Beyond Theology: On the Design of Domain-Specific Modeling Languages

Bran Selic
 President, Malina Software Corp. CANADA

Abstract

As a technical discipline in its infancy, lacking the support of either
established know-how or theoretical foundations, software engineering is
often characterized by acrimonious and emotional debates on the merits and
demerits of this or that technical approach or method. This is not without
value, in the sense that opposition to an idea can be used constructively as
a stimulus for its improvement. However, on the negative side, this often
leads to polarization, such that ideas may be either accepted totally and
uncritically or, alternatively, totally and ruthlessly rejected. A
particularly fallow debate of this type which has raged in recent years
pertains to the "right" way of designing domain-specific modeling languages
(DSMLs).

In this talk, we attempt to separate the emotional issues from the technical
ones and examine the topic from an objective perspective - inasmuch as this
is feasible by someone who is a contemporary of and participant in the
debate. The intent is to approach the issue constructively rather than
antagonistically. To that end, we first examine a fuller spectrum of
requirements for DSMLs; requirements that extend beyond just the obvious
technical aspects. This helps us generate a set of criteria that can be used
in evaluating which approach to use in which circumstances. Next, we examine
the two seemingly opposing methods, as well as hybrid combinations of the
two, and discuss their strong and weak points. In particular, we examine
some of the specifics of the much-maligned profile mechanism defined by the
OMG for their MOF-based languages. Finally, we outline a proposal for the
study of a theory of DSML design.