I'm no expert but this is how I would use modules instead of IoC container.

gistfile1.rb

class Nails
  def to_s
    "nails"
  end
end

class Glue
  def to_s
    "glue"
  end
end

class PowerSaw
  def initialize(power_source)
    @power_source = power_source
  end

  def to_s
    "#{@power_source} power saw"
  end
end

class HandSaw
  def to_s
    "hand saw"
  end
end

class Electricity
  def to_s
    "electrical"
  end
end

class Carpenter
  def build_something
    puts "Building something in wood with #{fixings} and #{tool}"
  end
end

module ManualToolsModule
  def tool
    HandSaw.new
  end

  def fixings
    Glue.new
  end
end 

module PowerToolsModule
  def power_source
    Electricity.new
  end

  def tool
    PowerSaw.new power_source
  end

  def fixings
    Nails.new
  end
end 

# reopening the class to mix the module in
class Carpenter
  include ManualToolsModule
end

# or (alt registration which replaces first registration)
class Carpenter
  include PowerToolsModule
end

Carpenter.new.build_something

Andy I'm relatively new to ruby, so these are not words of a seasoned rubyist.

1. Monkey patching does feel dirty and yea can be seen as ocp violation in this example.
2. Dynamic languages are simple, in that if the dependency is created elsewhere I.e. no .new in the consumer then you are decoupled and are bound only by the methods and attrs consumed on the dependency as an interface.
3. So as per point 2, you are decoupled from creation and you can swap in alternate dependencies as long as they quack like a duck. This includes mocks and stubs for testing.

I think everything is loosely coupled by the dynamic nature and the new instance creation is nowhere near the consumer. The module itself in this example could be refined, each method defined within simple acting like a factory method that you might see in the IoC container. The part that feels dirty is the monkey patching at a class level, where it would be nice to do this upon the instantiation on a per instance basis.

I still feel there is some value to be had from an IOC container and am not convinced otherwise myself. I'd be eager to work with you on this experiment as I too find it interesting, to see what the most idiomatic way would be to achieve your goals. Instance Lifetime management outside of the class is something I miss in ruby.

IoC + param injection is a pattern which really helps keep classes clean, tiny and single purposes, but 'ruby is like play dough' so by creating new instances outside of a container, favouring poor mans DI is no big deal as testing in isolation is totally possible by replacing a Class with a mock or stub version. This always bites you in the ass for testability in staticly typed languages. I think this is the major argument agains the need for a container.

I totally get what you are saying here. I guess it's a mindset thing to some degree. Are classes loosely coupled in a dynamic language by default because it's easier to change the behaviour compared to a static language.

I wonder if IoC is more popular in status languages because it is the only way to get nice loosely coupled classes whereas in dynamic languages there are other options that are easy and so people take that approach even if these methods are not strictly IoC/loosely coupled as seen by a purest.

Just a word on testability. For me, being able to mock/test easily is not the goal of IoC but a nice side affect of doing it. Obviously in ruby we can mock anything using the language features but my feeling is that we do this because it's the easy way rather than the "right" way.