Code review is an important activity in software development, which offers benefits such as improving code quality, reducing defects and distributing knowledge. Tencent, as a giant company, hosts a great number of proprietary software projects that are only open to specific internal developers. Since these proprietary projects receive up to 100,000 of newly submitted code changes per month, it is extremely needed to automatically recommend code reviewers. To this end, we first conduct an empirical study on a large scale of proprietary projects from Tencent, to understand their characteristics and how code reviewer recommendation approaches work on them. Based on the derived findings and implications, we propose a new approach named Camp that recommends reviewers by considering their collaboration and expertise in multiple projects, to fit the context of proprietary software development. The evaluation results show that Camp can achieve higher scores on proprietary projects across most metrics than other state-of-the-art approaches, i.e., Revfinder, CHREV, Tie and Comment Network and produce acceptable performance scores for more projects. In addition, we discuss the possible directions of code reviewer recommendation.

Methods in programs must be accurately named to facilitate source code analysis and comprehension. With the evolution of software, method names may be inconsistent with their implemented method bodies, leading to inaccurate or buggy method names. Debugging method names remains an important topic in the literature. Although researchers have proposed several approaches to suggest accurate method names once the method bodies have been modified, two main drawbacks remain to be solved: there is no analysis of method name structure, and the programming context information is not captured efficiently. To resolve these drawbacks and suggest more accurate method names, we propose a novel automated approach based on the analysis of the method name structure and lexical analysis with the programming context information. Our approach first leverages deep feature representation to embed method names and method bodies in vectors. Then, it obtains useful verb-tokens from a large method corpus through structural analysis and noun-tokens from method bodies through lexical analysis. Finally, our approach dynamically combines these tokens to form and recommend high-quality and project-specific method names. Experimental results over 2111 Java testing methods show that the proposed approach can achieve a Hit Ratio, or Hit@5, of 33.62% and outperform the state-of-the-art approach by 14.12% in suggesting accurate method names. We also demonstrate the effectiveness of structural and lexical analyses in our approach.

Information Retrieval (IR) techniques have been widely utilized by a growing number of software maintenance activities. However, there is a mismatch between source code lexicon (especially identifiers) and vocabulary in software artifacts, leading to the inefficiency of IR techniques. Consequently, it is essential to normalize identifiers, whose aim is to parse identifiers into several natural language terms. Identifier splitting significantly impacts on the effectiveness of identifier normalization. Even though researchers have proposed several approaches to split identifiers, three main drawbacks remain to be resolved, including without considering morphemes, over-splitting, and under-splitting. In this paper, we propose a new Character-level Hybrid-granularity Identifier Splitting approach CHIS to resolve the three drawbacks and better split identifiers. CHIS combines the Bidirectional Encoder Representation from Transformers (BERT) and Conditional Random Fields (CRF) to train a deep learning model to split identifiers. In addition, CHIS further employs a pre-processing component and a post-processing component to resolve the morpheme acquisition drawback and the over-splitting as well as the under-splitting drawbacks respectively, thus further improving its performance. Specifically, in the pre-processing component, CHIS obtains and labels the most frequent subwords of the training identifiers as morphemes through the Byte Pair Encoding (BPE) algorithm and the sequence labeling algorithm. In the post-processing component, CHIS iteratively merges and splits the splitting results obtained by the deep learning model to resolve the over-splitting and under-splitting drawbacks. We conduct extensive experiments to show the effectiveness of CHIS. Experimental results show that CHIS achieves the Accuracy of 0.943 on average and outperforms the state-of-the-art approach by 0.085 on average. In addition, the effectiveness of the pre-processing and post-processing components of CHIS are also validated.

During the process of software development and maintenance, developers would regularly refactor existing source code to improve efficiency and maintainability. Among various code refactoring activities, method renaming often happens within the whole project evolution process. To perform method renaming, developers should first identify the exact methods that should be renamed, which is generally tedious and error-prone through manual analysis. Towards this end, researchers have proposed some approaches to automatically recommend candidate methods for renaming. To further improve the performance of existing techniques, in this paper, we propose a novel approach that fully leverages historical code changes and overlapping relationships among code entities to identify renaming opportunities for methods. Specifically, we first embed methods into vectors and incorporate overlapping relationships among code entities by using different attention heads in a deep learning network. Then, we apply these obtained vectors to train a classifier to predict potential renaming opportunities for methods. Finally, we utilize historical renaming activities of related code entities to further refine the predicted results. Experimental results on 114,398 methods from 10 open source Java projects show that our approach could outperform the state-of-the-art approach by achieving an average F-measure of 80.02%. To better validate the effectiveness of our approach, we also explore the performance of some major components of our approach. For example, we find that employing related code entities help to improve the performance of our approach by 40.40% in terms of the average F-measure.