Structuring data via behavioural synthesis

This thesis furthers work in the dynamic analysis of programs by examining how to derive data structure definitions from the behavioural aspects of programs. We refer to this technique as behavioural synthesis. In particular we first present a general data typing mechanism, called a dynamic abstract data type (DADT), whose instances act as proxies for other data types in a computation. Each instance of a DADT has the ability to reconfigure its internal representation and detect the context in which it is used. This is followed by an example of a DADT that represents the union of other abstract data types whose interfaces are not necessarily disjoint; techniques for specializing their use in a program are also demonstrated. The specializations are realized as code in the same language as the program, promoting program evolution. Furthermore, the code generation is based on source code locations, does not require explicit annotations and is performed interactively. Finally, we show a method of deriving the composition and relationship of classes through augmentation of a language runtime. The approach permits the use of objects without class definitions. Behavioural synthesis is presented as a design exploration aid to support the notion of using a typical programming language as a specification language and may be viewed as extending the duality between program and data to a duality between program structure and data structure.