A controlled experiment investigation of an object-oriented design heuristic for maintainability

Abstract

The study presented in this paper is a controlled experiment, aiming at investigating the impact of a design heuristic, dealing with the `god class' problem, on the maintainability of object-oriented designs. In other words, we wish to better understand to what extent a specific design heuristic contributes to the quality of designs developed. The experiment has been conducted using undergraduate students as subjects, performing on two system designs using the Coad & Yourdon method. The results of this study provide evidence that the investigated design heuristic: (a) affects the evolution of design structures; and (b) considerably affects the way participants apply the inheritance mechanism.

Introduction

In the last decade or more, the object-oriented (OO) paradigm has gained a broad acceptance within the software community, the industry, and the software development organizations of any size. Hence, OO methodologies, languages and development environments, have been developed supporting this technology. It has mainly been made popular by C++, and now even more so by Java. Jones (1994) has noticed the rapid growth of OO technology, since 1994. Even recently, a survey carried out by Goodley (1999), has indicated the continuing growth in the popularity of Java as an OO development language.

Much of the literature asserts that substantial gains, such as increased understandability, productivity, quality, ease of modification, reuse, should accrue from using OO analysis, design, coding, and reusable components. Nevertheless, these benefits are mostly based on intuition and not on empirical evidence. Intuition may provide a starting point, but it needs to be backed up with empirical evidence. According to Basili and Burgess (1995), experimentation has shown that intuition about software in many cases is wrong. Jones (1994) as well, has identified a lack of empirical evidence to support the claims accredited to OO technology, like improved productivity and quality, defect removal efficiency, and even more so reusability.

For this reason, over recent years, there has been a growing interest in empirical evaluation. In a review (Deligiannis et al., 2002), examining how experimentation has been carried out in the OO technology, the evidence does not support the claim that OO techniques always provide the benefits accredited to them. A case study performed by Hatton (1998) on corrective maintenance issues, indicated a number of concerns whether OO technology has met its claims. Considering the performance and strategies of programmers new to OO technology it was found that OO concepts were not easy to learn and use quickly (Hillegersberg et al., 1995).

Clearly, more empirical research is needed to investigate these claims; examining in particular the efficacy and effectiveness with which the various OO features are applied. One particular area that warrants thorough investigation is considered the impact of design heuristics on one quality factor, namely maintainability. The reasons leading us to this decision are: (a) the maintenance phase is considered the most costly phase of system life cycle (Meyer, 1997; Hatton, 1998). Subsequently, investigations aiming to reducing it should be of increased interest for the empirical research community; and (b) design heuristics, concentrating cumulative knowledge mainly based on experience and practice (Booch, 1995), might provide useful assistance to achieve this goal.

The study presented in this paper is a controlled experiment, carried out using students as participants, which investigates the impact of a single design heuristic on quality of OO designs, captured by two important quality factors of design assessment, namely understandability, and maintainability. The paper is structured as follows. Section 2 describes the details of the experiment. Section 3 summarizes the data collected and presents the data analysis results. Section 4 identifies and discusses possible threats to the validity of the study. Finally, Section 5, presents the conclusions and future research targets. The structure of the paper follows the structure of a Briand et al.'s paper (2001).