Type slicing: An accurate object oriented slicing based on sub-statement level dependence graph

Abstract

Abstract Context Program slicing is very useful in program analysis and software engineering. It computes the slice, which is a part of program and contains all the statements related to the given slicing criterion. The more accurate a slicing technique could be, the smaller the slice is. Objective This paper aims to improve the current slicing accuracy for object-oriented programs. The slicing accuracy is mainly related to three factors, the dependency graph (which extracts the inner relationships of source code), the slicing criterion (which determines the slicing requirement), and the slicing algorithm (which computes the slice for the criterion from the dependency graph). Method Our method consists of three parts. First, we present a Sub-Statement Level Dependence Graph (SSLDG), which computes finer-grained dependences for object-oriented programs than mostly-used statement level graph. Second, we present a new type slicing criterion called (Sub-statement) Type Slicing Criterion (STSC), which supports the user to specify not only the statement and object variable, but also the type of object among its polymorphic types. At last, a corresponding slicing algorithm called (Sub-statement) Type Slicing (STS) is designed to perform the slicing process. Results We implement STS on Java programs as MyJavaSlicer, and run it with ten open source projects and random slicing criteria. The results show that STS slicing algorithm as well as SSLDG would make slices 35.90% smaller than traditional two-phase slicing; additionally using STSC would make the slices 48.4% further smaller; STSC also helps traditional two-phase slices reduced by 56.90%. Conclusions STS could provide more accurate slices than traditional two-phase slicing, and it also runs faster on most cases; STSC helps specify the slicing requirement, and roughly halves the size of slices for both slicing algorithms.