(An) approach to subject-oriented software composition based on relation model

Abstract

Current software composition techniques build large software from existing software components but suffer from the continuous evolution after composition. In current composition techniques, the most recognizable composition paradigms addressing this problem are component-based software development (CBSD), aspect-oriented software development (AOSD) and role modeling. Those three composition paradigms have their own strengths and weaknesses. Black box composition used in CBSD provides simple and safe modularization for its strong information hiding, which is, however, the main obstacle for a black box composite to evolve later. AOP (Aspect-Oriented Programming) of AOSD enhances maintainability and comprehensibility by modularizing concerns crosscutting multiple components but lacks the support for the hierarchical and external composition of aspects themselves and compromises the important software engineering principles such as encapsulation, which is almost perfectly supported in black box composition. Role and role model has been recognized to have many similarities with CBSD and AOP but they have significant differences with those composition techniques as well. Role model without global architecture has weakness in handling crosscutting concerns, which can evolve later. Subject-oriented composition means software composition techniques supporting the evolution of software composition according to subjects. Subject defined in this dissertation includes related objects, their states, and crosscutting concerns over them. In this dissertation, a subject-oriented software composition technique named relation model which is based on representational abstraction of the relationship between component instances is introduced. Relation model supports the simple, elegant, and dynamically evolvable software composition with extended predicate logics defined in terms of component instances and provides the synergy of the above composition techniques. Relation model can ...