No abstract

Correctness of compilation is important for the reliability of software. New techniques to guarantee correctness do not verify the compiler itself, but check for each compiled program whether it is correctly translated. Following these ideas, we developed an approach in which checking is realized as proof checking within a formal specification and verification framework. Based on formal specifications of source and target language and a translation predicate, compilers produce, in addition to the target program c, a proof that c is correct w.r.t. its source program. This proof can be checked independently of the compiler by the framework. Thus, it can be used as a translation certificate. The paper describes the overall approach and applies it to a simple translation scenario. Specification and verification is done within the theorem prover Isabelle/HOL. To show the flexibility of the approach, we present two different proof techniques for translation correctness.

Java bytecode verification is traditionally performed by a polynomial time dataflow algorithm. We investigate an alternative based on reducing bytecode verification to model checking. Despite an exponential worst case time complexity, model checking type-correct bytecode is polynomial in practice when carried out using an explicit state, onthe-fly model checker like Spin. We investigate this theoretically and experimentally and explain the practical advantages of this alternative.

No abstract