Static Infinite Wait Anomaly Detection in Polynomial Time

In nite wait anomalies associated with a barrier rendezvous model (e.g., Ada) can be divided into two classes: stalls and deadlocks. Although precise static deadlock detection is NP-hard, we present two polynomial time algorithms which operate on a statically derivable program representation, the sync graph, to certify a useful class of programs free of deadlocks. We identify three conditions local to any deadlocked tasks, and a fourth global condition on all tasks, which must occur in the sync graph of any program which can deadlock. Again, exact checking of the local conditions is NP-hard; the algorithms check them using conservative approximations. Certifying stall freedom is intractable for programs with conditional branching, including loops. We give program transforms which may help alleviate this di culty.