h2. Rails Serializers

This guide describes how to use Active Model serializers to build non-trivial JSON services in Rails. By reading this guide, you will learn:

* When to use the built-in Active Model serialization
* When to use a custom serializer for your models
* How to use serializers to encapsulate authorization concerns
* How to create serializer templates to describe the application-wide structure of your serialized JSON
* How to build resources not backed by a single database table for use with JSON services

This guide covers an intermediate topic and assumes familiarity with Rails conventions. It is suitable for applications that expose a
JSON API that may return different results based on the authorization status of the user.

endprologue.

h3. Serialization

By default, Active Record objects can serialize themselves into JSON by using the `to_json` method. This method takes a series of additional
parameter to control which properties and associations Rails should include in the serialized output.

When building a web application that uses JavaScript to retrieve JSON data from the server, this mechanism has historically been the primary
way that Rails developers prepared their responses. This works great for simple cases, as the logic for serializing an Active Record object
is neatly encapsulated in Active Record itself.

However, this solution quickly falls apart in the face of serialization requirements based on authorization. For instance, a web service
may choose to expose additional information about a resource only if the user is entitled to access it. In addition, a JavaScript front-end
may want information that is not neatly described in terms of serializing a single Active Record object, or in a different format than.

In addition, neither the controller nor the model seems like the correct place for logic that describes how to serialize an model object
*for the current user*.

Serializers solve these problems by encapsulating serialization in an object designed for this purpose. If the default +to_json+ semantics,
with at most a few configuration options serve your needs, by all means continue to use the built-in +to_json+. If you find yourself doing
hash-driven-development in your controllers, juggling authorization logic and other concerns, serializers are for you!

h3. The Most Basic Serializer

A basic serializer is a simple Ruby object named after the model class it is serializing.

<ruby>
class PostSerializer
  def initialize(post, scope)
    @post, @scope = post, scope
  end

  def as_json
    { post: { title: @post.name, body: @post.body } }
  end
end
</ruby>

A serializer is initialized with two parameters: the model object it should serialize and an authorization scope. By default, the
authorization scope is the current user (+current_user+) but you can use a different object if you want. The serializer also
implements an +as_json+ method, which returns a Hash that will be sent to the JSON encoder.

Rails will transparently use your serializer when you use +render :json+ in your controller.

<ruby>
class PostsController < ApplicationController
  def show
    @post = Post.find(params[:id])
    render json: @post
  end
end
</ruby>

Because +respond_with+ uses +render :json+ under the hood for JSON requests, Rails will automatically use your serializer when
you use +respond_with+ as well.

h4. +serializable_hash+

In general, you will want to implement +serializable_hash+ and +as_json+ to allow serializers to embed associated content
directly. The easiest way to implement these two methods is to have +as_json+ call +serializable_hash+ and insert the root.

<ruby>
class PostSerializer
  def initialize(post, scope)
    @post, @scope = post, scope
  end

  def serializable_hash
    { title: @post.name, body: @post.body }
  end

  def as_json
    { post: serializable_hash }
  end
end
</ruby>

h4. Authorization

Let's update our serializer to include the email address of the author of the post, but only if the current user has superuser
access.

<ruby>
class PostSerializer
  def initialize(post, scope)
    @post, @scope = post, scope
  end

  def as_json
    { post: serializable_hash }
  end

  def serializable_hash
    hash = post
    hash.merge!(super_data) if super?
    hash
  end

private
  def post
    { title: @post.name, body: @post.body }
  end

  def super_data
    { email: @post.email }
  end

  def super?
    @scope.superuser?
  end
end
</ruby>

h4. Testing

One benefit of encapsulating our objects this way is that it becomes extremely straight-forward to test the serialization
logic in isolation.

<ruby>
require "ostruct"

class PostSerializerTest < ActiveSupport::TestCase
  # For now, we use a very simple authorization structure. These tests will need
  # refactoring if we change that.
  plebe = OpenStruct.new(super?: false)
  god   = OpenStruct.new(super?: true)

  post  = OpenStruct.new(title: "Welcome to my blog!", body: "Blah blah blah", email: "tenderlove@gmail.com")

  test "a regular user sees just the title and body" do
    json = PostSerializer.new(post, plebe).to_json
    hash = JSON.parse(json)

    assert_equal post.title, hash.delete("title")
    assert_equal post.body, hash.delete("body")
    assert_empty hash
  end

  test "a superuser sees the title, body and email" do
    json = PostSerializer.new(post, god).to_json
    hash = JSON.parse(json)

    assert_equal post.title, hash.delete("title")
    assert_equal post.body, hash.delete("body")
    assert_equal post.email, hash.delete("email")
    assert_empty hash
  end
end
</ruby>

It's important to note that serializer objects define a clear interface specifically for serializing an existing object.
In this case, the serializer expects to receive a post object with +name+, +body+ and +email+ attributes and an authorization
scope with a +super?+ method.

By defining a clear interface, it's must easier to ensure that your authorization logic is behaving correctly. In this case,
the serializer doesn't need to concern itself with how the authorization scope decides whether to set the +super?+ flag, just
whether it is set. In general, you should document these requirements in your serializer files and programatically via tests.
The documentation library +YARD+ provides excellent tools for describing this kind of requirement:

<ruby>
class PostSerializer
  # @param [~body, ~title, ~email] post the post to serialize
  # @param [~super] scope the authorization scope for this serializer
  def initialize(post, scope)
    @post, @scope = post, scope
  end

  # ...
end
</ruby>

h3. Attribute Sugar

To simplify this process for a number of common cases, Rails provides a default superclass named +ActiveModel::Serializer+
that you can use to implement your serializers.

For example, you will sometimes want to simply include a number of existing attributes from the source model into the outputted
JSON. In the above example, the +title+ and +body+ attributes were always included in the JSON. Let's see how to use
+ActiveModel::Serializer+ to simplify our post serializer.

<ruby>
class PostSerializer < ActiveModel::Serializer
  attributes :title, :body

  def initialize(post, scope)
    @post, @scope = post, scope
  end

  def serializable_hash
    hash = attributes
    hash.merge!(super_data) if super?
    hash
  end

private
  def super_data
    { email: @post.email }
  end

  def super?
    @scope.superuser?
  end
end
</ruby>

First, we specified the list of included attributes at the top of the class. This will create an instance method called
+attributes+ that extracts those attributes from the post model.

NOTE: Internally, +ActiveModel::Serializer+ uses +read_attribute_for_serialization+, which defaults to +read_attribute+, which defaults to +send+. So if you're rolling your own models for use with the serializer, you can use simple Ruby accessors for your attributes if you like.

Next, we use the attributes methood in our +serializable_hash+ method, which allowed us to eliminate the +post+ method we hand-rolled
earlier. We could also eliminate the +as_json+ method, as +ActiveModel::Serializer+ provides a default +as_json+ method for
us that calls our +serializable_hash+ method and inserts a root. But we can go a step further!

<ruby>
class PostSerializer < ActiveModel::Serializer
  attributes :title, :body

private
  def attributes
    hash = super
    hash.merge!(email: post.email) if super?
    hash
  end

  def super?
    @scope.superuser?
  end
end
</ruby>

The superclass provides a default +initialize+ method as well as a default +serializable_hash+ method, which uses
+attributes+. We can call +super+ to get the hash based on the attributes we declared, and then add in any additional
attributes we want to use.

NOTE: +ActiveModel::Serializer+ will create an accessor matching the name of the current class for the resource you pass in. In this case, because we have defined a PostSerializer, we can access the resource with the +post+ accessor.

h3. Associations

In most JSON APIs, you will want to include associated objects with your serialized object. In this case, let's include
the comments with the current post.

<ruby>
class PostSerializer < ActiveModel::Serializer
  attributes :title, :body
  has_many :comments

private
  def attributes
    hash = super
    hash.merge!(email: post.email) if super?
    hash
  end

  def super?
    @scope.superuser?
  end
end
</ruby>

The default +serializable_hash+ method will include the comments as embedded objects inside the post.

<javascript>
{
  post: {
    title: "Hello Blog!",
    body: "This is my first post. Isn't it fabulous!",
    comments: [
      {
        title: "Awesome",
        body: "Your first post is great"
      }
    ]
  }
}
</javascript>

Rails uses the same logic to generate embedded serializations as it does when you use +render :json+. In this case,
because you didn't define a +CommentSerializer+, Rails used the default +as_json+ on your comment object.

If you define a serializer, Rails will automatically instantiate it with the existing authorization scope.

<ruby>
class CommentSerializer
  def initialize(comment, scope)
    @comment, @scope = comment, scope
  end

  def serializable_hash
    { title: @comment.title }
  end

  def as_json
    { comment: serializable_hash }
  end
end
</ruby>

If we define the above comment serializer, the outputted JSON will change to:

<javascript>
{
  post: {
    title: "Hello Blog!",
    body: "This is my first post. Isn't it fabulous!",
    comments: [{ title: "Awesome" }]
  }
}
</javascript>

Let's imagine that our comment system allows an administrator to kill a comment, and we only want to allow
users to see the comments they're entitled to see. By default, +has_many :comments+ will simply use the
+comments+ accessor on the post object. We can override the +comments+ accessor to limit the comments used
to just the comments we want to allow for the current user.

<ruby>
class PostSerializer < ActiveModel::Serializer
  attributes :title. :body
  has_many :comments

private
  def attributes
    hash = super
    hash.merge!(email: post.email) if super?
    hash
  end

  def comments
    post.comments_for(scope)
  end

  def super?
    @scope.superuser?
  end
end
</ruby>

+ActiveModel::Serializer+ will still embed the comments, but this time it will use just the comments
for the current user.

NOTE: The logic for deciding which comments a user should see still belongs in the model layer. In general, you should encapsulate concerns that require making direct Active Record queries in scopes or public methods on your models.

h3. Customizing Associations

Not all front-ends expect embedded documents in the same form. In these cases, you can override the
default +serializable_hash+, and use conveniences provided by +ActiveModel::Serializer+ to avoid having to
build up the hash manually.

For example, let's say our front-end expects the posts and comments in the following format:

<plain>
{
  post: {
    id: 1
    title: "Hello Blog!",
    body: "This is my first post. Isn't it fabulous!",
    comments: [1,2]
  },
  comments: [
    {
      id: 1
      title: "Awesome",
      body: "Your first post is great"
    },
    {
      id: 2
      title: "Not so awesome",
      body: "Why is it so short!"
    }
  ]
}
</plain>

We could achieve this with a custom +as_json+ method. We will also need to define a serializer for comments.

<ruby>
class CommentSerializer < ActiveModel::Serializer
  attributes :id, :title, :body

  # define any logic for dealing with authorization-based attributes here
end

class PostSerializer < ActiveModel::Serializer
  attributes :title, :body
  has_many :comments

  def as_json
    { post: serializable_hash }.merge!(associations)
  end

  def serializable_hash
    post_hash = attributes
    post_hash.merge!(association_ids)
    post_hash
  end

private
  def attributes
    hash = super
    hash.merge!(email: post.email) if super?
    hash
  end

  def comments
    post.comments_for(scope)
  end

  def super?
    @scope.superuser?
  end
end
</ruby>

Here, we used two convenience methods: +associations+ and +association_ids+. The first,
+associations+, creates a hash of all of the define associations, using their defined
serializers. The second, +association_ids+, generates a hash whose key is the association
name and whose value is an Array of the association's keys.

The +association_ids+ helper will use the overridden version of the association, so in
this case, +association_ids+ will only include the ids of the comments provided by the
+comments+ method.

h3. Special Association Serializers

So far, associations defined in serializers use either the +as_json+ method on the model
or the defined serializer for the association type. Sometimes, you may want to serialize
associated models differently when they are requested as part of another resource than
when they are requested on their own.

For instance, we might want to provide the full comment when it is requested directly,
but only its title when requested as part of the post. To achieve this, you can define
a serializer for associated objects nested inside the main serializer.

<ruby>
class PostSerializer < ActiveModel::Serializer
  class CommentSerializer < ActiveModel::Serializer
    attributes :id, :title
  end

  # same as before
  # ...
end
</ruby>

In other words, if a +PostSerializer+ is trying to serialize comments, it will first
look for +PostSerializer::CommentSerializer+ before falling back to +CommentSerializer+
and finally +comment.as_json+.

h3. Overriding the Defaults

h4. Authorization Scope

By default, the authorization scope for serializers is +:current_user+. This means
that when you call +render json: @post+, the controller will automatically call
its +current_user+ method and pass that along to the serializer's initializer.

If you want to change that behavior, simply use the +serialization_scope+ class
method.

<ruby>
class PostsController < ApplicationController
  serialization_scope :current_app
end
</ruby>

You can also implement an instance method called (no surprise) +serialization_scope+,
which allows you to define a dynamic authorization scope based on the current request.

WARNING: If you use different objects as authorization scopes, make sure that they all implement whatever interface you use in your serializers to control what the outputted JSON looks like.

h3. Using Serializers Outside of a Request

The serialization API encapsulates the concern of generating a JSON representation of
a particular model for a particular user. As a result, you should be able to easily use
serializers, whether you define them yourself or whether you use +ActiveModel::Serializer+
outside a request.

For instance, if you want to generate the JSON representation of a post for a user outside
of a request:

<ruby>
user = get_user # some logic to get the user in question
PostSerializer.new(post, user).to_json # reliably generate JSON output
</ruby>

If you want to generate JSON for an anonymous user, you should be able to use whatever
technique you use in your application to generate anonymous users outside of a request.
Typically, that means creating a new user and not saving it to the database:

<ruby>
user = User.new # create a new anonymous user
PostSerializer.new(post, user).to_json
</ruby>

In general, the better you encapsulate your authorization logic, the more easily you
will be able to use the serializer outside of the context of a request. For instance,
if you use an authorization library like Cancan, which uses a uniform +user.can?(action, model)+,
the authorization interface can very easily be replaced by a plain Ruby object for
testing or usage outside the context of a request.

h3. Collections

So far, we've talked about serializing individual model objects. By default, Rails
will serialize collections, including when using the +associations+ helper, by
looping over each element of the collection, calling +serializable_hash+ on the element,
and then grouping them by their type (using the plural version of their class name
as the root).

For example, an Array of post objects would serialize as:

<plain>
{
  posts: [
    {
      title: "FIRST POST!",
      body: "It's my first pooooost"
    },
    { title: "Second post!",
      body: "Zomg I made it to my second post"
    }
  ]
}
</plain>

If you want to change the behavior of serialized Arrays, you need to create
a custom Array serializer.

<ruby>
class ArraySerializer < ActiveModel::ArraySerializer
  def serializable_array
    serializers.map do |serializer|
      serializer.serializable_hash
    end
  end

  def as_json
    hash = { root => serializable_array }
    hash.merge!(associations)
    hash
  end
end
</ruby>

When generating embedded associations using the +associations+ helper inside a
regular serializer, it will create a new <code>ArraySerializer</code> with the
associated content and call its +serializable_array+ method. In this case, those
embedded associations will not recursively include associations.

When generating an Array using +render json: posts+, the controller will invoke
the +as_json+ method, which will include its associations and its root.