Anders B. Sandholm
Domain-specific language design is an emerging trend in software engineering. The idea is to express the analysis of a particular problem domain through the design of a programming language tailored to solving problems within that domain. There are many established examples of domain-specific languages, such as LaTeX, flex, and yacc. The novelty is to use such techniques for solving traditional software engineering task, where one would previously construct some collection of libraries within a general-purpose language. In Part I, we start by presenting the design of a domain-specific language for programming Web services. The various domain-specific aspects of Web service programming, such as, Web server runtime system design, document handling, and concurrency control, are treated.
Having treated many of the Web related topics in some detail, we turn to the particular ideas and central design issues involved in the development of our server-side runtime system. We build a runtime system that not only shares the advantage of traditional CGI-based services, namely that of portability, we also overcome many of the known disadvantages, such as intricate and tedious programming and poor performance.
We then turn to the use of dynamic Web documents, which in our language relies heavily on the use of static program analysis. Static program analysis allows one to offer static compile-time techniques for predicting safe and computable approximations to the set of values or behavior arising dynamically at runtime when executing a program on a computer. Some of the traditional applications of program analysis are avoidance of redundant and superfluous computations. Another aspect is that of program validation, where one provides guarantees that the analyzed code can safely be executed. We present an example of such an analysis as explained in the following. Many interactive Web services use the CGI interface for communication with clients. They will dynamically create HTML documents that are presented to the client who then resumes the interaction by submitting data through incorporated form fields. This protocol is difficult to statically type-check if the dynamic document are created by arbitrary script code using ``printf'' statements. Previous proposals have suggested using static document templates which trades flexibility for safety. We propose a notion of typed, higher-order templates that simultaneously achieve flexibility and safety. Our type system is based on a flow analysis of which we prove soundness. We also present an efficient runtime implementation that respects the semantics of only well-typed programs.
Having dealt with dynamic Web documents, we turn to the topic of control synthesis. In the design of safety critical systems it is of great importance that certain guarantees can be made about the system. We describe in Part II of this thesis a method for synthesizing controllers from logical specifications. First, we consider the development of the BDD-based tool Mona for translating logical formulae into automata and give examples of its use in small applications. Then we consider the use of Mona for control synthesis, that is, we turn to the use of Mona as a controller generator. First, in the context of our domain specific language for generating Web services, where concurrency control aspects are treated using our control synthesis technique. Second, in the context of LEGO robots, where we show how controllers for autonomous LEGO robots are generated using our technique and actually implemented on the physical LEGO brick.
Finally, in Part III we consider the problem of sequentiality and full abstraction in denotational semantics. The problem of finding an abstract description of sequential functional computation has been one of the most enduring problems of semantics. The problem dates from a seminal paper of Plotkin, who pointed out that certain elements in Scott models are not defineable. In Plotkin's example, the ``parallel or'' function cannot be programmed in PCF, a purely functional, sequential, call-by-name language. Moreover, the function causes two terms to be distinct denotationaly even though the terms cannot be distinguished by any program. The problem of modeling sequentiality is enduring because it is robust. For instance, changing the reduction strategy from call-by-name to call-by-value makes no difference. When examples like parallel or do not exist, the denotational model is said to be fully abstract. More precisely, full abstraction requires an operational definition of equivalence (interchangeability in all programs) to match operational equivalence. It has been proved that there is exactly one fully abstract model of PCF. Until recently, all descriptions of this fully abstract model have used operational semantics. New constructions using logical relations have yielded a more abstract understanding of Milner's model.
We consider in Part III how to adapt and extend the logical relations model for PCF to a call-by-value setting. More precisely, we construct a fully abstract model for the call-by-value, purely functional, sequential language FPC.
Available as PostScript,