A Method for Improving the Precision and Coverage of Atomicity Violation Predictions

摘要: Atomicity violations are the most common non-deadlock concurrency bugs, which have been extensively studied in recent years. Since detecting actual occurrences of atomicity is extremely hard and exhaustive testing a multi-threaded program general impossible, many predictive methods proposed, make error predictions based on small number instrumented interleaved executions. Predictive often tradeoffs between precision coverage. An over-approximate method ensures coverage but lacks thus may report large false bugs. under-approximate can miss significant real This paper presents post-prediction analysis for improving prediction results obtained through over-approximation while achieving better than that under-approximation. Our analyzes filters by evaluating subset read-after-write relationships without enforcing all them as existing under-approximation methods. static predicted traces from dynamic instrumentation C/C++ executable, faster replaying ensuring precision.