Are Human Rules Necessary? Generating Reusable APIs with CoT Reasoning and In-Context Learning

Nowadays, more and more developers resort to Stack Overflow for solutions (e.g., code snippets) when they encounter technical problems. Although domain experts provide huge amounts of valuable solutions in Stack Overflow, these code snippets are often difficult to reuse directly. Developers have to digest the information within relevant posts and make necessary modifications, and the whole solution-seeking process can be time-consuming and tedious. To facilitate the reuse of Stack Overflow code snippets, Terragni et al. first explored transforming a code snippet in Stack Overflow into a well-formed method API ( A pplication P rogram I nterface) by using a rule-based approach, named APIzator. The reported performance of their approach is promising, however, after our in-depth analysis of their experiment results, we find that (1) 92.5% of APIs generated by APIzator are pointless and thus are difficult to use in practice. This is because the method name generated by APIzator (extracting verb + object ) can rarely represent the method’s functionality, which can hardly be claimed as meaningful/reusable APIs. (2) The authors manually summarized a number of rules to identify parameter variables and return statements for Java methods. These hand-crafted rules are extremely complex and sophisticated, and the manual rule design process is labor-intensive and error-prone. Moreover, since these rules are designed for Java, they can hardly be extended to other programming languages. Inspired by the great potential of Large Language Models (LLMs) for solving complex coding tasks, in this paper, we propose a novel approach, named Code 2API, to automatically perform APIzation for Stack Overflow code snippets. Code 2API does not require additional model training or any manual crafting rules and can be easily deployed on personal computers without relying on other external tools. Specifically, Code 2API guides the LLMs through well-designed prompts to generate well-formed APIs for given code snippets. To elicit knowledge and logical reasoning from LLMs, we used c hain- o f- t hought (CoT) reasoning and few-shot in-context learning, which can help the LLMs fully understand the APIzation task and solve it step by step in a manner similar to a developer. Our evaluations show that Code 2API achieves a remarkable accuracy in identifying method parameters (65%) and return statements (66%) equivalent to human-generated ones, surpassing the current state-of-the-art approach, APIzator, by 15.0% and 16.5% respectively. Moreover, compared with APIzator, our user study demonstrates that Code 2API exhibits superior performance in generating meaningful method names, even surpassing the human-level performance, and developers are more willing to use APIs generated by our approach, highlighting the applicability of our tool in practice. Finally, we successfully extend our framework to the Python dataset, achieving a comparable performance with Java, which verifies the generalizability of our tool.