Fix lns init phase + constraint handler mix for minizinc subsolvers

- Fix ConstraintHandlerMix for minizinc subsolvers (and add test)
- Raise errors when lns has no initial_solution_provider but does not
  skip the initial phase or when it does not return a solution

discrete_optimization/generic_tools/lns_tools.py

@@ -326,15 +326,25 @@ def solve(
 
         self.init_model(**kwargs)
 
-        if (
-            not skip_initial_solution_provider
-            and self.initial_solution_provider is not None
-        ):
+        if skip_initial_solution_provider:
+            best_objective = (
+                float("inf") if sense == ModeOptim.MINIMIZATION else -float("inf")
+            )
+            store_lns = None
+            stopping = False
+        else:
+            if self.initial_solution_provider is None:
+                raise ValueError(
+                    "`initial_solution_provider` cannot be None "
+                    "if `skip_initial_solution_provider` is False."
+                )
             store_lns = self.initial_solution_provider.get_starting_solution()
             store_lns = self.post_process_solution.build_other_solution(store_lns)
             init_solution, objective = store_lns.get_best_solution_fit()
             if init_solution is None:
-                satisfy = False
+                raise RuntimeError(
+                    "`initial_solution_provider` + `post_process_solution` gave no solution."
+                )
             else:
                 satisfy = self.problem.satisfy(init_solution)
             logger.debug(f"Satisfy Initial solution {satisfy}")
@@ -348,12 +358,6 @@ def solve(
             best_objective = objective
             # end of step callback: stopping?
             stopping = callbacks_list.on_step_end(step=0, res=store_lns, solver=self)
-        else:
-            best_objective = (
-                float("inf") if sense == ModeOptim.MINIMIZATION else -float("inf")
-            )
-            store_lns = None
-            stopping = False
 
         result_store: ResultStorage
         lsn_contraints: Optional[Iterable[Any]] = None
@@ -590,7 +594,10 @@ def adding_constraint_from_results_store(
         self.last_index_param = choice
         self.status[self.last_index_param]["nb_usage"] += 1
         logger.debug(f"Status {self.status}")
-        return ch.adding_constraint_from_results_store(solver, result_storage, **kwargs)
+        constraints = ch.adding_constraint_from_results_store(
+            solver=solver, result_storage=result_storage, **kwargs
+        )
+        return constraints
 
     def remove_constraints_from_previous_iteration(
         self, solver: SolverDO, previous_constraints: Iterable[Any], **kwargs: Any

tests/rcpsp/solvers/test_lns.py

@@ -6,10 +6,15 @@
 from discrete_optimization.generic_rcpsp_tools.solvers.lns_cp import (
     LnsCpMznGenericRcpspSolver,
 )
+from discrete_optimization.generic_rcpsp_tools.solvers.lns_cp.neighbor_builder import (
+    mix,
+)
 from discrete_optimization.generic_tools.callbacks.early_stoppers import TimerStopper
 from discrete_optimization.generic_tools.cp_tools import ParametersCp
+from discrete_optimization.generic_tools.lns_cp import LnsCpMzn
 from discrete_optimization.rcpsp.parser import get_data_available, parse_file
 from discrete_optimization.rcpsp.problem import RcpspProblem
+from discrete_optimization.rcpsp.solvers.cp_mzn import CpRcpspSolver
 
 
 @pytest.mark.parametrize(
@@ -34,3 +39,26 @@ def test_lns_solver(file_name):
     )
     sol, fit = results.get_best_solution_fit()
     assert rcpsp_problem.satisfy(sol)
+
+
+@pytest.mark.parametrize(
+    "file_name",
+    ["j1201_1.sm"],
+)
+def test_mix_constraints_handlers(file_name):
+    files_available = get_data_available()
+    file = [f for f in files_available if file_name in f][0]
+    rcpsp_problem: RcpspProblem = parse_file(file)
+
+    constraints_handler = mix(rcpsp_problem=rcpsp_problem)
+    solver = LnsCpMzn(
+        problem=rcpsp_problem,
+        subsolver=CpRcpspSolver(problem=rcpsp_problem),
+        constraint_handler=constraints_handler,
+    )
+    solver.init_model()
+    res = solver.solve(
+        nb_iteration_lns=5, skip_initial_solution_provider=True, time_limit_subsolver=5
+    )
+    sol, fit = res.get_best_solution_fit()
+    assert rcpsp_problem.satisfy(sol)