Originally posted by: glebB

I try to figure out a reliable way to query solution status in C. There are many status flags for different outcomes and their documentation is skinny. And I know that has been changing between CPLEX versions. Is there a sustainable way to query the result states (OPT, FEAS, INFEAS, UNBND, INForUNBND, UNKNOWN, ERROR), similar to IloAlgorithm? At the moment I use the following switch and not sure if it's correct:

   /* Converting the status. */
   switch(cplexStatus) {
     case CPXMIP_OPTIMAL:
     case CPXMIP_OPTIMAL_TOL:
       s = Status::OPT;
       break;
     case CPXMIP_INFEASIBLE:
//      case CPXMIP_OPTIMAL_INFEAS:
     case CPXMIP_INForUNBD:
       s = Status::UNSAT;
       break;
     case CPXMIP_SOL_LIM:
     case CPXMIP_NODE_LIM_FEAS:
     case CPXMIP_TIME_LIM_FEAS:
     case CPXMIP_FAIL_FEAS:
     case CPXMIP_MEM_LIM_FEAS:
     case CPXMIP_ABORT_FEAS:
     case CPXMIP_FAIL_FEAS_NO_TREE:
       s = Status::SAT;
       break;
     case CPXMIP_UNBOUNDED:
     case CPXMIP_ABORT_RELAXATION_UNBOUNDED:
       s = Status::UNBND;
       break;
     case CPXMIP_ABORT_INFEAS:
     case CPXMIP_FAIL_INFEAS:
       s = Status::ERROR;
       break;
     default:
       s = Status::UNKNOWN;
   }

Thanks

What may help is the CPXsolninfo() function. This gives you a quick way to assess whether the problem was found primal/dual feasible and what type of solution is available. Other than that, I don't see anything better than enumerating the CPLEX status values.

Originally posted by: glebB

Currently doing as follows, using CPXgetsolnpoolnumsolns():
 

   /* Converting the status. */
   switch(cplexStatus) {
     case CPXMIP_OPTIMAL:
       s = Status::OPT;
       wrap_assert(CPXgetsolnpoolnumsolns(env, lp), "Optimality reported but pool empty?", false);
       break;
     case CPXMIP_INFEASIBLE:
       s = Status::UNSAT;
       break;
//      case CPXMIP_OPTIMAL_INFEAS:
     case CPXMIP_INForUNBD:
       s = Status::UNSATorUNBND;
       break;
     case CPXMIP_SOL_LIM:
     case CPXMIP_NODE_LIM_FEAS:
     case CPXMIP_TIME_LIM_FEAS:
     case CPXMIP_FAIL_FEAS:
     case CPXMIP_MEM_LIM_FEAS:
     case CPXMIP_ABORT_FEAS:
     case CPXMIP_FAIL_FEAS_NO_TREE:
       s = Status::SAT;
       wrap_assert(CPXgetsolnpoolnumsolns(env, lp), "Feasibility reported but pool empty?", false);
       break;
     case CPXMIP_UNBOUNDED:
       s = Status::UNBND;
       break;
//     case CPXMIP_ABORT_INFEAS:
     case CPXMIP_FAIL_INFEAS:
       s = Status::ERROR;
       break;
     default:
//      case CPXMIP_OPTIMAL_TOL:
//      case CPXMIP_ABORT_RELAXATION_UNBOUNDED:
       if (CPXgetsolnpoolnumsolns (env, lp))
         s = Status::SAT;
       else
        s = Status::UNKNOWN;
   }

Another question, is there any primal bound? Sometimes it is possible to obtain such a bound even without having any primal solution.

CPLEX does not compute a primal bound unless the model has proven to be feasible (i.e., there is an incumbent solution). It does that because the presence of a primal bound would suggest feasibility and that was not proven yet. If you know a primal bound and want CPLEX to use this to cut off nodes then you can tell this bound to CPLEX via parameters CPX_PARAM_CUTLO or CPX_PARAM_CUTUP.