make setting and getting the objective constant a part of the LQOI interface

[rdeits]

First step for #43

[codecov-io]

Codecov Report

Merging #44 into master will increase coverage by 0.05%.
The diff coverage is 88.88%.

@@            Coverage Diff             @@
##           master      #44      +/-   ##
==========================================
+ Coverage   75.29%   75.35%   +0.05%     
==========================================
  Files          13       13              
  Lines        1263     1266       +3     
==========================================
+ Hits          951      954       +3     
  Misses        312      312

Impacted Files	Coverage Δ
src/solve.jl	0% <ø> (ø)	⬆️
src/solver_interface.jl	53.33% <100%> (+11.66%)	⬆️
src/objective.jl	88.09% <83.33%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d6957f3...4c8289a. Read the comment docs.

[rdeits]

Gurobi.jl tests also pass for me with LQOI set to this branch.

[rdeits]

I tried updating Gurobi.jl to overload the functions introduced here by setting ObjCon, but i think I may have found a bug in Gurobi itself. While setting ObjCon before solving the model works once, further modifications to it have no effect on the objective value returned by later solves. You should be able to reproduce this without needing this particular branch:

model = GurobiOptimizer();
MOI.empty!(model)
MOIU.loadfromstring!(model,"""
    variables: x
    maxobjective: 1.0x
    c1: 1.0x <= 1.0
""")
x = MOI.get(model, MOI.VariableIndex, "x")

Gurobi.set_dblattr!(model.inner, "ObjCon", 1.0)  # this works
Gurobi.update_model!(model.inner)
MOI.optimize!(model)
# This assertion passes
@assert MOI.get(model, MOI.ObjectiveValue()) == 2.0  # (1.0 * 1.0 + 1.0)

Gurobi.set_dblattr!(model.inner, "ObjCon", 2.0)  # this has no effect
Gurobi.update_model!(model.inner)
MOI.optimize!(model)
# This assertion fails:
@assert MOI.get(model, MOI.ObjectiveValue()) == 3.0  # (1.0 * 1.0 + 2.0)

[odow]

Do you have python/gurobipy installed? Otherwise I can have a look when I get back to my desk (edit: to see if it is a gurobi bug)

[rdeits]

Yeah, I'm trying it in python now

[rdeits]

Hm, I'm not sure how to reproduce this in Python, since the API is totally different. I tried doing:

from __future__ import print_function
import gurobipy as grb

m = grb.Model("lp")
x = m.addVar(name="x")
obj = 1.0 * x + 1.0
m.setObjective(obj, sense=grb.GRB.MAXIMIZE)
m.addConstr(1.0 * x <= 1.0)
m.optimize()
print("objval:", m.getObjective().getValue())

m.setObjective(1.0 * x + 2.0)
m.optimize()
print("objval:", m.getObjective().getValue())

which works just fine, but that involves replacing the entire objective expression, which isn't what we're doing in Julia.

[rdeits]

In Julia, I get the correct result if I also set the "Obj" attribute (to the same value it had before):

model = GurobiOptimizer();
MOI.empty!(model)
MOIU.loadfromstring!(model,"""
    variables: x
    maxobjective: 1.0x
    c1: 1.0x <= 1.0
""")
x = MOI.get(model, MOI.VariableIndex, "x")

Gurobi.set_dblattr!(model.inner, "ObjCon", 1.0)  # this works
Gurobi.update_model!(model.inner)
MOI.optimize!(model)
# This assertion passes
@assert MOI.get(model, MOI.ObjectiveValue()) == 2.0  # (1.0 * 1.0 + 1.0)

Gurobi.set_dblattr!(model.inner, "ObjCon", 2.0) 
Gurobi.set_dblattrarray!(model.inner, "Obj", 1, 1, ones(1))
Gurobi.update_model!(model.inner)
MOI.optimize!(model)
@assert MOI.get(model, MOI.ObjectiveValue()) == 3.0  # (1.0 * 1.0 + 2.0)

[rdeits]

Ok, I can reproduce the issue in Python:

from __future__ import print_function
import gurobipy as grb

m = grb.Model("lp")
x = m.addVar(name="x")
obj = 1.0 * x + 1.0
m.setObjective(obj, sense=grb.GRB.MAXIMIZE)
m.addConstr(1.0 * x <= 1.0)
m.optimize()
print("ObjVal:", m.getAttr("ObjVal"))

m.setAttr("ObjCon", 2.0)
m.optimize()
print("ObjVal:", m.getAttr("ObjVal"))

[odow]

Have you sent them an issue (https://groups.google.com/forum/#!newtopic/gurobi)? Otherwise I'll do it now.

[rdeits]

I just did (but it looks like my message is still being held for moderation)

[odow]

Since @joaquimg is happy, would you like me to merge this now or wait for Gurobi?

[rdeits]

I'd say go ahead and merge. The issue with Gurobi is annoying but not related to this PR, and I think this is a good change nonetheless.