Evaluating CAVM: A new search-based test data generation tool for C

Abstract

We present CAVM (pronounced “ka-boom”), a new search-based test data generation tool for C. CAVM is developed to augment an existing commercial tool, CodeScroll, which uses static analysis and input partitioning to generate test data. Unlike the current state-of-the-art search-based test data generation tool for C, Austin, CAVM handles dynamic data structures using purely search-based techniques. We compare CAVM against CodeScroll and Austin using 49 C functions, ranging from small anti-pattern case studies to real world open source code and commercial code. The results show that CAVM can cover branches that neither CodeScroll nor Austin can, while also exclusively achieving the highest branch coverage for 20 of the studied functions.