Towards Testing a Verifying Compiler

In this paper, we present our approach on testing a particular verification system that is industrially used to generate mathematical proofs of the correctness of C programs. Normally, the tools used in such a verification process are seldomly verified nor thoroughly tested, and their correctness is taken for granted. Our approach to obtain assurance in such tools does not rely on the knowledge of their internal details and enables regular users of these tools to write test cases for them. Those tests are then assessed using our domain-specific axiomatization coverage that measures the impact of the axiomatization, which is an integral component of the verification process. Furthermore, we explore several sources of test cases, as the risk of constructing buggy test cases is high due to the input domain’s complexity.