Scalable program analysis through proof caching (doctoral symposium)

Abstract

Despite the remarkable advances attained by the SMT community in the last decade, solving complex formulas still represents the main bottleneck to the scalability of program analysis techniques. Recent research work has shown that formulas generated during program analysis recur, and such redundancy can be captured and exploited by means of caching frameworks to avoid repeating complex queries to solvers. Although current approaches show that reusing formulas syntactically can indeed reduce the impact of SMT solvers on program analysis, they still suffer from being logic-dependent, and performing poorly on huge sets of heterogenous formulas. The core idea of our approach is to go beyond merely syntactical caching frameworks by designing a caching framework that is able to reuse proofs instead of formulas. In fact, even formulas that are syntactically different can share solutions. We aim to study the recurrence of proofs across heterogeneous formulas, and to define a technique to efficiently retrieve such proofs. We plan to exploit a suitable distance function that measures the amount of proofs shared by two formulas to allow the efficient retrieval of candidate proofs within a potentially large space of proofs. In this paper, we present the problem, draft the core idea, discuss the early results and present our research plans.