Understanding the interplay between the logical and structural coupling of software classes

Abstract

During the lifetime of object-Oriented (OO) software systems, new classes are added to increase functionality, also increasing the inter-dependencies between classes. Logical coupling depicts the change dependencies between classes, while structural coupling measures source code dependencies induced via the system architecture. The relationship or dependency between logical and structural coupling have been debated in the past, but no large study has confirmed yet their interplay. In this study, we have analysed 79 open-source software projects of different sizes to investigate the interplay between the two types of coupling. First, we quantified the overlapping or intersection of structural and logical class dependencies. Second, we statistically computed the correlation between the strengths of logical and structural dependencies. Third, we propose a simple technique to determine the stability of OO software systems, by clustering the pairs of classes as "stable" or "unstable", based on their co-change pattern. The results from our statistical analysis show that although there is no strong evidence of a linear correlation between the strengths of the coupling types, there is substantial evidence to conclude that structurally coupled class pairs usually include logical dependencies. However, not all co-changed class pairs are also linked by structural dependencies. Finally, we identified that only a low proportion of structural coupling shows excessive instability in the studied OSS projects.

A design structure matrix approach for measuring co-change-modularity of software products

“…Based on their experiments they conclude "any metric which uses dependencies alone to pass judgment on the evolvability of a piece of Java software is thus unreliable". Their claim is further substantiated by the work of Ajienka and Capiluppi [1], and Oliva and Gerosa [19]. In particular, Cafeo et al [6] observed a strong correlation between co-changes and dependencies.…”

“…They measured the stability of the system in terms of the propagation cost and clustered cost measures of MacCormack et al [16]. As pointed out in [1,6,12], we still do not have a deep appreciation of the nature of the static dependencies that translate into co-change relations. However, since co-changes capture logical and hidden dependencies [11] as well as a portion of static dependencies [6,12], we use them in our study of software modularity.…”

“…This motivated Gall, Hajek and Jazayeri [11], and others [7,22] to define coupling and cohesion in terms of co-change relations. Geipel and Schweitzer [12] and the subsequent work [1,6,19] shows that the nature of the interplay between static dependencies and co-changes is not completely understood although reasonably strong correlation exists between the two. Nevertheless, co-changes encompass a wide array of dependencies [6,11,12].…”

Software Components’ Coupling Detection for Software Reusability

“…An empirical study is presented in [18] that analyzed the coupling among number of open source software projects to identify two types of coupling; logical coupling and structural coupling. This study aims at to determine the interplay between the two types of coupling, the coupling strength between classes, and the level of stability between the coupled classes as stable or unstable.…”

An Entropy-Based Metric of Developer Contribution in Open Source Development and Its Application to Fault-Prone Program Analysis*

“…In order to support a successful quality management of OSS products, there have been studies focusing on software repositories which store a lot of data in regard to the software development and maintenance-these studies are referred to as the mining software repositories (MSR) 4 . In MSR studies, there have been many useful methods and empirical reports: data of fault fixes in the past are utilized to predict faulty programs 5,6 ; Co-change relationships among source files are analyzed and applied to a successful maintenance support 7,8 ; Commits to code repositories (code changes) are analyzed to evaluate the impacts of those changes on the future changes 9,10 , and those commits are examined to assess the risk that their changes introduce new faults 11,12 , etc.…”