Making a DSL in Ruby

Posted on | 520 words | ~3mins

The main goal of writing such thing as a DSL is ussualy helping yourself to not repeat the same code over and over (keeping code DRY). Tradeoff for this approach is ussualy a piece of unreadable and complex code with simple purpose: To handle this DSL and translate it to Ruby language. Lets take a quick example:

rule :apache do enable 80 enable 8080 enable 443 name ’example.com' directory ‘/var/www/example.com’ update 120 end

This DSL is self describing. It creates a configuration rules for Apache. However this example also introduce some basic mistakes which people ussualy do when designing DSL. First of all, the original idea of writing such thing is to keep code DRY. Now look again on example above. The word ’enable’ is used there three times, which definitely isn’t DRY. The proper syntax should be:

rule :apache do enable 80, 8080, 443 end

Now, the ugly next thing on the DSL sample above is that it’s not descriptive enough. If you’re spending the time on writing code that will handle example above, lets take few extra minutes and think about the right methods names. For example the property ’name’ says nothing about its suppose. Perhaps it’s name of the rule or the domain name. This automatically force further implementors to look inside documentation, which is ‘wrong’. Lets refactor this example once again:

rule :apache do enable 80, 8080, 443 domain ’example.com’ root_path ‘/var/www/example.com’ update_interval 120 end

The last nasty nit of our DSL example is the fact that reader (you ;-) have no clue about what the value of ‘update_interval’ is set to. Is the ‘120’ value specified in minutes? Or perhaps in seconds? This sort of questions are ussualy wrong and confuse people. How about using something like this:

rule :apache do update_every 2.minutes end

Since every class in Ruby is open, it’s easy enough to add methods like ‘.minutes’ or ‘.seconds’ to Integer class.

Now, when we’re happy with the architecture of our DSL, lets implement it in Ruby. As you may guess, when there is word ‘do’ in Ruby it ussualy means a ‘block’. You should get familiar with this Ruby feature ASAP, otherwise the code below will look cryptic for you :-)

First we’re going to implement the ‘Rule’ class. It’s a good practise to use class name same as the root of DSL.

class Rule attr_accestor :name, :ports, :domain, :root_path, :update_interval

def initialize(name, &block) @name = name instance_eval(&block) if block_given? end

def enable(*ports) @ports = ports.map end

def domain(name) @name = name end

def root_path(name) @root_path = name end

def update_interval(interval) @update_interval = interval end

end

Now, if you look on the code above, the first thing that will hurt your eyes is that methods for setting variables is not really DRY. We’re repeating the same pattern in setting instance variables. Now, lets refactor this code a little bit:

class Rule attr_accestor :name, :ports (@@config_attrs = [ :domain, :root_path, update_interval ]).each do |a| attr_accessor a end

def initialize(name, &block) @name = name instance_eval(&block) if block_given? end

def enable(*ports) @ports = ports.map end

def method_missing(name, *args) if name.member_of(@@config_attrs) send(name+"=", args.first) end end

end