Dependent Types for Program Understanding

Weakly-typed languages such as Cobol often force programmers to represent distinct data abstractions using the same low-level physical type. In this paper, we describe a technique to recover implicitly-defined data abstractions from programs using type inference. We present a novel system of dependent types which we call guarded types, a path-sensitive algorithm for inferring guarded types for Cobol programs, and a semantic characterization of correct guarded typings. The results of our inference technique can be used to enhance program understanding for legacy applications, and to enable a number of type-based program transformations.