The three most costly and difficult tasks in maintaining a large software system involve

• identifying the code relevant to a particular feature,

• localizing software faults effectively, and

• conducting intelligent and economic regression testing.

Documentation for a large system may not exist, or even if it does exist it may be incomplete, difficult to comprehend or not updated. Moreover, those with responsibility for maintenance may not have participated in the original design and/or implementation, or may lack a thorough understanding of how the entire system works. Perhaps a faster and a more efficient way of

• identifying the important segments that are unique to a feature or common to a group of features,

• tracking down faults to files, functions, lines of code, decisions or even definition-use pairs of certain variables,

• selecting a representative subset of all regression tests to ensure that the system still behaves the same way as the previous version, except where changes are expected,

• determining where the performance bottlenecks lie, and

• generating additional tests for testing new functionalities and also guaranteeing the coverage of new code,

is to let the system speak for itself. We will present an integrated solution to these problems by analyzing the dynamic behavior of the system with the aid of software visualization and heuristic guidance. The Telcordia's Software Visualization and Analysis Toolsuite will be used to demonstrate how programmers, testers, and maintainers can benefit from our approach.

For more information send email to ewong@research.telcordia.com