Support HSL solvers

[amontoison]

Related to #69
@PierreMartinon @ocots

[ocots]

@amontoison Are we ok that with your proposal then the user won't have to do using HSL or import HSL_jll like here: https://github.com/0Yassine0/COTS.jl/blob/main/Benchmark/GoddadModeles.ipynb

[amontoison]

Yes, the users don't need to do using HSL or import HSL_jll.
We check that HSL solvers are installed or not here and try to use by default.
"mumps" is used as fallback if we can't use a linear solver.

[amontoison]

@amontoison Are we ok that with your proposal then the user won't have to do using HSL or import HSL_jll like here: https://github.com/0Yassine0/COTS.jl/blob/main/Benchmark/GoddadModeles.ipynb

@ocots
You did a typo in your notebook. You used ma27 instead of ma57 for some benchmarks:

sol_ma57 = OptimalControl.solve(RocketOC,init=init,output_file="rocketOC.out",linear_solver="ma27",hsllib=HSL_jll.libhsl_path);
bOC_ma57 = @benchmark (OptimalControl.solve(RocketOC,init=init,display=false,linear_solver="ma27",hsllib=HSL_jll.libhsl_path));

[amontoison]

Two tests are failing locally but I don't think that it's related to this PR:

Test: abstract OCP definition
:goddard :max_rf :abstract :constr: Test Failed at /home/alexis/Bureau/git/CTDirect.jl/test/suite/abstract_ocp.jl:83
  Expression: ≈(sol3.objective, 1.0125, rtol = 0.01)
   Evaluated: 1.002093564754221 ≈ 1.0125 (rtol=0.01)

Stacktrace:
 [1] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:672 [inlined]
 [2] macro expansion
   @ ~/Bureau/git/CTDirect.jl/test/suite/abstract_ocp.jl:83 [inlined]
 [3] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:1577 [inlined]
 [4] top-level scope
   @ ~/Bureau/git/CTDirect.jl/test/suite/abstract_ocp.jl:82
Test: all constraint types
Test: discrete continuation
INFO: normalizing given time grid...
:global_variable :warm_start: Test Failed at /home/alexis/Bureau/git/CTDirect.jl/test/suite/continuation.jl:124
  Expression: ≈(mean(iter_list), 17, rtol = 0.01)
   Evaluated: 18.0 ≈ 17 (rtol=0.01)

Stacktrace:
 [1] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:672 [inlined]
 [2] macro expansion
   @ ~/Bureau/git/CTDirect.jl/test/suite/continuation.jl:124 [inlined]
 [3] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:1577 [inlined]
 [4] top-level scope
   @ ~/Bureau/git/CTDirect.jl/test/suite/continuation.jl:114

[ocots]

Two tests are failing locally but I don't think that it's related to this PR:

Test: abstract OCP definition
:goddard :max_rf :abstract :constr: Test Failed at /home/alexis/Bureau/git/CTDirect.jl/test/suite/abstract_ocp.jl:83
  Expression: ≈(sol3.objective, 1.0125, rtol = 0.01)
   Evaluated: 1.002093564754221 ≈ 1.0125 (rtol=0.01)

Stacktrace:
 [1] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:672 [inlined]
 [2] macro expansion
   @ ~/Bureau/git/CTDirect.jl/test/suite/abstract_ocp.jl:83 [inlined]
 [3] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:1577 [inlined]
 [4] top-level scope
   @ ~/Bureau/git/CTDirect.jl/test/suite/abstract_ocp.jl:82
Test: all constraint types
Test: discrete continuation
INFO: normalizing given time grid...
:global_variable :warm_start: Test Failed at /home/alexis/Bureau/git/CTDirect.jl/test/suite/continuation.jl:124
  Expression: ≈(mean(iter_list), 17, rtol = 0.01)
   Evaluated: 18.0 ≈ 17 (rtol=0.01)

Stacktrace:
 [1] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:672 [inlined]
 [2] macro expansion
   @ ~/Bureau/git/CTDirect.jl/test/suite/continuation.jl:124 [inlined]
 [3] macro expansion
   @ ~/julia/julia-1.10.4/share/julia/stdlib/v1.10/Test/src/Test.jl:1577 [inlined]
 [4] top-level scope
   @ ~/Bureau/git/CTDirect.jl/test/suite/continuation.jl:114

I do not have this error...

[ocots]

C'est pénible, à cause de notre registre perso, j'ai l'impression que les tests passent pas. J'ai essayé d'updater les fichiers yml mais ça n'a aucun rapport. Il faudrait pas merge directement dans main mais dans une branche hsl de CTDirect.jl.

J'ai redirigé vers hsl. J'accepte puis test de merge dans main.

[amontoison]

Est-ce que tu as installé la version officielle de HSL_jll.jl sur ta machine?
Peut-être que ça change la solution et les tests ne passent que lorsqu'on utilise mumps.

[ocots]

Non j'ai rien installé. Du coup, tout se fait avec mumps. Du coup faudrait inspecter ça.

@PierreMartinon Point intéressant !

[PierreMartinon]

Non j'ai rien installé. Du coup, tout se fait avec mumps. Du coup faudrait inspecter ça.

@PierreMartinon Point intéressant !

Gouzi ? J'ai rien fait moi !

Ce que je peux dire c'est que le test sur mean(iter_list) compare le nombre moyen d'iterations sur une serie de problemes (continuation discrete). Ce n'est peut-etre pas tres robuste a cause de l'arrondi donc je pourrai faire une comparaison sur des flottants au lieu d'entiers.

(note: en pleins travaux ces jours-ci, donc pas super reactif)

[jbcaillau]

Gouzi ? @ocots tu as quand même pas mergé alors que les tests passaient pas (@amontoison dis qqchose merde 🤭)

[ocots]

Les tests passent car on ne teste pas avec autre chose que Mumps.

[jbcaillau]

Dans le notebook je vois des tests avec les solveurs HSL : ça fonctionne ou pas ? L'install est documentée qqpart ?

[ocots]

Oui Yassine a des résultats avec les solveurs HSL. Pour l'installation voir control-toolbox/CTBenchmarks.jl#3