This thesis is concerned with verifying the correctness of programs written in GP 2 (for Graph Programs), an experimental, nondeterministic graph manipulation language, in which program states are graphs, and computational steps are applications of graph transformation rules. GP 2 allows for visual programming at a high level of abstraction, with the programmer freed from manipulating low-level data structures and instead solving graph-based problems in a direct, declarative, and rule-based way. To verify that a graph program meets some specification, however, has been -- prior to the work described in this thesis -- an ad hoc task, detracting from the appeal of using GP 2 to reason about graph algorithms, high-level system specifications, pointer structures, and the many other practical problems in software engineering and programming languages that can be modelled as graph problems. This thesis describes some contributions towards the challenge of verifying graph programs, in particular, Hoare logics with which correctness specifications can be proven in a syntax-directed and compositional manner.
We contribute calculi of proof rules for GP 2 that allow for rigorous reasoning about both partial correctness and termination of graph programs. These are given in an extensional style, i.e. independent of fixed assertion languages. This approach allows for the re-use of proof rules with different assertion languages for graphs, and moreover, allows for properties of the calculi to be inherited: soundness, completeness for termination, and relative completeness (for sufficiently expressive assertion languages).
We propose E-conditions as a graphical, intuitive assertion language for expressing properties of graphs -- both about their structure and labelling -- generalising the nested conditions of Habel, Pennemann, and Rensink. We instantiate our calculi with this language, explore the relationship between the decidability of the model checking problem and the existence of effective constructions for the extensional assertions, and fix a subclass of graph programs for which we have both. The calculi are then demonstrated by verifying a number of data- and structure-manipulating programs.
We explore the relationship between E-conditions and classical logic, defining translations between the former and a many-sorted predicate logic over graphs; the logic being a potential front end to an implementation of our work in a proof assistant.
Finally, we speculate on several avenues of interesting future work; in particular, a possible extension of E-conditions with transitive closure, for proving specifications involving properties about arbitrary-length paths.