Avoid these traps when nesting Ruby modules
Modules (and classes) are meant to be nested. Code fragments like
ActiveRecord::RecordNotFound are so common that we don't think twice about them. But buried within Ruby's nesting implementation - and Rails' autoload system - are a few traps that can cause your code to fail in strange and wonderful ways. In this post, we'll discuss the origin of these traps and how you can avoid them.
What is a constant?
This post is about modules, but to understand those we need to understand constants. In most languages, constants are only used to store little bits of data, like in the example below:
# These are simple constants MAX_RETRIES = 5 DEFAULT_LANGUAGE = "en"
But in Ruby, classes and modules are also constants. I've written a little example to demonstrate this. There are three constants in the module below: a number, a class and a nested module. When you access a nested class or module, ruby finds it by using the same rules it would use with a simple numeric constant.
module MyModule MY_FAVORITE_NUMBER = 7 # Classes are constants class MyClass end # So are modules module MyEmbeddedModule end end puts MyModule.constants.inspect # => [:MY_FAVORITE_NUMBER, :MyClass, :MyEmbeddedModule]
Often, modules have access to the constants defined in their parents. That's why you can write code like this:
module X MARCO = "polo" module Y def self.n puts MARCO end end end X::Y.n() # => "polo"
But you'd be wrong if you thought that this parent/child relationship was what allows Y to access X's constant MARCO.
A common problem
If we rewrite the code above in a slightly different way, something surprising happens. Y can no longer access X::MARCO. What the heck is going on here?
module A MARCO = "polo" end module A::B def self.n puts MARCO # => uninitialized constant A::B::MARCO (NameError) end end A::B.n()
It turns out that the "inheritance" of a parent's constants by the child isn't due to the parent/child relationship. It's lexical. That means that it's based on the structure of your code, not on the structure of the objects your code is building.
If you'd like to get a deeper understanding of how Ruby searches for nested constants, it's worth checking out the
This function returns an array of objects that make up the "search path" for constants in a given scope. Let's examine the nesting for our previous examples.
In our first example, we see that the nesting is
[A::B, A]. This means that if we use the constant MARCO, Ruby will look for it first in
A::B, and then in
module A MARCO = "polo" module B def self.n puts Module.nesting.inspect # => [A::B, A] puts MARCO # => "polo" end end end
In the second example, we see that the nesting only includes
A. Even though B is a "child" of
A, the way I've written the code doesn't show them as nested, so for this purpose they might as well not be.
module A MARCO = "polo" end module A::B def self.n puts Module.nesting.inspect # => [A::B] puts MARCO # => uninitialized constant A::B::MARCO (NameError) end end
Rails autoload complications
Have you ever noticed that you don't have to include files when you use Rails? If you want to use a model, you just use it.
This is possible because Rails implements an autoload system. It uses
Module.const_missing to detect when you try to reference a constant that hasn't been loaded. It then loads the files it believes should contain the constant. This works most of the time, but there's a catch.
Rails assumes that a module always has the largest possible nesting. It assumes that module A::B::C will have a nesting of
[A::B::C, A::B, A]. If it doesn't you'll get unexpected behavior.
In the code below, module B shouldn't be able to access
A::MARCO. In normal Ruby, it wouldn't be able to because its nesting is [A::B]. So you should get an exception. But Rails' autoload doesn't throw an exception. Instead, it returns
# a.rb module A MARCO = "polo" end # a/b.rb module A::B def self.n puts MARCO # => "polo" end end # some_controller.rb A::B.n()
All of this is a lot to think about. I prefer to avoid thinking where possible, so I try to stay away from the module
A::B syntax. I can't think of a case where I would intentionally want to manipulate the module nesting. If you know of any, I'd love to hear about them!