Code mobility and Java RMI

Today, more programmers than ever before are charged with the development and main­ tenance of distributed object­oriented applications. These tasks are dif cult to tackle alone, as distribution introduces myriad complications and opportunities for failure that are not encountered in centralised programs. Fortunately, the programmer is not alone. Several frameworks exist that attempt to transparently take care of many of the dif cul­ ties and small annoyances of distributed application development. However, by automating the activities required by distributed programs, such as mar­ shaling data for network transfer, we make it harder to understand the behaviour of these programs at source­code level. To rectify this, this thesis gives a formal semantics for a distributed object­oriented language. We call this language DJ, and it models key fea­ tures of the Java Remote Method Invocation framework. Importantly, it facilitates the marshaling of executable code as well as data, allowing the programmer to create higher­ order functions that can be stored and used as and when needed by an application. A signi cant contribution of this thesis is establishing type safety for this language. This requires the speci cation of a non­trivial number of invariant properties that dis­ tributed applications must preserve, capturing important sanity conditions such as en­ suring that local references do not leak to remote sites. We develop a theory of behavioural equivalence for DJ. Given as a set of semantically sound transformation rules that can be applied to source code and run­time terms, this theory judges two programs to be equivalent if they can be transformed into one an­ other. We then apply our technique to establishing the correctness of arguably typical distributed program optimisations. We conclude this thesis by proposing an extension to DJ, called DJ+, that models the concurrency features of Java. Type safety for this language is established using extended invariant properties.