Modulation - Explicit Dependency Management for Ruby

Modulation | mɒdjʊˈleɪʃ(ə)n | Music - a change from one key to another in a piece of music.

INSTALL | GUIDE | EXAMPLES | RDOC

Modulation provides an alternative way of organizing your Ruby code. Modulation lets you explicitly import and export declarations in order to better control dependencies in your codebase. Modulation helps you refrain from littering the global namespace with a myriad modules, or complex multi-level nested module hierarchies.

Using Modulation, you will always be able to tell where a class or module comes from, and you'll have full control over which parts of a module's code you wish to expose to the outside world. Modulation can also help you write Ruby code in a functional style, minimizing boilerplate code.

Note: Modulation is not a replacement for RubyGems. Rather, Modulation is intended for managing dependencies between source files inside your Ruby applications. Though it does support loading gems that were written using Modulation, it is not intended as a comprehensive solution for using third-party libraries.

Rationale

You're probably asking yourself "what the ****?" , but when your Ruby app grows and is split into multiple files loaded using #require, you'll soon hit some issues:

  • Once a file is #required, any class, module or constant in it is available to any other file in your codebase. All "globals" (classes, modules, constants) are loaded, well, globally, in a single namespace. Name conflicts are easy in Ruby.
  • To avoid class name conflicts, classes need to be nested under a single hierarchical tree, sometime reaching 4 levels or more. Just look at Rails.
  • Since a #required class or module can be loaded in any file and then made available to all files, it's easy to lose track of where it was loaded, and where it is used.
  • There's no easy way to hide implementation-specific classes or methods. Yes, there's #private, #private_constant etc, but by default everything is #public!
  • Writing reusable functional code requires wrapping it in modules using class << self, def self.foo ..., extend self or include Singleton (the pain of implementing singletons in Ruby has been discussed before.)

There's a recent discussion on the Ruby bug tracker regarding possible solutions to the problem of top-level name collision. Hopefully, the present gem could contribute to an eventual "official" API.

Personally, I have found that managing dependencies with #require in large codebases is... not as elegant or painfree as I would expect from a first-class development environment. I also wanted to have a better solution for writing in a functional style.

So I came up with Modulation, a small gem (less than 300 LOC) that takes a different approach to organizing Ruby code: any so-called global declarations are hidden unless explicitly exported, and the global namespace remains clutter-free. All dependencies between source files are explicit, visible, and easy to understand.

Features

  • Provides complete isolation of each module: constant definitions in one file do not leak into another.
  • Enforces explicit exporting and importing of methods, classes, modules and constants.
  • Supports circular dependencies.
  • Supports default exports for modules exporting a single class or value.
  • Modules can be lazy loaded to improve start up time and memory consumption.
  • Modules can be reloaded at runtime without breaking your code in wierd ways.
  • Allows mocking of dependencies for testing purposes.
  • Can be used to write gems.
  • Facilitates unit-testing of private methods and constants.
  • Can load all source files in directory at once.

Installing Modulation

You can install the Modulation as a gem, or add it in your Gemfile:

$ gem install modulation

Organizing your code with Modulation

Modulation builds on the idea of a Ruby Module as a "collection of methods and constants". Using modulation, each Ruby source file becomes a module. Modules usually export method and constant declarations (usually an API for a specific, well-defined functionality) to be shared with other modules. Modules can also import declarations from other modules. Anything not exported remains hidden inside the module and normally cannot be accessed from the outside.

Each source file is evaluated in the context of a newly-created Module instance, with some additional methods for introspection and miscellaneous operations such as hot reloading.

Exporting declarations

Any class, module or constant be exported using #export:

export :User, :Session

class User
...
end

class Session
...
end

A module may also expose a set of methods without using class << self, for example when writing in a functional style:

seq.rb

export :fib, :luc

def fib(n)
  (0..1).include?(n) ? n : (fib(n - 1) + fib(n - 2))
end

def luc(n)
  (0..1).include?(n) ? (2 - n) : (luc(n - 1) + luc(n - 2))
end

app.rb

require 'modulation'
Seq = import('./seq')
puts Seq.fib(10)

Importing declarations

Declarations from another module can be imported using #import:

require 'modulation'
Models = import('./models')
...

user = Models::User.new(...)

...

Alternatively, a module interested in a single declaration from another module can use the following technique:

require 'modulation'
User = import('./models')::User
...

user = User.new(...)

Note about paths: module paths are always relative to the file calling the #import method, just like #require_relative.

Importing all source files in a directory

To load all source files in a directory you can use #import_all:

import_all('./ext') # will load ./ext/kernel.rb, ./ext/socket.rb etc

Groups of modules providing a uniform interface can also be loaded using #import_map:

API = import_map('./math_api') #=> hash mapping filenames to modules
API.keys #=> ['add', 'mul', 'sub', 'div']
API['add'] #=> add module

The #import_map takes an optional block to transform hash keys:

API = import_map('./math_api') { |name, mod| name.to_sym }
API.keys #=> [:add, :mul, :sub, :div]
API[:add] #=> add module

Importing methods into classes and modules

Modulation provides the #extend_from and #include_from methods to include imported methods in classes and modules:

module Sequences
  extend_from('./seq.rb')
end

Sequences.fib(5)

# extend integers
require 'modulation'
class Integer
  include_from('./seq.rb')

  def seq(kind)
    send(kind, self)
  end
end

5.seq(:fib)

The #include_from method accepts an optional list of symbols to import:

class Integer
  include_from './seq.rb', :fib
end

5.fib

Default exports

A module may wish to expose just a single class or constant, in which case it can use #export_default:

user.rb

export_default :User

class User
  ...
end

app.rb

require 'modulation'
User = import('./user')
User.new(...)

The default exported value can also be defined directly thus:

config.rb 

export_default(
  host: 'localhost',
  port: 1234,
  ...
)

app.rb

require 'modulation'
config = import('./config')
db.connect(config[:host], config[:port])

Accessing a module's root namespace from nested modules within itself

The special constant MODULE allows you to access the containing module from nested modules or classes. This lets you call methods defined in the module's root namespace, or otherwise introspect the module:

export :AsyncServer

# Await a promise-like callable
def await
  calling_fiber = Fiber.current
  p = ->(v = nil) {calling_fiber.resume v}
  yield p
  Fiber.yield
end

class AsyncServer < SomeTCPServer
  def async_read
    MODULE.await {|p| on_read {|data| p.(data)}}
  end
end

Accessing the global namespace

If you need to access the global namespace inside a module just prefix the class name with double colons:

class ::GlobalClass
  ...
end

::ENV = { ... }

what_is = ::THE_MEANING_OF_LIFE

Unit testing modules

Methods and constants that are not exported can be tested using the #__expose! method. Thus you can keep implementation details hidden, while being able to easily test them:

parser.rb

export :parse

def parse(inp)
  split(inp).map(&:to_sym)
end

# private method
def split(inp)
  inp.split(',').map(&:strip)
end

test_seq.rb

require 'modulation'
require 'minitest/autorun'

Parser = import('../lib/parser').__expose!

class FibTest < Minitest::Test
  def test_that_split_trims_split_parts
    assert_equal(%w[abc def ghi], Parser.split(' abc ,def , ghi  '))
  end
end

Mocking dependencies

Modules loaded by Modulation can be easily mocked when running tests or specs, using Modulation.mock:

require 'minitest/autorun'
require 'modulation'

module MockStorage
  extend self

  def get_user(user_id)
    {
      user_id: user_id,
      name: 'John Doe',
      email: '[email protected]'
    }
  end
end

class UserControllerTest < Minitest::Test
  def test_user_storage
    Modulation.mock('../lib/storage', MockStorage) do
      controller = UserController.new
      ...
    end
  end
end

Lazy Loading

Modulation allows the use of lazy-loaded modules - loading of modules only once they're needed by the application, in similar fashion to Module#auto_load. To lazy load modules use the #auto_import method, which takes a constant name and a path:

export :foo

auto_import :BAR, './bar'

def foo
  # the bar module will only be loaded once this method is called
  MODULE::BAR
end

Lazy-loaded constants must always be qualified. When referring to a lazy-loaded constant from the module's top namespace, use the MODULE namespace, as shown above.

The #auto_import method can also take a hash mapping constant names to paths. This is especially useful when multiple concerns are grouped under a single namespace:

export_default :SuperNet

module SuperNet
  auto_import(
    HTTP1:      './http1',
    HTTP2:      './http2',
    WebSockets: './websockets'
  )
end

Reloading modules

Modules can be reloaded at run-time for easy hot code reloading:

require 'modulation'
SQL = import('./sql')
...
SQL.__reload!

Another way to reload modules is using Modulation.reload, which accepts a module or a filename:

require 'filewatcher'

FileWatcher.new(['lib']).watch do |fn, event|
  if(event == :changed)
    Modulation.reload(fn)
  end
end

When a module is reloaded, its entire content - constants and methods - will be replaced. That means that any code using that module could continue to use it without even being aware it was reloaded, providing its API has not changed.

Reloading of modules with default exports is also possible. Modulation will
extend the exported value with a #__reload! method. The value will need to be reassigned:

require 'modulation'
settings = import('settings')
...
settings = settings.__reload!

Writing gems using Modulation

Modulation can be used to write gems, providing fine-grained control over your gem's public APIs and letting you hide any implementation details. In order to allow loading a gem using either #require or #import, code your gem's main file normally, but add require 'modulation/gem' at the top, and export your gem's main namespace as a default export, e.g.:

require 'modulation/gem'

export_default :MyGem

module MyGem
  ...
  MyClass = import('my_gem/my_class')
  ...
end

Importing gems using Modulation

Gems written using modulation can also be loaded using #import. If modulation does not find the module specified by the given relative path, it will attempt to load a gem by the same name. It is also possible to load specific files inside modules by specifying a sub-path:

require 'modulation'
MyFeature = import 'my_gem/my_feature'

Note: Since there's not much of a point in #importing gems that do not use Modulation to export symbols, Modulation will refuse to import any gem that does not depend on Modulation.

Coding style recommendations

  • Import modules into constants, not variables:
  Settings = import('./settings')
  • Place your exports at the top of your module, followed by #requires, followed by #imports:
  export :foo, :bar, :baz

  require 'json'

  Core = import('./core')

  ...

Why you should not use Modulation

  • Modulation is not production-ready.
  • Modulation is not thread-safe.
  • Modulation doesn't play well with rdoc/yard.
  • Modulation (probably) doesn't play well with Marshal.
  • Modulation (probably) doesn't play well with code-analysis tools.