Why do we return void for initializers?

[aaronmallen]

Hello I was curious why the documentation repeatedly returns void for initializers instead of instance

[ParadoxV5]

When you instantiate a class, you call the class method new (e.g., User.new).
Class#new is the one that returns instance.

The initializer we typically override is the instance method initialize, which is merely a callback Class#new triggers.
initializes seldom explicitly return self – think about all the def initialize blocks you’ve written.

#initialize is often overridden, but specifying a return type can be inconvenient for type checking.
⸺ #2008

[ksss]

One reason is that initialize is a private method, and code should not depend on the return value of private methods.

---

Another reason concerns issues when defining the initialize method in a subclass.

For example, before the change in #2008, the definition of BasicObject was as follows:

# rbs
class BasicObject
  def initialize: () -> nil
end

Now, suppose you define a class that inherits from BasicObject:

# rb
class Foo < BasicObject
  def initialize
    @foo = :default
  end
end

This code results in a type error because the return types of initialize (nil vs. Symbol) conflict. There are two ways to fix this:

1. Return nil.

# rb
class Foo < BasicObject
  def initialize
    @foo = :default
    nil
  end
end

2. Define the type for Foo#initialize.

# rbs
class Foo < BasicObject
  def initialize: () -> Symbol
end

I don’t find either solution practical. Therefore, I believe the return type of initialize should be untyped or void.

To more actively convey the intent that the return type should not be used, I specify void.

This idea has now been implemented as a Cop in RuboCop and is being actively used.

RBS/Style/InitializeReturnType