A new modification of summary-based analysis method for large software system testing

Abstract

The automated testing tools becoming a frequent practice require thorough computer-aided testing of large software systems, including system inter-component interfaces. To achieve a good coverage, one should overcome scalability problems of different methods of analysis. These problems arise from impossibility to analyze all the execution paths. The objective of this research is to build a method for inter-procedural analysis, which efficiency enables us to analyse large software systems (such as Android OS codebase as a whole) for a reasonable time (no more than 4 hours). This article reviews existing methods of software analysis to detect their potential defects. It focuses on the symbolic execution method since it is widely used both in static analysis of source code and in hybrid analysis of object files and intermediate representation (concolic testing). The method of symbolic execution involves separation of a set of input data values into equivalence classes while choosing an execution path. The paper also considers advantages of this method and its shortcomings. One of the main scalability problems is related to inter-procedural analysis. Analysis time grows rapidly if an inlining method is used for inter-procedural analysis. So this work proposes a summary-based analysis method to solve scalability problems. Clang Static Analyzer, an open source static analyzer (a part of the LLVM project), has been chosen as a target system. It allows us to compare performance of inlining and summary-based inter-procedural analysis. A mathematical model for preliminary estimations is described in order to identify possible factors of performance improvement.

Keywords

• C++
• static code analysis
• symbolic execution
• interprocedural analysis
• summary-based method
• Clang Static Analyzer