Evaluating Clone Detection Techniques

In the last decade, several researchers have investigated techniques to detect duplicated code in programs exceeding hundreds of thousands lines of code. All of these techniques have known merits and deficiencies, but as of today, little is known on where to fit these techniques into the software maintenance process. This paper compares three representative detection techniques (simple line matching, parameterized matching, and metric fingerprints) by means of five small to medium cases and analyses the differences between the reported matches. Based on this experiment, we conclude that (1) simple line matching is best suited for a first crude overview of the duplicated code; (2) metric fingerprints work best in combination with a refactoring tool that is able to remove duplicated subroutines; (3) parameterized matching works best in combination with more fine-grained refactoring tools that work on the statement level.