The Study Of Clone Code Detection Based On LLVM

The fast change in requirement leads to the short period of code implementation.Though clone code make the development convenient,it also reduces the maintainability of program.The research on the Code Clone Detecting Technology is quite magnificent to Code Development,Code Maintenance,Defect Checking,Software Rights Protection,etc.The feature of detection method based on Program Dependency Graph is accurate,but the high cost and too much matching information make it inapplicable for checking on the large scale.A new clone code detection method based on Dependency Graph and LLVM is put forward as followed,combined with the research on traditional method.(a)A method is proposed to simplify complex Program Dependency Graph into simple one.(b)This paper proposes a method of splitting and matching in Simple dependency graph,which accelerates speed and promote accuracy by fixing operation in the scope and matching gist.(c)This paper proposes a method of filtering clone fragment of low-similarity to get complete clone code.The method combines the Machine Learning Theory and design merging algorithm,solving the problem of noncontinuous results and promoting accuracy.(d)According to the content as mentioned above,the design on the clone detecting frame is achieved.The experimental results show that:(a)Similarity Threshold has a great influence on the result and prediction model can get accurate Similarity Threshold.(b)The parallel algorithm accelerate the speed of calculation effectively.(c)The whole algorithm mentioned can promote the 26% of recall ratio on average,compared with the test results of mainstream software in different ways of clone.