Code Search on Bytecode for Mobile App Development

Code search is an essential task in software development. Existing code search engines often rely on open-source software code repositories to build their search database. However, in mobile app development, high quality code is rarely available as open-source. In this paper, we propose SUSHI, a novel code search system for mobile app development that exploits the vast amount of high-quality, production-ready code from millions of mobile apps publicly available on online app stores. SUSHI extracts names and usages of APIs directly from apps' bytecode and learns vector based representations of technical terms and statistical models of API usages. Then, it uses those vectors and models to calculate the relevance of queries and code snippets collected from mobile apps or other source code corpata if compatible.