Open Components £ [position Paper]

Traditional binary component models do not support the application-speci c customization of components at integration-time . This problem cannot be solved by introducing new component languages, because it is rooted in the properties of executable, binary code and not in the implementation language of the component. In this paper we suggest using \open components" instead of binary component models. This novel approach uses source code as the distribution form of components and allows a exible, tool-supported customization with respect to the application environment. The result is a better chance to reuse a component through adaptable interfaces and a very eÆcient implementation of the composed software product. 1. MOTIVATION A key factor for the success of component systems, especially in resource restricted application domains, is the overhead imposed by the component framework. Component software should not require signi cantly more resources than a noncomponent based implementation. In traditional binary component models, especially those which are independent of the implementation language and thus have to use mechanisms like function argument marshaling, the communication over component boundaries is signi cantly more expensive than a similar plain languagelevel procedure call. Arising from this, real-world components are often limited in their granularity as many small components would increase the overall communication overhead. This work has been partly supported by the German Research Council (DFG), grant no. SCHR 603/1-1 and SCHR 603/2. the point in time where several components, which might come from di erent vendors, are composed to form an application Designing the inner structure of a component is a trade-o between exibility with respect to all possible applications and optimality with respect to the speci c contexts the component could be used in. The problem becomes more obvious by looking at a simple example ( gure 1).