Effective race detection for event-driven programs (invited talk abstract)

Like shared-memory multi-threaded programs, event-driven programs are susceptible to non-deterministic errors that are hard to reproduce and debug. Mobile application platforms and the web require writing event-driven code, motivating the need for better tools to detect these types of errors. In the first part of this talk, we will show how techniques from data race detection for multi-threaded programs can be applied to detect these bugs, by defining an appropriate notion of a happens-before relation and memory accesses for the underlying platform. While extant data race detectors can be applied to event-driven programs based on the above, the pervasive use of ad hoc synchronization in such programs can lead a race detector to report too many false positives. Race detection for such programs also faces a scalability challenge, as a large number of short-running event handlers can overwhelm standard vector-clock-based techniques. We present a novel race detector EventRacer that addresses both of these challenges. We introduce race coverage, a systematic method for exposing ad hoc synchronization and other key races to the user, significantly reducing false positives. We also present an efficient connectivity algorithm that uses chain decomposition to dramatically decrease the overhead of vector clocks. We show that for client-side web applications, our techniques lead to large performance and precision improvements over state-of-the-art techniques.