IRDiff：Source Code Differencing based on LLVM IR

• 1、Computer School, Beijing University of Posts and Telecommunications, Beijing 100876

Abstract：Abstract: Source Code differencing is one of the most important research issues in multi-version program analysis. Existing work, including text differencing or abstract syntax tree-based code analysis, is often affected by code movement, making the comparison inaccurate due to the misalignment of the code. This paper proposed a new approach of source code difference analysis based on LLVM intermediate representation language (LLVM IR). The two versions of the source code are as input transformed to the LLVM IR. Then based on the syntax structure, control flow and data flow information represented by LLVM IR to find the isomorphic nodes in the two versions of the control flow graph. A series of isomorphic sub-graphs and a set of difference nodes are obtained, and the code difference between two versions of the program is obtained. By using LLVM IR\'s hierarchical structure and control flow information, the problem of code misalignment caused by code movement can be avoided, and the accuracy of code variance analysis can be improved by combining control dependency, data dependence and type information, etc. Based on this algorithm, we implemented a code variance analysis tool, called IRDiff, and used the SIR benchmark for experimental evaluation. Experiments show that the tool can better analyze the code differences between different versions of C / C ++ programs.

Keywords： source evolution analysis control flow graph program comprehension graph differencing intermediate