Under the hood how named_scope works

Following code was tested with ruby 1.8.7 and Rails 2.x .

Rails recently added named_scope feature and it is a wonderful thing. If you don't know what named_scope is then you can find out more about it here .

This article is not about how to use named_scope. This article is about how named_scope does what it does so well.

Understanding with_scope

ActiveRecord has something called with_scope which is not associated with named_scope. The two are entirely separate thing. However named_scope relies on the workings on with_scope to do its magic. So in order to understand how named_scope works first let's try to understand what with_scope is.

with_scope let's you add scope to a model in a very extensible manner.

def self.all_male
  with_scope(:find => {:conditions => "gender = 'm'"}) do
    all_active
  end
end

def self.all_active
  with_scope(:find => {:conditions => "status = 'active'"}) do
    find(:first)
  end
end

# User.all_active
# SELECT * FROM "users" WHERE (status = 'active') LIMIT 1

# User.all_male
# SELECT * FROM "users" WHERE ((gender = 'm') AND (status = 'active')) LIMIT 1

We can see that when User.all_male is called, it internally calls all_active method and the final sql has both the conditions.

with_scope allows nesting and all the conditions nested together are used to form one single query. And named_scope uses this feature of with_scope to form one single query from a lot of named scopes.

Writing our own named_scope called mynamed_scope

The best way to learn named_scope is by implementing the functionality of named_scope ourselves. We will build this functionality incrementally. To avoid any confusion we will call our implementation mynamed_scope.

To keep it simple in the first iteration we will not support any lambda operation. We will support simple conditions feature. Here is a usage of mynamed_scope .

class User < ActiveRecord::Base
  mynamed_scope :active, :conditions => {:status =>  'active'}
  mynamed_scope :male, :conditions => {:gender => 'm'}
end

We expect following queries to provide right result.

User.active
User.male
User.active.male
User.male.active

Let's implement mynamed_scope

At the top of user.rb add the following lines of code

module ActiveRecord
  module MynamedScope
    def self.included(base)
      base.extend ClassMethods
    end

    module ClassMethods
      def mynamed_scope(name,options = {})
        puts "name is #{name}"
      end
    end

  end
end
ActiveRecord::Base.send(:include, ActiveRecord::MynamedScope)

Now in script/console if we do User then the code will not blow up.

Next we need to implement functionalities so that mynamed_scope creates class methods like active and male.

What we need is a class where each mynamed_scope could be stored. If 7 mynamed_scopes are defined on User then we should have a way to get reference to all those mynamed_scopes. We are going to add class level attribute myscopes which will store all the mynamed_scopes defined for that class.

def myscopes
  read_inheritable_attribute(:myscopes) || write_inheritable_attribute(:myscopes, {})
end

This discussion is going to be tricky.

We are storing all mynamed_scope information in a variable called myscopes. This will contain all the mynamed_scopes defined on User.

However we need one more way to track the scoping. When we are executing User.active then the active mynamed_scope should be invoked on the User. However when we perform User.male.active then the mynamed_scope active should be performed in the scope of User.male and not directly on User.

This is really crucial. Let's try one more time. In the case of User.active the condition that was supplied while defining the mynamed_scope active should be acted on User directly. However in the case of User.male.active the condition that was supplied while defining mynamed_scope active should be applied on the scope that was returned by User.male .

So we need a class which will store proxy_scope and proxy_options.

class Scope
  attr_reader :proxy_scope, :proxy_options
  def initialize(proxy_scope, options)
    @proxy_scope, @proxy_options = proxy_scope, options
  end
end # end of class Scope

Now the question is when do we create an instance of Scope class. The instance must be created at run time. When we execute User.male.active, until the run time we don't know the scope object active has to work upon. It means that User.male should return a scope and on that scope active will work upon.

So for User.male the proxy_scope is the User class. But for User.male.active, mynamed_scope 'active' gets (User.male) as the proxy_scope.

Also notice that proxy_scope happens to be the value of self.

Base on all that information we can now write the implementation of mynamed_scope like this.

def mynamed_scope(name,options = {})
  name = name.to_sym
  myscopes[name] = lambda { |proxy_scope| Scope.new(proxy_scope,options) }

  (class << self; self end).instance_eval do
    define_method name do
      myscopes[name].call(self)
    end
  end
end

At this point of time the overall code looks like this.

module ActiveRecord
  module MynamedScope
    def self.included(base)
      base.extend ClassMethods
    end

    module ClassMethods

      def myscopes
        read_inheritable_attribute(:myscopes) || write_inheritable_attribute(:myscopes, {})
      end

      def mynamed_scope(name,options = {})
        name = name.to_sym
        myscopes[name] = lambda { |proxy_scope| Scope.new(proxy_scope,options) }

        (class << self; self end).instance_eval do
          define_method name do
            myscopes[name].call(self)
          end
        end
      end

      class Scope
        attr_reader :proxy_scope, :proxy_options
        def initialize(proxy_scope, options)
          @proxy_scope, @proxy_options = proxy_scope, options
        end
      end # end of class Scope

    end # end of module ClassMethods

  end # endof module MynamedScope
end
ActiveRecord::Base.send(:include, ActiveRecord::MynamedScope)


class User < ActiveRecord::Base
  mynamed_scope :active, :conditions => {:status =>  'active'}
  mynamed_scope :male, :conditions => {:gender => 'm'}
end

On script/console

>> User.active.inspect
  SQL (0.000549)    SELECT name
 FROM sqlite_master
 WHERE type = 'table' AND NOT name = 'sqlite_sequence'

=> "#<ActiveRecord::MynamedScope::ClassMethods::Scope:0x203201c @proxy_scope=User(id: integer, gender: string, status: string, created_at: datetime, updated_at: datetime), @proxy_options={:conditions=>{:status=>"active"}}>"
>>

What we get is an instance of Scope. What we need is a way to call sql statement at this point of time.

But calling sql can be tricky. Remember each scope has a reference to the proxy_scope before it. This is the way all the scopes are chained together.

What we need to do is to start walking through the scope graph and if the previous proxy_scope is an instance of scope then add the condition from the scope to with_scope and then go to the previous proxy_scope. Keep walking and keep nesting the with_scope condition until we find the end of chain when proxy_scope will NOT be an instance of Scope but it will be a sub class of ActiveRecord::Base.

One way of finding if it is an scope or not is to see if it responds to find(:all). If the proxy_scope does not respond to find(:all) then keep going back because in the end User will be able to respond to find(:all) method.

# all these two methods to Scope class
def inspect
  load_found
end

def load_found
  find(:all)
end

Now in script/console you will get undefined method find. That is because find is not implemented by Scope.

Let's implement method_missing.

def method_missing(method, *args, &block)
  if proxy_scope.myscopes.include?(method)
    proxy_scope.myscopes[method].call(self)
  else
    with_scope :find => proxy_options do
      proxy_scope.send(method,*args)
    end
  end
end

Statement User.active.male invokes method 'male' and since method 'male' is not implemented by Scope, we don't want to call proxy_scope yet since this method 'male' might be a mynamed_scope. Hence in the above code a check is done to see if the method that is missing is a declared mynamed_scope or not. If it is not a mynamed_scope then the call is sent to proxy_scope for execution. Pay attention to with_scope. Because of this with_scope all calls to proxy_scope are nested.

However Scope class doesn't implement with_scope method. However the first proxy_scope ,which will be User in our case, implements with_scope method. So we can delegate with_scope method to proxy_scope like this.

delegate :with_scope,  :to => :proxy_scope

At this point of time the code looks like this

module ActiveRecord
  module MynamedScope
    def self.included(base)
      base.extend ClassMethods
    end

    module ClassMethods

      def myscopes
        read_inheritable_attribute(:myscopes) || write_inheritable_attribute(:myscopes, {})
      end

      def mynamed_scope(name,options = {})
        name = name.to_sym
        myscopes[name] = lambda { |proxy_scope| Scope.new(proxy_scope,options) }

        (class << self; self end).instance_eval do
          define_method name do
            myscopes[name].call(self)
          end
        end
      end

      class Scope
        attr_reader :proxy_scope, :proxy_options
        delegate :with_scope,  :to => :proxy_scope
        def initialize(proxy_scope, options)
          @proxy_scope, @proxy_options = proxy_scope, options
        end

        def inspect
          load_found
        end

        def load_found
          find(:all)
        end

        def method_missing(method, *args, &block)
          if proxy_scope.myscopes.include?(method)
            proxy_scope.myscopes[method].call(self)
          else
            with_scope :find => proxy_options do
              proxy_scope.send(method,*args)
            end
          end
        end

      end # end of class Scope

    end # end of module ClassMethods

  end # endof module MynamedScope
end
ActiveRecord::Base.send(:include, ActiveRecord::MynamedScope)


class User < ActiveRecord::Base
  mynamed_scope :active, :conditions => {:status =>  'active'}
  mynamed_scope :male, :conditions => {:gender => 'm'}
end

Let's checkout the result in script/console

>> User.active
SELECT * FROM "users" WHERE ("users"."status" = 'active')
>> User.male
SELECT * FROM "users" WHERE ("users"."gender" = 'm')
>> User.active.male
SELECT * FROM "users" WHERE (("users"."gender" = 'm') AND ("users"."status" = 'active'))
>> User.male.active
SELECT * FROM "users" WHERE (("users"."status" = 'active') AND ("users"."gender" = 'm'))

# you can also see count
>> User.active.count
SELECT count(*) AS count_all FROM "users" WHERE ("users"."status" = 'active')
=> 2

named_scope supports a lot more things than what we have shown. named_scope supports passing lambda instead of conditions and it also supports joins and extensions.

However in the process of building mynamed_scope we got to see the workings of the named_scope implementation.