Systems created within the aspect-oriented paradigm (AOP) are difficult for programmers to understand fully. AOP suggests moving crosscutting concerns scattered throughout class code into an individual module, known as an aspect. The process of aspect weaving injects the crosscutting concern back into class code at specific locations, known as joinpoints. A side effect of the weaving process is aspect interference-when aspect code creates unexpected results at a joinpoint. Therefore, developing an understanding of locations that could either cause or exhibit aspect interference problems is essential to developing an interference-free AOP system. This study used the interference potential (IP) and interference causality potential (ICP) metrics, and derived a new metric called total interference potential (TIP), to classify areas of potential interference problems. In addition, the project performs a hierarchical agglomerative clustering using the three metrics. Experiments conducted on two AOP systems identified clusters within each program that could cause or exhibit aspect interference problems. Results showed the merit of using clustering analysis as a technique to locate portions of a system to review or alter to prevent interference problems.

Software design continues to evolve from the structured programming paradigm of the 1970s and 1980s and the object-oriented programming (OOP) paradigm of the 1980s and 1990s. The functional decomposition design methodology used in these paradigms reduced the prominence of non-functional requirements, which resulted in scattered and tangled code to address non-functional elements. Aspect-oriented programming (AOP) allowed the removal of crosscutting concerns scattered throughout class code into single modules known as aspects. Aspectization resulted in increased modularity in class code, but introduced new types of problems that did not exist in OOP. One such problem was aspect interference, in which aspects meddled with the data flow or control flow of a program. Research has developed various solutions for detecting and addressing aspect interference using formal design and specification methods, and by programming techniques that specify aspect precedence. Such explicit specifications required practitioners to have a complete understanding of possible aspect interference in an AOP system under development. However, as system size increased, understanding of possible aspect interference could decrease. Therefore, practitioners needed a way to increase their understanding of possible aspect interference within a program. This study used clustering analysis to locate potential aspect interference within an aspect-oriented program under development, using k-means partitional clustering. Vector space models, using two newly defined metrics, interference potential (IP) and interference causality potential (ICP), and an existing metric, coupling on advice execution (CAE), provided input to the clustering algorithms. Resulting clusters were analyzed via an internal strategy using the R-Squared, Dunn, Davies-Bouldin, and SD indexes. The process was evaluated on both a smaller scale AOP system (AspectTetris), and a larger scale AOP system (AJHotDraw). By seeding potential interference problems into these programs and comparing results using visualizations, this study found that clustering analysis provided a viable way for detecting interference problems in aspect-oriented software. The ICP model was best at detecting interference problems, while the IP model produced results that were more sporadic. The CAE clustering models were not effective in pinpointing potential aspect interference problems. This was the first known study to use clustering analysis techniques specifically for locating aspect interference.