MKorat : a novel approach for memorizing the Korat search and some potential applications

Writing logical constraints that describe properties of desired inputs enables an effective approach for systematic software testing, which can find many bugs. The key problem in systematic constraint-based testing is efficiently exploring very large spaces of all possible inputs to enumerate desired valid inputs. The Korat technique provides an effective solution to this problem. Korat uses desired input properties written as imperative predicates and implements a backtracking search that prunes large parts of the input space and enumerates all non-isomorphic inputs within a given bound on input size. Despite the effectiveness of Korat’s pruning, systematically creating and running large numbers of tests can be costly in practice. Previous work introduced parallel test generation and execution using Korat to make it more practical. We build on a specific algorithm, SEQ-ON, introduced in previous work for equi-distancing candidate inputs, which allows re-execution of Korat for input generation using parallel workers with evenly distributed workload. Our key insight is that the Korat search typically encounters many consecutive candidates that are all invalid inputs and such invalid ranges of candidates can be memoized succinctly to optimize re-execution of Korat. We introduce a novel approach for memoizing Korat’s checking of consecutive invalid candidates, embody the approach into three new techniques based on SEQ-ON, evaluate the techniques using a standard suite of data structure subjects to show the efficacy of our approach, and show some potential applications of it in two new application domains for Korat. We believe our work opens a promising new direction to optimize solving of imperative constraints and using them in novel application domains.