Yak Serializers
One Stop Hypermedia Shopping
We did the shaving for you
Yaks is a tool for turning your domain models into Hypermedia resources.
There are at the moment a number of competing media types for building Hypermedia APIs. These all add a layer of semantics on top of a low level serialization format such as JSON or XML. Even though they each have their own design goals, the core features mostly overlap. They typically provide a way to represent resources (entities), and resource collections, consisting of
- Data in key-value format, possibly with composite values
- Embedded resources
- Links to related resources
- Outbound links that have a specific relation to the resource
They might also contain extra control data to specify possible future interactions, not unlike HTML forms.
These different media types for Hypermedia clients and servers base themselves on the same set of internet standards, such as RFC4288 Media types, RFC5988 Web Linking, RFC6906 The "profile" link relation and RFC6570 URI Templates.
Concepts
Yaks is a processing pipeline, you create and configure the pipeline, then feed data through it.
yaks = Yaks.new do
default_format :hal
rel_template 'http://api.example.com/rels/{rel}'
format_options(:hal, plural_links: [:copyright])
namespace ::MyAPI
json_serializer do |data|
MultiJson.dump(data)
end
end
yaks.call(data) # => JSON
Yaks performs this serialization in three steps
- It maps your data to a
Yaks::Resource
- It formats the resource to a syntax tree representation
- It serializes to get the final output
For JSON types, the "syntax tree" is just a combination of Ruby primitives, nested arrays and hashes with strings, numbers, booleans, nils.
A Resource is an abstraction shared by all output formats. It can contain key-value attributes, RFC5988 style links, and embedded sub-resources.
To build an API you create a "mapper" for each type of object you want to represent. Yaks takes care of the rest.
For all configuration options see Yaks::Config::DSL.
See also the API Docs on rdoc.info
Mappers
Say your app has a Post
object for blog posts. To serve posts over your API, define a PostMapper
class PostMapper < Yaks::Mapper
link :self, '/api/posts/{id}'
attributes :id, :title
has_one :author
has_many :comments
end
Configure a Yaks instance and start serializing!
yaks = Yaks.new
yaks.call(post)
or a bit more elaborate
yaks = Yaks.new do
default_format :json_api
rel_template 'http://api.example.com/rels/{rel}'
format_options(:hal, plural_links: [:copyright])
end
yaks.call(post, mapper: PostMapper, format: :hal)
Attributes
Use the attributes
DSL method to specify which attributes of your model you want to expose, as in the example above. You can override the load_attribute
method to change how attributes are fetched from the model.
For example, if you are representing data that is stored in a Hash, you could do
class PostHashMapper < Yaks::Mapper
attributes :id, :body
# @param name [Symbol]
def load_attribute(name)
object[name]
end
end
The default implementation will first try to find a matching method for an attribute on the mapper itself, and will then fall back to calling the actual model. So you can add extra 'virtual' attributes like so :
class CommentMapper < Yaks::Mapper
attributes :id, :body, :date
def date
object.created_at.strftime("at %I:%M%p")
end
end
Filtering
You can override #attributes
, or #associations
.
class SongMapper
attributes :title, :duration, :lyrics
has_one :artist
has_one :album
def minimal?
env['HTTP_PREFER'] =~ /minimal/
end
# @return Array<Yaks::Mapper::Attribute>
def attributes
return super.reject {|attr| attr.name.equal? :lyrics } if minimal?
super
end
# @return Array<Yaks::Mapper::Association>
def associations
return [] if minimal?
super
end
end
Links
You can specify link templates that will be expanded with model attributes. The link relation name should be a registered IANA link relation or a URL. The template syntax follows RFC6570 URI templates.
class FooMapper < Yaks::Mapper
link :self, '/api/foo/{id}'
link 'http://api.foo.com/rels/comments', '/api/foo/{id}/comments'
end
To prevent a link to be expanded, add expand: false
as an option. Now the actual template will be rendered in the result, so clients can use it to generate links from.
To partially expand the template, pass an array with field names to expand. e.g.
class ProductMapper < Yaks::Mapper
link 'http://api.foo.com/rels/line_item', '/api/line_items?product_id={product_id}&quantity={quantity}', expand: [:product_id]
end
# "_links": {
# "http://api.foo.com/rels/line_item": {
# "href": "/api/line_items?product_id=273&quantity={quantity}",
# "templated": true
# }
# }
You can pass a symbol instead of a template, in that case the symbol will be used as a method name on the object to retrieve the link. You can override this behavior just like with attributes.
class FooMapper < Yaks::Mapper
link 'http://api.foo.com/rels/go_home', :home_url
# by default calls object.home_url
def home_url
object.setting('home_url')
end
end
Associations
Use has_one
for an association that returns a single object, or has_many
for embedding a collection.
Options
:mapper
: Use a specific for each instance, will be derived from the class name if omitted (see Policy vs Configuration):collection_mapper
: For mapping the collection as a whole, this defaults to Yaks::CollectionMapper, but you can subclass it for example to add links or attributes on the collection itself:rel
: Set the relation (symbol or URI) this association has with the object. Will be derived from the association name and the configured rel_template if ommitted:link_if
: Conditionally render the association as a link. A:href
option is required
class ShowMapper < Yaks::Mapper
has_many :events, href: '/show/{id}/events', link_if: ->{ events.count > 50 }
end
Calling Yaks
Once you have a Yaks instance, you can call it with call
(serialize
also works but might be deprecated in the future.) Pass
it the data to be serialized, plus options.
:env
a Rack environment, see next section:format
the format to be used, e.g.:json_api
. Note that if the Rack env contains anAccept
header which resolves to a recognized format, then the header takes precedence:mapper
the mapper to be used. Will be inferred if omitted:item_mapper
When rendering a collection, the mapper to be used for each item in the collection. Will be inferred from the class of the first item in the collection if omitted.
Rack env
When serializing, Yaks lets you pass in an env
hash, which will be made available to all mappers.
yaks = Yaks.new
yaks.call(foo, env: my_env)
class FooMapper
attributes :bar
def bar
if env['something']
#...
end
end
end
The env hash will be available to all mappers, so you can use this to pass around context. In particular context related to the current HTTP request, e.g. the current logged in user, which is why the recommended use is to pass in the Rack environment.
If env
contains a HTTP_ACCEPT
key (Rack's way of representing the Accept
header), Yaks will return the format that most closely matches what was requested.
Namespace
Yaks by default will find your mappers for you if they follow the naming convention of appending 'Mapper' to the model class name. This (and all other "conventions") can be easily redefined though, see below. If you have your mappers inside a module, use namespace
.
module API
module Mappers
class PostMapper < Yaks::Mapper
#...
end
end
end
yaks = Yaks.new do
namespace API::Mappers
end
If your namespace contains a CollectionMapper
, Yaks will use that instead of Yaks::CollectionMapper
, e.g.
module API
module Mappers
class CollectionMapper < Yaks::CollectionMapper
link :profile, 'http://api.example.com/profiles/collection'
end
end
end
You can also have collection mappers based on the type of members the collection holds, e.g.
module API
module Mappers
class LineItemCollectionMapper < Yaks::CollectionMapper
link :profile, 'http://api.example.com/profiles/line_items'
attributes :total
def total
collection.inject(0) do |memo, line_item|
memo + line_item.price * line_item.quantity
end
end
end
end
end
Yaks will automatically detect and use this collection when serializing an array of LineItem
objects.
Custom attribute/link/subresource handling
When inheriting from Yaks::Mapper
, you can override map_attributes
, map_links
and map_resources
to skip (or augment) above methods, and instead implement your own custom mechanism. These methods take a Yaks::Resource
instance, and should return an updated resource. They should not alter the resource instance in-place. For example
class ErrorMapper < Yaks::Mapper
link :profile, '/api/error'
def map_attributes(resource)
attrs = {
http_code: 500,
message: object.to_s,
type: object.class.name.underscore
}
case object
when AllocationException
attrs[:http_code] = 422
when ActiveRecord::RecordNotFound
attrs[:http_code] = 404
attrs[:type] = "record_not_found"
end
resource.update_attributes(attrs)
end
end
Resources and Serializers
Yaks uses an intermediate "Resource" representation to support multiple output formats. A mapper turns a domain model into a Yaks::Resource
. A serializer (e.g. Yaks::Serializer::Hal
) takes the resource and outputs the structure of the target format.
Since version 0.4 the recommended API is through Yaks.new {...}.serialize
. This will give you back a composite value consisting of primitives that have a mapping to JSON, so you can use your favorite JSON encoder to turn this into a character stream.
my_yaks = Yaks.new
hal = my_yaks.call(model)
puts JSON.dump(hal)
There are at least a handful of JSON libraries and implementations for Ruby out there, with different trade-offs. Yaks does not impose an opinion on which one to use
HAL
This is the default. In HAL one decides when building an API which links can only be singular (e.g. self), and which are always represented as an array. Yaks defaults to singular as I've found it to be the most common case. If you want specific links to be plural, then configure their rel href as such.
hal = Yaks.new do
format_options :hal, plural_links: ['http://api.example.com/rels/foo']
end
CURIEs are not explicitly supported (yet), but it's possible to use them with some effort, see examples/hal01.rb
for an example.
The line between a singular resource and a collection is fuzzy in HAL. To stick close to the spec you're best to create your own singular types that represent collections, rather than rendering a top level CollectionResource.
JSON-API
default_format :json_api
JSON-API has no concept of outbound links, so these will not be rendered. Instead the key will be inferred from the mapper class name by default. This can be changed per mapper:
class AnimalMapper
key :pet
end
Or the policy can be overridden:
yaks = Yaks.new do
derive_type_from_mapper_class do |mapper_class|
piglatinize(mapper_class.to_s.sub(/Mapper$/, ''))
end
end
Collection+JSON
default_format :collection_json
Subresources aren't mapped because Collection+JSON doesn't really have that concept, and the other way around templates and queries don't exist (yet) in Yaks.
More formats
Are planned... at the moment HAL is the only format I actually use, so it's the one that's best supported. Adding formats that follow the resource=(attributes, links, subresources) structure or a subset thereof is straightforward. More features, e.g. forms/actions such as used in Mason might be added in the future.
Hooks
It is possible to hook into the Yaks pipeline to perform extra processing steps before, after, or around each step. It also possible to skip a step.
yaks = Yaks.new do
# Automatically give every resource a self link
after :map, :add_self_link do |resource|
resource.add_link(Yaks::Resource::Link.new(:self, "/#{resource.type}/#{resource.attributes[:id]}"))
end
# Skip serialization, so Ruby primitives come back instead of JSON
# This was the default before versions < 0.5.0
skip :serialize
end
Policy over Configuration
It's an old adage in the Ruby/Rails world to have "Convention over Configuration", mostly to derive values that were not given explicitly. Typically based on things having similar names and a 1-1 derivable relationship.
This saves a lot of typing, but for the uninitiated it can also create confusion, the implicitness makes it hard to follow what's going on.
What's worse, is that often the Configuration part is skipped entirely, making it very hard to deviate from the Golden Standard.
There is another old adage, "Policy vs Mechanism". Implement the mechanisms, but don't dictate the policy.
In Yaks whenever missing values need to be inferred, like finding an unspecified mapper for a relation, this is handled by a policy object. The default is Yaks::DefaultPolicy
, you can go there to find all the rules of inference. Single rules of inference can be redefined directly in the Yaks configuration:
yaks = Yaks.new do
derive_mapper_from_object do |model|
# ...
end
derive_type_from_mapper_class do |mapper_class|
# ...
end
derive_mapper_from_association do |association|
# ...
end
derive_rel_from_association do |mapper, association|
# ...
end
end
You can also subclass or create from scratch your own policy class
class MyPolicy < DefaultPolicy
#...
end
yaks = Yaks.new do
policy MyPolicy
end
Primitives
For JSON based formats, the "syntax tree" is merely a structure of Ruby primitives that have a JSON equivalent. If your mappers return non-primitive attribute values, you can define how they should be converted. For example, JSON has no notion of dates. If your mappers return these types as attributes, then Yaks needs to know how to turn these into primitives. To add extra types, use map_to_primitive
Yaks.new do
map_to_primitive Date, Time, DateTime do |date|
date.iso8601
end
end
This can also be used to transform alternative data structures, like those from Hamster, into Ruby arrays and hashes. Use call()
to recursively turn things into primitives.
Yaks.new do
map_to_primitive Hamster::Vector, Hamster::List do |list|
list.map do |item|
call(item)
end
end
end
Yaks by default "primitivizes" symbols (as strings), and classes that include Enumerable (as arrays).
Real World Usage
Yaks is used in production by Ticketsolve. You can find an example API endpoint here.
Get in touch if you like to see your name and API here.
Demo
You can find an example app at Yakports, or browse the HAL api directly using the HAL browser.
Acknowledgment
The mapper syntax is largely borrowed from ActiveModel::Serializers, which in turn closely mimics the syntax of ActiveRecord models. It's a great concise syntax that still offers plenty of flexibility, so to not reinvent the wheel I've stuck to the existing syntax as far as practical, although there are several extensions and deviations.
Lightweight
Yaks is a lean library. It only depends on a few other tiny libraries (inflection, concord, uri_template). It has no core extensions (monkey patches). There is deliberately no built-in "integration" with existing frameworks, since the API is simply enough. You just call it.
If this approach sounds appealing, have a look at microrb.com.
Is it any good
How to contribute
Run the tests, the examples, try it with your own stuff and leave your impressions in the issues. Or discuss on API-craft.
To fix a bug
- Fork the repo
- Fix the bug, add tests for it
- Push it to a named branch
- Add a PR
To add a feature
- Open an issue as soon as possible to gather feedback
- Same as above, fork, push to named branch, make a pull-request
Yaks uses Mutation Testing. Run rake mutant
and look for percentage coverage. In general this should only go up.
License
MIT License (Expat License), see LICENSE