Arie van Deursen Paul Klint Joost Visser
CWI, P.O. Box 94079, 1090 GB Amsterdam, The Netherlands
Domain-Speciﬁc Languages are used in software engineering in order to enhance quality, ﬂexibility, and timely delivery of
softwaresystems, by taking advantageof speciﬁc properties of a particularapplication domain. This surveycoversterminology,
risks and beneﬁts, examples, design methodologies, and implementation techniques of domain-speciﬁc languages as used for
the construction and maintenance of software systems. Moreover, it covers an annotated selection of 75 key publications in the
area of domain-speciﬁc languages.
1998 ACM Computing Classiﬁcation System: D.3
Keywords and Phrases: Example DSLs, DSL design, DSL implementation, survey.
Note: Work carried out under project SEN 1.5, Domain-Speciﬁc Languages, sponsored by the Telematica Instituut.
In all branches of science and engineering one can distinguish between approaches that are generic and those that are speciﬁc.
A generic approach provides a general solution for many problems in a certain area, but such a solution may be suboptimal. A
speciﬁc approach provides a much better solution for a smaller set of problems. One of the incarnations of this dichotomy in
computer science is the topic of this survey: domain-speciﬁc languages versus generic programming languages.
Of course, this is not a new topic. The older programming languages (Cobol, Fortran, Lisp) all came into existence as
dedicatedlanguages for solving problemsin a certain area (respectivelybusiness processing, numeric computationandsymbolic
processing). Gradually they have evolved into general purpose languages and over and over again the need for more specialized
language support to solve problems in well-deﬁned application domains has resurfaced. Over time, the following solutions
have been tried:
Subroutine libraries contain subroutines that perform related tasks in well-deﬁned domains like, for instance, differential
equations, graphics, user-interfaces and databases. The subroutine library is the classical method for packaging reusable
Object-oriented frameworks and component frameworks continue the idea of subroutine libraries. Classical libraries
have a ﬂat structure, and the application invokes the library. In object-oriented frameworks it is often the case that the
framework is in control, and invokes methods provided by the application-speciﬁc code [42, 32].
A domain-speciﬁc language (DSL) is a small, usually declarative, language that offers expressive power focused on a
particular problem domain. In many cases, DSL programs are translated to calls to a common subroutine library and the
DSL can be viewed as a means to hide the details of that library.
Although many domain-speciﬁc languages have been designed and used over the years, the systematic study of domain-
speciﬁc languages has only started more recently. This survey provides an inventory of the ﬁeld and covers references to
research that deals with the following topics: terminology (Section 2), risks and opportunities (Section 3), example DSLs
(Section 4), DSL design methodology (Section 5), and DSL implementation strategies (Section 6). The papers listed are
annotated with summaries, which in turn are cross-referenced to related papers.
The question what exactly is a domain-speciﬁc language is subject to debate. We propose the following deﬁnition:
A domain-speciﬁc language (DSL) is a programming language or executable speciﬁcation language that offers,
through appropriate notations and abstractions, expressive power focused on, and usually restricted to, a particular
The key characteristic of DSLs according to this deﬁnition is their focussed expressive power.
Our deﬁnition inherits the vagueness of one of its deﬁning terms: problem domain. Rather than attempting to deﬁne this
volatile notion as well, we list and categorize a number of domains for which DSLs have actually been built in Section 4.
Moreover, we refer to , which contains an interesting discussion contrasting a “domain as the real world” point of view
as adopted in the artiﬁcial intelligence community, with a “domain as a set of systems” approach, as used in the systematic
software reuse research community.
DSLs are usually small, offering only a restricted suite of notations and abstractions. In the literature they are also called
micro-languages and little languages . Sometimes, however, they contain an entire general-purpose language (GPL) as a
sublanguage, thus offering domain-speciﬁc expressive power in addition to the expressive power of the GPL. This situation
occurs when DSLs are implemented as embedded languages (see Section 6). Languages such as Cobol or Fortran, which could
be viewed as languages tailored towards the domain of business and scientiﬁc programming, respectively, are generally not
regarded as DSLs, because they are not small and because their expressive power is not restricted to these domains.
Domain-speciﬁc languages are usually declarative. Consequently, they can be viewed as speciﬁcation languages, as well
as programming languages. Many DSLs are supported by a DSL compiler which generates applications from DSL programs.
In this case, the DSL compiler is referred to as application generator in the literature , and the DSL as application-speciﬁc
language. Other DSLs, such as YACC  or ASDL , are not aimed at programming (specifying) complete applications,
but rather at generating libraries or components. Also, DSLs exist for which execution consists in generating documents (T
or pictures (PIC ). A common term for DSLs geared towards building business data processing systems is 4th Generation
Related to domain-speciﬁc programming is end-user programming, which happens when end-users perform simple pro-
gramming tasks using a macro or scripting language. A typical example is spreadsheet programming using the Excel macro-
3 Risks and Opportunities
Adopting a DSL approach to software engineering involves both risks and opportunities. The well-designed DSL manages to
ﬁnd the proper balance between these two. The beneﬁts of DSLs include:
DSLs allow solutions to be expressed in the idiom and at the level of abstraction of the problem domain. Consequently,
domain experts themselves can understand, validate, modify, and often even develop DSL programs.
DSL programs are concise, self-documenting to a large extent, and can be reused for different purposes .
DSLs enhance productivity, reliability, maintainability [24, 47], and portability .
DSLs embody domain knowledge, and thus enable the conservation and reuse of this knowledge.
DSLs allow validation and optimization at the domain level [6, 13, 55].
DSLs improve testability following approaches such as .
The disadvantages of the use of a DSL are:
The costs of designing, implementing and maintaining a DSL.
The costs of education for DSL users.
The limited availability of DSLs .
The difﬁculty of ﬁnding the proper scope for a DSL.
The difﬁculty of balancing between domain-speciﬁcity and general-purpose programming language constructs.
The potential loss of efﬁciency when compared with hand-coded software.
Comparisons of the DSL approach to other approaches to software generation are made in [20, 22, 47]. In  the costs
and beneﬁts of DSLs are analyzed from the perspective of software maintenance. In , DSLs are categorized as one of the
main approaches to software reuse, and a detailed comparison is made to other reuse techniques.
4 Example DSLs
Literally hundreds of DSLs are in existence today. Of these, only a subset is actually described in the software engineering or
programming language literature. Best-known are classical examples like PIC, SCATTER, CHEM, LEX, YACC, and Make,
which are described in . Other well-known examples are SQL, BNF, and HTML. We have included references to various
example domain-speciﬁc languages. Their domains can be grouped into the following areas:
Financial products[12, 22, 24], behavior control andcoordination [9, 10], software architectures, and databases .
Description and analysis of abstract syntax trees [77, 19, 51], video device driver speciﬁcations , cache coherence
protocols , data structures in C , and operating system specialization .
Web computing [14, 35, 4, 33], image manipulation , 3D animation , and drawing .
String and tree languages for model checking , communication protocols , telecommunication switches , and
signature computing .
Simulation [2, 13], mobileagents , robotcontrol , solving partial differential equations, and digital hardware
A collection of several papers on DSLs can be found in .
5 DSL Design Methodology
The development of a domain-speciﬁc language typically involves the following steps (see [17, 24]):
(1) Identify the problem domain. (2) Gather all relevant knowledge in this domain. (3) Cluster this knowledge in a
handful of semantic notions and operations on them. (4) Design a DSL that concisely describes applications in the
(5) Construct a library that implements the semantic notions. (6) Design and implement a compiler that translates DSL
programs to a sequence of library calls.
(7) Write DSL programs for all desired applications and compile them.
The aim of the analysis steps (1) through (4) is to build up a thorough understanding of the underlying application domain.
Guidelines for acquiring such an understanding are provided by the research area of domain analysis which investigates ways
of modeling domains. Following , a domain analyst is a person who examines the needs and requirements of a collection
of systems which seem “similar”. Neighborsemphasizes that this is work that only can be done by a person who has built many
systems for different customers in the same problem area. The domain analyst is like a systems analyst, except that the goal is
to support the development of families of related systems, not just one-of-a-kind productions .
Domain engineering  refers to the activity of systematically modeling domains. Domain engineering originates from
research in the area of software reuse, and can be used when constructing domain-speciﬁc reusable libraries, frameworks or
languages. A recent domain engineering survey is provided by [20, Chapter 3]. Several domain engineering methodologies
exists, of which ODM (Organizational Domain Modeling [69, 70]), FODA (Feature-Oriented Domain Analysis ), and
DSSA (Domain-Speciﬁc Software Architectures ) are best known.
Strongly related to domain engineering is the notion of program families which are sets of similar programs [52, 18]. At
Lucent, a systematic approach to the development of families is in use, the Family-Oriented Abstraction, Speciﬁcation and
Translation (FAST) approach, which has been successfully applied to over 25 different domains . Program families are
in turn related to software product lines. These emphasize features shared by all products, and are focused on the needs of a
selected market [21, 53, 78].
A prerequisite to developing a DSL is mature domain knowledge. For that reason, a DSL is viewed as the ﬁnal and
most mature phase of the evolution of an object-oriented application framework [66, 22]. For the same reason, the existence of
legacy systems implementing domainconcepts will be of use when developinga DSL forthat domain . Reverseengineering
techniques may be used to distill domain knowledge from such legacy systems — an overview of such techniques is provided
by [16, 25].
6 DSL Implementation
The implementation steps (5) and (6) of the previous section can be carried out using several approaches:
Interpretation or compilation
This is the classical approach to implementing a new language. Standard compiler tools [1, 7] can be used, or tools dedicated
to the implementation of DSLs like Draco , ASF+SDF , Kephera , Kodiyak , design by selection , or
The main advantage of building a compiler or interpreter is that the implementation is completely tailored towards the
DSL and no concessions are necessary regarding notation, primitives and the like. Also, error detection, static analysis, and
optimizations can be done at the domain level, for example using an effect system as in .
Clearly, an important problem is the cost of building such a compiler or interpreter from scratch, and the lack of reuse from
other (DSL) implementations, although some DSL tool sets (for example InfoWiz ) are particularly designed to overcome
As an alternative to implementing a DSL from scratch, a DSL can be implemented by extending a given base language.
For instance,  describes an extension of (a restricted version of) a general-purposelanguage with domain-speciﬁc constructs.
The main advantage of this approach is that all features of the base language remain available and need not be re-implemented.
When implementing domain-speciﬁc extensions of a base language, the implementation of the base language can be reused
in three different ways:
Embedded languages / domain-speciﬁc libraries
In this approach, existing mechanisms such as deﬁnitions for functions or operators with user-deﬁned syntax are used to build
a library of domain-speciﬁc operations. The syntactic mechanisms of the base language are used to express the idiom of the
An advantage of this approach is that the compiler or interpreter of the base language is reused as is for the DSL. The
main limitation is in the expressiveness of the syntactic mechanisms in the base language. In many cases, the optimal domain-
speciﬁc notation has to be compromised to ﬁt the limitations of the base language. Typical examples of this approach are 
(a robot control language embedded in Haskell) and  (a PIC-like drawing language embedded in ML). The concept of
domain-speciﬁc embedded language was coined by Hudak .
Preprocessing or macro processing
In this approach the new constructs are translated to statements in the base language by a preprocessor. The main advantage
of this approach is simplicity. Its main disadvantage is that static checking and optimization are not done at the domain level.
Consequently, generated code is error prone, and the user is provided with feedback on these errors at the level of the base
language, or only at run-time.
Extensible compiler or interpreter
This approach is similar to the previous one, but the preprocessing phase is now integrated in the compiler. The advantage is
that more type checking and better optimization is possible. This approach is taken by [30, 74]. The Tcl  interpreter is also
a prime example: it has been extended for dozens of domains.
Apart from building a dedicatedDSL compiler or interpreter,or reusing the implementation of an underlyingbase language,
other implementation techniques may be used. For instance, in aspect-oriented programming  a DSL is used to describe an
aspect of a system’s behavior that is orthogonal to its main functionality. An aspect weaver is then used to generate domain-
speciﬁc code and merge it with the main code.
7 Concluding Remarks
This survey on domain-speciﬁc languages covered covered terminology, risks and opportunities, example DSLs, and design
and implementation issues, listing relevant references for each of these topics. The references themselves are annotated with a
summary of the most important results discussed in each paper.
For up to date information on the topic of domain-speciﬁc languages, we refer to the series of DSL conferences organized
by USENIX [64, 27], which most likely will have successors in the years to come.
Another valuable source of up to date information may be the web. A searchable domain engineering bibliography, with
abstracts, is available at http://www.iese.fhg.de/pubs_and_links/spl/bibliography/. An online bibliographyon the topic
of generative programming can be found at http://home.t-online.de/home/Ulrich.Eisenecker/gpref.htm. Finally, http:
//www.irisa.fr/compose/dsl/ provides a survey of domain-speciﬁc languages in general.
This research was sponsored by the Dutch Telematica Instituut, project DSL (see also http://www.cwi.nl/projects/
An earlier version of this paper appeared as Domain-Speciﬁc Languages: An Annotated Bibliography in ACM SIGPLAN
We would like to thank Jan Heering from CWI for many useful remarks.
 A.V. Aho, R. Sethi, and J.D. Ullman. Compiler: Principles, Techniques and Tools. Addison-Wesley, 1986.
Standard text on compiler construction.
 M. Antoniotti and A. G¨oll¨u. SHIFT and SMART-AHS: A language for hybrid system engineering modeling and simula-
tion. In Ramming , pages 171–182.
Describes the language SHIFT for hybrid system simulation. Main application area is trafﬁc simulation. Implemented by
translation to C and a run-time library with solvers for various kinds of differential equations.
 G. Arango. Domain analysis: From art form to engineering discipline. In Fifth International Workshop on Software
Speciﬁcation and Design, pages 152–159, May 1989. Appeared as ACM SIGSOFT Engineering Notes 14(3).
Outlines a framework to synthesize domain analysis methods, and to compare between different methods. The paper
advocates an incremental, evolving approach towards developing domain models.
 D. Atkins, T. Ball, G. Bruns, and K. Cox. Mawl: A domain-speciﬁclanguage for form-based services. In DSL-IEEE ,
pages 334–346. An earlier version appeared in .
Describes the language Mawl thatis intended forimplementing form-based information services fordifferent devices (web
browser, interactive voice response service). The main contributions of this language are: (1) separation of user-interface
code and service logic, (2) static type checking, (3) device-independence,(4) automatic generation of low-level CGI code,
(5) automatic generation of HTML templates, and (6) automatic generation of usage statistics.
 D. R. Barstow. Domain-speciﬁc automatic programming. IEEE Transactions on Software Engineering, SE-11(11):1321–
36, November 1985.
Envisions a framework for stepwise synthesis of domain-speciﬁc applications from informal speciﬁcations. The frame-
work applies search techniques to explore possible reﬁnements of an initial speciﬁcation, given a base of domain and
programming knowledge (facts and heuristics).
 A. Basu, M. Hayden, G. Morrisett, and T. von Eicken. A language-based approach to protocol construction. In Kamin
, pages 1–15.
Reports on the design and implementation of Promela++, a DSL for protocol construction and validation. Promela++
adds domain-speciﬁc constructs to restricted C, and supports validation and optimization on the domain-level.
 J. L. Bentley. Programming pearls: Little languages. Communications of the ACM, 29(8):711–721,August 1986.
Demonstrates and advocates the use of “little languages”. Takes PIC as an example, as well as a number of little lan-
guages from which PIC input is generated (SCATTER, CHEM), and little languages that were used to implement PIC
(LEX, YACC, Make). Contrasts three approaches: interactive systems, subroutine libraries, and little languages. Dis-
cusses DSL design principles.
 J. A. Bergstra, J. Heering, and P. Klint, editors. Algebraic Speciﬁcation. ACM Press/Addison-Wesley, 1989.
Introduces the Syntax Deﬁnition Formalism SDF, the Algebraic Speciﬁcation Formalism ASF, and their combination,
ASF+SDF, which can be used to describe the syntax and semantics of (domain-speciﬁc) languages. Contains several
language deﬁnition case studies. See also [12, 23].
 J.A. Bergstra and P. Klint. The discrete time TOOLBUS—a software coordination architecture. Science of Computer
Programming, 31:205–229, 1998.
Describes how a language based on process algebra is used in the TOOLBUS coordination architecture for building
heterogeneous, distributed software systems. See also .
 F. Bertrand and M. Augeraud. BDL: A specialized language for per-object reactive control. In DSL-IEEE , pages
347–362. An earlier version appeared in .
Many object-oriented languages contain only implicit constraints on the order of application of the methods in a class.
This paperintroduces the Behaviour Description Language (BDL) which uses a process-oriented notation to describe this
ordering. BDL is translated to C, with ESTEREL as intermediary. The resulting C code is linked with a C++ program and
acts as controller for the execution of C++ classes. See also .
 D. Bonachea, K. Fisher, A. Rogers, and F. Smith. Hancock: A language for processing very large-scale data. In DSL-99
, pages 163–176.
Describes the language Hancock that is intended for signature computations on the data collected from telephone calls.
A signature is a user proﬁle with applications ranging from fraude detection to marketing. Typical issues are the large
volume of data, the complex traversal patterns of these data and the different levels of precision for signatures. Hancock
is translated to C combined with several run-time libraries. The major beneﬁt of this DSL is a separation of concerns
(traversal patterns, efﬁciency, signature computations).As a result programmers can concentrate on the signature compu-
tation, since the other concerns are taken care of by the DSL compiler. The major reason to design a DSL (as opposed to
using a library) were the traversal patterns that cannot be captured in a library. The paper concludes with a description
of the design process used.
 M. van den Brand, A. van Deursen, P. Klint, S. Klusener, and E. van der Meulen. Industrial applications of ASF+SDF.
In M. Wirsing and M. Nivat, editors, Algebraic Methodology and Software Technology (AMAST ’96), volume 1101 of
Lecture Notes in Computer Science, pages 9–18. Springer-Verlag, 1996.
Provides an overview of some industrial applicationsof the language prototyping environmentASF+SDF. The RISLA case
study, involving a language for describing ﬁnancial products, is discussed in considerable detail, covering pure RISLA,
modular RISLA, and RISQUEST, a language for generating questionnaires used when composing new products. From a
modular RISLA product description, COBOL code is generated for accessing a library of COBOL functions providing
operations on cash ﬂows, balances, intervals, and the like. See also [22, 24].
 D. Bruce. What makes a good domain-speciﬁc language? APOSTLE, and its approach to parallel discrete event simula-
tion. In Kamin , pages 17–35.
Discusses the design of a DSL for parallel discrete event simulation. On the basis of this experience a number of observa-
tions are made regarding DSL design principles. Most notably, the use of a strong effect system is advocated to do static
checking on the domain level, and to determine applicability of optimizations.
 L. Cardelli and R. Davies. Service combinators forweb computing. In DSL-IEEE, pages309–316. An earlier version
appeared in .
Access to the resources of the World-Wide Web is usually obtained though manual browsers. Service combinators are
intended for writing programs that reproduce human browsing behaviour, including reactions to slow transmission rates
and various kinds of failure. Based on a concurrent programming model, the paper gives both an informal and formal
treatment of a DSL for Web computing.
 S. Chandra, , B. Richards, and J. R. Larus. Teapot: A domain-speciﬁc language for writing cache coherence protocols. In
DSL-IEEE , pages 317–333. An earlier version appeared in .
The problem of cache coherence occurs when local replica of shared data are made in a distributed system in order
to improve its scalability and performance. Writing the code to support coherence protocols is error-prone. This paper
describes experience with the language Teapot for describing these protocols. Teapot programs can be translated to (1)
C code that implements the protocol, or (2) input for an automatic veriﬁer. Two case studies and overall experience with
this approach are discussed.
 E.J. Chikofsky and J.H. Cross. Reverse engineeringand design recovery: A taxonomy. IEEE Software, 7(1):13–17,1990.
Overview of reverse engineering techniques, which also can be used to distill domain knowledge from legacy system. See
also [25, 70].
 J. C. Cleaveland. Building application generators. IEEE Software, pages 25–33, July 1988.
Uses the term “application generators” to refer to DSL compilers. Gives a compiler generator architecture diagram.
Describes relationships between roles of customers, domain engineers and system engineers. Lists pros and cons of
application generators. Describes “Stage”, an application-generator development tool. Describes a methodology for
building an application generator.
 J. Coplien, D. Hoffman, and D. Weiss. Commonality and variability in software engineering. IEEE Software, pages
37–45, November/December 1998.
A software family is a set of similar systems with possibly many different variations. Scope, commonality, and variability
(SCV) analysis gives software engineers a systematic way of thinking about and identifying the product family they are
creating. The paper describes the Family-Oriented Abstraction, Speciﬁcation, and Translation (FAST) approach, which
has been used with immediate payoff in over 25 domains at Lucent Technologies.
 R. F. Crew. ASTLOG: A language for examining abstract syntax trees. In Ramming , pages 229–242.
Introduces a Prolog-based query language for analyzing abstract syntax trees of C/C++ programs.
 K. Czarnecki and U. Eisenecker. Generative Programming: Methods, Techniques and Applications. Addison-Wesley,
1999. To appear.
Gives a comprehensive discussion of a range of programming techniques that involve some sort of code generation step,
such as aspect-oriented, subject-oriented, and adaptive programming, composition ﬁlters, anddomain-speciﬁc languages.
Chapter 3 of this book provides a survey of domain-engineering methods.
 J.-M. DeBaud and K. Schmid. A systematic approach to derive the scope of software product lines. In 21st International
Conference on Software Engineering, ICSE-99, pages 34–43. ACM, 1999.
Argues that economic motives should be used for scoping software product lines, rather than more traditional domain
engineering methods. The paper proposes PuLSE, which iteratively reﬁnes business objectives towards more operational
 A. van Deursen. Domain-speciﬁc languages versus object-oriented frameworks: A ﬁnancial engineering case study. In
Smalltalk and Java in Industry and Academia, STJA’97, pages 35–39. Ilmenau Technical University, 1997.
Contrasts domain-speciﬁc languages with object-oriented frameworks by comparing two projects in the ﬁnancial engi-
neering domain: RISLA (DSL) and the ET++SwapsManager (OO framework). See also .
 A. van Deursen, J. Heering, and P. Klint, editors. Language Prototyping: An Algebraic Speciﬁcation Approach, volume 5
of AMAST Series in Computing. World Scientiﬁc Publishing Co., 1996.
Describes the use of ASF+SDF as a meta-language for the speciﬁcation of syntax and semantics. After introducing
ASF+SDF, a number of language speciﬁcation case studies are presented, and various styles for writing language speci-
ﬁcations are illustrated. Moreover, different techniques for generating tools from these are presented. See also .
 A. van Deursen and P. Klint. Little languages: Little maintenance? Journal of Software Maintenance, 10:75–92, 1998.
Domain-speciﬁc languages (DSLs) have the potentialto make software maintenancesimpler: domain-expertscan directly
use the DSL to make required routine modiﬁcations. At the negative side, however, more substantial changes may become
more difﬁcult: such changes may involve altering the domain-speciﬁc language. This will require compiler technology
knowledge, which not every commercial enterprise has easily available. The paper describes and uses the experience
of the RISLA language for interest rate products to discuss the role of DSLs in software maintenance, the opportunities
introduced by using them, and techniques for controlling the risks involved. See also .
 A. van Deursen, P. Klint, and C. Verhoef. Research issues in software renovation. In J.-P. Finance, editor, Fundamental
Approachesto Software Engineering, FASE99, volume 1577 of Lecture Notes in Computer Science, pages 1–23.Springer-
Overview of parsing, transformation, and program understanding techniques that can be used when searching for domain
knowledge in legacy systems. See also [16, 70].
 T. B. Dinesh, M. Haveraaen, and J. Heering. An algebraic programming style for numerical software and its optimization.
Technical Report SEN-R9844, CWI, 1998. ACM CoRR Preprint Server cs.SE/9903002 (March 1999). Submitted to
Discusses a domain-speciﬁc programming style for the domain of partial differential equations, using an expression style
directly obtained from the underlying algebraic theory. The use of this style permits optimizations beyond the scope of
current compiler optimizations.
 Proceedings of the second USENIXConference on Domain-SpeciﬁcLanguages. USENIX Association, October 3–5 1999.
 Special issue on domain-speciﬁc languages. IEEE Transactions on Software Engineering, 25(3), May/June 1999.
 C. Elliott. An embedded modeling language approach to interactive 3D and multimedia animation. In DSL-IEEE ,
pages 291–308. An earlier version appeared in .
Describes a multi-media extension for Haskell and discusses the merits of Haskell as basis for domain-speciﬁcextensions.
 D. R. Engler. Interface compilation: Steps toward compiling program interfaces as languages. In DSL-IEEE , pages
387–400. An earlier version appeared in .
Describes the extensible ANSI C compiler framework MAGIK, which allows the dynamic incorporation of user-deﬁned
compiler extensions.The extensions can transform, optimize or inspect the generated intermediate representation.The ap-
proach gives safe access to compiler internals and supports full optimization of application-speciﬁc language extensions.
Implemented on top of lcc. See also .
 R. E. Faith, L. S. Nyland, and J. F. Prins. Khepera: A system for rapid implementation of domain speciﬁc languages. In
Ramming , pages 243–55.
Presents Khepera, a tool kit for rapid implementation and long-term maintenance of DSLs via source-to-source transfor-
mation separated into three phases: parsing, AST transformation, and pretty-printing.
 M. E. Fayad and D. C. Schmidt. Object-oriented application frameworks. Communications of the ACM, 40(10):32–38,
Introduction to a special issue on (domain-speciﬁc) object-oriented frameworks, which are deﬁned as reusable, semi-
complete applications that can be specialized to produce custom applications. Covers classiﬁcation, strengths and weak-
nesses, and future trends. See also .
 M. Fern´andez, D. Suciu, and I. Tatarinov. Declarative speciﬁcation of data-intensive web sites. In DSL-99 , pages
Covers a query language to describe data-intensive web sites. Three programming tasks are distinguished to build such
sites: accessing and integrating the data available in the site, building the site’s structure, and generating the HTML
representation of the site. The solution proposed is a declarative query language (StruQL) to deﬁne the site’s content and
structure, a template language to deﬁne the HTML representation and an extension of the query language with functions
to describe dynamic behaviour and to promote reusability of queries. Reengineering an existing AT&T web site using
this approach has resulted in less, more maintainable, code with more functionality. The initial learning curve of the new
language is more than compensated for by the advantages gained.
 M. Fromherz, V. Gupta, and V. Saraswat. cc — A generic framework for domain-speciﬁc languages. In Kamin ,
Proposes cc, a family of languages for concurrent constraint programming, as a framework for DSL construction. Two
approaches are explained by example: building a DSL on top of cc, and extending cc with domain-speciﬁc constructs.
 M. Fuchs. Domain speciﬁc languages for ad hoc distributed applications. In Ramming , pages 27–36.
The current architecture of the Web is based on a client/server model in which most of the computation is done at the
server side, while the client side is a browser that only displays the results of server computations. SGML/XML is used
as meta-language for describing the interactions between heterogeneous agents on the Web. Essentially, a grammar is
deﬁned of all possible interactions and this grammar steers the behaviour of each agent. See  for a fully process-based
approach to this problem.
 R. Gray. Agent Tcl: A transportable agent system. In J. Mayﬁeld and T. Finnin, editors, Proceedings of the CIKM Work-
shop on Intelligent Information Agents, Fourth International Conference on Information and Knowledge Management
(CIKM’95), December 1995.
Describes an extension of Tcl  for mobile agents.
 S. Z. Guyer and C. Lin. An annotation language for optimizing software libraries. In DSL-99 , pages 39–52.
A language is presented for annotating C libraries with information that is exploited by an optimizing compiler. Domain-
speciﬁc information is conveyed by annotations that in effect deﬁne (i) a dataﬂow analysis problem on the various library
procedures, and (ii) procedure specializations that are to be triggered by the outcome of the analysis. The approach aims
at giving libraries some of the compiler support enjoyed by DSLs.
 R. M. Herndon and V. A. Berzins. The realizable beneﬁts of a language prototyping language. IEEE Transactions on
Software Engineering, SE-14:803–809, 1988.
Discusses language prototyping tools (LPT) in general, as well as the speciﬁc LPT Kodiyak. Lists application areas of
LPTs and beneﬁts of applying them. Gives a brief description of Kodiyak and reports on experience with it.
 E. Horowitz, A. Kemper, and B. Narasimhan. A survey of application generators. IEEE Software, pages 40–54, January
Surveysa number of databasequery and updatelanguages, as prime examples of application generators(DSL compilers),
and hypothesizes a ‘generic’ database language. Discusses the possibilities of combining such a language with a general
purpose language. Outlines AdaRel, an extension of Ada with relational database programming constructs.
 P. Hudak. Building domain-speciﬁc embedded languages. ACM Computing Surveys, 28(4es), December 1996.
Argues that a DSL is the “ultimate abstraction”, capturing precisely the semantics ofthe applicationdomain, but also that
designing and implementing languages is difﬁcult and resists evolution. Proposes the notion of embedded DSLs, which
inherit the infrastructructure from some other language, and discusses the importance of modular monadic interpreters,
instrumentation, and partial evaluation.
 J. Jennings and E. Beuscher. Verischemelog: Verilog embedded in Scheme. In DSL-99 , pages 123–134.
Verilog, a digitalhardware design language,is extendedwith facilities for generatingand manipulatinghardware descrip-
tions by embedding it into the general purpose language Scheme. The extended language features early error detection
and high customizability.
 R. E. Johnson and B. Foote. Designing reusable classes. Journal of Object-Oriented Programming, 1(2):22–35, 1988.
Introduced the notion of object-oriented frameworks. A framework is deﬁned as a set of classes that embodies an abstract
design for solutions to a family of related problems, and supportsreuse at a larger granularity thanclasses. In a white-box
framework, application-speciﬁcbehavior is obtained via method overriding or by adding new methods to the framework’s
classes. In a black-box,support for extensibility is provided by deﬁning interfaces for componentsthat can beplugged into
the framework via object composition, thus better hiding the implementation details of the framework. See also [32, 66]
 S. Kamin, editor. DSL ’97 – First ACM SIGPLAN Workshop on Domain-Speciﬁc Languages, in Association with POPL
’97, Paris, France, January 1997. University of Illinois Computer Science Report.
 S. Kamin and D. Hyatt. A special-purpose language for picture-drawing. In Ramming , pages 297–310.
Describes FPIC, a reconstruction of the original PIC embedded in ML.
 K. C. Kang, S. G. Cohen, J. A. Hess, W. E. Novak, and A. S. Peterson. Feature-oriented domain analysis (FODA)
feasibility study. Technical Report CMU/SEI-90-TR-21, Software Engineering Institute, Carnegie Mellon University,
FODA is a domain engineering approach emphasizing feature analysis. A feature is deﬁned as a prominent, user-visible
characteristic of a software system. FODA aims at building up a feature model, consisting of a features diagram (hierar-
chical decomposition of mandatory, alternative, or optional features), feature deﬁnitions, composition rules for features,
and a rationale for features indicating the trade-offs. See also [20, 70]
 G. Kiczales, J. Irwin, J. Lamping, J.-M. Loingtier, C. Lopes, C. Maeda, and A. Mendhekar. Aspect oriented programming.
In Kamin , pages 75–88.
Presents a novel programming technique, called aspect-oriented programming (AOP). This technique consists in describ-
ing each aspect (e.g. basic functionality, communication, coordination) of a system’s behaviour in a (little) language
that allows it to be expressed in its most natural form. An aspect weaver merges these separate aspect descriptions into
a single, efﬁcient program. An important beneﬁt of AOP is that it allows high-level domain-speciﬁc programming for
performance-critical domains. See also 
 R. B. Kieburtz, L. McKinney, J. M. Bell, J. Hook, A. Kotov, J. Lewis, D. P. Oliva, T. Sheard, I. Smith, and L. Walton. A
software engineering experiment in software component generation. In Proceedings of the 18th International Conference
on Software Engineering ICSE-18, pages 542–553. IEEE, 1996.
Reports the results of an experiment in which a template-based approach and a DSL approach to software generation
were compared. Several subjects were monitored while performing a number of developmentand maintenance tasks using
alternatively template technology and DSL technology. Flexibility, productivity, reliability, and usability were measured.
The DSL approach scored better on all counts.
 N. Klarlund and M. I. Schwartzbach. A domain-speciﬁc language for regular sets of strings and trees. In DSL-IEEE ,
pages 378–386. An earlier version appeared in .
Describes design and implementation of FIDO, a language to express large ﬁnite-state automata on large alphabets.
Typical application is in veriﬁcation and model checking.
 C. W. Krueger. Software reuse. ACM Computing Surveys, 24(2):131–183,June 1992.
Categorizes, describes and compares existing approaches to software reuse, among which DSLs (or application gener-
ators). Compared to the other approaches DSLs reduce the intellectual effort required to obtain an executable system
from its speciﬁcation. Limited availability and difﬁculty of building DSLs of optimal speciﬁcity/generality are listed as
disadvantages of DSLs.
 D. A. Ladd and J. C. Ramming. Two applicationlanguages in softwareproduction. In USENIXVeryHigh LevelLanguages
Symposium Proceedings, pages 169–178, October 1994.
Describes how PRL5, an application-oriented, declarative language used to maintain the integrity of databases in the
AT&T 5ESS telecommunications switch, evolved from an earlier, imperative domain-speciﬁc language, PRL, which in
turn replaced a combination of English and C. The constraint descriptions expressed in PRL5 can be used in more than
one way, whereas a program to check constraints is useful only for performing that particular computation. A key lesson
is that domain-speciﬁc languages should not be designed to describe computation, but to express useful facts from which
one or more computations can be derived.
 D. Leijen and E. Meijer. Domain speciﬁc embedded compilers. In DSL-99 , pages 109–122.
Explains how a DSL (SQL is taken as example) can be embedded in Haskell by (i) coding an abstract syntax of the DSL
as a Haskell datatype (ii) writing a code generator in Haskell that maps the abstract syntax to the concrete syntax, and
(iii) making Haskell call an external server which compiles and executes the generated DSL code.
 F. van der Linden, editor. Development and Evolution of Software Architectures for Product Families, volume 1429 of
Lecture Notes in Computer Science. Springer-Verlag, 1998.
Proceedings of a workshop originating from the ESPRIT ARES project, which investigates software architectures for
families of embedded systems.
 R. R. Macala, L. D. Sutckey, and D. C. Gross. Managing domain-speciﬁcproduct-line development. IEEE Software, 13,
Describes recommendations and lessons learned from managing a reusability project at Boeing in the area of real-time
training systems for ﬂight crews. Product-line development separates the software-developmentprocess into two separate
life cycles: domain engineering, which aims to create reusable assets, and application engineering, which ﬁelds systems
using those assets. Lessons learned include that product-line development demands careful strategic planning, a mature
development process, and the ability to overcome organizational resistance.
 N. Medvidovic and D. S. Rosenblum. Domains of concern in software architectures and architecture description lan-
guages. In Ramming , pages 199–212.
Gives a categorization of DSLs for describing software architectures.
 V. Menon and K. Pingali. A case for source-level transformations in MATLAB. In DSL-99 , pages 53–66.
Three kindsof source-to-sourcetransformationsfor optimizingMATLAB programsare proposed and shown to be effective.
The transformations yield performance beneﬁts additional to those obtained by (optimizing) compilation, and may be
useful for other DSLs that are high-level, untyped, and interpreted.
 L. Nakatani and M. Jones. Jargons and infocentrism. In Kamin , pages 59–74.
Describesand advocatesthe development of DSLs as jargons: domain-speciﬁcextensions of a tiny common base language.
According to a newprogrammingparadigm(infocentrism)the applicationsemantics forthese jargonscanbe programmed
by providing actions for the constructs speciﬁc to the jargon only; the traversal semantics is inherited from the base
language. Because all jargons share the base syntax and semantics, it is easy to combine and reuse their deﬁnitions as
well as their tools. The InfoWiz technology which supports the development of jargons is discussed.
 L. H. Nakatani, M. A. Ardis, R. G. Olsen, and P. M. Pontrelli. Jargons for domain engineering. In DSL-99 , pages
Discusses the use of jargons (see ) in the domain of conﬁguration control.
 J. M. Neighbors. The Draco approachto constructing software from reusable components. IEEE Transactions on Software
Engineering, SE-10(5):564–74, September 1984.
The Draco approachstarts by capturing domainanalysis information in a DSL. The objectsand operations of this DSL are
reﬁned into various DSLs of lower levels of abstraction, and ﬁnally into executable languages. These reﬁnements capture
design information (implementation decisions). The Draco system supports the developmentand reuse of constellationsof
DSLs and reﬁnements. It offers tactics for reﬁnement selection as well as automatic consistency checking of the resulting
 J. K. Ousterhout. Scripting: Higher level programming for the 21st century. IEEE Computer, March 1998.
Discusses scripting languages, such as Perl, Tcl, and Visual Basic, which are designed for gluing applications, assuming
the existence of a set of components that just need to be connected together. Emphasizes that scripting languages should
be typeless and interpreted.
 J. Peterson and G. Hager. Monadic robotics. In DSL-99 , pages 95–108.
Discusses the importance of monads in the implementation of tasks in Frob (see ), which help to achieve modularity
 J. Peterson, P. Hudak, and C. Elliott. Lambda in motion: Controlling robots with Haskell. In PADL’99, volume 1551 of
LNCS, pages 91–105, 1999.
Describes two domain-speciﬁc extensions of Haskell: Frob a language for robot control and Fran a language for reactive
 P. Pfahler and U. Kastens. Language design and implementation by selection. In Kamin , pages 97–108.
A language design system is presented which allows a user to design a DSL by selecting language features from menus.
After selection, an implementation of the DSL can be generated. The system relies on domain designers to provide a
deﬁnition of the design space, as well as speciﬁcation components for all possible language features.
 C. Pu, A. Black, C. Cowan, J. Walpole, and C. Consel. Microlanguages for operating system specialization. In Kamin
, pages 49–57.
Discusses the use of DSLs in the domain of operating system specialization. A high-level DSL is envisioned to describe
application behavior, which will be compiled into a low-level DSL describing customized operating system behavior.
 J. C. Ramming, editor. Proceedings of the USENIX Conference on Domain-Speciﬁc Languages, Berkeley, CA, Octo-
ber 15–17 1997. USENIX Association.
 J. Reichwein, G. Rothermel, and M. Burnett. Slicing spreadsheets: An integrated methodology for spreadsheet testing
and debugging. In DSL-99 , pages 25–38.
Building on techniques for dynamic program slicing and program dicing, a fault localization technique for incremental
spreadsheet debugging is developed. Using various kinds of visual clues, the technique is integrated into a spreadsheet
 D. Roberts and R. Johnson. Evolve frameworks into domain-speciﬁc languages. In 3rd International Conference on
Pattern Languages, Allerton Park, Ill., September 1996.
Discusses 9 stages of framework development. An object-oriented framework evolves gradually, starting from three ex-
amples, moving via a white-box framework, component library, pluggable objects, to a black-box framework. The ﬁnal,
and most mature, stage is when the domain knowledge is sufﬁciently stable to merit the development of a domain-speciﬁc
language or visual builder to access the framework.
 P.H. Salus, editor. Little Languages, volume III of Handbook of Programming Languages. MacMillan, 1998.
This bookcontainsa collection ofmostly reprints and only a few original papers describing DSLs. It contains, for instance,
papers like Little Languages (Bentley ), A system for typesettting mathematics: EQN (Kernighan and Cherry), and
an overview of the Documenter’s Workbench (Akkerhuis) covering TROFF and several DSLs for describing graphics,
chemical formulae, and the like. Other chapters cover AWK, SED, SQL, TCL/TK, PERL and PYTHON. The most original
papers are a survey of DSLs and domain-speciﬁc extension languages by Hudak and an elaborate description of Little
Music Languages by Langston.
 T. Sheard, Z. Benaissa, and E. Pasalic. DSL implementation using staging and monads. In DSL-99 , pages 81–94.
Discusses how the use of staging (separating compile-time computations from run time ones) and monads (for capturing
effects and actions of the target code) lead to a simple, reusable, controlable, and correct DSL methodology.
 M. Simos. Organization domainmodeling (ODM): Formalizingthe core domainmodelinglife cycle. In M.Samadzeh and
M. Zand, editors, Proceedings of the Symposium on Software Reusability SSR’95, pages 196–205, August 1995. ACM
Software Engineering Notes.
Summarizes the key elements of the ODM domain engineering methodology. The full description is given in .
 M. Simos, D. Creps, C. Klinger, L. Levine, and D. Allemang. Organization domain modelling (ODM) guidebook version
2.0. Technical Report STARS-VC-A025/001/00, Synquiry Technologies, Inc, 1996.
A comprehensive description of the ODM approach to domain engineering. The three main ODM steps are: (1) plan the
domain, selecting objectives, stakeholders, and a set of boundary decisions to scope the domain. (2) model the domain,
building a domain lexicon, and describing the concepts and features, as well as their commonalities and variabilities. (3)
(optional) engineer an asset base of components by combining features and customers in novel ways. ODM emphasizes
existing (legacy) software systems as valuable sources of domain knowledge. It takes the “domain as a set of systems”
point of view, rather than the “domain as the real world” viewpoint.
 E. G. Sirer and B. N. Bershad. Using production grammars in software testing. In DSL-99 , pages 1–14.
Describes lava, a DSL for specifying production grammars. These are used to generate sentences over a language, for the
purpose of testing tools implementing that language. Experience with lava demonstrates that a special purpose language
for production grammars can bring high coverage, simplicity, manageability, and structure to the testing effort. Observe
that the production grammar approach can also be used to for testing DSL-tools.
 Y. Smaragdakis and D. Batory. DiSTiL: A transformation library for data structures. In Ramming , pages 257–270.
Describes DiSTiL, a DSL for describing container data structures in C, implemented on top of MicroSoft’s Intentional
Programming (IP) system.
 D. E. Stevenson and M. M. Fleck. Programming language support for digitized images or, the monsters in the closet. In
Ramming , pages 271–284.
Describes the image manipulation language Envision, implemented as an extension of Scheme.
 J. M. Stichnoth and T. Gross. Code composition as an implementation language for compilers. In Ramming , pages
Describes the ANSI C compiler framework Catacomb that supports code composition. By providing user-deﬁned code
templates (describing new language constructs such as parallel array assignment) and a ﬁxed code composition mecha-
nism inside the compiler, new constructs can be implemented in the same way as standard ones. See also .
 R. N. Taylor, W. Tracz, and L. Coglianese. Software development using domain-speciﬁc software architectures. ACM
SIGSOFT Software Engineering Notes, 20(5):27–37, 1995.
Provides the material used for a course on DSSA, Domain-Speciﬁc Software Architectures, which aims at the reduction
in time and cost of producing speciﬁc application systems within a supported domain. The paper covers key examples,
architecture representation formalisms, domain engineering, and the DSSA process. See also [20, 70]
 S. A. Thibault, R. Marlet, and C. Consel. Domain-speciﬁc languages: From design to implementationapplication to video
device drivers generation. In DSL-IEEE , pages 363–377. An earlier version appeared in .
A video card stores and displays images on a computer display. Each card is programmed by similar, but highly vendor-
speciﬁc, instructions. The authors exploit this similarity by designing a DSL for specifying drivers for video cards in the
context of the XFree86 implementation of X windows. This Graphic Adaptor Language is implemented in two stages: a C
library provides a low level abstract machine that is used by an interpreter for the DSL. The Tempo partial evaluator for
C is used to eliminate the overhead of interpretation and of the generality of the abstract machine. Includes a discussion
of the merits of the DSL approach in this domain.
 D. C. Wang, A. W. Appel, J. L. Korn, and C. S. Serra. The Zephyr abstract syntax description language. In Ramming
, pages 213–28.
Presents the Abstract Syntax Description Language (ASDL). Reports the implementation of a tool that converts ASDL
descriptions into C, C++, Java, or ML code. The generated code deﬁnes data-structures corresponding to this abstract
syntax as well as functions for reading and writing abstract terms to a standard ﬂattened representation. ASDL has been
used to respecify the compiler intermediate format SUIF.
 J. Withey. Investment analysis of software assets for product lines. Technical Report CMU/SEI-96-TR-010, Software
Engineering Institute, 1996.
Presents a model for analyzing the expected beneﬁts from investing in domain-speciﬁc software product lines. One of the
key concepts is economy of scope, which is a condition where fewer inputs (such as effort and time) are needed to produce
a greater variety of outputs. By contrast, economy of scale is achieved where fewer inputs are needed to produce greater
quantities of a single output.