Introducing reference semantics via refinement

Two types of semantics have been given to object-oriented formal specification languages. Value semantics denote a class by a set of values representing its objects. Reference semantics denote a class by a set of references, or pointers, to values representing its objects. While adopting the former facilitates formal reasoning, adopting the latter facilitates transformation to object-oriented code. In this paper, we propose a combined approach using value semantics for abstract specification and reasoning, and then refining to a reference semantics before transforming specification to code.

A partial-correctness semantics for modelling assembler programs

Previous work on formally modelling and analysing program compilation has shown the need for a simple and expressive semantics for assembler level programs. Assembler programs contain unstructured jumps and previous formalisms have modelled these by using continuations, or by embedding the program in an explicit emulator. We propose a simpler approach, which uses techniques from compiler theory in a formal setting. This approach is based on an interpretation of programs as collections of program paths, each of which has a weakest liberal precondition semantics. We then demonstrate, by example, how we can use this formalism to justify the compilation of block-structured high-level language programs into assembler.