Automatic Inference of Heap Properties Exploiting Value Domains

Abstract

Effective static analyses of heap-manipulating programs need to track precise information about the heap structures and the values computed by the program. Most existing heap analyses rely on manual annotations to precisely analyze general and, in particular, recursive, heap structures. Moreover, they either do not exploit value information to obtain more precise heap information or require more annotations for this purpose. In this paper, we present a combined heap and value analysis that infers complex invariants for recursive heap structures such as lists and trees, including relations between value fields of heap-allocated objects. Our analysis uses a novel notion of edge-local identifiers to track value information about the source and target of a pointer, even if these are summary nodes. With each potential pointer in the heap, our analysis associates value information that describes in which states the pointer may exist, and uses this information to improve the precision of the analysis by pruning infeasible heap structures. Our analysis has been implemented in the static analyzer Sample; experimental results show that it can automatically infer invariants for data structures, for which state-of-the-art analyses require manual annotations.