Sonntag, 1. Februar 2015

Template method in Ruby!

Sharing business logic by inheritance is a common pattern in object oriented programming. But sometimes the piece of logic has very little variances, like:
class Bike
  def start
    "Check 2 tyres. Bike is started!"
  end
end

class Car
  def start
    "Check 4 tyres. Start gasoline engine. Car is started!"
  end
end
and:
Bike.new.start
=> "Check 2 tyres. Bike is started!"
Car.new.start
=> "Check 4 tyres. Start gasoline engine. Car is started!"
Quite static. Both classes seem to have something in common (they can be started and have a number of tyres to be checked before). But at the same point, they are specialized by the number of tyres and in having an engine to be started or none.
The template method pattern is a reasonable option to deal with such small variances. It is one of those patterns defined by the Gang of Four and fairly simple to implement.
The pattern is based on inheritance for the similarities (the template method), whereas the variances are separated by extracting them into appropriate methods, so called hook methods. The template method drives the bit that needs to vary, but it does so by making calls to abstract methods, which are then supplied by the concrete subclasses.
The template pattern is absolutely reasonable, when:
  1. inheritance is reasonable
  2. the logic equality is way more distinct than its variance
The example in a templated manner:
class Vehicle
  def start
    todos = ["Check #{tyres} tyres."]
    todos << "Start #{fuel} engine." if fuel
    todos << "#{self.class} is started!"
    todos.join(' ')
  end
private
  def tyres
    raise 'Called abstract method: tyres'
  end

  def fuel
    raise 'Called abstract method: fuel'
  end
end

class Car < Vehicle
private
  def tyres
    4
  end

  def fuel
    gasoline
  end
end

class Bike < Vehicle
private
  def tyres
    2
  end

  def fuel
    nil
  end
end
Since Ruby does not supply abstract methods, defining exception methods in the base class is common practice in Ruby world.
Even though the generalized start method (template method) now looks more complex, it dries out the repeated similarities and is way more flexible for further subclasses.
Please note the template method pattern only makes sense, when the algorithms are way more alike than different.
Even the refactored result can be improved. Abstract methods are the more static typed language way. It makes more sense for the base class Vehicle to simply supply a default implementation of these methods for the convenience of its subclasses:
class Vehicle
  def start
    todos = ["#{self.class} is started!"]
    todos.unshift "Start #{fuel} engine." if fuel
    todos.unshift "Check #{tyres} tyres." unless tyres.to_i.zero?
    todos.join(' ')
  end
private
  attr_reader :fuel, :tyres
end

class Car < Vehicle
private
  def tyres
    4
  end

  def fuel
    gasoline
  end
end

class Bike < Vehicle
  def tyres
    2
  end
end
and:
Bike.new.start
=> "Check 2 tyres. Bike is started!"
Car.new.start
=> "Check 4 tyres. Start gasoline engine. Car is started!"
Creating a new subclass can take advantage of the existing method structure required by the base class:
class Rocket < Vehicle
private
  def fuel
    "oxydizer"
  end
end
It works like a charm:
Rocket.new.start
=> "Start oxydizer engine. Rocket is started!"
Further articles of interest:

Supported by Ruby 2.1.1

Sonntag, 25. Januar 2015

Bitmask the boolean attributes!

Some database tables suffer from excessive boolean attributes. Especially when most are likely to be NULL.
Merging the boolean attributes into one single bitmasked integer attribute is a solution. Then the integer value represents multiple flags (booleans) bitwise. Metaphorical spoken: several boolean attributes are stacked into the value of one integer attribute
The advantages for bitmasking boolean attributes are:
  1. much lower attribute baggage: 1 integer (usually 4 bytes) can take up to 32 boolean attributes
  2. flexible database structure (a new boolean attribute does not require a database migration)
  3. readable semantic values (the values are meaningful compared to the boolean true/ false)
  4. dynamic access to the bitmask values, which feels more Rubyish
Although there are various pros, the downsides also have to be considered. Bitmasking definitely does not fit for every use case, like most patterns. Abusing it introduces more pain than relief, because it also:
  1. reduces readability in the database layer (the stacked integer value is hiding the meaningful attribute names)
  2. introduces a (very small) Ruby layer for processing the stacked values (back and forth)
  3. means to work around attribute representations, which rely on the database data type (like simple_form gem)
The gem bitmask_attribute lightens the workload for dealing with the bitmasking logic. It provides a nice API to deal with. So instead of a migration with only a few boolean attributes:
create_table :people do |t| 
  t.string :name
  t.boolean :product_owner, null: false, default: false
  t.boolean :developer, null: false, default: false
  t.boolean :scrum_master, null: false, default: false
  t.boolean :sales, null: false, default: false
end 
the same model can be refactored to:
create_table :people do |t| 
  t.string :name
  t.integer :roles
end 
The appropriate model:
class Person < ActiveRecord::Base
  bitmask :roles, 
    as: [:scrum_master, :product_owner, :developer, :sales]
end
and by now the bitmask can be used:
Person.values_for_roles
=> [:scrum_master, :product_owner, :developer, :sales]
person = Person.create name: 'Bob', 
  roles: [:scrum_master, :developer]
person.roles
=> [:scrum_master, :developer]
person.roles << :sales
person.roles
=> [:scrum_master, :developer, :sales]
person.roles? :developer
=> true
If the bitmask has to be represented by check boxes in the Ruby on Rails view:
<% Person.values_for_roles.each do |role| %>
  <%= check_box_tag 'person[roles][]', role, 
        @person.roles.include?(role), id: role %>
  <%= label_tag role, role.to_s.humanize %>
<% end %>
generates the HTML:








Using bitmask_attribute requires to think about the pros and cons in the forefront, because migrating the production data back to boolean attributes is awkward. Furthermore, when it comes to boolean attributes representing several states (especially when the combination of boolean attributes respresent those states), a state machine should be considered.
Further articles of interest:

Supported by Ruby 2.1.1 and Ruby on Rails 3.2.19

Sonntag, 18. Januar 2015

Filter the named parameters!

Named parameters are great for reducing complexity of method parameters in Ruby. Filtering the named parameters for relevance, by limiting the keys to the expected few, makes sense in some cases. Especially, if logic is bound to the API keywords.
An example Team class API could be:
class Team
  def initialize name, roles={}
    @name = name
    @roles = roles
  end

  def roles
    @roles.keys
  end
end
Team.new('Scrum', product_owner: 'Alice', 
  scrum_master: 'Bob').roles
=> [:product_owner, :scrum_master]
looks plausible.
But:
Team.new('Scrum', product_owner: 'Alice', 
  sales: 'Sarah').roles
=> [:product_owner, :sales]
is definitely not a Scrum team.
Every time the API expects certain named parameters, those have to be filtered, like:
class Team
  def initialize name, roles={}
    @name = name
    @roles = [:product_owner, :scrum_master, :developer].
      each_with_object(Hash.new) { |role, hash| 
        hash[role] = roles[role] if roles.has_key?(role) 
      }
  end

  def roles
    @roles.keys
  end
end
Team.new('Scrum', product_owner: 'Alice', 
  sales: 'Sarah').roles
=> [:product_owner]
Please note, that :sales was filtered out, like any would have been except :product_owner, :scrum_master or :developer.
The Hash filtering logic is likely supposed to be repeated. And that is why extending the Hash is plausible.
Ruby on Rails already has in their ActiveSupport gem. The Hash#slice does exactly filter a Hash for keys. So the example in Ruby On Rails would rather look like:
class Team
  def initialize name, roles={}
    @name = name
    @roles = roles.slice :product_owner, :scrum_master, :developer
  end

  def roles
    @roles.keys
  end
end
Team.new('Scrum', product_owner: 'Alice', 
  sales: 'Sarah').roles
=> [:product_owner]
Furthermore Hash#slice! even replaces the hash itself with the given keys. But beware it returns a hash with the removed key/ value pairs.
Further articles of interest:

Supported by Ruby 2.1.1 and Ruby on Rails 4.1.8

Sonntag, 11. Januar 2015

Customize the Rails FormBuilder!

Tolerating as less as possible Ruby logic in the Ruby on Rails views is a basic pattern.
But instead of putting the logic into a Helpers method or Decorator class, moving it into a customized FormBuilder is worth a consideration.
Especially when it:
  1. is tied to a form
  2. is repeated
  3. generates tags (or even multi tag widgets)
Lots of Ruby on Rails views suffer from Ruby blocks with only logic inside (generating the over and over same HTML tag structure) or iterations like:
<% form_for @recipe do |f| %>
  <% (1..10).each do |rating| %>
    <%= f.radio_button :rating, rating %>
    <%= f.label "rating_#{rating}", rating %>
  <% end %>
<% end %>
What if the view could be refactored to:
<% form_for @recipe, builder: Forms::CollectionFormBuilder do |f| %>
  <%= f.labeled_radio_button_group 1..10, :rating %>
<% end %>
Please note the defined option builder:, pointing to the customized FormBuilder.
The better readability is obvious, aside from the less coding. It can be achieved by creating a new class (collection_form_builder.rb) in lib/forms:
module Forms
  class CollectionFormBuilder < ActionView::Helpers::FormBuilder
    def labeled_radio_button_group collection, method, options={}
      collection.inject(''.html_safe) { |html, value|
        checked = object.send(method).eql? value
        options[:id] = "#{method}_#{value}"
        html += @template.radio_button_tag("#{@object_name}[#{method}]", value, checked, options) + 
        @template.label_tag(options[:id], value)
      }   
    end 
  end 
end
by inheriting from ActionView::Helpers::FormBuilder. Thus some instance variables stated by FormBuilder#form_for are available:
  1. @object (the object assigned to form_for itself, like @recipe)
  2. @object_name (the objects name, like "recipe")
  3. @template (the current view (an instance of ActionView::Base); this object provides all methods available in your view)
  4. @options (the options assigned to form_for)
  5. @proc (the block assigned to form_for)
The @template object is coupled with concat (for output) and @proc (which provides the binding from your view). And it provides access to all standard ActionView::Base helper methods, including the custom helper methods (e.g. defined in ApplicationHelper). Furthermore partials can be rendered with @template:
@template.render partial: "fancy_widget", 
                 locals:  { object: @object }
Besides overwriting existing FormBuilder methods can make sense:
module Forms
  class LabeledFormBuilder < ActionView::Helpers::FormBuilder
    def check_box method, options={}, checked_value="1", unchecked_value="0"
      text = options.delete :text
      @template.content_tag(:label) {
        super(method, options, checked_value, unchecked_value) +
        (text or checked_value).to_s.html_safe
      }
    end
  end 
end
which generates the typical check box tag embraced by a label tag:
<% form_for @recipe, builder: Forms::LabeledFormBuilder do |f| %>
  <%= f.check_box :published, text: 'Published', id: nil %>
<% end %>
resulting in HTML:

Further articles of interest:

Supported by Ruby 2.1.1 and Ruby on Rails 4.1.8

Sonntag, 4. Januar 2015

Splat in Ruby method parameters!

Some methods have to deal with collections. The method definition for assigning a collection should not look like:
class Recipe
  def initialize ingredients=[]
    @ingredients = ingredients
  end
end
because sending the message then is as awful as:
Recipe.new ['Banana', 'Chocolate', 'Almonds']
or even worse:
Recipe.new ['Banana']
But the splat (*) operator in Ruby can solve it. The code can be refactored to:
class Recipe
  def initialize *ingredients
    @ingredients = ingredients.flatten.sort
  end

  def ingredients
    @ingredients.join(', ')
  end
end
and sending the message again:
recipe = Recipe.new 'Banana', 'Chocolate', 'Almonds'
recipe.ingredients
=> "Almonds, Banana, Chocolate"
or assigning just one object:
Recipe.new 'Banana'
Even assigning an Array works, because Array#flatten flattens the collection into an one dimensional array:
ingredients = %w(Banana Chocolate Almonds)
recipe = Recipe.new ingredients
=> #<Recipe:0x00000003de66a8 @ingredients=["Almonds", "Banana", "Chocolate"]>
In general the splat (*) operator can be used to split an Array:
one, two, three = *[1, 2, 3]
one
=> 1
two
=> 2
three
=> 3
and to collect an Array:
*numbers = 1, 2, 3
numbers
=> [1, 2, 3]
With the help of the Ruby splat (*) operator method parameters can be variable from zero up to many and handled as an Array internally. No archaic Array assignments anymore. And flatten the collection parameter with Array#flatten increases the flexibility of the API method.
Further articles of interest:

Supported by Ruby 2.1.1