Enhancing design models with composition properties: a software maintenance study

Abstract

A considerable part of software design is dedicated for the composition of two or more modules. The implication is that changes made later in the implementation often require some reasoning about module composition properties. However, these properties are often not explicitly specified in design artefacts. Moreover, they cannot be easily inferred from the source code either. As a result, implicit composition properties may represent a major source of software maintenance complexity. This fact is particularly true with the advent of post object-oriented techniques, which are increasingly providing advanced mechanisms to enable flexible module composition. However, there is little empirical knowledge on how design models with explicitly-specified composition properties can improve software maintenance tasks. This paper reports an experiment that analyses the impact of design models enriched with composition properties on system maintenance. Explicit composition modelling was achieved in the experiment through a conservative approach, i.e., a specific set of additional UML stereotypes dedicated to model composition properties. The experiment involved 28 participants, who were asked to realize four maintenance tasks using UML design models with different levels of module composition details. Our findings suggested that explicit composition modelling contributed to better results in the realization of program change tasks, regardless of the developers' expertise. Users of composition-enhanced models consistently yielded better results when compared to users of plain UML models. The use of explicit composition specification also led to an average increase of 44.7% in the quality of changes produced by the participants