Model Counting for Complex Data Structures

We extend recent approaches for calculating the probability of program behaviors, to allow model counting for complex data structures with numeric fields. We use symbolic execution with lazy initialization to compute the input structures leading to the occurrence of a target event, while keeping a symbolic representation of the constraints on the numeric data. Off-the-shelf model counting tools are used to count the solutions for numerical constraints and field bounds encoding data structure invariants are used to reduce the search space. The technique is implemented in the Symbolic PathFinder tool and evaluated on several complex data structures. Results show that the technique is much faster than an enumeration-based method that uses the Korat tool and also highlight the benefits of using the field bounds to speed up the analysis.