API analytics for curating static analysis rules

Abstract

Use of third-party library APIs is pervasive, but can be error-prone. API-usage errors can be detected via static analysis if specifications of correct usage are available, but manually creating such specifications is a bottleneck. We showcase a semi-automated "big code" solution, where we use large code corpora to mine patterns in API usage, and ask human experts to perform analytics on those patterns to create static analysis rules.