Abstract

This empirical study analyzes changes in C++ source code which occurred between two releases of an industrial software product and compares them with entities and relations available in object-oriented modeling techniques. The comparison offers increased understanding of what changes can and cannot be described using such object models. The goals were to investigate if t he object model in this particular project is either abstract and stable or detailed and sensitive to change, and whether or not changes made to the C++ source code are visible in the object model . Four metrics for characterization of change are formally defined and used, namely correctness, completeness, compliance, and visibility factor. T he major finding is that even though many of the classes are changed, the majority of these changes turn out to be invisible in the object model. That is, changes made on the source code level are of a finer granularity than available in common object modeling concepts. This may explain why object models seem to be of little use in release-oriented development.

Citations