Incremental data flow analysis in a structured program editor

Abstract

For many years, the generally accepted methods for solving global data flow analysis problems (GDFAPs) have relied on the bit vector paradigm [Alle 70] [Kenn 81]. In these algorithms, the information for each variable is represented as a position in a bit vector. These vectors are propagated from node to node during the analysis of the program. This paper investigates a different paradigm for solving GDFAPs. In this strategy, the data flow solution for each variable is computed independently. A new algorithm for solving some GDFAPs is presented here. While it is expected that this algorithm will run slower in the batch compiling environment, there are conditions under which this algorithm can be expected to be superior to those in the bit vector class.