Scaling up algorithmic debugging with virtual execution trees

Abstract

Declarative debugging is a powerful debugging technique that has been adapted to practically all programming languages. However, the technique suffers from important scalability problems in both time and memory. With realistic programs the huge size of the execution tree handled makes the debugging session impractical and too slow to be productive. In this work, we present a new architecture for declarative debuggers in which we adapt the technique to work with incomplete execution trees. This allows us to avoid the problem of loading the whole execution tree in main memory and solve the memory scalability problems. We also provide the technique with the ability to debug execution trees that are only partially generated. This allows the programmer to start the debugging session even before the execution tree is computed. This solves the time scalability problems. We have implemented the technique and show its practicality with several experiments conducted with real applications.