Tool-based language development

Abstract

This paper shows how to tackle the problem of ever larger languages and the problem of combining multiple languages. The central idea is to build a foundation for language definitions that uses a metamodel-driven approach with support of language engineering tools. Thus development and further evolution of languages will be transformed from the traditional paper- and -pen style into an engineering activity. This allows an early detection of problems within the language definition and increases the reliability of the language and the supporting tools.

Introduction

Developing high quality software in a cost effective manner is crucial for the competitiveness of software industry, and it continues to be a technological challenge. Today, a diversity of languages and notations is used, depending on the development phase and the viewpoints to be expressed. However, the use of several notations in the same project often leads to ambiguities and inconsistencies, jeopardizing the overall success of the project. To overcome this problem and to achieve the full benefit of the languages, they have to be integrated both syntactically and semantically.

An additional problem arises from the complexity of single languages. A monolithic style of a language definition (in SDL-2000 [1] for example), not only makes it hard for the user to fully understand and apply the language, but also aggravates the development of supporting tools. Five years after the finalization of the SDL-2000 Recommendation there is still not one tool available that fully supports SDL-2000. A hierarchical language definition is a promising way forward. It has to consist of a language core and a series of enhancing building blocks, accompanied with a clear list of compliance points that integrate the single part syntactically and semantically.

The development of SDL-2000—especially with the definition of the formal semantics—has already shown that it is possible and feasible to use software engineering tools in language development. This leads to a smaller number of errors and inconsistencies in the language standard and results in a prototypical set of tools that support this language.

Moreover, there is a trend towards languages that are tailored for special problem domains, and there is also the trend to integrate different languages. In the OMG (Object Management Group) context this is done by using the extension mechanisms of UML [2]; this is either the definition of UML profiles or the definition of new MOF (Meta Object Facility) based metamodels [3]. Tools that support the definition and application of this kind of language are largely missing.

It has been shown that it is possible to combine languages whose grammar is primarily based on a (context-free) syntax with languages that are based on metamodels [4]. But in general this requires a considerable amount of work in comparison with purely metamodel based languages, where very generic solutions can be achieved. To overcome this problem a transformation of the grammar based definition into a metamodel is necessary. In [5] an approach is given, in which an automatic generation of a initial metamodel from a grammar is combined with semi-automatic model transformations and manual refinements.

Until today, most approaches for integration of languages are tailored to two concrete languages (such as SDL and MSC, SDL and ASN.1), and they are not reusable. If the integration of another language is required, many steps of the previously done, work, have to be repeated. Therefore we propose a general framework for a metamodel based integration of languages.

This general framework is the key to the integration of different languages and their building blocks. It will allow the definition of languages on all modelling and metamodelling levels, including a proper exchange format for models. This implies also that languages are defined in a modular fashion, including proper interfaces. A language core, defined in such a way, can be extended in different ways, and thus cover further and more language specific concepts; it is sufficiently flexible to support future language add-ins.

Ultimately, the availability of a common semantic middleware with the possibility of modular language description paves the way for a new methodology of language design: Languages can be customized from a set of building blocks on a case by case basis, depending, for instance, on the application context and the preferences of the system developer.

We will implement this strategy within the SMILE¹ project. Our SMILE project targets all levels of the OMG four-layer metamodel architecture; it uses a common basic representation on all levels. This representation is called FORM after its main construct Form.² This lays the ground for executable UML with “metadata facilities” and makes it possible to specify how two neighbouring levels fit together.

In order to integrate our approach with traditional grammar based language definitions of SDL and other ITU languages, we developed supporting tools that transform them into a metamodel based definition. This is accompanied by a proposal for the hierarchical division of a language into a core language and building blocks of enhanced features. Such a solution does not only help to manage the complexity of the language itself, but it also supports the development of dedicated tools.

Different solutions for the combination of the language core with the enhancement building blocks are shown. These are not only applicable for the definition of a single language but also for the combination with other languages.

This paper is structured as follows: Section 2 gives an introduction into metamodelling and MDA, as well as into the combination of models. Afterwards we give a short description of the SMILE project to show how it solves the integration problems. We discuss the first steps made in Section 4, namely the basic FORM representation and the transformation of a grammar based language definition into a metamodel. In Section 5, we provide a deeper discussion of approaches for the combination and integration of different languages. Finally, we give some conclusions in Section 6.