Dr. Liu Mingwei (刘名威)

Fixing crashes is challenging, and developers often discuss their encountered crashes and refer to similar crashes and solutions on online Q&A forums (e.g., Stack Overflow). However, a crash often involves very complex contexts, which includes different contextual elements, e.g., purposes, environments, code, and crash traces. Existing crash solution recommendation or general solution recommendation techniques only use an incomplete context or treat the entire context as pure texts to search relevant solutions for a given crash, resulting in inaccurate recommendation results. In this work, we propose a novel crash solution knowledge graph (KG) to summarize the complete crash context and its solution with a graph-structured representation. To construct the crash solution KG automatically, we propose to leverage prompt learning to construct the KG from SO threads with a small set of labeled data. Based on the constructed KG, we further propose a novel KG-based crash solution recommendation technique KG4CraSolver, which precisely finds the relevant SO thread for an encountered crash by finely analyzing and matching the complete crash context based on the crash solution KG. The evaluation results show that the constructed KG is of high quality and KG4CraSolver outperforms baselines in terms of all metrics (e.g., 13.4%-113.4% MRR improvements). Moreover, we perform a user study and find that KG4CraSolver helps participants find crash solutions 34.4% faster and 63.3% more accurately.

Library migration, which replaces the current library with a different one to retain the same software behavior, is common in software evolution. An essential part of this is finding an analogous API for the desired functionality. However, due to the multitude of libraries/APIs, manually finding such an API is time-consuming and error-prone. Researchers created automated analogical API recommendation techniques, notably documentation-based methods. Despite potential, these methods have limitations, e.g., incomplete semantic understanding in documentation and scalability issues. In this study, we present KGE4AR, a novel documentation-based approach using knowledge graph (KG) embedding for recommending analogical APIs during library migration. KGE4AR introduces a unified API KG to comprehensively represent documentation knowledge, capturing high-level semantics. It further embeds this unified API KG into vectors for efficient, scalable similarity calculation. We assess KGE4AR with 35,773 Java libraries in two scenarios, with and without target libraries. KGE4AR notably outperforms state-of-the-art techniques (e.g., 47.1%-143.0% and 11.7%-80.6% MRR improvements), showcasing scalability with growing library counts.

Recently, many large language models (LLMs) have been proposed, showing advanced proficiency in code generation. Meanwhile, many efforts have been dedicated to evaluating LLMs on code generation benchmarks such as HumanEval. Although being very helpful for comparing different LLMs, existing evaluation focuses on a simple code generation scenario (i.e., function-level or statement-level code generation), which mainly asks LLMs to generate one single code unit (e.g., a function or a statement) for the given natural language description. Such evaluation focuses on generating independent and often small-scale code units, thus leaving it unclear how LLMs perform in real-world software development scenarios. To fill this knowledge gap, we make the first attempt to evaluate LLMs in a more challenging code generation scenario, i.e., classlevel code generation. Compared with existing code generation benchmarks, it better reflects real-world software development scenarios due to it comprising broader contextual dependencies and multiple, interdependent units of code. We first manually construct the first class-level code generation benchmark ClassEval of 100 class-level Python code generation tasks with approximately 500 person-hours. Based on the new benchmark ClassEval, we then perform the first study of 11 state-of-the-art LLMs on class-level code generation. Based on our results, we find that all LLMs perform much worse on class-level code generation compared to the methodlevel. While GPT models still dominate other LLMs on class-level code generation, the performance rankings of other models on method-level code generation no longer holds for class-level code generation. Besides, most models (except GPT models) perform better when generating the class method by method; and they have the limited ability of generating dependent code. Based on our findings, we call for software engineering (SE) researchers’ expertise to build more LLM benchmarks based on practical and complicated software development scenarios

Unit testing plays an essential role in detecting bugs in functionally-discrete program units (e.g., methods). Manually writing high-quality unit tests is time-consuming and laborious. Although the traditional techniques are able to generate tests with reasonable coverage, they are shown to exhibit low readability and still cannot be directly adopted by developers in practice. Recent work has shown the large potential of large language models (LLMs) in unit test generation. By being pre-trained on a massive developer-written code corpus, the models are capable of generating more human-like and meaningful test code. In this work, we perform the first empirical study to evaluate the capability of ChatGPT (i.e., one of the most representative LLMs with outstanding performance in code generation and comprehension) in unit test generation. In particular, we conduct both a quantitative analysis and a user study to systematically investigate the quality of its generated tests in terms of correctness, sufficiency, readability, and usability. We find that the tests generated by ChatGPT still suffer from correctness issues, including diverse compilation errors and execution failures (mostly caused by incorrect assertions); but the passing tests generated by ChatGPT almost resemble manually-written tests by achieving comparable coverage, readability, and even sometimes developers’ preference. Our findings indicate that generating unit tests with ChatGPT could be very promising if the correctness of its generated tests could be further improved. Inspired by our findings above, we further propose ChatTester, a novel ChatGPT-based unit test generation approach, which leverages ChatGPT itself to improve the quality of its generated tests. ChatTester incorporates an initial test generator and an iterative test refiner. Our evaluation demonstrates the effectiveness of ChatTester by generating 34.3% more compilable tests and 18.7% more tests with correct assertions than the default ChatGPT. In addition to ChatGPT, we further investigate the generalization capabilities of ChatTester by applying it to two recent open-source LLMs (i.e., CodeLlama-Instruct and CodeFuse) and our results show that ChatTester can also improve the quality of tests generated by these LLMs.

Recent advancements in large language models (LLMs) have demonstrated impressive capabilities in code translation, typically evaluated using benchmarks like CodeTransOcean and RepoTransBench. However, dependency-free benchmarks fail to capture real-world complexities by focusing primarily on simple function-level translations and overlooking repository-level context (e.g., dependencies). Full-repository translation benchmarks significantly exceed the current capabilities of existing models, resulting in performance bottlenecks that fail to provide actionable insights for guiding model development. Furthermore, existing benchmarks do not account for the scenario of incrementally translating new or modified modules from the source to the target language, which demands careful handling of repositorylevel contexts such as dependencies, cross-module references, and architectural divergence. Moreover, LLMs’ effectiveness in translating to newer, low-resource languages like Rust remains largely underexplored. To address these gaps, we introduce RustRepoTrans, the first repository-level context code translation benchmark targeting incremental translation, comprising 375 tasks translating into Rust from C, Java, and Python. Using this benchmark, we evaluate seven representative LLMs, analyzing their errors to assess limitations in complex translation scenarios. Among them, DeepSeek-R1 performs best with 51.5% Pass@1, excelling in both basic functionality and additional translation abilities, such as noise robustness and syntactical difference identification. However, even DeepSeek-R1 experiences a 22.2% performance drop (Pass@1 from 73.7% to 51.5%) when handling repositorylevel context compared to previous benchmarks without such context. Meanwhile, we propose a set of more fine-grained evaluation metrics and an enhanced evaluation framework, enabling a more comprehensive analysis of LLMs’ performance in repository-level context code translation tasks to provide finegrained insights that can effectively inform the development of code translation techniques.