benchmark tools fail for CUTEst problems

[vepiteski]

Even if I added finalize(problem) in solve_problems, CUTE complains: ERROR: LoadError: CUTEst: call cutest_finalize on current model first

[abelsiqueira]

Can you show the code? Was that finalize or cutest_finalize?

[vepiteski]

It was finalize, at this stage, a problem can be any abstractNLPModel and I expected it to be overloaded with the specific version for any model. I also added printskip to eventually print skipped problems.

function solve_problems(solver :: Function, problems :: Any; prune :: Bool=true, printskip :: Bool = false, kwargs...)
  display_header()
  nprobs = length(problems)
  verbose = nprobs ≤ 1
  stats = -ones(nprobs, 3)
  k = 0
  for problem in problems
    try  
      (f, g, h) = solve_problem(solver, problem, verbose=verbose; kwargs...)
      k = k + 1
      stats[k, :] = [f, g, h]
    catch e
      isa(e, SkipException) || rethrow(e)
      printskip && @printf("%-15s  %8d          ncons = %-8d  %13s   skipped\n",problem.meta.name, 
problem.meta.nvar, problem.meta.ncon, unconstrained(problem.meta) ? "unconstrained" : 
"constrained")
    end
    finalize(problem)
  end
  return prune ? stats[1:k, :] : stats
end

[abelsiqueira]

Ah ok, looks like finalize is the default for Julia (http://docs.julialang.org/en/release-0.5/stdlib/base/), not specifically for NLPModels. cutest_finalize is not a finalizer yet. It should be easy to change that, though. I'm currently debugging another thing, then I'll go to that.

[abelsiqueira]

Closed by #91