Demand-Driven Static Backward Program Slicing Based on Predicated Code Block Graphs

Static backward program slicing is a technique to compute the set of program statements, predicates and inputs that might affect the value of a particular variable at a program location. The importance of this technique comes from being an essential part of many critical areas such as program maintenance, testing, verification, debugging, among others. The state-of-art slicing approach collects all the data- and control-flow information in the source code before the slicing, but not all the collected information are used for computing the slice. Thus, this approach causes a significant amount of unnecessary computations, particularly for slicing large industrial systems, where unnecessary computations lead to wastage of a considerable amount of processing time and memory. Moreover, this approach often suffers from scalability issues. The demand-driven slicing approaches aim at solving this problem by avoiding unnecessary computations. However, some of these approaches trade precision for performance, whereas others are not entirely demand-driven, particularly for addressing unstructured programs, pointer analysis, or inter-procedural cases. This thesis presents a new demand-driven slicing approach that addresses well-structured, unstructured, and inter-procedural programs. This approach has four distinct features, each of which prevents a special type of unnececessary computations. The effectiveness and correctness of the proposed approach are verified using experimental evaluation. In addition, the thesis proposes an approach that can compute on the fly the control dependencies in unstructured programs