A pattern-based foundation for language-driven software engineering

This work brings together two fundamental ideas for modelling, programming and analysing software systems. The first idea is of a methodological nature: engineering software by systematically creating and relating languages. The second idea is of a technical nature: using patterns as a practical foundation for computing. The goal is to show that the systematic creation and layering of languages can be reduced to the elementary operations of pattern matching and instantiation and that this pattern-based approach provides a formal and practical foundation for language-driven modelling, programming and analysis. The underpinning of the work is a novel formalism for recognising, deconstructing, creating, searching, transforming and generally manipulating data structures. The formalism is based on typed sequences, a generic structure for representing trees. It defines basic pattern expressions for matching and instantiating atomic values and variables. Horizontal, vertical, diagonal and hierarchical operators are different ways of combining patterns. Transformations combine matching and instantiating patterns and they are patterns themselves. A quasiquotation mechanism allows arbitrary levels of meta-pattern functionality and forms the basis of pattern abstraction. Path polymorphic operators are used to specify fine-grained search of structures. A range of core concepts such as layering, parsing and pattern-based computing can naturally be defined through pattern expressions. Three language-driven tools that utilise the pattern formalism showcase the applicability of the pattern-approach. Concat is a self-sustaining (meta-)programming system in which all computations are expressed by matching and instantiation. This includes parsing, executing and optimising programs. By applying its language engineering tools to its own meta-language, Concat can extend itself from within. XMF (XML Modeling Framework) is a browser-based modellingand meta-modelling framework that provides flexible means to create and relate modelling languages and to query and validate models. The pattern functionality that makes this possible is partly exposed as a schema language and partly as a JavaScript library. CFR (Channel Filter Rule Language) implements a language-driven approach for layered analysis of communication in complex networked systems. The communication on each layer is visible in the language of an “abstract protocol” that is defined by communication patterns.