Abstract
This dissertation presents a new debugging assistant called a Debugging Critic. As an alternative to an automated debuggin oracle, the debugging critic evaluates hypotheses about fault locations. If it cannot confirm that the code at the hypothesized location contains a fault, it formulates an alternative hypothesis about the location of a faulty statement or the location of omitted statements. The debugging critic derives knowledge of possible locations of a fault that manifested itself under a given test case from failure symptoms. Therefore, it can operate without a line-by-line specification and a knowledge base of faults.
A prototype of our debugging critic has been implemented as an extension of an existing debugger, Spyder. An experiment with Spyder shows that programmers debug two to four times faster on the average with the debugging critic than without it. Ninety-two percent of the critic's users recommend the critic as an extension of conventional debuggers.
The research iin this dissertation contributes to debugging and critic systems. In debugging, our experiment shows that an active debugging assistant can effectively improve debugging performance. Another contribution is our approach to evaluate and formulate hypotheses about fault locations, especially the locations of omitted statements. In critic systems, our contribution is the use of questions to prevent information and non-insulting criticism.
