SBHDetector: A Fuzzy-Based Hybrid Approach to Detect Renaming and Shifting Between Versions

Introduction

Mining Software Repositories (MSR) is widely used during the software evolution and maintenance phase for Version-Controlled Applications (VCA). During the evolution of code base with time, Software Repositories (SR) records the historical information in respect of versions/revisions for VCA (Agrawal et al., 2020). Analysis of these SR for different versions helps in identification of code clones (Roy et al., 2007) and change patterns that further enhances understandability, knowledge transfer and tracking similarities/changes in collaborative work. GIT, Open Source Distributed Version Control System (OSDVCS) is the most frequently used tool for collaborative work (GIT, n.d.). There are three challenges that restrict these MSR approaches in using SR-

• •

  Low performance for larger than necessary slices of Revision History: SR incorporates the changes in artifacts of software code shared among multiple developers. Usually, these changes are pushed/pulled using the diff and merge techniques to/from remote repository that may clutter the SR. These changes are recorded at coarse granularity (at file level) resulting in larger than necessary slices of Revision History (RH) (Zhu et al., 2020). Example- In order to extract changes at finer granularities (class and method level), it requires preprocessing for History Slicing using developer scripts that may be cumbersome and error-prone (Higo et al., 2020).

• •

  Restructuring may lead to unnecessary splitting of RH esp. where similarity index is low between the entities: The sequence of changes may involve major or minor refactoring along with recording of change-type such as renaming, moving or splitting but restricted with limitations of default similarity (Higo et al., 2020). Example- If similarity percentage of an entity before and after revision falls below 50% then GIT accept the change as creation of entity in Successive Version (SV) and removal of entity from preceding Version (PV). Due to this, it leads to unnecessary splitting of RH for that entity which makes the entity un-trackable for previous versions.

• •

  Textual similarity lacks in identification of changes at edge level: GIT is content addressable (Git-Internals-Git-Objects, n.d.), thus offers textual similarity in software code. However, existing approaches treats software code as graphs where changes may occur both at node and edge level (Störrle, 2015). Due to this, change in any reference to a particular object or hierarchy of interrelated object or change in number of calls to an object or dependencies etc. are not recorded in RH.

Therefore, changes in artifacts are missing at finer granularities and edge level leading to incomplete RH (Tang et al., 2022). Moreover, inaccuracy arises from the splitting of RH that breaks the link for changes in all PV by treating them as new entities with fresh RH. Due to this incomplete or inaccurate RH, it is very difficult to track these entities for MSR approaches (Agrawal et al., 2020) making it unreliable and time-consuming (Grund et al., n.d.). In order to overcome these challenges, the authors proposed SBHDetector, a technique with fuzzy based hybrid approach for the detection of similarities/changes between the versions. It classifies the similarities/changes in coarse & finer granularities even at edge level accounting both renaming and shifting of entities when empirically validated using eight Subject Systems (SS). The classification model has been built with high precision and recall in all the cases to identify similarities/changes where Random Forest outperforms in majority SS. Thus, it proves the generalizability of proposed approach and helps to improve productivity, reusability and maintainability with respect to VCA.