PHALANX: a graph-theoretic framework for test case prioritization

Test case prioritization for regression testing can be performed using different metrics (e.g., statement coverage, path coverage) depending on the application context. Employing different metrics requires different prioritization schemes (e.g., maximum coverage, dissimilar paths covered). This results in significant algorithmic and implementation complexity in the testing process associated with various metrics and prioritization schemes. In this paper, we present a novel approach to the test case prioritization problem that addresses this limitation. We devise a framework, Phalanx, that identifies two distinct aspects of the problem. The first relates to metrics that define ordering relations among test cases; the second defines mechanisms that implement these metrics on test suites. We abstract the information into a test-case dissimilarity graph -- a weighted graph in which nodes specify test cases and weighted edges specify user-defined proximity measures between test cases. We argue that a declustered linearization of nodes in the graph results in a desirable prioritization of test cases, since it ensures that dissimilar test cases are applied first. We explore two mechanisms for declustering the test case dissimilarity graph -- Fiedler (spectral) ordering and a greedy approach. We implement these orderings in Phalanx, a highly flexible and customizable testbed, and demonstrate excellent performance for test-case prioritization. Our experiments on test suites available from the Subject Infrastructure Repository (SIR) show that a variety of user-defined metrics can be easily incorporated in Phalanx.