Inferring Repository File Structure Modifications Using Nearest-Neighbor Clone Detection

During the re-engineering of legacy software systems, a good knowledge of the history of past modifications on the system is important to recover the design of the system and transfer its functionalities. In the absence of a reliable revision history, development teams often rely on system experts to identify hidden history and recover software design. In this paper, we propose a new technique to infer the history of repository file modifications of a software system using only past released versions of the system. The proposed technique relies on nearest-neighbor clone detection using the Manhattan distance. We performed an empirical evaluation of the technique using Tomcat, JHotDraw and Adempiere SVN information as our oracle of file operations, and obtained an average precision of 97% and an average recall of 98%. Our evaluation also highlighted the phenomena of implicit Moves, which are, Moves between a system's versions, that are not recorded in the SVN repository. In the absence of revision history and software experts, development teams can make use of the proposed technique during the re-engineering of their legacy systems.