Highs_getRowsByRange. So maybe this is related to #207.

[odow]

It's the same cause as #207.

I fixed the normal case in #226, but I didn't fix the dual ray case.