Static Analysis of Concurrency Constructs in Higher-Order Programs

Detecting and debugging concurrency bugs is particularly hard. The complexity of concurrent programs is huge due to the exponential number of possible interleavings between the multiple units of execution of those programs. Having help from the computer to detect concurrency bugs — or prove their absence — can substantially reduce debugging time. In this dissertation, we investigate the use of abstract interpretation to detect race conditions and deadlocks in concurrent programs. We use the PCESK machine, as described by Might and Van Horn, to compute an over-approximation of the set of states reachable by a concurrent program. We provide a number of benchmarks to analyze the influence of garbage collection and state subsumption on the performance of this machine in terms of state space size and analysis time. We define several analyses that use the computed set of reachable states to deduce relevant properties of a concurrent program. Race condition analysis and deadlock analysis are the most notable analyses we develop. Although both these analyses are unsound and thus might miss some defects, they exhibit high precision and recall, allowing them to detect a good portion of bugs in relatively short programs. This dissertation shows that it is possible to use the PCESK machine for building useful analyses to detect race conditions and deadlocks. The main limitation of this approach is the size of the state spaces involved in the analyses. We therefore investigate the PCESKL machine, a variant of the PCESK machine that uses first-class locks instead of cas (compare-and-swap) as a synchronization primitive. We find that the PCESKL machine leads to simplified analyses and reduced state space size, resulting in reduced analysis time. This machine is therefore more suited for analyzing longer and more complex programs.