General purpose languages should be metalanguages

In his paper, The Next 700 Programming Languages, the late Landin writes that "most programming languages are partly a way of expressing things in terms of other things and partly a basic set of given things." Landin tries to separate the general purpose aspects of a language from the problem specific aspects. Instead of hundreds of languages with ad-hoc differences, Landin proposes one general-purpose language with many different sets of primitive operations for different problem domains.

However, achieving this separation is not so easy: the domain specific parts cannot always be neatly packaged up in a set of primitives, but instead may require complex syntactic and semantics structures. In the last decade or so we made significant steps towards this vision through the development domain-specific embedded languages (DSELs). We discovered how to trick several general purpose languages into making software libraries look like domain-specific languages. As a result, we now develop DSELs that enjoy many of the benefits of a real languages: they perform domain-specific type checking, they provide custom syntactic abstractions, and they optimize themselves to achieve high efficiency. Most importantly, unlike real languages, DSELs nicely inter-operate with each other through the underlying general purpose language.

The main limitation of today's DSELs is that they leak: the underlying general purpose language peeks through at inopportune times. This is not surprising; these general purpose languages were not designed to be metalanguages for building DSELs. In this talk I argue that general purpose languages should be designed to be metalanguages and I identify some features that could help our general purpose languages become metalanguages.