Approximate Code Search in Program Histories

Very often a defect must be corrected not only in the current version of a program at one particular place but in many places and many other versions -- possibly even in different development branches. Consequently, we need a technique to efficiently locate all approximate matches of an arbitrary defective code fragment in the program´s history as they may need to be fixed as well. This paper presents an approximate whole-program code search in multiple releases and branches. We evaluate this technique for real-world defects of various large and realistic programs having multiple releases and branches. We report runtime measurements and recall using varying levels of allowable differences of the approximate search.