A Novel Lightweight Virtual Machine Based Decompiler to Generate C/C++ Code with High Readability

As a key part of reverse engineering, decompilation plays a very important role in software security and maintenance. Many decompilation techniques and tools have been developed while all of them have defects in different aspects. For example, IDA Hex rays generates pseudocodes with poor readability, and Boomerang is unable to identify composite structures such as Classes and multidimensional arrays. In this paper we present C-Decompiler, an integrated C/C++ decompiler based on lightweight virtual machine, which is capable to analyze the data flow especially the data dependency across basic blocks with high accuracy. C-Decompiler is able to recognize composite structures and libraries, such as Standard Template Library. Our experimental results show that on average C-Decompiler has the highest reduction rate of 55.91% and lowest expansion rate of 55.79%, which prove that the code decompiled by C-Decompiler is most similar to the original code in comparison with other existing decompilers such as Boomerang and IDA Hex rays. Furthermore, C-Decompiler is able to recognize all the functions without any false positive nor false negative, and it is convincing to support that the code decompiled by CDecompiler is of high readability by the least redundancy and most accuracy.