Predicting Function Changes by Mining Revision History

Software is consistently changing and evolving to new circumstances. Modifications to software do not always involve changes to a single, well-encapsulated module. Software developers are often faced with modification task that involve changes to source code artifacts such as function and comments that are spread across the code base. Developer must ensure that related entities are updated accordingly to be consistent with changes. In this paper, we propose hybrid approach that combines the best of data mining and impact analysis techniques to improve the overall performance (precision and recall) of change propagation heuristics. Our aim is to investigate the Function co-change in large software systems over period of time by utilizing various heuristics such as Function dependencies and History. It is not always every heuristic is good predictor for each entity. Therefore, we augment our effort to provide recommendations to programmers based on the prediction of heuristics that are best for entity needed to be changed. In this paper, we identify the best performing change propagation heuristic based on empirical case study, using a large open source system PostgreSQL database. This database consists of 31,000 functions and 1,493 files over the period of 12 years.