The Constraint Database Approach to Software Verification

Based on constraint database techniques, we present a new approach to software verification. This new approach has some similarity to abstract interpretation that uses various widening operators; therefore, we call the new approach l-u widening. We show that our l-u widening leads to a more precise over-approximation of the invariants in a program than comparable previously proposed widening operators based on difference-bound matrices, although l-u widening can be computed as efficiently as the other widening operators. We show that constraint database techniques can compute non-convex program invariants too. Finally, we give a compact representation of addition-bound matrices, which generalize difference-bound matrices.