Predictive Runtime Analysis of Multithreaded Programs

We present a technique to predict property violations in multi-threaded programs from successful executions. An appealing aspect of our technique is that it is entirely automatic; another is that no special simulation or modeling infrastructure is needed. All the user needs to do is to provide the multi-threaded system and the property to check. An observer is automatically generated from the property and an instrumentation procedure based on vector clocks automatically modifies the program to emit relevant events to the observer. By making intensive use of a dynamically computed generalized “happens-before” causal partial order that is refined with control-flow and data-flow dependency information obtained apriory via static analysis of the program, the observer is able to build from one concrete execution trace a set of abstract execution traces. Those abstract execution traces correspond to concrete executions that have not necessarily happened but “were close to happen” and could indeed happen in another execution of the system under a different thread scheduling. The predictive aspect of our technique comes from the fact that some of those executions may violate the property. If that is the case, a counter-example is provided. The technique has been implemented in the context of Java and has been shown to be useful via a series of experiments.