A specification-based approach to reasoning about pointers

This paper explains how a uniform, specification-based approach to reasoning about component-based programs can be used to reason about programs that manipulate pointers. No special axioms, language semantics, global heap model, or proof rules for pointers are necessary. We show how this is possible by capturing pointers and operations that manipulate them in the specification of a software component. The proposed approach is mechanizable as long as programmers are able to understand mathematical specifications and write assertions, such as loop invariants. While some of the previous efforts in reasoning do not require such mathematical sophistication on the part of programmers, they are limited in the kinds of properties they can prove about programs that use pointers. We illustrate the idea using a "Splice" operation for linked lists, which has been used previously to explain other analysis techniques. Not only can the proposed approach be used to establish shape properties given lightweight specifications, but also it can be used to establish total correctness given more complete specifications.