Textual Domain-Specific language (DSL) development with Microsoft Visual Studio

Question

I did some searches on developing a DSL in visual studio. At the beginning, I found out there is a Visualization and Modeling SDK for VS2010. It has a tool called DSL tool. But it seems that it is only for graphical DSL development.

Then I saw some posts saying "Oslo" is a tool for developing textual DSL, which "was" a Microsoft product - Microsoft no longer supports the tool. http://blogs.msdn.com/b/modelcitizen/archive/2010/09/22/update-on-sql-server-modeling-ctp-repository-modeling-services-quot-quadrant-quot-and-quot-m-quot.aspx

Therefore, I would like to know if I want to develop a textual DSL, what tool is the best? What do you think if I implement a DSL parser making use of F# powerpack with FSLex and FSYacc?

Answer 1

I am currently developing several external text-based DSLs using FsLex/FsYacc. I was using a hand parser, but I find the FsLex/FsYacc much easier to maintain in the design stage.

FsLex/FsYacc are not as sophisticated as ANTLR, but since most DSLs are fairly simple, FsLex/FsYacc are a perfectly sound choice for use within Visual Studio. And keeping DSLs simple is a good thing, since they are intended to be restricted and simple to learn.

I find Martin Fowler's book to be a good resource, less for the examples and details than as an encyclopedia of DSL ideas. His discussion of useability and other design aspects of DSLs is also worth reading. As Toumas indicated, it does not cover either F# or functional languages. Mr. Fowler writes that he lacked the experience in those subjects to bring the book to market in a timely way.

Having praised FsLex/FsYacc, I do still wish someone would write a good ANTLR back-end for F#. :)

-Neil

Answer 2

I am a fan of embedded DSLs, a la

http://lorgonblog.wordpress.com/2010/04/15/using-vs2010-to-edit-f-source-code-and-a-little-logo-edsl/

http://lorgonblog.wordpress.com/2010/04/16/fun-with-turtle-graphics-in-f/

where you just use leverage F# syntax with some good function names and possibly other syntax cleverness (lists, workflows, ...) to get code that "looks like maybe it is another language" but is actually just F#.

But yes, for external DSLs, you just need a grammar/parser/etc tool chain, and either FsLex/FsYacc, or maybe ANTLR or FParsec are various choices. (I don't have enough experience with any of these to know trade-offs among them.)

Answer 3

Since having made my earlier post, I have also bought and read parts of Terence Parr’s book “Language Implementation Patterns.” It is excellent, though quite a bit more technical than Martin Fowler’s book (with some additional material it could be a “Dragon Book” for the new millennium). The examples are strongly based in Java and ANTLR, but the text is the main thing, so the book is useful regardless of one’s language development environment.

Interestingly, there is little overlap between the two books. Martin Fowler’s book does a good job of covering the design and implementation of basic DSLs, such as those used for specification and configuration, while Terence Parr’s book is more technical and covers the realm extending all the way up through more sophisticated languages and byte-code machines. I recommend both if you can budget for them, otherwise, either is an excellent choice within its given domain.

Answer 4

Martin Fowler has a new book about DSL:s. Sadly, it won't discuss much about Microsoft's tooling nor functional languages.

Microsoft no longer support the graphical tool "Quadrant", but MGrammar is still supported and integrated to SQL server, right? MGrammar is the "DSL making language".

Still, I would say that functional languages (read: F#) are the way to go.

This book has a simple example of how to make a DSL with F#: http://www.manning.com/petricek/ and also Google finds many other good references about this topic.

Answer 5

Try MBase, but it only worth using if your DSL is complicated enough to require an efficient compiler and a PEG grammar. Otherwise FsYacc is more than enough.

Answer 6

Our DMS Software Reengineering Toolkit is designed to handle arbitrary DSLs (I happen to be the architect).

Most people think if you have a parser you have enough, and it is technically true, in the same sense that if you have transistors you can build a computer.

In my experience you want a lot more than just a parser: you need ways to build symbol tables so that your generator knows what the meaning of a particular identifier is, means to analyze the specification, ways to easily encode your translation and to apply optimizations to generated results.

DMS provides all these capabilities to support building DSLs. And in that sense, it goes much beyond F#.