Towards Semantic-Based Aspect Interaction Detection

Interactions between dependent or conflicting aspects are a wellknown problem with aspect-oriented development. These interactions are potentially dangerous and can lead to unexpected or incorrect results when aspects are composed. To date, there have been very few attempts to address this issue at the modeling level. We present a new approach for detecting interactions that is based on lightweight semantic annotations of aspect models. Each aspect model is annotated with domain-specific markers and a separate goal model describes how semantic markers from different domains influence each other. When aspect models are composed, the composed model is inspected for any semantic markers that potentially conflict. This is achieved by propagating values through the goal model to see which goals (typically nonfunctional properties) are satisficed by the composition and which are not. The technique can be used both to highlight potential aspect conflicts and to tradeoff aspects. We illustrate the approach using two aspect techniques.