Measuring Similarity Of Large Software Systems Based On Metric Space Range Search

The occurrences of similar codes in different software systems are the foundation of understanding and refactoring the original product. From a software engineering's perspective, the detection of similar codes can help us to test the systems, to facilitate the original system's refactoring, to promote version maintenance based on the measured discrepancy, as well as to gauge a quantitative similarity result which will help us to manage a project more efficiently. It is easy to see that the research of software similarity detection is of great importance in a practical way.This paper focuses on the similarity detection of large software systems. The existing software similarity detection technology embodies different definition assumption, and there's no firm framework to assist the analysis of large software systems' similarity detection. This paper gives analysis to the present similarity detection technologies, and promotes the metric detection approach, targeting very large software systems. After picking some proper software system metrics, a formal and quantitative definition of software similarity is laid out. Then the existing metric space range search algorithms are browsed on a theoretical basis and a fitted one comes out. A practical code similarity detector's design is given, and experiments are conducted on some large simulated sample date sets. For those important parameters in software metric space similarity detection, analysis from experiments result points out their effect on the detection result and operating efficiency, which will give a strong reference point for software similarity detection. Herein this paper forms a complete framework for large software systems similarity detection.