[Bug-glpk] Dualp simplex method classifies an unbounded problem as "no primal feasible"

[Price, David T]

I am testing GLPK 4.36. I tried a test case that I have also used with
GLPK 4.19.
 
Maximize 
       40 * x1 + 60 * x2
subject to
   0 <=     x1
   0 <=               x2
  70 <= 2 * x1 +      x2
  40 <=     x1 +      x2
  90 <=     x1 +  3 * x2
 
This problem is primal feasible. It is unbounded. 

In GLPK 4.19, after running the dualp simplex method (GLP_DUALP), the
solution status is unbounded (GLP_UNBND). This is what I expected.

In GLPK 4.36, after running the dualp simplex method (GLP_DUALP), the
solution status is infeasible (GLP_INFEAS). Why doesn't the code find a
feasible solution?

David T. Price  (address@hidden) 
Principal 
Global Markets Group Analytics US 
Bank of America, Chicago 
312-234-2519