Minimizing the sum of the infeasabilities is working
well. Since I know that a subset of the constraints (the section
sizes) are the only possible causes of infeasability, the non-zero infeasability
var (there appears to always be exactly one in practice) tells me what
kind of memory is lacking.
> > This question probably reveals my poor grasp
of MIP solving, but here
> > goes:
> > I'm trying to use glpsol in MIP mode as part of a software build
> > specifically a locate step. The problem being solved is
to assign various
> > code sections into various physical memories, and all the constraints
> > the problem map nicely to integer linear programming. That
> > fine. However, I'm having a difficult time producing meaningfull
> > output in the event that the constraints are infeasable (which
> > problem means there's not enough physical memory space to locate
> > sections). What I would like to get as output in that case is
> > mutually infeasable constraints, or the most infeasable constraint,
> > really anything that would allow me to generate some vaguely
> > output when the locate fails.
> As you suspected, the difficulty is not peculiar to GLPK,
> it's an aspect of MILPs and even LPs.
> Try minimizing the sum of the infeasibiliities.
> If you can, it might be useful to tell glpsol to print
> a solution whenever it finds a new and improved one.
> A set of original constraints which are tight or violated
> is a set of mutually infeasible constraints.