Doctoral thesis

Test generation for high coverage with abstraction refinement and coarsening (ARC)

    30.04.2014

140 p

Thèse de doctorat: Università della Svizzera italiana, 2014

English Testing is the main approach used in the software industry to expose failures. Producing thorough test suites is an expensive and error prone task that can greatly benefit from automation. Two challenging problems in test automation are generating test input and evaluating the adequacy of test suites: the first amounts to producing a set of test cases that accurately represent the software behavior, the second requires defining appropriate metrics to evaluate the thoroughness of the testing activities. Structural testing addresses these problems by measuring the amount of code elements that are executed by a test suite. The code elements that are not covered by any execution are natural candidates for generating further test cases, and the measured coverage rate can be used to estimate the thoroughness of the test suite. Several empirical studies show that test suites achieving high coverage rates exhibit a high failure detection ability. However, producing highly covering test suites automatically is hard as certain code elements are executed only under complex conditions while other might be not reachable at all. In this thesis we propose Abstraction Refinement and Coarsening (ARC), a goal oriented technique that combines static and dynamic software analysis to automatically generate test suites with high code coverage. At the core of our approach there is an abstract program model that enables the synergistic application of the different analysis components. In ARC we integrate Dynamic Symbolic Execution (DSE) and abstraction refinement to precisely direct test generation towards the coverage goals and detect infeasible elements. ARC includes a novel coarsening algorithm for improved scalability. We implemented ARC-B, a prototype tool that analyses C programs and produces test suites that achieve high branch coverage. Our experiments show that the approach effectively exploits the synergy between symbolic testing and reachability analysis outperforming state of the art test generation approaches. We evaluated ARC-B on industry relevant software, and exposed previously unknown failures in a safety-critical software component.
Language
  • English
Classification
Computer science
License
License undefined
Identifiers
Persistent URL
https://susi.usi.ch/usi/documents/318427
Statistics

Document views: 49 File downloads:
  • 2014INFO018.pdf: 7