Code search with input/output queries: Generalizing, ranking, and assessment

Abstract

In this work we generalize, improve, and extensively assess our semantic source code search engine through which developers use an input/output query model to specify what behavior they want instead of how it may be implemented. Under this approach a code repository contains programs encoded as constraints and an SMT solver finds encoded programs that match an input/output query. The search engine returns a list of source code snippets that match the specification.The initial instantiation of this approach showed potential but was limited. It only encoded single-path programs, reported just complete matches, did not rank the results, and was only partly assessed. In this work, we explore the use of symbolic execution to address some of these technical shortcomings. We implemented a tool, Satsy, that uses symbolic execution to encode multi-path programs as constraints and a novel ranking algorithm based on the strength of the match between an input/output query and the program paths traversed by symbolic execution. An assessment about the relevance of Satsy’s results versus other search engines, Merobase and Google, on eight novice-level programming tasks gathered from StackOverflow, using the opinions of 30 study participants, reveals that Satsy often out-performs the competition in terms of precision, and that matches are found in seconds.