Convergence debugging

Abstract

This paper proposes a new practical automatic debugging method, called Convergence Debugging, which isolates a set of test cases that converge on the internal root cause of a failure. This method involves evaluating the debugging effectiveness of a set of test case by utilizing a new measure of code-level distance between a set of debug test cases and the test case that caused the failure. The same distance measure could also be used to select a set of debug test cases that maximize debugging effectiveness. In order to gain insights into the root cause of a failure, the debug test cases could be used to analyze the differences between the failed test case of interest and the debug test cases.An industrial-strength tool, called Diversity Analyzer, for programs written in C, C++, C#, Java and VB in the Microsoft .NET environment, is used to experiment with the effectiveness of Convergence Debugging in locating faults in industrial software.