Chapter

Verification, Model Checking, and Abstract Interpretation

Volume 3385 of the series Lecture Notes in Computer Science pp 164-180

Shape Analysis by Predicate Abstraction

  • Ittai BalabanAffiliated withLancaster UniversityNew York University
  • , Amir PnueliAffiliated withLancaster UniversityNew York University
  • , Lenore D. ZuckAffiliated withLancaster UniversityUniversity of Illinois at Chicago

* Final gross prices may vary according to local VAT.

Get Access

Abstract

The paper presents an approach for shape analysis based on predicate abstraction. Using a predicate base that involves reachability relations between program variables pointing into the heap, we are able to analyze functional properties of programs with destructive heap updates, such as list reversal and various in-place list sorts. The approach allows verification of both safety and liveness properties. The abstraction we use does not require any abstract representation of the heap nodes (e.g. abstract shapes), only reachability relations between the program variables.

The computation of the abstract transition relation is precise and automatic yet does not require the use of a theorem prover. Instead, we use a small model theorem to identify a truncated (small) finite-state version of the program whose abstraction is identical to the abstraction of the unbounded-heap version of the same program. The abstraction of the finite-state version is then computed by BDD techniques.

For proving liveness properties, we augment the original system by a well-founded ranking function, which is abstracted together with the system. Well-foundedness is then abstracted into strong fairness (compassion). We show that, for a restricted class of programs that still includes many interesting cases, the small model theorem can be applied to this joint abstraction.

Independently of the application to shape-analysis examples, we demonstrate the utility of the ranking abstraction method and its advantages over the direct use of ranking functions in a deductive verification of the same property.