LinQ is often understood in terms of introducing a Domain Specific Language to work with data to C# and .Net in general. The fact is:it is not, and there is a considerable difference between LinQ syntax nature and a DSL. The problem is that DSL definition is blur enough to take anything interesting or cool under it!
Back to the beginning. When faced to doing software for a business that has some moving parts, a good practice is to make a generalization or an abstraction of the fixed part and express what is moving through some sort of configurations. That is in some respect what most of enterprise frameworks do when implementing the fixed part and letting you nicely only express what varies for your context. This way you are almost only concerned about your part since the framework itself is supposed to be steady and well abstracted and as a result implicit in your context. This experience hasn’t been nice enough when XML used to be involved. As someone that defends multi-paradigm design I could argue that XML can play a best fit in some contexts but this is out of the scope of this post. XML charset is not optimal for a lot of cases, so why not creating our own charset, and here comes DSLs.
Of course here I am only talking about the rise of DSLs in the enterprise and not the DSLs that FP been doing for ages, and I’ll get back to this one later.
Now using my favorite meta magical super dynamic language, I’d like to construct my customized embedded(or internal) domain specific language, and I’ll make my best to be English like/business like/ host language unlike to be really specific. And that is where my brains blocks! How can I get grasp of that thing, that ‘lil’ language? It might be a personal problem I have because of the nature of my profession as a consultant: I switch contexts often, and… well I can only hope that it is 1:designed, 2: designed considering practices, rules and well established… did I say patterns?
P for Patterns :
Another way of providing a solution to the problem of generality/variability is use a catalogue of non-formal generalization of recurring solutions to recurrent problems that provide a rich knowledge base of experience about tackling the kind of problem. Being of non-formal nature means that patterns aren’t provided as a framework but rather as a set of vocabulary and description that has its power in being adaptive and flexible. In my own experience, what I found particularly interesting when working with code that uses patterns is the way they communicate intent. If I find “Strategy” postfix I start searching for the other parts of the patterns and then start building on that. Providing this shared vocabulary helps me getting operational quite fast and help me scan code faster in a more “visual” way.
The good thing also about patterns is that I am still playing under the rules of my preferred hosting language. I know very well doing two very important things with it: combining and abstraction. A programming language is a set primitives together with a way to combine them and a mechanism for abstraction. What about my lil DSL? well it depends… Did I make sure to break any host language logic in there? Did I fight the intuitive nice “.”? Or did I decide that to implement my DSL for scratch? A programming language without means of abstraction results inevitably in long flat copy/paste configurations files and please don’t argue with me about their maintainability. What is the problem? The problem is that I am not a programming language designer, and to create a programming language no matter how simple and small it is, you need to have programming language design skills, so do you?
Abstraction and Communication for People:
Something that really drew my attention in functional programming languages like Haskell is their solid formal abstractions used for expressing computational patterns. These abstractions are packaged into a framework, and have a clear identity defined by a set of rules and are useable whenever semantics match. Take monads for example. It doesn’t matter if you are working with state, a database, the web, the screen or a robitic arm: you just use the do notation. The do notation in Haskell is a convenient way of working with a sound abstraction that is THE MONAD. And all what you need to do for your framework is to verify that you satisfy the Monad laws and then you can safely use that syntax for it. This sharing of well identified semantics of the abstraction help developers reuse a well established mental model. LinQ syntax is the same, the dot “.” operator is the same, F#’s “|>” operator is the same and examples are a lot. And I tend to prefer these kind of abstractions as they don’t sacrifice readability for developers for “readability” for users. I guess they have both the communicative property of Design Patterns yet they are formal and are offered as a concrete framework. Then of course you can name them DSLs, if you insist :) .
Now, if you are from the functional programming camp, and you have been doing embedded domain specific language using Lisp or Haskell for long time, then I apologize for wasting your time. I know you guys know best about doing abstractions, and you have lessons to tell if we’d only listen. And for those interested in some examples in EDSLs from the functional world, Paul Hudak goes through the design of a language for reactive animation in his book, Lennart Augustsson discussed EDSLs in Haskell on his blog.
Conclusion
Though an attractive option of abstraction, DSLs trade off readability for developers for “potential” readability of users. Haskell monads, F#’s “|>” operator and LinQ are examples of sound formal abstraction that can be offered as frameworks (code). These abstractions can save developers time being in a form of a recurring pattern that can be used in different contexts that match semantically, thus reusing developers’ understanding of the abstraction’s mental model.
