Departmental Papers (CIS)

Document Type

Conference Paper

Date of this Version

March 2004


Postprint version. Published in Lecture Notes in Computer Science, Volume 2988, Tools and Algorithms for the Construction and Analysis of Systems, (TACAS 2004), pages 467-481.
Publisher URL:


Model checking of linear temporal logic (LTL) specifications with respect to pushdown systems has been shown to be a useful tool for analysis of programs with potentially recursive procedures. LTL, however, can specify only regular properties, and properties such as correctness of procedures with respect to pre and post conditions, that require matching of calls and returns, are not regular. We introduce a temporal logic of calls and returns (CARET) for specification and algorithmic verification of correctness requirements of structured programs. The formulas of CARET are interpreted over sequences of propositional valuations tagged with special symbols call and ret. Besides the standard global temporal modalities, CARET admits the abstract-next operator that allows a path to jump from a call to the matching return. This operator can be used to specify a variety of non-regular properties such as partial and total correctness of program blocks with respect to pre and post conditions. The abstract versions of the other temporal modalities can be used to specify regular properties of local paths within a procedure that skip over calls to other procedures. CARET also admits the caller modality that jumps to the most recent pending call, and such caller modalities allow specification of a variety of security properties that involve inspection of the call-stack. Even though verifying context-free properties of pushdown systems is undecidable, we show that model checking CARET formulas against a pushdown model is decidable. We present a tableau construction that reduces our model checking problem to the emptiness problem for a Büchi pushdown system. The complexity of model checking CARET formulas is the same as that of checking LTL formulas, namely, polynomial in the model and singly exponential in the size of the specification.



Date Posted: 21 December 2005