Toward an Execution Model for Component Software

The notion of component-oriented programming is largely based on a model of software distribution: if a certain software artifact can be distributed independently of other similar artifacts, and yet be smoothly integrated into an environment composed of such artifacts, then we call it a component. Unlike established areas of computer science, such as object-oriented programming, component-oriented programming has no coherent programming model associated with it at this time, nor is there a general agreement as to what the run-time structure of component-based systems should be. Basing a whole new technology on the “middle” ground of software distribution is at once strange and risky. A host of potential misunderstandings and incompatibilities will arise unless the existing model is soon extended twofold: On one hand, the mechanics of component interaction need to be brought semantically closer to programming languages. On the other hand, components need to be mapped onto appropriate run-time architectures. Our research is focused on the latter part of the problem: we are developing techniques that enable systems composed of independentlydistributed platform-independent software components to function as efficiently as monolithic application programs. A key element of our strategy is profile-guided run-time code generation, provided as a central system service, which allows to genuinely integrate dynamically loaded extensions with the existing code base.