I am a reviewer for Java One. I have about 350 project proposals to plow through and not enough time to give each of them justice.
One submitter proposed a talk on "Doing your own language". The outline talked about parsing, abstract syntax trees, and generating code. I flippantly commented "A DSL is not a DYOL. You use a DSL precisely because you DON'T want to write another parser."
Of course, what I had in mind was the recent trend of embedding DSLs in programming languages such as Groovy and Scala. (See this blog for Groovy.)
The submitter happened to be a reviewer in another Java One track, so he was able to see my comment, and he was not happy. He emailed me: "I feel strongly that this is mistaken. Most DSL's originate with domain experts, who become weary of using general-purpose languages to work within their domain concepts and patterns. Many of us are the beneficiaries of such efforts, but DSL's don't spring forth from nothingness."
Now here is my thinking. Technically speaking, the submitter is right. A DSL is simply a domain-specific language, and the term does not imply an implementation strategy. You can use a parser and code generator, as was customary in the past, or you can embed the language in a larger host language, by rigging the metaobject protocol, doing tricks with closures and operator overloading, or whatever.
I just think that the embedded approach should be the approach of choice when it is at all feasible. After all, the toolbuilders are busy giving us tools for JRuby, Groovy, and Scala. If your DSL is contained in one of these languages, you get to leverage those tools. If, on the other hand, you start a new language from scratch, you might get a marginally prettier syntax, but you now have to build up a whole tool chain.
Consider JavaFX Script. (Doesn't that just roll off your tongue?) It has a bunch of nice features for building GUIs, such as the bind and dur operators. But it is yet another programming language. People need to learn it, they need to build tools for it, they need to learn how to use those tools. Sadiya Hameed, one of my graduate students, is implementing the key features of JavaFX Script in Scala and Groovy. The syntax is a bit clunkier, but not by much. Either of them would have made a fine host for a FX DSL.
What do you think? Am I being naive? Is it just too awkward to shoehorn DSL syntax into the Procrustean bed of a host language? Is it crazy to have a program with ten mutually incomprehensible DSLs in Scala or Groovy? Or has the time come to say "no" to building more parsers?