Dynamic Analyses for Data-Race Detection

Abstract

Data races caused by unsynchronized accesses to shared data have long been the source of insidious errors in concurrent software. They are hard to identify during testing, reproduce, and debug. Recent advances in race detection tools show great promise for improving the situation, however, and can enable programmers to find and eliminate race conditions more effectively. This tutorial explores dynamic analysis techniques to efficiently find data races in large-scale software. It covers the theoretical underpinnings, implementation techniques, and reusable infrastructure used to build state-of-the-art data-race detectors (as well as analyses targeting other types of concurrency errors). The tutorial provides industrial case studies on finding data races and closes with a discussion of open research questions in this area.