Recovery of noncurrent variables in source-level debugging of optimized code

Abstract

Source-level debugging and code optimization are important tools of programming. A debugger has to supply precise information about the values of the variables at any run time error or breakpoint. Code optimization, on the other hand, removes dead stores and reorders code to save time and space, causing intermediate events to happen out of sequence. A thorough implementation of either limits the usefulness of the other. Hennessy, in his paper “Symbolic Debugging of Optimized Code” [2], suggested various techniques to handle this. However, he explicitly chose not to take advantage of the fact that many operations are reversible and many partial results will be avaiable in registers and temporaries. This paper considers these questions in order to extend Hennessy's techniques for recovery of noncurrent variables. The algorithms developed can detect most of the recoverable variables, and run in linear time. We also consider methods to limit our search to only a part of the dag.

This research was done at the Computer Science Department of the Pennsylvania State University.