A framework for monitoring program execution.

Abstract

Program execution monitors are used to improve human beings' understanding of program run-time behavior in a variety of important applications such as debugging, performance tuning, and the study of algorithms. Unfortunately, many program execution monitors fail to provide adequate understanding of program behavior, and progress in this area of systems software has been slow due to the difficulty of the task of writing execution monitors. In high-level programming languages the task of writing execution monitors is made more complex by features such as non-traditional control flow and complex semantics. Additionally, in many languages, such as the Icon programming language, a significant part of the execution behavior that various monitors need to observe occurs in the language run-time system code rather than the source code of the monitored program. This dissertation presents a framework for monitoring Icon programs that allows rapid development of execution monitors in the Icon language itself. Monitors have full source-level access to the target program with which to gather and process execution information, without intrusive modification to the target executable. In addition, the framework supports the monitoring of implicit run-time system behavior crucial to program understanding. In order to demonstrate its practicality, the framework has been used to implement a collection of program visualization tools. Program visualization provides graphical feedback about program execution that allows human beings to deal with volumes of data more effectively than textual techniques. Ideally, the user specifies program execution controls in such tools directly in the graphics used to visualize execution, employing the same visual language that is used to render the output. Some monitors that exhibit this characteristic are presented.