FESA: Fold- and Expand-Based Shape Analysis

A static shape analysis is presented that can prove the absence of NULLand dangling pointer dereferences in standard algorithms on lists, trees and graphs. It is conceptually simpler than other analyses that use symbolically represented logic to describe the heap. Instead, it represents the heap as a single graph and a Boolean formula. The key idea is to summarize two nodes by calculating their common points-to information, which is done using the recently proposed fold and expand operations. The force of this approach is that both, fold and expand , retain relational information between points-to edges, thereby essentially inferring new shape invariants. We show that highly precise shape invariants can be inferred using off-the-shelf SAT-solvers. Cheaper approximations may augment standard points-to analysis used in compiler optimisations.