Software developers usually rename identifiers and propagate the renaming based on the name binding of identifiers. Currently, software applications are usually developed using more than one language to enhance their functions and behaviors. Hence, when an identifier renaming is performed, it frequently affects more than one language in the multiple-language software applications. However, existing name binding approaches for identifiers only focus on a specific single language without considering the cross-language scenario. In this paper, we propose a cross-language name binding approach for the Java framework based on the deep learning model. Specifically, we first detect the potential name binding pairs via string matching. By analyzing the name binding pairs, the context, and the framework rules of identifiers, we extract several deep semantic features of identifiers and employ the BERT pre-trained model to recognize the name binding for unique identifiers, and further combine several classifiers to discriminate the name binding for duplicate identifiers. Our approach is evaluated on a manually constructed experimental dataset from 10 multiple-language projects. Experimental results demonstrate that our approach can achieve the average F-Measure of 85.14% in unique identifiers and 86.57% in duplicate identifiers, which significantly outperforms the baseline approaches. We also compare the performance of our approach against IntelliJ IDEA to further show its usefulness for developers in the real scenario.

Comments are summary descriptions of code snippets. When analyzing and maintaining programs, developers tend to read tidy comments rather than lengthy code. To prevent developers from misunderstanding the program or leading to potential bugs, ensuring the consistency and co-evolution of comments and the corresponding code is an integral development activity in practice. Nevertheless, when modifying code, developers sometimes neglect to update the relevant comments, resulting in inconsistency. Such comments may pose threats to the comprehension and maintenance of the software. In our study, we propose an overall approach named Context and Refactor based Comment Updater (CRCU). CRCU is a Just-In-Time (JIT) comment updater for specific commits. It takes a commit-id as input and updates all the method comments in this commit according to the code change. CRCU could be viewed as an optimization and augmentation of existing comment updaters, especially those that rely only on neural networks. Compared to the existing comment updaters, CRCU fully leverages the programming context and refactoring types of the modified methods to improve its performance. In addition, several customized enhancements in data pre-processing are introduced in CRCU to handle and filter out low-quality commits. We conduct extensive experiments to evaluate the effectiveness of CRCU. The evaluation results show that CRCU combined with the state-of-the-art approaches could improve the average Accuracy by 6.87% and reduce the developers’ edits by 0.298 on average.

Scale analysis plays a vital role in pedestrian detection. Conventional approaches usually directly concatenate multi-scale outputs, which is only capable of modeling first-order dependency among various scales. In contrast, this work proposes a novel scale-context modeling scheme by exploiting the highly nonlinear dependency among scales. The proposed scheme aggregates output response maps from mid-results of convolutional layers via a bi-directional recurrent sub-network. Therefore scale information could flow among different layers and implicit underlying dependency structure information in the scale space would be disclosed, which yields more consistency detection. Experimental results on Caltech Pedestrian detection benchmark demonstrate the superior detection (state-of-the-art miss rate of 8.56%) of the proposed method over prior art.