Add IoC to rails with minimal fuss and ceremony.
(Also check out shog for better rails logs)
Dependency injection helps to break explicit dependencies between objects making it much easier to maintain a single responsibility and reduce coupling in our class designs. This leads to more testable code and code that is more resilient to change.
Several have argued that the dynamic properties of Ruby make Dependency Injection frameworks irrelevant. Some argue that you can build in defaults and make them overridable, or just use module mixins.
Most of these counter arguments focus on testing, and given how easy it is to mock objects in Ruby, you don't really need a framework. If testing were the only virtue they'd be spot on. Despite its virtues DI doesn't come without its own problems. However for larger projects that you expect to be long-lived, a DI framework may help manage the complexity.
For a deeper background on Dependency Injection consider the Wikipedia article on the subject.
Assuming you've embraced the general concept of DI why would you want to use a framework. Lets consider the alternatives.
class Hunter
def weapon
@weapon ||= Weapon.new
end
def weapon=( value )
@weapon = value
end
end
In this scenario the Hunter class knows how to create a weapon and provides a sane default, but allows the dependency to be overridden if needed.
PROS
- Very simple to understand and debug
- Provides basic flexibility
- The dependency is clearly defined.
CONS
- Still coupled to a specific type of Weapon.
- If multiple classes use this approach and you decide to upgrade your armory, you'd have to modify every line that creates new weapons. The factory pattern can be used to address such a dependency.
- No global method of replacing a Weapon class with a specialized or augmented class. For example a ThreadLockedWeapon.
class Hunter
def initialize( weapon )
@weapon = weapon
end
end
Here Hunters can use any weapon and can be designed to an interface Weapon that does not have an implementation yet.
PROS
- Provides flexibility
- Work can proceed concurrently on Hunter and Weapon classes by different engineers on the team.
CONS
- Hard to reason about Hunters and Weapons as a whole.
- The dependency is not clearly defined - what is a weapon?
- It pushes the responsibility of constructing dependencies onto the consumer of the class. If the class is used in multiple places this becomes a maintenance chore when changes are required.
- It becomes tedious to use classes resulting in repeated boilerplate code that distracts from the primary responsibility of the calling code.
Using a good framework can help conserve the pros of each method while minimizing the cons. A DI framework works like an automatic factory system resolving dependencies cleanly like a factory but without all the effort to create custom factories.
A good framework should
- Make dependencies clear
- Require a minimal amount of configuration or ceremony
class Hunter
depend_on do
weapon Weapon
end
# or
attr_dependency :weapon, Weapon
end
Here the dependency is clearly defined - and even creates accessors for getting and setting the weapon. When a Hunter is created its dependencies are also created - and any of their dependencies and so on. Usage is equally simple
hunter = scorpion.fetch Hunter
hunter.weapon # => a Weapon
Overriding the kind of weapons used by hunters.
class Axe < Weapon; end
scorpion.prepare do
hunt_for Axe
end
hunter = scorpion.fetch Hunter
hunter.weapon # => an Axe
Overriding hunters!
class Axe < Weapon; end
class Predator < Hunter; end
scorpion.prepare do
hunt_for Predator
hunt_for Axe
end
hunter = scorpion.fetch Hunter
hunter # => Predator
hunter.weapon # => an Axe
Out of the box Scorpion does not need any configuration and will work immediately. You can hunt for any Class even if it hasn't been configured.
hash = Scorpion.instance.fetch Hash
hash # => {}
Scorpions feed their Scorpion Objects - any object that should be fed its dependencies when it is created. Simply include the Scorpion::Object module into your class to benefit from Scorpion injections.
class Keeper
include Scorpion::Object
depend_on do
lunch FastFood
end
end
class Zoo
include Scorpion::Object
depend_on do
keeper Zoo::Keeper
vet Zoo::Vet, lazy: true
end
# or with like attr_accessor
attr_dependency :keeper, Zook::Keeper
attr_dependency :vet, Zoo::Vet, lazy: true
end
zoo = scorpion.fetch Zoo
zoo.keeper # => an instance of a Zoo::Keeper
zoo.vet? # => false it hasn't been hunted down yet
zoo.vet # => an instance of a Zoo::Vet
zoo.keeper.lunch # => an instance of FastFood
All of your classes should be objects! And any dependency that is also a Object will be fed.
A good scorpion should be prepared to hunt. An effort that describes what the scorpion hunts for and how it should be found. Scorpion uses Classes and Modules as the primary means of identifying dependency in favor of opaque labels or strings. This serves two benefits:
- The type of object expected by the dependency is clearly identified making it easier to understand what the concrete dependencies really are.
- Types (Classes & Modules) explicitly declare the expected behavioral contract of an object's dependencies.
Most scorpion hunts will be for an instance of a specific class (or a more derived class). In the absence of any configuration, Scorpion will simply create an instance of the specific class requested.
scorpion.fetch Hash # => Hash.new
scorpion.prepare do
hunt_for Object::HashWithIndifferentAccess
end
scorpion.fetch Hash # => Object::HashWithIndifferentAccess.new
Modules can be hunted for in two ways.
- If a Class has been prepared for hunting that includes the module, it will be used to satisfy requests for that module
- If no Class is found, the Module itself will be returned.
module Sharp
module_function
def poke; self.class.name end
end
class Sword
include Sharp
end
poker = scorpion.fetch Sharp
poker.poke # => "Module"
scorpion.prepare do
hunt_for Sword
end
poker = scorpion.fetch Sharp
poker.poke # => "Sword"
Sometimes resolving the correct dependencies is a bit more dynamic. In those cases you can use a builder block to hunt for dependency.
class Samurai < Sword; end
class Broad < Sword; end
scorpion.prepare do
hunt_for Sword do |scorpion|
scorpion.spawn Random.rand( 2 ) == 1 ? Samurai : Broad
end
end
Objects may also define their own .create methods that receive a scorpion and arguments.
class City
def self.create( scorpion, name )
klass =
if name == "New York"
BigCity
else
SmallCity
end
scorpion.new klass, name
end
def initialize( name )
@name = name
end
end
class BigCity < City; end
class SmallCity < City; end
For really complex dependencies you may want to delegate the effort to retrieve
the dependencies to another type - a factory module for example. Scorpion
allows you to delegate hunting dependency using the :with
option.
module ChocolateFactory
module_function
def call( scorpion, *args, &block )
case args.first
when Nuget then scorpion.spawn Snickers, *args, &block
when Butterscotch then scorpion.spawn Butterfinger, *args, &block
when Coconut then scorpion.spawn Garbage, *args, &block
end
end
end
scorpion.prepare do
hunt_for Candy, with: ChocolateFactory
end
scorpion.fetch Candy, Nuget.new #=> Snickers.new Nugget.new
Any object that responds to #call( scorpion, *args, &block )
can be used as
a hunting delegate.
Scorpion allows you to capture dependency and feed the same instance to everyone that asks for a matching dependency.
DI singletons are different then global singletons in that each scorpion can have a unique instance of the class that it shares with all of its objects. This allows, for example, global variable like support per-request without polluting the global namespace or dealing with thread concurrency issues.
class Logger; end
scorpion.prepare do
capture Logger
end
scorpion.fetch Logger # => Logger.new
scorpion.fetch Logger # => Previously captured logger
Captured dependencies are not shared with child scorpions (for example when conceiving scorpions from a Nest). To share captured dependency with children use
share
.
A scorpion nest is where a mother scorpion lives and conceives young - duplicates of the mother but maintaining their own state. The scorpion nest is used by the Rails integration to give each request its own scorpion.
All preparation performed by the mother is shared with all the children it conceives so that configuration is established when the application starts.
nest.prepare do
hunt_for Logger
end
scorpion = nest.conceive
scorpion.fetch Logger # => Logger.new
Scorpion provides simple integration for rails controllers to establish a scorpion for each request.
# user_service.rb
class UserService
def find( username ) ... end
end
# config/initializers/nest.rb
require 'scorpion'
Scorpion.prepare do
capture UserService # Share with all the objects that are spawned in _this_ request
share do
capture Logger # Share with every request
end
end
# application_controller.rb
require 'scorpion'
class ApplicationController < ActionController::Base
depend_on do
users UserService, lazy: true
end
end
# users_controller.rb
class UsersController < ApplicationController
def show
user = users.find( "batman" )
logger.write "Found a user: #{ user }"
end
end
Simliar to support for controllers, Scorpion provides support for dependency injection into ActiveJob objects.
# avatar_job.rb
class AvatarJob < ActiveJob::Base
depend_on do
users UserService, lazy: true
logger Logger
end
def perform( id )
user = users.find( id )
logger.write "Found a user: #{ user }"
end
end
Scorpion enhances ActiveRecord models to support resolving dependencies from a scorpion and sharing that scorpion with all associations.
Consider using a SOA framework like Shamu for managing complex resource relationships.
class User < ActiveRecord::Base
depend_on do
credentials Service::Auth::Credentials
end
def check_password( password )
credentials.check encoded_password, password
end
end
class SessionsController < ActionController::Base
def create
user = User.with_scorpion( scorpion ).find params[:id]
user = scorpion( User ).find params[:id]
sign_in if user.check_password( params[:password] )
end
end
- Fork it ( https://github.com/phallguy/scorpion/fork )
- Create your feature branch (
git checkout -b my-new-feature
) - Commit your changes (
git commit -am 'Add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create a new Pull Request
Copyright (c) 2015 Paul Alexander