Identifying Test Conditions for Software Maintenance

As a software system goes through maintenance, it is necessary to test that the system’s behaviour has changed (only) as expected. For each program change, this requires the identification of a set of test conditions that together provide some measure of confidence about the effects of the change. In this paper, we propose a technique to automatically identify the set of conditions to completely test for the effects of an atomic program change at some identified ‘target’ points in the program. This work builds upon our work to identify the conditions for influence between program points using a program dependence graph representation. The set of conditions thus identified can be used to either select tests from an existing test-suite, or to produce new tests that do not belong to the test-suite, thus helping in maintaining the test-suite as the system evolves. Some results of using our technique on sample programs are also reported. We also briefly discuss how the basic idea behind our technique can have potential applications in areas such as impact analysis and program understanding.