A brutally simple take on the observer pattern.
Flirt acts as a single point to which events can be sent (published) and listeners can be registered (subscribed), in order to promote extreme decoupling between components.
Add this line to your application's Gemfile:
gem 'flirt'
And then execute:
$ bundle
Or install it yourself as:
$ gem install flirt
###To publish an event:
event_data = { fruit: 'apple' }
Flirt.publish :picked, event_data
###To subscribe:
class MyListener
def initialize
Flirt.subscribe self, :picked, with: :picked_callback
end
def picked_callback(event_data)
puts "The #{event_data[:fruit]} has been picked"
end
end
###To unsubscribe:
Flirt.unsubscribe self, :picked, with: :picked_callback
Syntactic sugar for subscription and unsubscription has been provided in the form of a module:
class MyListener
include Flirt::Listener
def initialize
subscribe_to :picked, with: :picked_callback
end
def before_destroy
unsubscribe_from :picked, with: :picked_callback
end
def picked_callback(event_data)
puts "The #{event_data[:fruit]} has been picked"
end
end
or even:
class MyListener
extend Flirt::Listener
subscribe_to :picked, with: :picked_callback
def self.picked_callback(event_data)
puts "The #{event_data[:fruit]} has been picked"
end
end
unsubscribe_from
can technically be used in the class context, but probably doesn't have as much use.
###Flirt defaults to 'enabled'. Switch Flirt off:
Flirt.disable
And back on again:
Flirt.enable
Enabled status affects publishing only, listeners can still be added and will be remembered. No listeners will be removed.
###Disable only a set of events:
Flirt.disable only: [:pancake_cooked, :picked]
###Enable only a set of events:
Flirt.enable only: [:topping_added, :pancake_flipped]
Disabling and enabling sets of events is not cumulative. The new set of events will overwrite all previous calls. For example:
Flirt.disable only: [:pancake_cooked, :picked]
Flirt.disable only: [:flipped]
The above code will leave only :flipped
disabled.
Flirt.enable only: [:flipped, :picked]
Flirt.enable only: [:flipped]
The above code will leave only :flipped
enabled.
Calling Flirt.enable
or Flirt.disable
will clear any previously set enabled or disabled events.
###Clear all listeners:
Flirt.clear
This operation cannot be undone. Cleared listeners would have to be re-added one by one.
##Motivation
Ruby projects (and Rails projects in particular) can easily become a mess of tightly coupled code. The Rails framework almost encourages the idea of 'fat models', with features like callbacks (often touching other models and thus coupling themselves) and concerns (that use modules to hide code that often should really have its own class) to name but two.
The observer (or pub/sub) pattern can help decouple code, decompose large classes and allow for smaller classes with a single purpose. This promotes easy testing, readability, maintainability and eventually stability of your code.
So why use Flirt?
Flirt gives you just enough to use and test the pub/sub pattern with the minimum of cruft. The number of objects is kept to a minimum for speed and ease of debugging. The extendable Listener module has only the two methods you need to use.
With such a simple syntax, it's easy to understand what Flirt is doing when you revisit your code again in three months.
For basic use there is no set-up necessary beyond requiring the gem.
Events can only be represented as symbols, avoiding problems with string/symbol confusion.
Only one parameter can be passed as event data, encouraging structured data objects (hashes, structs or class instances) rather than unreadable parameter lists.
Only one object (Flirt itself) can be listened to, reducing the danger of implicit coupling between publishers and subscribers. Subscribers listen to events, not objects.
This means events are fired in a deterministic way, without over-obfuscating the control flow for debug tools. You can depend on listeners being called before, for example, the end of a controler call. If you wish to delegate a task to a worker thread (like Sidekiq for example) it's easy enough to do in a listener and you're not tied to any particular implementation.
Seriously, why use any other pub/sub system when you could be flirting instead?
It can help project organisation if you keep your listeners in one place (in Rails you could use app/listeners) and minimise the logic in the listener class itself. Delegate to service objects when possible:
class MyListener
extend Flirt::Listener
subscribe_to :cooked, with: :cooked_callback
def self.cooked_callback(pancake)
ToppingAdder.new(pancake).add(:blueberry)
end
end
This will make finding your listeners very easy. It also decouples the service object completely from the event process, meaning it can be used in other contexts and testing is a breeze.
This style of coding promotes the Single Responsibility Principle: "A class should only have one reason to change." The class only requires amending if the event specification changes.
It may be useful to use an object to keep a repository of event symbols or define a class to specify an event:
class Events
FLIPPED = :flipped
COOKED = :cooked
...
end
...
Flirt.publish Events::COOKED, pancake
This eliminates invisible errors due to typos in event symbols and gives IDEs a fighting chance of giving you auto-complete for event names. This style of coding promotes DRY principles: "Every piece of knowledge must have a single, unambiguous, authoritative representation within a system."
Alternatively, an event class might look something like this but will obviously depend on the project:
class PouredEvent
attr_accessor :millilitres, :time
def initialize(opts = {})
self.millilitres = opts.fetch(:millilitres)
self.time = opts.fetch(:time)
end
def self.name; :poured; end
end
...
Flirt.publish PouredEvent.name, PouredEvent.new(millilitres: 20, time: Time.now)
The clear
, enable
and disable
features are provided to aid testing. If you find yourself reaching for them in production code you're probably using the wrong pattern, or you may need to re-think your architecture.
Have a look at decorators if you need to add different functionality to a model depending on where it's called.
Alternatively, change the location in the code where you publish your events. A useful move in Rails is from the model to the controller, to avoid admin or other background updates triggering events that should only be based on user actions. This kind of move can also help break some event or callback loops, where a side effect of one event causes another event to fire and vice versa. ActiveRecord save and validate callbacks (and thus events based on them) are particularly prone to this.
If you find yourself creating and throwing away multiple listener objects, Ruby will still keep references to those objects for the lifetime of the application. This can result in memory leaks, as Ruby will not be able to garbage collect the listeners. To avoid this situation, ensure you unsubscribe from any events when you're done with them. Alternatively, on subscription you can request that Flirt uses weak references:
Flirt.subscribe object, :flipped, with: :flipped_callback, weakref: true
This means that garbage collection can target the listener. Be careful though, as the listener will be garbage collected unless you keep a reference to it somewhere else. This kind of code would be problematic:
def set_listener
Flirt.subscribe TossedListener.new, :tossed, with: :tossed_callback, weakref: true
end
This will lead to intermittent bugs as nothing keeps a reference to TossedListener.new
. The listener will work until garbage collection kicks in.
Flirt defaults to using strong references to ensure consistent behaviour.
While no set-up is required for basic Flirt use, there is a quirk of Rails autoloading that could cause issues.
If you wish to use listeners on a class level, you must make sure the class is loaded. If the class definition has not been required, the listener will not be registered.
Rails auto-loads classes when they're first referenced so unless you take an extra step, class-level listeners won't ever be loaded.
Placing the following code in an initializer will ensure that anything you place in the #{Rails.root}/app/listeners
directory will get loaded when the Rails framework boots up:
# /config/initializers/flirt.rb
Dir["#{Rails.root}/app/listeners/**/*.rb"].each {|file| require file }
##Testing
In order to test units of behaviour, you probably want to disable Flirt and test each unit of your code in isolation. You can always enable Flirt or individual events in your test file for integration tests.
If you're creating listeners on the fly (useful for visual applications like games and dialogue based utilities), you'll want to clear Flirt before each test to stop callbacks building up. If you're setting listeners up once (by using listeners on the class level for example, useful for applications like REST APIs and web back-ends like Rails), then you don't want to clear them for each test.
If you want to use both types of listener at once, testing may be more complex. You may want to re-think your architecture, but you could set up listeners thus:
class FlipListener
extend Flirt::Listener
def self.subscribe
subscribe_to Events::FLIPPED, with: :flip
end
subscribe
def self.flip(pancake)
PancakeFlipper.new(pancake).flip
end
end
Clear Flirt for each test, then switch on individual listeners in the appropriate tests using the listener's equivalent of:
FlipListener.subscribe
If you're using RSpec, you probably want to disable and clear Flirt in the before block in spec/spec_helper.rb
:
RSpec.configure do |config|
config.before(:each) do
Flirt.disable
# Flirt.clear # unless you're using solely class-based listeners.
end
...
If you're using MiniTest, adding this to your initialisation code will disable or clear Flirt before each test:
module FlirtMinitestPlugin
def before_setup
super
Flirt.disable
# Flirt.clear # unless you're using solely class-based listeners.
end
end
class MiniTest::Unit::TestCase
include FlirtMinitestPlugin
end
- Fork it ( https://github.com/[my-github-username]/flirt/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