[Help-glpk] Re: testing for optimal solution

[Andrew Makhorin]

> I'm converting some PLI models from XPress/Mosel to GNU Mathprog using
> the glpsol tool provided with GLPK as part of an academic work. Until
> now, I found success with all issues, but one:
> The model I received is able to do tests for the optimal solution with
> something (pseudo) like this:
> 
> if var_that_holds_the_optiomal_solution = SOME_THING
>  {
>     print all the results found
>  }
> else
> {
>     print "no solution found"
> }
> 
> I found nowhere a description saying that such variables that hold
> internal value of the solver are available from the model. Can I do
> that ? If I do, is there a list of all kinds of variables like this
> that might help me in the future?
> 
> I do need that, because theses programs will be run on a testsuite
> together with the old Mosel ones, and they both are expected to return
> an unified output.

Unfortunately this feature is not implemented yet. If it is vital,
you can change the glpsol source code. Just replace line 871, file
glpsol.c

      if (mpl != NULL && mpl_has_solve_stmt(mpl))

by the following

      if (mpl != NULL && mpl_has_solve_stmt(mpl) &&
          lpx_get_status(lp) == LPX_OPT)

Then, if optimal solution does not exist or was not found, glpsol will
not call the mathprog postsolver.

Andrew Makhorin