On the Use of Properties in Java Applications

Abstract

When building software systems, developers have to weigh the benefits of using one specific solution approach against the risks and costs of using another one. This process is not random. Certain preferences, architectural styles, and solution domain pressures create systematic biases that we can measure in order to assess their impact on the system being built and the underlying development process itself. In this paper we explore, whether the getter and setter methods in Java give rise to a bias also. Getter and setter methods, called "properties", are perceived commonplace and considered by some as a threat to data encapsulation. However, little empirical evidence exists that can reliably inform us about the real impact of the use of properties in Java. For this reason, we examined 102 open-source Java systems and discovered that properties are employed much more carefully than one might expect. Contrary to some folklore, developers use properties not just to gain access to an object's private state, but in a systematic and responsible manner and, in general, consistent with the domain requirements of the developed software system.

On the Application of Inequality Indices in Comparative Software Analysis

“…Yet, herein also lies one of our greatest challenges: the proper interpretation of software metrics data [5], [6]. We collect software metrics with the aim to improve productivity and quality of software development [3].…”

“…We also observe a critical Theil index value of 2.4 that, when compared with the associated Gini coefficient, effectively represents perfect inequality. The Gini coefficient allows direct comparison between different software systems [5], [6], [10], whereas, the Theil index supports parsing differences in inequality within groups and between group components [24]. We can therefore study, for example, how the package structure of a Java software system affects its maintainability and future evolution.…”

“…RQ3 What are the expected values of the Gini coefficient and the Theil index, and how do these indices support the study and analysis of evolving software systems? The Gini coefficient lies within the interval [0, 1) [12], but for software metrics data it usually assumes values between 0.4 and 0.75 [5], [6], [10]. The Theil index, on the other hand, falls into the interval [0, log n] [12], [24].…”

“…We can employ inequality analysis in software engineering too [5], [6], [10], [17]- [22]. The two most prominent candidates for measuring the inequality in distribution of software metrics data include the Gini coefficient [23] and the Theil index [24], which are both socio-economic entropic inequality indices.…”

“…However, metrics are usually defined at artifact level (e.g., class, method, or package) and, in order to allow for meaningful interpretation, we need to apply some appropriate aggregation techniques to reason about software at system level [7]. Standard central tendency techniques are generally not applicable, as software metrics data does not follow a normal distribution [5], [6]. There is no "probable error of a mean" or expected value [8] in software.…”