Rethinking Soot for summary-based whole-program analysis

Whole-program static analysis has been extensively studied and widely used in the past few decades. For modern object-oriented programs, scalability has become an important issue for using whole-program analysis in real-world tools. In addition, the ever-increasing size of libraries (e.g., the JDK library) and frameworks exacerbates the scalability problems. To achieve the desired level of analysis performance, an effective approach could be to generate and apply analysis summary information for library methods. In this paper, such an approach is referred to as a summary-based whole-program analysis. The challenges for this technique are twofold: (1) carefully designed abstractions and algorithms are needed to create and use client-independent and analysis-specific library summary information; and (2) support for summary generation and application should be effectively incorporated into existing analysis infrastructures. This paper focuses on the second challenge. It uses Soot, a widely-used program analysis framework for Java, as a vehicle to explore some of the important issues in providing analysis infrastructures with capabilities for summary-based analysis. Experimental studies are presented to show that significant savings can potentially be achieved by making a whole-program alias analysis summary-based. To actually achieve these savings, the paper proposes to extend Soot to add support for summary-based analysis. Finally, a brief discussion of the required framework extensions is presented.