Incremental data-flow analysis algorithms

Abstract

An incremental update algorithm modifies the solution of a problem that has been changed, rather than re-solving the entire problem. ACINCF and ACINCB are incremental update algorithms for forward and backward data-flow analysis, respectively, based on our equations model of Allen-Cocke interval analysis. In addition, we have studied their performance on a “nontoy” structured programming language L. Given a set of localized program changes in a program written in L, we identify a priori the nodes in its flow graph whose corresponding data-flow equations may be affected by the changes. We characterize these possibly affected nodes by their corresponding program structures and their relation to the original change sites, and do so without actually performing the incremental updates . Our results can be refined to characterize the reduced equations possibly affected if structured loop exit mechanisms are used, either singly or together, thereby relating richness of programming language usage to the ease of incremental updating.