Concurrent Predicates: Finding and Fixing the Root Cause of Concurrency Violations

To reduce the complexity of debugging multithreaded programs, researchers have developed compile- and run-time techniques that automatically detect concurrency bugs. These techniques can identify a wide range of shared memory errors, but are sometimes impractical because they produce many false positives making it difficult to triage and reproduce specific bugs. To address these concerns, we introduce a control structure, called concurrent predicate (CP), which allows programmers to single out a specific bug by specifying the conditions that must be satisfied for the bug to be triggered. Using bugs from a test suite of 23 programs, applications from RADBench, and TBoost.STM, we show how CP is used to diagnose and reproduce such bugs that could not otherwise be reproduced using similar techniques.