NSF Award
0238697
This Small Business Innovation Research Phase II project aims to make it much easier to create unit-level regression tests for Java programs. Their benefits are clear, but existing techniques for creating them are flawed because they are difficult to apply to existing code, and tool support requires modification of the target code. The innovation is to instrument a gold-standard version of the module of interest so that when a client application executes, all events that cross the boundary to the module are intercepted and logged to a file. Later, after the unit has been modified or extended, and without any further need for the client application, the events can be reconstructed and fed to the unit. The results are checked for consistency with the log, and discrepancies flagged as faults. This makes it much easier for a user to create a test suite for a module. The approach is made feasible by using Aspect-Oriented Programming, and object mocking. The research challenges are: how to devise techniques for tolerating permitted changes in the target module, and how to reduce the chances of a single failure triggering a profusion of cascading failures. The use of advanced static analysis techniques, including dependence analysis, is the key to solving these problems.<br/><br/>If successful, this system will help software development organizations reduce the cost of development and maintenance of their software assets while at the same time increasing its quality. It will help increase assurance of safety-critical software, such as in medical equipment, or flight-control systems, thus reducing the risk of damage to property and loss of life.
