Incremental re-execution of programs

Abstract

Interpreters replace the edit/compile/run cyle with edit/run. Dynamic computing environments, like spreadsheets, shorten this still more to just edit. So-called "Visiprog" environments, such as Maryland's XED, permit developing normal imperative programs in a dynamic computing environment, XED and similar environments, because they show the results of executing a program after every (reasonable) editing step, raise the issue of efficient incremental execution. Incremental execution optimizations are also applicable to any programming situation, including batch/cards, in which nearly the same program is run many times on nearly the same data. However, the requirement of remembering large amounts of internal state between runs make incremental exectution most natural for interpreted languages. This paper examines some algorithms for incremental execution. Based on the frequency of typical program editing changes, we predict the importance of optimizing certain kinds of incremental execution. We also examine actual speedups obtained in executing programs after subjecting them to these simulated incremental edits under these optimizations. The speedups range from factors of 1.1 to near 10. Finally, we discuss the feasibility of including these optimizations in an actual dynamic computing environment like XED, and in more traditional programming environments.