Practical detection of concurrency issues at coding time

Abstract

We have developed a practical static checker that is designed to interactively mark data races and deadlocks in program source code at development time. As this use case requires a checker to be both fast and precise, we engaged a simple technique of randomized bounded concrete concurrent interpretation that is experimentally effective for this purpose. Implemented as a tool for C# in Visual Studio, the checker covers the broad spectrum of concurrent language concepts, including task and data parallelism, asynchronous programming, UI dispatching, the various synchronization primitives, monitor, atomic and volatile accesses, and finalizers. Its application to popular open-source C# projects revealed several real issues with only a few false positives.