An Empirical Approach to Analysis of Similarities between Software Failure Regions
Abstract
Previous authors have postulated that faults are related to each other and testers have tried to exploit the effect. However, the evidence and applications have been largely anecdotal. This thesis uses an analytical derivation of software failure regions to develop a quantitative metric of the relationship of one fault to another. This metric is then applied in an empirical study of a population of failure regions derived from faults used in a previous experiment. The failure regions were analyzed for clustering behavior using graph theory techniques. The goal of this study is to be able to use information about known faults in a program as a means of finding other faults in the same program. This study provides strong evidence that failure regions have a tendency to form clusters. Further, two specific characteristics of failure regions that lead to cluster formation are identified: shared bounding conditions (the Identical dimension) and shared variables that appear in different contexts (the Coincidental dimension). The nature of the clusters formed by these two dimensions are markedly different. The Identical dimension clusters are small, isolated, and strongly connected. The Coincidental dimension clusters are larger and more loosely connected. Software testing implications of failure region clustering behavior are discussed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1991
- Accession Number
- ADA246208
Entities
People
- Lelon L. Ginn
Organizations
- Naval Postgraduate School