(A) bottom-up pointer analysis using the update history

Abstract

Pointer analysis is an important part for the source code analysis of C programs. Most of pointer analysis algorithms are developed as top-down analyses, where the analyses are performed from callers to callees. Since such a scheme disables the analysis to operate on incomplete programs, existing top-down pointer analyses are not readily applicable to the modular software development process where individual software components are separately developed and subsequently linked with other components. Bottom-up pointer analyses can address this problem by performing analyses from callees to callers. However, analyzing the behavior of a procedure without utilizing any information on callers makes the different pointer behaviors of a procedure for different calls difficult to be traced precisely. In this thesis, we propose a bottom-up and flow- and context-sensitive pointer analysis algorithm based on a new bottom-up pointer analysis domain named the update history. The update history can not only abstract memory states of a procedure independently of the information on aliases between memory locations, and but also keep information on the order of side effects performed. Such characteristics of the update history not only enable pointer analyses to be formulated as bottom-up analyses, but also help bottom-up pointer analyzers to effectively identify killed side effects and relevant alias contexts, both of which are the main contributions of this thesis. Our bottom-up and flow- and context-sensitive pointer analysis is formulated as an inference algorithm of a type system for the update history. Then we formally prove the soundness of our pointer analysis algorithm, which has not been satisfactorily investigated for this kind of problem previously, using the well-developed type soundness machinery from the type system framework. The experiments performed on a pilot implementation show that our approach is cost-effective in a sense that the precision was im...