C for Yourself: Comparison of Front-End Techniques for Formal Verification

Abstract

With the improvement of hardware and algorithms, the main challenge of software model checking has shifted from pure algorithmic performance toward supporting a wider set of input programs. Successful toolchains tackle the problem of parsing a wide range of inputs in an efficient way by reusing solutions from existing compiler technologies such as Eclipse CDT or LLVM. Our experience suggests that well-established techniques in compiler technology are not necessarily beneficial to model checkers and sometimes can even hurt their performance. In this paper, we review the tools mature enough to participate in the Software Verification Competition in terms of the employed analysis and frontend techniques. We find that successful tools do exhibit a bias toward certain combinations. We explore the theoretical reasons and suggest an adaptable approach for model checking frameworks. We validate our recommendations by implementing a new frontend for a model checking framework and show that it indeed benefits some of the algorithms.