A Combined Testing and Verification Approach for Software Reliability

摘要

Automatic and manual software verification is based on applying mathematical methods to a model of the software. Modeling is usually done manually, thus it is prone to modeling errors. This means that errors found in the model may not correspond to real errors in the code, and that if the model is found to satisfy the checked properties, the actual code may still have some errors. For this reason, it is desirable to be able to perform some consistency checks between the actual code and the model. Exhaustive consistency checks are usually not possible, for the same reason that modeling is necessary. We propose a methodology for improving the throughput of software verification by performing some consistency checks between the original code and the model, specifically, by applying software testing. In this paper we present such a combined testing and verification methodology and demonstrate how it is applied using a set of software reliability tools. We introduce the notion of a neighborhood of an error trace, consisting of a tree of execution paths, where the original error trace is one of them. Our experience with the methodology shows that traversing the neighborhood of an error is extremely useful in locating its cause. This is crucial not only in understanding where the error stems from, but in getting an initial idea of how to redesign the code. We use as a case study a robot control system, and report on several design and modeling errors found during the verification and testing process.