Solve status, infeasiblity, and unboundedness

[IainNZ]

• Right now we place a box around the variables because otherwise the initial solve with no cutting planes is unbounded.
  • One way is to be clever with the rays - probably the right thing to do
  • Another way is relax the box at some point - but I feel like this could still go wrong
• I don't think there is any handling right now of infeasibility in the cutting plane problem
• If you have an empty uncertainty set I think it makes the solution "crazy"
• Checking whether the reformulation is empty would require one cutting plane solve...

[vgupta1]

A somewhat related issue is that it's fairly easy to specify an uncertainty set which is empty (i.e. the constraints defining it are infeasible.) More often than not this leads to a unbounded master problem, and it's pain to locate. It would be nice to be able to run a check (user requests it) that the uncertainty set is non-empty....

[IainNZ]

Yes I had been thinking about that but didn't have a good solution - making it user checkable is a great idea

[IainNZ]

(as you can see I have ton of debug flags in there right now, all undocumented)

[IainNZ]

I'm stress testing this right now before the 0.1 release, adding new tests and making things more consistent.