> BTW, glp_intopt allows solving pure lp (i.e. having no integer > variables), so glp_simplex needs not to be used at all. Would this > simplify the logic?
I think this was an effort to avoid the glp_intopt overhead when not necessary. However, a few quick test with miplib problems with only binary and continuous variables (where glp_simplex is used instead of glp_intopt in do_refine) indicate that the slow-down when using glp_intopt unconditionally is minor - less that 1% in most cases, while in a few cases it seems faster. As an aside, I only get the first solution in feasibility pump compared to what you showed earlier: ... Applying FPUMP heuristic... Pass 1 Solution found by heuristic: -46.3368032777 Pass 1 Pass 2 Pass 3 Pass 4 Pass 5 ... Do you have any other changes in your version? Best Regards, Chris Matrakidis _______________________________________________ Bug-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/bug-glpk
