Adapting C++ Exception Handling to an Extended COM Exception Model

The reliability of a software system is to a high degree determined by the reliability of the software components comprising the system [KriMat97]. For a software component to be reliable it must be able recover various exceptions or failures that may arise at run-time, i.e. it must be fault tolerant. Code for exception detection, handling, and signalling often amounts to a substantial portion of the code. This is especially the case in a multi-language environment, mixing different exception models, and requiring transformation between different exception representations. To allow flexible system configuration and system evolution, e.g. making changes to a component, replacing a component with another, or adding new components, each of the components should have minimum dependency on the other components. Loose coupling between components is also important with respect to exception handling and the mechanisms used. As argued in [Cri95], a termination mechanism mixes well with the information hiding principles underlying data abstraction, and consequently also component based systems. The focus should be on the local transition and consistent local state of the component and on exceptions that may arise during a state transition. It is the responsibility of the developer of the component to detect, handle, and signal exceptions based on the state of the component, the arguments of the call, and the results from lowerlevel components. In software development project there are always conflicts between different requirements, and the development team has to make trade-offs between time-to-market, functional requirements, and non-functional requirements, like reliability. The resources used on reliability are often scarce, hence it is important to concentrate the effort on those exception situations that contributes to the overall reliability of the system. Whenever reliability is sacrificed for other requirements, the failure situations should be clearly distinguishable from correct behaviour and easily traceable in the code. If the system reliability turns out to be unacceptable it should be possible to increase the overall reliability by improving individual components of the system, i.e. taking control over more of the failure situations. This paper is based on work done in the BRIX team in DNV IT Solutions. The main purpose of this group is to provide the various application development teams with common solutions to software technical needs. Most BRIX solutions are based on Microsoft's COM technology and C++ as implementation language. In section 2 the differences between C++ exception handling and COM exception handling are briefly outlined. Section 3 describes the BRIX language independent exception model, which is based on the COM exception model. In section four we present how the BRIX exception model is supported at C++ level. Finally, in section 4 we draw some conclusions.