Predicting All Data Race Pairs for a Specific Schedule (extended version)

TL;DR

This paper introduces an efficient method combining a linear time vector clock algorithm and a quadratic post-processing step to predict all data race pairs in a program trace, improving scalability and accuracy over previous approaches.

Contribution

It presents a novel refined vector clock algorithm and a quadratic post-processing method to comprehensively predict schedulable data race pairs, reducing false positives.

Findings

Scales to real-world program traces

Predicts all schedulable data race pairs

Reduces false positives in race detection

Abstract

We consider the problem of data race prediction where the program's behavior is represented by a trace. A trace is a sequence of program events recorded during the execution of the program. We employ the schedulable happens-before relation to characterize all pairs of events that are in a race for the schedule as manifested in the trace. Compared to the classic happens-before relation, the schedulable happens-before relations properly takes care of write-read dependencies and thus avoids false positives. The challenge is to efficiently identify all (schedulable) data race pairs. We present a refined linear time vector clock algorithm to predict many of the schedulable data race pairs. We introduce a quadratic time post-processing algorithm to predict all remaining data race pairs. This improves the state of the art in the area and our experiments show that our approach scales to real-world examples. Thus, the user can systematically examine and fix all program locations that are in a race for a particular schedule.