Add batched modification methods

[joaquimg]

Modify problems bit by bit can be surprisingly slow in some solvers (like Xpress). Batch modifications like this solve the problems. I was working on a model that changes from 750s to 150s build+modify+solve time after using batch modifications.

rhs would be weird in plural as rhss, I don't know what to do about it.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.41%. Comparing base (1aedb44) to head (48012e9).
Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3716      +/-   ##
==========================================
+ Coverage   98.37%   98.41%   +0.03%     
==========================================
  Files          43       43              
  Lines        5737     5813      +76     
==========================================
+ Hits         5644     5721      +77     
+ Misses         93       92       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[blegat]

Looks good, only minor comments

[pedroripper]

Hey everyone!

I tried to use this new dispatch for set_normalized_coefficient but got an error... here is a MWE

using JuMP
using HiGHS

model = Model(HiGHS.Optimizer)

@variable(model, x[1:3] >= 0)
@constraint(model, c[i in 1:3], x[i] == i)

constraint_coeffs = Vector{JuMP.ConstraintRef}()
variables = Vector{JuMP.VariableRef}()
coefficients = Vector{Float64}()

for i in 1:3
    push!(constraint_coeffs, c[i])
    push!(variables, x[i])
    push!(coefficients, 5)
end

set_normalized_coefficient(constraint_coeffs, variables, coefficients)

And when I am more specific about the type of the constraint reference vector, it works

using JuMP
using HiGHS

model = Model(HiGHS.Optimizer)

@variable(model, x[1:3] >= 0)
@constraint(model, c[i in 1:3], x[i] == i)

constraint_coeffs =  Vector{JuMP.ConstraintRef{JuMP.Model, MOI.ConstraintIndex{MOI.ScalarAffineFunction{Float64}, MOI.EqualTo{Float64}}, ScalarShape}}()
variables = Vector{JuMP.VariableRef}()
coefficients = Vector{Float64}()

for i in 1:3
    push!(constraint_coeffs, c[i])
    push!(variables, x[i])
    push!(coefficients, 5)
end

set_normalized_coefficient(constraint_coeffs, variables, coefficients)

[odow]

@joaquimg I'll fix the error messages.

@pedroripper this is expected behavior because of the abstract Vector{ConstraintRef}.

[franckgaga]

@joaquimg @odow @blegat I just wanted to thank you for this excellent contribution, you are doing amazing work!

ModelPredictiveControl.jl uses OSQP.jl by default and it also has this issue. I raised an issue (#3566) in which I was suspecting a type instability for the bad performance. It was not the culprit at all. Using the new set_normalized_rhs and set_objective_coefficient methods with vectors drastically improved the performance ! 🥳