Slicing Based Code Recommendation for Type Based Instance Retrieval

Abstract

It is common for developers to retrieve an instance of a certain type from another instance of other types. However, it is quite often that developers do not exactly know how to retrieve the instance although they know exactly what they need (the instance to re retrieved, also known as the target instance) and where it could be retrieved (i.e., the source instance). Such kind of instance retrieval is popular and thus their implementations, in different forms, are often publicly available on the Internet. Consequently, a number of approaches have been proposed to retrieve such implementations (code snippets) and release developers from reinventing such snippets. However, the performance of such approaches deserves further improvement. To this end, in this paper, we propose a slicing based approach to recommending code snippets that could retrieve the target instance from the source instance. The approach works as follows. First, from a large code base, it retrieves methods that contain the source instance and the target instance. Second, for each of these methods, it locates the target instances, and extracts related code snippets that generate the target instances by backward code slicing. Third, from the extracted code snippets, it removes those that do not contain the source instance. Fourth, it merges code snippets whose corresponding target instances are at parallel execution paths. Fifth, it removes duplicate code snippets. Finally, it ranks the resulting code snippets, and presents the top ones. We implement the approach as an Eclipse plugin called TIRSnippet. We also evaluate it with real type based instance retrieval queries. Evaluation results suggest that compared to the state-of-the-art approaches, the proposed approach improves the precision and recall by 8.8%, and 25%, respectively.