A Predictive Model for Exception Handling

Abstract

The exception handling mechanism has been one of the most used reliability tools in programming languages for over four decades. Nearly all modern languages have some form of “try-catch” model for exception handling and encourage its use. Nevertheless, this model has not seen significant change, even in the face of new challenges, such as concurrent programming and the advent of reactive programming. As it stands, the current model is reactive, rather than proactive—exceptions are raised, caught, and handled. We propose an alternative exception handling model—PreX—where exceptions are no longer caught but, rather, predicted and prevented. Online Failure Prediction techniques generally work at a very high level, showing potential for prediction of program crashes. However, these techniques have never been at the hands of the programmers as an effective tool to improve software quality. By applying recent advances in Online Failure Prediction, PreX aims to fully prevent exceptions, bringing failure prediction techniques to a much more fine-grained level that the programmer can control. Predicting exceptions enables a range of preventive measures that enhance the reliability and robustness of a system, offering new revitalization strategies to developers.