Skip to content


Subversion checkout URL

You can clone with
Download ZIP
tree: 592dd89751
Fetching contributors…

Cannot retrieve contributors at this time

1073 lines (613 sloc) 49.207 kB

This is a detailed guide to the internal workings of Sprockets. Hopefully with this information, somebody else besides Josh Peek, Sam Stephenson, Yehuda Katz and (partially) myself can begin to understand how Sprockets works.

Sprockets Initialization

To first understand Sprockets, we must first understand how it hooks into Rails. Just like with the Active Record and Action Pack components of Rails, Sprockets too has its own railtie. This is included by the require "rails/all" call in config/application.rb or if you don't have that then it must be required explicitly with require "sprockets/railtie". This railtie is actually kept inside the actionpack gem itself, rather than the sprockets gem.

The first thing this file does is define the Sprockets module and then inside that defines three autoload'd modules.

rails: actionpack/lib/sprockets/railtie.rb, 4 lines, beginning line 1

module Sprockets
  autoload :Helpers, "sprockets/helpers"
  autoload :LazyCompressor, "sprockets/compressors"
  autoload :NullCompressor, "sprockets/compressors"

This file then sets up the normal Railtie stuff, such as configuration:

rails: actionpack/lib/sprockets/railtie.rb, 2 lines, beginning line 7

class Railtie < ::Rails::Railtie
  config.default_asset_host_protocol = :relative

This configuration option is caught by the config options method_missing which will then store this setting in the configuration hash for the application. It is retreivable by calling simply Rails.application.config.default_asset_host_protocol again.

Next, this Railtie defines some rake tasks:

rails: actionpack/lib/sprockets/railtie.rb, 3 lines, beginning line 10

rake_tasks do
  load "sprockets/assets.rake"

This sprockets/assets.rake file provides the assets:precompile and assets:clean Rake tasks we will see later on in this internals guide.

After that, the Railtie defines an initializer, which will be appended onto the list of initializers currently at this point of Rails. Currently, this initializer will run directly after ActiveResource's initializers have run. Inside this initializer, there's a check first if Rails should even bother with loading sprockets at all:

rails: actionpack/lib/sprockets/railtie.rb, 2 lines, beginning line 15

config = app.config
next unless config.assets.enabled

This is the Rails.application.config.assets.enabled flag. If it is set to a non-truthy value then this initializer will stop right there and then. By default though, it's enabled and so it will continue.

After this point, then sprockets is required and a new Sprockets::Environment object is set up:

rails: actionpack/lib/sprockets/railtie.rb, 3 lines, beginning line 18

require 'sprockets'

app.assets = do |env|

The initialize method in Sprockets::Environment will receive the application's root. The initialize method in Sprockets::Environment is quite long and sets up a lot of the functionality. It begins by creating a new Hike::Trail object out of the root of the application that was passed in and setting a default logger:

sprockets: lib/sprockets/environment.rb, 5 lines, beginning line 20

def initialize(root = ".")
  @trail =

  self.logger =$stderr)
  self.logger.level = Logger::FATAL

This then sets up a new class based of the context class, defines a digest class and defaults the versioning to a blank string:

sprockets: lib/sprockets/environment.rb, 6 lines, beginning line 27

@context_class =

# Set MD5 as the default digest
require 'digest/md5'
@digest_class = ::Digest::MD5
@version = ''

This context class is used by the bundled asset feature in sprockets, which we'll see in greater detail later in this guide. The @digest_class variable is used to determine what digest class should be used to provide unique identifiers for precompiled assets, such as those generated with rake assets:precompile. Finally, @version will be used to provide a manually overridable string for the assets versions. We can change this in config.assets.version in config/application.rb to expire all our assets manually.

Next, the initialize method sets up more default values:

sprockets: lib/sprockets/environment.rb, 5 lines, beginning line 34

@mime_types        = {}
@engines           = Sprockets.engines
@preprocessors     = { |h, k| h[k] = [] }
@postprocessors    = { |h, k| h[k] = [] }
@bundle_processors = { |h, k| h[k] = [] }

We'll see what the mime types, pre-processors, post-processors and bundle processors are in just a bit, but first let's see what Sprockets.engines is. This method is defined in lib/sprockets/engines.rb which is loaded with the lib/sprockets.rb file that was required by the Railtie. The lib/sprockets/engines.rb file defines the Sprockets::Engines module and defines the engines method like this:

sprockets: lib/sprockets/engines.rb, 8 lines, beginning line 41

def engines(ext = nil)
  if ext
    ext = Sprockets::Utils.normalize_extension(ext)

When this method is called with no arguments, like in Sprockets::Environment's initialize method, it will return a duplicated object of the engines currently registered with Sprockets. These engines are the templating engines that Sprockets will make use of in the asset pipeline. These should not be confused with the "engines" that Rails has. The ones for Sprockets are templating engines, where the ones in Rails are more like miniature applications.

Now, it may seem like that there are no engines registered with Sprockets at the moment, there actually is. At the bottom of the lib/sprockets/engines.rb file, the Sprockets module is extended by the Engines module contained within (this is how the engines method is then provided for Sprockets) and then these default engines are registered:

sprockets: lib/sprockets/engines.rb, 22 lines, beginning line 76

  extend Engines
  @engines = {}

  # Cherry pick the default Tilt engines that make sense for
  # Sprockets. We don't need ones that only generate html like HAML.

  # Mmm, CoffeeScript
  register_engine '.coffee', Tilt::CoffeeScriptTemplate

  # JST engines
  register_engine '.jst',    JstProcessor
  register_engine '.eco',    EcoTemplate
  register_engine '.ejs',    EjsTemplate

  # CSS engines
  register_engine '.less',   Tilt::LessTemplate
  register_engine '.sass',   Tilt::SassTemplate
  register_engine '.scss',   Tilt::ScssTemplate

  # Other
  register_engine '.erb',    Tilt::ERBTemplate
  register_engine '.str',    Tilt::StringTemplate

All of these engines inherit from Tilt::Template which will be used to render these templates. The register_engine method is also defined within the lib/sprockets/engines.rb file and goes like this:

sprockets: lib/sprockets/engines.rb, 4 lines, beginning line 63

def register_engine(ext, klass)
  ext = Sprockets::Utils.normalize_extension(ext)
  @engines[ext] = klass

This method calls Sprockets::Utils.normalize_extension to, uh, normalize the extension by doing this:

sprockets: lib/sprockets/utils.rb, 8 lines, beginning line 58

def self.normalize_extension(extension)
  extension = extension.to_s
  if extension[/^\./]

This way, you can call register_engine and pass it an extension for that template with or without the dot prefix and this method will prefix the dot. Once normalize_extension has done its thing, then register_engine finishes by adding a new key to the @engines hash with the new extension and the specified class.

Going back to lib/sprockets/environment.rb now, and the next thing that happens is that these engines are added to the trail:

sprockets: lib/sprockets/environment.rb, 3 lines, beginning line 40

@engines.each do |ext, klass|
  add_engine_to_trail(ext, klass)

The add_engine_to_trail method is defined in lib/sprockets/processing.rb beginning like this:

sprockets: lib/sprockets/processing.rb, 2 lines, beginning line 270

def add_engine_to_trail(ext, klass)

The @trail object was originally set up earlier in the initialize method for our new Sprockets::Environment object, and is a Hike::Trail object. Therefore, this append_extension method is defined within Hike and not Sprockets. It is defined within lib/hike/trail.rb very simply:

hike: lib/hike/trail.rb, 4 lines, beginning line 84

def append_extensions(*extensions)
alias_method :append_extension, :append_extensions

This is so that Hike will be able to find files with the given extensions when they are searched for later on in this process.o

Now that we know what the Sprockets.engines method does, we've still got the remainder of the initialize method for Sprockets::Environment to figure out. The next two lines in this method register mime types for CSS and JavaScript:

sprockets: lib/sprockets/environment.rb, 2 lines, beginning line 44

register_mime_type 'text/css', '.css'
register_mime_type 'application/javascript', '.js'

This method works very similarly to the register_engine method we saw earlier, which was defined in lib/sprockets/engines.rb'. Theregister_mime_typemethod is defined inlib/sprockets/mime.rb` like this:

sprockets: lib/sprockets/mime.rb, 4 lines, beginning line 29

def register_mime_type(mime_type, ext)
  ext = Sprockets::Utils.normalize_extension(ext)
  @mime_types[ext] = mime_type

This calls normalize_extension again which will of course prefix the extension with a dot if it didn't have one. In this case though, there are dots. A new key is added to the @mime_types hash with this new extension with the mime_type being its value.

Next in the Sprockets::Environment#initialize method, the register_preprocessor method is called:

sprockets: lib/sprockets/environment.rb, 2 lines, beginning line 47

register_preprocessor 'text/css', DirectiveProcessor
register_preprocessor 'application/javascript', DirectiveProcessor

The DirectiveProcessor is the class that is responsible for parsing the *= require 'blah' and //= require 'other_blah' directives in our JavaScript and CSS files. We'll see the inner workings of this when we are going through the process of serving an asset.

The register_preprocessor method is a little more complicated than the register_engine and register_mime_types method, and it is defined within lib/sprockets/processing.rb:

sprockets: lib/sprockets/processing.rb, 13 lines, beginning line 90

def register_preprocessor(mime_type, klass, &block)

  if block_given?
    name  = klass.to_s
    klass = do
      @name      = name
      @processor = block


First, the expire_index! method is called. This method is defined in lib/sprockets/environment.rb and does the following:

sprockets: lib/sprockets/environment.rb, 5 lines, beginning line 87

def expire_index!
  # Clear digest to be recomputed
  @digest = nil
  @assets = {}

This method ensures that our index is at a pristine state, where the digest has not yet been computed and there have been no assets served.

After the pre-processor has been registered, a single post-processor is registered:

sprockets: lib/sprockets/environment.rb, 1 lines, beginning line 50

register_postprocessor 'application/javascript', SafetyColons

This class is responsible for adding semi-colons to the end of each file before they are concatenated into a single bundle. Without this, it could lead to JavaScript syntax errors.

Next, a bundle processor is added for CSS files:

sprockets: lib/sprockets/environment.rb, 1 lines, beginning line 51

register_bundle_processor 'text/css', CharsetNormalizer

Bundle processors are run after the assets are concatenated. This one searches for @charset definitions in CSS files, keeps the first one and strips out the rest. Otherwise, multiple charset specifications will lead to undesired results. For more information, check out the comments on the Sprockets::CharsetNormalizer class.

After register_bundle_processor runs, expire_index! is run again just for good measure and the new object is yielded if block is given to this method, which it is.

sprockets: lib/sprockets/environment.rb, 3 lines, beginning line 53


yield self if block_given?

When yield is called, it will evaluate the block it is given using the code specified back in actionpack/lib/sprockets/railtie.rb:

rails: actionpack/lib/sprockets/railtie.rb, 8 lines, beginning line 20

app.assets = do |env|
  env.logger  = ::Rails.logger
  env.version = ::Rails.env + "-#{config.assets.version}"

  if config.assets.cache_store != false
    env.cache = ActiveSupport::Cache.lookup_store(config.assets.cache_store) || ::Rails.cache

This block takes the object that is given by yield and sets the logger to be the Rails.logger and the version to be a combination of the Rails environment's name and the config.assets.version setting, which is "1.0" by default in config/application.rb. Next, it sets up a cache for the assets (only if config.assets.cache_store is not exactly false), using the value specified in config.assets.cache_store or alternatively using the Rails.cache settings.

So as we can see here, by default the Sprockets environment will use the same logger and cache as the application itself, but we can configure these if we please.

That's the end of the "sprockets.environment" initializer now. The next thing the Railtie does is add a hook for when Action View is loaded using these lines:

rails: actionpack/lib/sprockets/railtie.rb, 7 lines, beginning line 29

ActiveSupport.on_load(:action_view) do
  include ::Sprockets::Helpers::RailsHelper

  app.assets.context_class.instance_eval do
    include ::Sprockets::Helpers::RailsHelper

The on_load method is used to add hooks for certain components used within a Rails application. In this usage, when Action View is loaded then the Sprockets::Helpers::RailsHelper module is included into ActionView::Base. Also inside this block, the environment's context_class is referenced (remember, it's a new anonymous class inheriting from Sprockets::Context) and the Sprockets::Helpers::RailsHelper module is included on that also.

Finally in the Railtie an after_initialize hook is defined. It begins like this:

rails: actionpack/lib/sprockets/railtie.rb, 5 lines, beginning line 42

config.after_initialize do |app|
  next unless app.assets
  config = app.config

  config.assets.paths.each { |path| app.assets.append_path(path) }

The after_initialize hook is skipped if app.assets is set to false, but by default this is not the case and so let's assume it's not. Next, there's a config variable setup so that the code doesn't have to make continual references to app.config and can get away with just writing config instead. Finally in the above example, the config.assets.paths collection is iterated through with each path being used in an append_path call on the app.assets object, which is the Sprockets::Environment object that was set up earlier.

The config.assets.paths are set up inside of Rails at railties/lib/rails/engine.rb using these lines:

rails: railties/lib/rails/engine.rb, 5 lines, beginning line 542

initializer :append_assets_path do |app|

When the Rails application is initialized, The asset directories inside of vendor/assets, lib/assets and app/assets paths will be added to config.assets.paths, but only if these directories exist.

The append_path method on instances of Sprockets::Environment is defined in lib/sprockets/trail.rb and does the following:

sprockets: lib/sprockets/trail.rb, 4 lines, beginning line 38

def append_path(path)

This calls the expire_index to clear the index, as it is potentially adding new assets, and then calls append_path on the @trail object, which is an instance of Hike::Trail. The append_path method for Hike::Trail is defined like this:

hike: lib/hike/trail.rb, 4 lines, beginning line 67

def append_paths(*paths)
alias_method :append_path, :append_paths

The self.paths in this case is an array of paths, and so push will just add those paths to the end of the array.

The next line in the Sprockets Railtie checks to see if the config.assets.compress setting is set to a truthy value. By default, it isn't in the development environment but is in the production environment.

rails: actionpack/lib/sprockets/railtie.rb, 11 lines, beginning line 48

if config.assets.compress
  # temporarily hardcode default JS compressor to uglify. Soon, it will work
  # the same as SCSS, where a default plugin sets the default.
  unless config.assets.js_compressor == false
    app.assets.js_compressor = { expand_js_compressor(config.assets.js_compressor || :uglifier) }

  unless config.assets.css_compressor == false
    app.assets.css_compressor = { expand_css_compressor(config.assets.css_compressor) }

If the setting is truthy then it will determine if there is a js_compressor setting or a css_compressor setting. For the js_compressor setting, the expand_js_compressor method is used, which is defined inside the railtie like this:

rails: actionpack/lib/sprockets/railtie.rb, 15 lines, beginning line 70

def expand_js_compressor(sym)
  case sym
  when :closure
    require 'closure-compiler'
  when :uglifier
    require 'uglifier'
  when :yui
    require 'yui/compressor'

The js_compressor setting takes a symbol which can be one of :closure, :uglifier or :yui, and then will require the required files and return the class used for compression. The gems that provide these files must be specified within the Gemfile of the application before they can be loaded.

The cs_compressor method is similar, except it only takes one value:

rails: actionpack/lib/sprockets/railtie.rb, 9 lines, beginning line 86

def expand_css_compressor(sym)
  case sym
  when :yui
    require 'yui/compressor'

The only currently supported CSS compressor is the YUI compressor. These compressor objects are then passed to inside a block. The purpose of this is to provide a way to compress content if a compressor is specified, or otherwise falback to a Sprockets::NullCompressor class which just returns the content as-is. We can see the code for this in actionpack/lib/sprockets/compressors.rb:

rails: actionpack/lib/sprockets/compressors.rb

module Sprockets
  class NullCompressor
    def compress(content)

  class LazyCompressor
    def initialize(&block)
      @block = block

    def compressor
      @compressor ||= ||

    def compress(content)

Next up in this after_initialize hook, the routes are prepended with a route to the assets using these lines:

rails: actionpack/lib/sprockets/railtie.rb, 3 lines, beginning line 60

app.routes.prepend do
  mount app.assets => config.assets.prefix

The routes object here is the same routes object normally returned by Rails.application.routes, famously used in declaring routes for an application in config/routes.rb. The append method on this object is a new feature of Rails 3.1 and just like its name implies it will append a set of routes to the beginning of the routing table. This means that the /assets route will now be the first route that is matched in the application. The app.assets object is the Sprockets::Environment object set up before, and the config.assets.prefix value defaults to /assets/.

Finally in the after_initialize block, there's a check to see if Rails is performing caching which does this:

rails: actionpack/lib/sprockets/railtie.rb, 3 lines, beginning line 64

if config.action_controller.perform_caching
  app.assets = app.assets.index

The index method on Sprockets::Environment does this:

sprockets: lib/sprockets/environment.rb, 3 lines, beginning line 63

def index

The Index class's purpose is to provide a cached version for the environment, which makes the calls to the file system much faster as it will be caching the location of the assets, rather than looking them up each time they are requested. Of course, in the development environment the perform_caching setting is set to false and so it's disabled, but in production it will be enabled.

And that's the Sprockets::Railtie class covered. This section has described how Sprockets attaches to Rails and provides the beginnings of the asset pipeline. One of the features we saw inside this Railtie was that it included modules into ActionView::Base. The purpose of this is to provide overrides for methods such as javascript_include_tag, image_tag and stylesheet_link_tag so that they use the asset pipeline, rather than the default Rails helpers which do not.

Let's take a look at these now.

Sprockets Asset Helpers

Within Rails 3.1, the behaviour of javascript_include_tag and stylesheet_link_tag are modified by the actionpack/lib/sprockets/rails_helper.rb file which is required by actionpack/lib/sprockets/railtie.rb, which itself is required by actionpack/lib/action_controller/railtie.rb and so on, and so forth.

The Sprockets::Helpers::RailsHelper is included into ActionView through the process described in my earlier Sprockets Railtie Setup internals doc. Once this is included, it will override the stylesheet_link_tag and javascript_include_tag methods originally provided by Rails itself. Of course, if assets were disabled (Rails.application.config.assets.enabled = false) then the original Rails methods would be used and JavaScript assets would then exist in public/javascripts, not app/assets/javascripts. Let's just assume that you're using Sprockets.

Let's take a look at the stylesheet_link_tag method from Sprockets::Helpers::RailsHelper. The javascript_include_tag method is very similar so if you want to know how that works, just replace stylesheet_link_tag with javascript_include_tag using your mind powers and I'm sure you can get the gist of it.

What stylesheet_link_tag / javascript_include_tag does

This method begins like this:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 4 lines, beginning line 37

def stylesheet_link_tag(*sources)
  options = sources.extract_options!
  debug = options.key?(:debug) ? options.delete(:debug) : debug_assets?
  body  = options.key?(:body)  ? options.delete(:body)  : false

The first argument for stylesheet_link_tag is a splatted sources which means that this method can take a list of stylesheets or manifest files and will process each of them. The method also takes some options, which are extracted out on the first line of this method using extract_options!. The two currently supported are debug and body.

The debug option will expand any manifest file into its contained parts and render each file individually. For example, in a project I have here, this line:

<%= stylesheet_link_tag "application" %>

When a request is made to this page that uses this layout that renders this file, it will be printed as a single line:

<link href="/assets/application.css" media="screen" rel="stylesheet" type="text/css"> 

Even though the file it's pointing to contains directives to Sprockets to include everything in app/assets/stylesheets:

*= require_self
*= require_tree .

What sprockets is doing here is reading this manifest file and compiling all the CSS assets specified into one big fuck-off file and serving just that instead of the *counts* 13 CSS files I've got in that directory.

This helper then iterates through the list of sources specified and first dives in to checking the debug option. If debug is set to true for this though, either through options[:debug] being passed or by debug_assets? evaluating to true, this will happen:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 5 lines, beginning line 42

sources.collect do |source|
  if debug &amp;&amp; asset = asset_paths.asset_for(source, 'css') { |dep|
      super(dep.to_s, { :href =&gt; asset_path(dep, 'css', true, :request) }.merge!(options))

The super method here will call the stylesheet_link_tag method defined in ActionView::Helpers::AssetTagHelper. This is the default stylesheet_link_tag method that would be called if we didn't have Sprockets enabled.

The debug_assets? method is defined as a private method further down in this file:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 7 lines, beginning line 59

def debug_assets?
  Rails.application.config.assets.allow_debugging &amp;&amp;
   (Rails.application.config.assets.debug ||
    params[:debug_assets] == '1' ||
    params[:debug_assets] == 'true')

If Rails.application.config.assets.allow_debugging is set to true and Rails.application.config.assets.debug is true or the debug_asset parameter in the request is either '1' or 'true' then the assets will be debugged. There may be a case where params doesn't exist, and so this method rescues a potential NoMethodError that could be thrown. Although I can't imagine a situation in Rails where that would ever be the case.

Back to the code within stylesheet_link_tag, this snippet will get all the assets specified in the manifest file, iterate over each of them and render a stylesheet_link_tag for each of them, ensuring that :debug is set to false for them.

It's important to note here that the CSS files that the original app/assets/stylesheets/application.css points to can each be their own manifest file, and so on and so forth.

If the debug option isn't specified and debug_assets? evaluates to false then the else for this if will be executed:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 3 lines, beginning line 47

  super(source.to_s, { :href =&gt; asset_path(source, 'css', body, :request) }.merge!(options))

This will render just the one line, rather than expanding the dependencies of the stylesheet. This calls the asset_path method which is defined like this:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 5 lines, beginning line 53

def asset_path(source, default_ext = nil, body = false, protocol = nil)
  source = source.logical_path if source.respond_to?(:logical_path)
  path = asset_paths.compute_public_path(source, 'assets', default_ext, true, protocol)
  body ? "#{path}?body=1" : path

(WIP: I don't know what logical_path is for, so let's skip over that for now. In my testing, source has always been a String object).

This method then calls out to asset_paths which is defined at the top of this file:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 11 lines, beginning line 9

def asset_paths
  @asset_paths ||= begin
    config     = self.config if respond_to?(:config)
    config   ||= Rails.application.config
    controller = self.controller if respond_to?(:controller)
    paths =, controller)
    paths.asset_environment = asset_environment
    paths.asset_prefix      = asset_prefix

This method (obviously) initializes a new instance of the RailsHelper::AssetPaths class defined later in this file, passing through the config and controller objects of the current content, which would be the same self.config and self.controller available within a view.

This RailsHelper::AssetPaths inherits behaviour from ActionView::AssetPaths, which is responsible for resolving the paths to the assets for vanilla Rails. The RailsHelper::AssetPaths overrides some of the methods defined within its superclass, though.

The asset_environment method is defined also in this file:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 3 lines, beginning line 80

def asset_environment

The assets method called inside the asset_environment method returns a Sprockets::Index instance, which we'll get to later.

The asset_prefix is defined just above this:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 3 lines, beginning line 73

def asset_prefix

The next method is compute_public_path which is called on this new RailsHelper::AssetPaths instance. This is defined simply:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 3 lines, beginning line 87

def compute_public_path(source, dir, ext=nil, include_host=true, protocol=nil)
  super(source, asset_prefix, ext, include_host, protocol)

This calls back to the compute_public_path within ActionView::AssetsPaths (actionpack/lib/action_view/asset_paths.rb) which is defined like this:

rails: actionpack/lib/action_view/asset_paths.rb, 20 lines, beginning line 14

# Add the extension +ext+ if not present. Return full or scheme-relative URLs otherwise untouched.
# Prefix with &lt;tt&gt;/dir/&lt;/tt&gt; if lacking a leading +/+. Account for relative URL
# roots. Rewrite the asset path for cache-busting asset ids. Include
# asset host, if configured, with the correct request protocol.
# When include_host is true and the asset host does not specify the protocol
# the protocol parameter specifies how the protocol will be added.
# When :relative (default), the protocol will be determined by the client using current protocol
# When :request, the protocol will be the request protocol
# Otherwise, the protocol is used (E.g. :http, :https, etc)
def compute_public_path(source, dir, ext = nil, include_host = true, protocol = nil)
  source = source.to_s
  return source if is_uri?(source)

  source = rewrite_extension(source, dir, ext) if ext
  source = rewrite_asset_path(source, dir)
  source = rewrite_relative_url_root(source, relative_url_root) if has_request?
  source = rewrite_host_and_protocol(source, protocol) if include_host

This method, unlike those in Sprockets, actually has decent documentation.

In this case, let's keep in mind that source is going to still be the "application" string from stylesheet_link_tag rather than a uri. The conditions for matching a uri are in the is_uri? method also defined in this file:

rails: actionpack/lib/action_view/asset_paths.rb, 3 lines, beginning line 41

def is_uri?(path)
  path =~ %r{^[-a-z]+://|^cid:|^//}

Basically, if the path matches a URI-like fragment then it's a URI. Who would have thought? `"application" is quite clearly NOT a URI and so this will continue to the rewrite_extension method.

The rewrite_extension method is actually overridden in Sprockets::Helpers::RailsHelpers::AssetPaths like this:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 7 lines, beginning line 122

def rewrite_extension(source, dir, ext)
  if ext &amp;&amp; File.extname(source).empty?

This method simply appends the correct extension to the end of the file (in this case, ext is set to "css" back in stylesheet_link_tag) if it doesn't have one already. If it does, then the filename will be left as-is. The source would now be "application.css".

Next, the rewrite_asset_path is used and this method is also overridden in Sprockets::Helpers::RailsHelpers::AssetPaths:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 10 lines, beginning line 111

def rewrite_asset_path(source, dir)
  if source[0] == ?/
    source = digest_for(source) if performing_caching?
    source = File.join(dir, source)
    source = "/#{source}" unless source =~ /^\//

If the source argument (now "application.css", remember?) begins with a forward slash, it's returned as-is. If it doesn't, then the digest_for method is called, but only if performing_caching? evaluates to true. This is determined like this:

rails: actionpack/lib/sprockets/helpers/rails_helper.rb, 3 lines, beginning line 131

def performing_caching?
  config.action_controller.present? ? config.action_controller.perform_caching : config.perform_caching

In the development environment, the config.action_controller.perform_caching value is set to false by default and so this digest_for line will not be run. The rewrite_asset_path method then joins the dir and source together to get a string such as "assets/application.css" which then has a forward slash prefixed to it by the next line of code.

This return value then bubbles its way back up through compute_public_path to asset_path and finally back to the stylesheet_link_tag method where it's then specified as the href to the link tag that it renders, with help from the stylesheet_link_tag from ActionView::Helpers::AssetTagHelper.

And that, my friends, is all that is involved when you call stylesheet_link_tag within the development environment. Now let's look at what happens when this file sis requested.

Asset Request Cycle

When an asset is requested in Sprockets it hits the small Rack application that sprockets has. This Rack application is mounted inside the config.after_initialize block inside the Sprockets::Railtie which is in actionpack/lib/sprockets/railtie.rb, using these three lines:

rails: actionpack/lib/sprockets/railtie.rb, 3 lines, beginning line 60

app.routes.prepend do
  mount app.assets =&gt; config.assets.prefix

The app object here is the same object we would get back if we used Rails.application, which would be an instance of our application class that inherits from Rails::Application. By calling .routes.prepend on that object, this Railtie places a new set of routes right at the top of our application's routes. In this case, it's just the one route which is mounting the app.assets object (a Sprockets::Index object) at config.assets.prefix, which by default is /assets.

This means that any request going to /assets will hit this Sprockets::Index object and invoke a call method on it. The Sprockets::Index class is fairly bare itself and doesn't define its own call method, but it inherits a lot of behaviour from Sprockets::Base. The Sprockets::Base class itself doesn't define a call method for it's instances either. However, when the Sprockets::Base is declared it includes a couple of modules:

sprockets: lib/sprockets/base.rb, 5 lines, beginning line 11

module Sprockets
  # `Base` class for `Environment` and `Index`.
  class Base
    include Digest
    include Caching, Processing, Server, Trail

It's the Server module here that provides this call method, which is defined within lib/sprockets/server.rb, beginning with these lines:

sprockets: lib/sprockets/server.rb, 5 lines, beginning line 22

def call(env)
  start_time =
  time_elapsed = lambda { (( - start_time) * 1000).to_i }

  msg = "Served asset #{env['PATH_INFO']} -"

This method accepts an env argument which is a Hash which represents the current Rack environment of the application, containing things such as headers set by previous pieces of Middleware as well as things such as the current request path, which is stored in ENV['PATH_INFO'].

These few lines define the methodology that this method uses to work out how long an asset has taken to compile. The final line in the above example is the beginning of the output that Sprockets will put into the Rails log once it is done.

Next, Sprockets checks for a forbidden request using these lines:

sprockets: lib/sprockets/server.rb, 4 lines, beginning line 28

# URLs containing a `".."` are rejected for security reasons.
if forbidden_request?(env)
  return forbidden_response

The comment describes acurrately enough what this method does, if the path contains ".." then it returns a forbidden_response. First, let's just see the simple code for forbidden_request?

sprockets: lib/sprockets/server.rb, 7 lines, beginning line 126

def forbidden_request?(env)
  # Prevent access to files elsewhere on the file system

The env["PATH_INFO"] method here is the request path that is requested from Sprockets, which would be /application.css at this point in time. If that path were to include two dots in a row, this forbidden_request? method would return true and the forbidden_response method would be called. The forbidden_response method looks like this:

sprockets: lib/sprockets/server.rb, 4 lines, beginning line 134

# Returns a 403 Forbidden response tuple
def forbidden_response
  [ 403, { "Content-Type" =&gt; "text/plain", "Content-Length" =&gt; "9" }, [ "Forbidden" ] ]

This response object is a standard three-part tuple that Rack expects, containing the HTTP status code first, a Hash of headers to present and finally an Array containing a String which represents the content for this response.

In this case, our request is /application.css and therefore will not trigger this forbidden_response to be called, falling to the next few of lines of this call method:

sprockets: lib/sprockets/server.rb, 4 lines, beginning line 33

# Mark session as "skipped" so no `Set-Cookie` header is set
env['rack.session.options'] ||= {}
env['rack.session.options'][:defer] = true
env['rack.session.options'][:skip] = true

In the case of sprockets, it does not care so much about the session information for a user, and so this is deferred and skipped with these lines.

Next, Sprockets gets to actually trying to find the asset that has been requested:

sprockets: lib/sprockets/server.rb, 6 lines, beginning line 38

# Extract the path from everything after the leading slash
path = env['PATH_INFO'].to_s.sub(/^\//, '')

# Look up the asset.
asset = find_asset(path)
asset.to_a if asset

At the beginning of this, Sprockets removes the trailing slash from the beginning of /application.css, turning it into just application.css. This path is then passed to the find_asset method, which should find our asset, if it exists. If it does not exist, then find_asset will return nil.

The find_asset method is defined in lib/sprockets/base.rb:

sprockets: lib/sprockets/base.rb, 9 lines, beginning line 95

def find_asset(path, options = {})
  pathname =

  if pathname.absolute?
    build_asset(attributes_for(pathname).logical_path, pathname, options)
    find_asset_in_path(pathname, options)

This method converts the path it receives, "application.css", into a new Pathname object for the ease that Pathname objects provide over strings for dealing with file-system-like things. This pathname object is then checked for absoluteness with absolute?, which will return false because in no reality is "application.css" an absolute path to anything. Therefore, this method falls to find_asset_in_path, defined inside lib/sprockets/trail.rb and begins like this:

sprockets: lib/sprockets/trail.rb, 8 lines, beginning line 90

def find_asset_in_path(logical_path, options = {})
  # Strip fingerprint on logical path if there is one.
  # Not sure how valuable this feature is...
  if fingerprint = attributes_for(logical_path).path_fingerprint
    pathname = resolve(logical_path.to_s.sub("-#{fingerprint}", ''))
    pathname = resolve(logical_path)

Here, Sprockets calls attributes_for which is set up back in lib/sprockets/base.rb using these simple lines:

sprockets: lib/sprockets/base.rb, 3 lines, beginning line 85

def attributes_for(path), path)

These lines aren't very informative, so let's take a look at what the AssetAttributes class's initialize method looks like:

sprockets: lib/sprockets/asset_attributes.rb, 4 lines, beginning line 11

def initialize(environment, path)
  @environment = environment
  @pathname = path.is_a?(Pathname) ? path :

This method takes the environment argument it's given, which is the Sprockets::Index object that we are currently dealing with and stores it in the @environment instance variable for safe keeping. It then takes the path, checks to see if it is a Pathname and if it isn't, it will convert it into one. The path argument passed in here is already going to be a Pathname object as that was set up in the find_asset method.

Now that the initialize method is done, we've now got a new Sprockets::AssetAttributes object. The next thing that happens is that path_fingerprint is called on this object. This method comes with a lovely comment explaining what it does:

sprockets: lib/sprockets/asset_attributes.rb, 8 lines, beginning line 115

# Gets digest fingerprint.
#     "foo-0aa2105d29558f3eb790d411d7d8fb66.js"
#     # =&gt; "0aa2105d29558f3eb790d411d7d8fb66"
def path_fingerprint
  pathname.basename(extensions.join).to_s =~ /-([0-9a-f]{7,40})$/ ? $1 : nil

As the comment quite accurately describes, this method will take the fingerprint, or the unique identifier from this asset and return it. If there isn't one, then it will simply return nil. In this case, our asset is still "application.css" and therefore doesn't contain a fingerprint and so this method will return nil.

In that case, the if statement's conditions in find_asset_in_path will return false and so it will fall to else to do its duty.

sprockets: lib/sprockets/trail.rb, 3 lines, beginning line 95

  pathname = resolve(logical_path)

Not too much magic here, this else just calls the resolve method which should return a value which is stored into pathname. The resolve method is also defined within this file and begins like this:

sprockets: lib/sprockets/trail.rb, 3 lines, beginning line 70

def resolve(logical_path, options = {})
  # If a block is given, preform an iterable search
  if block_given?

In this case, resolve is not being called with block and so the if statement's code is not run. The code inside the else though goes like this, and *does* callresolve with a block:

sprockets: lib/sprockets/trail.rb, 6 lines, beginning line 77

  resolve(logical_path, options) do |pathname|
    return pathname
  raise FileNotFound, "couldn't find file '#{logical_path}'"

Alright then, so let's take a closer look at what the if block_given? contains:

sprockets: lib/sprockets/trail.rb, 2 lines, beginning line 72

if block_given?
  args = attributes_for(logical_path).search_paths + [options]

In this case, we see our old friend attributes_for called again which is then handed the Pathname equivalent of "application.css" and so it returns a new AssetAttributes object for that again. Next, the search_paths method is called on it, which is defined in sprockets/lib/asset_attributes.rb like this:

sprockets: lib/sprockets/asset_attributes.rb, 11 lines, beginning line 27

def search_paths
  paths = [pathname.to_s]

  if pathname.basename(extensions.join).to_s != 'index'
    path_without_extensions = extensions.inject(pathname) { |p, ext| p.sub(ext, '') }
    index_path = path_without_extensions.join("index#{extensions.join}").to_s
    paths &lt;&lt; index_path


This method will return all the search paths that Sprockets will look through to find a particular asset. If this file is called "index" then the paths will only be the file that is being requested. If it's not, then it will extract the extensions from the path and build a new path called "application/index.css", adding that to the list of paths to search through.

It is done this way so that we can have folders containing specific groups of assets. For instance, for a "projects" resource we could have a "projects/index.css" file under app/assets/stylesheets and that would then specify directives or CSS for projects. This file would be includable from another sprockets-powered CSS file with simply //= require "projects" or with a stylesheet_link_tag "projects" in the layout. Sprockets will attempt to look for a file in the asset paths called "projects.css" and if it fails to find that then it will look for "projects/index.css" as a fallback.

That is what this method is doing, providing two possible solutions to finding the asset. In the case of our "application.css" request, the paths will be the Pathname objects of "application.css" and "application/index.css".

Now that we know what search_paths is going to assign to args, let's take a look at the next couple of lines:

sprockets: lib/sprockets/trail.rb, 3 lines, beginning line 74

  trail.find(*args) do |path|

The find method called on trail here will look for the specified paths, using any options that were passed to resolve as part of the args array. In this situation, there were no options passed in and so these options will just be an empty hash. The trail method is defined further down in this file very simply:

sprockets: lib/sprockets/trail.rb, 3 lines, beginning line 86

  def trail

This @trail instance variable is set up as the first thing in the initialize method for Sprockets::Environment:

sprockets: lib/sprockets/environment.rb, 2 lines, beginning line 20

def initialize(root = ".")
  @trail =

Where the root argument here is the root of the Rails application, exactly the same as Rails.root returns. The find method itself is defined within the Hike gem like this:

hike: lib/hike/trail.rb, 3 lines, beginning line 138

def find(*args, &amp;block)
  index.find(*args, &amp;block)

Where the index method is defined like this:

hike: lib/hike/trail.rb, 3 lines, beginning line 152

def index, paths, extensions, aliases)

Rake tasks

Cover assets:precompile and assets:clean here.

Jump to Line
Something went wrong with that request. Please try again.