Section 30.4. CONCLUSION

30.4. CONCLUSION

Our work on C-SAW has demonstrated the benefits of aspect-oriented software development (AOSD) and aspect weaving at different levels of abstraction. Applying aspect-oriented techniques at the level of domain modeling is known as aspect-oriented domain modeling (AODM). AODM yields several benefits that support improvements for exploring different design alternatives in domain-specific models. When these design alternatives are captured as constraints that crosscut the modeling boundaries, separating those constraints as aspects and then weaving them into a base model significantly improves the capabilities for changing properties of the base model. At the implementation level, generative techniques can be used to synthesize models into executable applications. Programming language weavers, such as AspectJ, are essential for customizing prefabricated components from the various concerns that are described in the model.

For the AODM approach to evolve to the next level of maturity, it is necessary to have an extensive library of reusable aspects. We believe that much research is still needed within the AOSD community in general regarding the idea of large-scale aspect reuse and composition.

30.4.1. Future Work

The current specification of modeling aspects is done using a textual language, such as the one shown in Listing 30-1. Future research will be conducted on the modeling representation science required to specify model aspects using graphical formalisms. These visual representation techniques and tools will permit the specification of aspects in a manner that is consistent with the abstraction used in the specification of the base model; that is, both aspects and models will be represented graphically within the same environment, rather than the current situation where aspects are specified textually. We are also investigating the feasibility of using the meta-weaver with other meta-modeling tools [17].

30.4.2. Related Work

There is an increasing interest among researchers toward applying advanced separation of concerns techniques to non-code artifacts [2]. In particular, AOSD techniques have been investigated at all levels of the development lifecycle, including requirements engineering [38]. Several researchers have investigated the application of AOSD concepts within the context of the UML [7, 10]. These efforts have yielded guidelines for describing crosscutting concerns at higher levels of abstraction. In this regard, they have common goals with the work described in this chapter. These efforts differ from our work, however, because we have been concentrating on the idea of building actual weavers for domain models.

Within the context of distributed real-time embedded systems (DRE), the C-SAW techniques described in this chapter are being integrated within the CoSMIC tool suite developed at Vanderbilt University [15]. CoSMIC extends the GME to provide a modeling environment to assist in the configuration and assembly of DRE component middlewarein particular, systems that have QoS requirements and are assembled from components constructed using concepts from Real-Time CCM [47]. Other research efforts related to CoSMIC and aspect weaving are the Virginia Embedded Systems Toolkit (VEST) from the University of Virginia [41] and the AIRES tool from the University of Michigan [20].

VEST is a toolkit that is built as a GME meta-model [30]. It supports modeling and analysis of real-time systems and introduces the notion of prescriptive aspects to specify programming language independent advice to a design. A distinction between VEST and our C-SAW is in the generalizabilty of the weaving process. C-SAW is constructed to work with any GME meta-model (including VEST itself), while the strength of VEST lies within real-time system specification. Additionally, the VEST prescriptive aspect language is not as rich as our ECL. According to Stankovic et al. [41], the structure of prescriptive aspect is limited to the form, "for <some conditional statement on a model property> change <other property>," which is comparatively less powerful than the ECL capabilities demonstrated in Listing 30-1. The AIRES tool also has a focus on modeling of Bold Stroke scenarios. The focus of AIRES is on analysis of real-time properties, but does not adopt an aspect-oriented approach to modeling.