SNIFF: A Search Engine for Java Using Free-Form Queries

Abstract

Reuse of existing libraries simplifies software development efforts. However, these libraries are often complex and reusing the APIs in the libraries involves a steep learning curve. A programmer often uses a search engine such as Google to discover code snippets involving library usage to perform a common task. A problem with search engines is that they return many pages that a programmer has to manually mine to discover the desired code. Recent research efforts have tried to address this problem by automating the generation of code snippets from user queries. However, these queries need to have type information and therefore require the user to have a partial knowledge of the APIs.We propose a novel code search technique, called SNIFF, which retains the flexibility of performing code search in plain English, while obtaining a small set of relevant code snippets to perform the desired task. Our technique is based on the observation that the library methods that a user code calls are often well-documented. We use the documentation of the library methods to add plain English meaning to an otherwise undocumented user code. The annotated user code is then indexed for the purpose of free-form query search. Another novel contribution of our technique is that we take a type-based intersection of the candidate code snippets obtained from a query search to generate a set of small and highly relevant code snippets.We have implemented SNIFF for Java and have performed evaluations and user studies to demonstrate the utility of SNIFF. Our evaluations show that SNIFF performed better than most of the existing online search engines as well as related tools.KeywordsSearch EngineUser QueryLonge Common SubsequenceUser CodeLonge Common SubsequenceThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.