Duplicate of a question I’ve recently asked on stackoverflow:
Consider the Ruby class Foo::Bar.
The convention is for the ‘Foo’ namespace to be a module, but it could
just as easily be a class:
module Foo; class Bar; end; end
Versus:
class Foo; class Bar; end; end
In the second case, Bar is not an inner class of Foo, it’s just
another constant defined on the singleton of Foo. The superclass is
Object in both cases and they only include the Kernel module. Their
ancestor chains are identical.
So, besides operations you could make with Foo depending on its class
(instantiate if a class, extend/include if a module), does the nature of
the namespace have any effect on Bar? Are there compelling reasons to
choose one many of name-spacing over another?
The only sort of weird things I see you can do are Foo::Bar.new.extend Foo and Foo.new.class::Bar respectively.
My own most-common use of a class defined in a class would be a helper
struct/class that is meant to be only used internally by the class:
class Foo
Bar = Struct.new(:something) do
def digest
puts "eating #{something}"
end
end
def eat(something)
Bar.new(something).digest
end
end
An answer there suggests the benefit of using modules as namespaces is
that ability to mix them into to other code and gain the ability to use
Constants defined in the mixed-in module without the namespace. To me,
that’s an edge case of a library that is meant to be mixed in. e.g of
all the Net:: libraries in ruby’s standard library, does anyone mix-in
Net? https://github.com/ruby/ruby/tree/trunk/lib/net