Roda

Roda is a routing tree web toolkit, designed for building fast and maintainable web applications in ruby.

Installation

$ gem install roda

Resources

Website: roda.jeremyevans.net
Source: github.com/jeremyevans/roda
Bugs: github.com/jeremyevans/roda/issues
Google Group: groups.google.com/group/ruby-roda
IRC: chat.freenode.net/#roda

Goals

Simplicity
Reliability
Extensibility
Performance

Simplicity

Roda is designed to be simple, both internally and externally. It uses a routing tree to enable you to write simpler and DRYer code.

Reliability

Roda supports and encourages immutability. Roda apps are designed to be frozen in production, which eliminates possible thread safety issues. Additionally, Roda limits the instance variables, constants, and methods that it uses, so that they do not conflict with the ones you use for your application.

Extensibility

Roda is built completely out of plugins, which makes it very extensible. You can override any part of Roda and call super to get the default behavior.

Performance

Roda has low per-request overhead, and the use of a routing tree and intelligent caching of internal datastructures makes it significantly faster than popular ruby web frameworks.

Usage

Here’s a simple application, showing how the routing tree works:

# cat config.ru
require "roda"

class App < Roda
  use Rack::Session::Cookie, :secret => ENV['SECRET']

  route do |r|
    # GET / request
    r.root do
      r.redirect "/hello"
    end

    # /hello branch
    r.on "hello" do
      # Set variable for all routes in /hello branch
      @greeting = 'Hello'

      # GET /hello/world request
      r.get "world" do
        "#{@greeting} world!"
      end

      # /hello request
      r.is do
        # GET /hello request
        r.get do
          "#{@greeting}!"
        end

        # POST /hello request
        r.post do
          puts "Someone said #{@greeting}!"
          r.redirect
        end
      end
    end
  end
end

run App.freeze.app

Here’s a breakdown of what is going on in the block above:

After requiring the library and subclassing Roda, the use method is called. This loads a Rack middleware into the current application.

The route block is called whenever a new request comes in. It is yielded an instance of a subclass of Rack::Request with some additional methods for matching routes. By convention, this argument should be named r.

The primary way routes are matched in Roda is by calling r.on, r.is, r.root, r.get, or r.post. Each of these “routing methods” takes a “match block”.

Each routing method takes each of the arguments (called matchers) that is given and tries to match it to the current request. If the method is able to match all of the arguments, it yields to the match block; otherwise, the block is skipped and execution continues.

r.on matches if all of the arguments match.
r.is matches if all of the arguments match and there are no further entries in the path after matching.
r.get matches any GET request when called without arguments.
r.get (when called with any arguments) matches only if the current request is a GET request and there are no further entries in the path after matching.
r.root only matches a GET request where the current path is /.

If a routing method matches and control is yielded to the match block, whenever the match block returns, Roda will return the Rack response array (containing status, headers, and body) to the caller.

If the match block returns a string and the response body hasn’t already been written to, the block return value will be interpreted as the body for the response. If none of the routing methods match and the route block returns a string, it will be interpreted as the body for the response.

r.redirect immediately returns the response, allowing for code such as r.redirect(path) if some_condition. If r.redirect is called without arguments and the current request method is not GET, it redirects to the current path.

The .freeze.app at the end is optional. Freezing the app avoids any possible thread safety issues inside the application at runtime, which shouldn’t be possible anyway. This generally should only be done in production mode. The .app is an optimization, which saves a few method calls for every request.

The Routing Tree

Roda is called a routing tree web toolkit because the way most sites are structured, routing takes the form of a tree (based on the URL structure of the site). In general:

r.on is used to split the tree into different branches.
r.is finalizes the routing path.
r.get and r.post handle specific request methods.

So, a simple routing tree might look something like this:

r.on "a" do           # /a branch
  r.on "b" do         # /a/b branch
    r.is "c" do       # /a/b/c request
      r.get do end    # GET  /a/b/c request
      r.post do end   # POST /a/b/c request
    end
    r.get "d" do end  # GET  /a/b/d request
    r.post "e" do end # POST /a/b/e request
  end
end

It’s also possible to handle the same requests, but structure the routing tree by first branching on the request method:

r.get do              # GET 
  r.on "a" do         # GET /a branch
    r.on "b" do       # GET /a/b branch
      r.is "c" do end # GET /a/b/c request
      r.is "d" do end # GET /a/b/d request
    end
  end
end

r.post do             # POST
  r.on "a" do         # POST /a branch
    r.on "b" do       # POST /a/b branch
      r.is "c" do end # POST /a/b/c request
      r.is "e" do end # POST /a/b/e request
    end
  end
end

This allows you to easily separate your GET request handling from your POST request handling. If you only have a small number of POST request URLs and a large number of GET request URLs, this may make things easier.

However, routing first by the path and last by the request method is likely to lead to simpler and DRYer code. This is because you can act on the request at any point during the routing. For example, if all requests in the /a branch need access permission A and all requests in the /a/b branch need access permission B, you can easily handle this in the routing tree:

r.on "a" do           # /a branch
  check_perm(:A)  
  r.on "b" do         # /a/b branch
    check_perm(:B)  
    r.is "c" do       # /a/b/c request
      r.get do end    # GET  /a/b/c request
      r.post do end   # POST /a/b/c request
    end
    r.get "d" do end  # GET  /a/b/d request
    r.post "e" do end # POST /a/b/e request
  end
end

Being able to operate on the request at any point during the routing is one of the major advantages of Roda.

Matchers

Other than r.root, the routing methods all take arguments called matchers. If all of the matchers match, the routing method yields to the match block. Here’s an example showcasing how different matchers work:

class App < Roda
  route do |r|
    # GET /
    r.root do
      "Home"
    end

    # GET /about
    r.get "about" do
      "About"
    end

    # GET /post/2011/02/16/hello
    r.get "post/:y/:m/:d/:slug" do |y, m, d, slug|
      "#{y}-#{m}-#{d} #{slug}" #=> "2011-02-16 hello"
    end

    # GET /username/foobar branch
    r.on "username/:username", :method=>:get do |username|
      user = User.find_by_username(username)

      # GET /username/foobar/posts
      r.is "posts" do
        # You can access user here, because the blocks are closures.
        "Total Posts: #{user.posts.size}" #=> "Total Posts: 6"
      end

      # GET /username/foobar/following
      r.is "following" do
        user.following.size.to_s #=> "1301"
      end
    end

    # /search?q=barbaz
    r.get "search" do
      "Searched for #{r['q']}" #=> "Searched for barbaz"
    end

    r.is "login" do
      # GET /login
      r.get do
        "Login"
      end

      # POST /login?user=foo&password=baz
      r.post do
        "#{r['user']}:#{r['password']}" #=> "foo:baz"
      end
    end
  end
end

Here’s a description of the matchers. Note that “segment”, as used here, means one part of the path preceded by a /. So, a path such as /foo/bar//baz has four segments: /foo, /bar, /, and /baz. The / here is considered the empty segment.

String

If a string does not contain a colon or slash, it matches a single segment containing the text of the string, preceded by a slash.

""    # matches "/"
"foo" # matches "/foo"
"foo" # does not match "/food"

If a string contains any slashes, it matches one additional segment for each slash:

"foo/bar" # matches "/foo/bar"
"foo/bar" # does not match "/foo/bard"

If a string contains a colon followed by any \w characters, the colon and remaining \w characters match any nonempty segment that contains at least one character:

"foo/:id" # matches "/foo/bar", "/foo/baz", etc.
"foo/:id" # does not match "/fo/bar"

You can use multiple colons in a string:

":x/:y" # matches "/foo/bar", "/bar/foo" etc.
":x/:y" # does not match "/foo", "/bar/"

You can prefix colons:

"foo:x/bar:y" # matches "/food/bard", "/fool/bart", etc.
"foo:x/bar:y" # does not match "/foo/bart", "/fool/bar", etc.

If any colons are used, the block will yield one argument for each segment matched containing the matched text, so:

"foo:x/:y" # matching "/fool/bar" yields "l", "bar"

Colons that are not followed by a \w character are matched literally:

":/a" # matches "/:/a"

Note that other than colons, strings do no handle regular expression syntax, the string is matched verbatim:

"\\d+(/\\w+)?" # matches "/\d+(/\w+)?"
"\\d+(/\\w+)?" # does not match "/123/abc"

Regexp

Regexps match one or more segments by looking for the pattern, preceded by a slash:

/foo\w+/ # matches "/foobar"
/foo\w+/ # does not match "/foo/bar"

If any patterns are captured by the Regexp, they are yielded:

/foo\w+/   # matches "/foobar", yields nothing
/foo(\w+)/ # matches "/foobar", yields "bar"

Symbol

Symbols match any nonempty segment, yielding the segment except for the preceding slash:

:id # matches "/foo" yields "foo"
:id # does not match "/"

Proc

Procs match unless they return false or nil:

proc{true}  # matches anything
proc{false} # does not match anything

Procs don’t capture anything by default, but they can do so if you add the captured text to r.captures.

Arrays

Arrays match when any of their elements match. If multiple matchers are given to r.on, they all must match (an AND condition). If an array of matchers is given, only one needs to match (an OR condition). Evaluation stops at the first matcher that matches.

Additionally, if the matched object is a String, the string is yielded. This makes it easy to handle multiple strings without a Regexp:

['page1', 'page2'] # matches "/page1", "/page2"
[]                 # does not match anything

Hash

Hashes allow easily calling specialized match methods on the request. The default registered matchers included with Roda are documented below. Some plugins add additional hash matchers, and the hash_matcher plugin allows for easily defining your own:

class App < Roda
  plugin :hash_matcher

  hash_matcher(:foo) do |v|
    # ...
  end

  route do |r|
    r.on :foo=>'bar' do
      # ...
    end
  end
end

:all

The :all matcher matches if all of the entries in the given array match, so

r.on :all=>[:a, :b] do
  # ...
end

is the same as:

r.on :a, :b do
  # ...
end

The reason it also exists as a separate hash matcher is so you can use it inside an array matcher, so:

r.on ['foo', {:all=>['foos', :id]}] do
end

would match /foo and /foos/10, but not /foos.

:method

The :method matcher matches the method of the request. You can provide an array to specify multiple request methods and match on any of them:

{:method => :post}             # matches POST
{:method => ['post', 'patch']} # matches POST and PATCH

false, nil

If false or nil is given directly as a matcher, it doesn’t match anything.

Everything else

Everything else matches anything.

Status codes

When it comes time to finalize a response, if a status code has not been set manually and anything has been written to the response, the response will use a 200 status code. Otherwise, it will use a 404 status code. This enables the principle of least surprise to work: if you don’t handle an action, a 404 response is assumed.

You can always set the status code manually, via the status attribute for the response.

route do |r|
  r.get "hello" do
    response.status = 200
  end
end

Verb Methods

As displayed above, Roda has r.get and r.post methods for matching based on the HTTP request method. If you want to match on other HTTP request methods, use the all_verbs plugin.

When called without any arguments, these match as long as the request has the appropriate method, so:

r.get do end

matches any GET request, and

r.post do end

matches any POST request

If any arguments are given to the method, these match only if the request method matches, all arguments match, and the path has been fully matched by the arguments, so:

r.post "" do end

matches only POST requests where the current path is /.

r.get "a/b" do end

matches only GET requests where the current path is /a/b.

The reason for this difference in behavior is that if you are not providing any arguments, you probably don’t want to also test for an exact match with the current path. If that is something you do want, you can provide true as an argument:

r.on "foo" do
  r.get true do # Matches GET /foo, not GET /foo/.*
  end
end

If you want to match the request method and do only a partial match on the request path, you need to use r.on with the :method hash matcher:

r.on "foo", :method=>:get do # Matches GET /foo(/.*)?
end

Root Method

As displayed above, you can also use r.root as a match method. This method matches GET requests where the current path is /. r.root is similar to r.get "", except that it does not consume the / from the path.

Unlike the other matching methods, r.root takes no arguments.

Note that r.root does not match if the path is empty; you should use r.get true for that. If you want to match either the the empty path or /, you can use r.get ["", true], or use the slash_path_empty plugin.

Note that r.root only matches GET requests. So, to handle POST / requests, use r.post ''.

Request and Response

While the request object is yielded to the route block, it is also available via the request method. Likewise, the response object is available via the response method.

The request object is an instance of a subclass of Rack::Request, with some additional methods.

If you want to extend the request and response objects with additional modules, you can use the module_include plugin.

Pollution

Roda tries very hard to avoid polluting the scope of the route block. This should make it unlikely that Roda will cause namespace issues with your application code. Some of the things Roda does:

The only instance variables defined by default in the scope of the route block are @_request and @_response. All instance variables in the scope of the route block used by plugins that ship with Roda are prefixed with an underscore.
The only methods defined (beyond the default methods for Object) are: call, env, opts, request, response, and session.
Constants inside the Roda namespace are all prefixed with Roda (e.g., Roda::RodaRequest).

Composition

You can mount any Rack app (including another Roda app), with its own middlewares, inside a Roda app, using r.run:

class API < Roda
  route do |r|
    r.is do
      # ...
    end
  end
end

class App < Roda
  route do |r|
    r.on "api" do
      r.run API
    end
  end
end

run App.app

This will take any path starting with /api and send it to API. In this example, API is a Roda app, but it could easily be a Sinatra, Rails, or other Rack app.

When you use r.run, Roda calls the given Rack app (API in this case); whatever the Rack app returns will be returned as the response for the current application.

If you have a lot of rack applications that you want to dispatch to, and which one to dispatch to is based on the request path prefix, look into the multi_run plugin.

multi_route plugin

If you are just looking to split up the main route block up by branches, you should use the multi_route plugin, which keeps the current scope of the route block:

class App < Roda
  plugin :multi_route

  route "api" do |r|
    r.is do
      # ...
    end
  end

  route do |r|
    r.on "api" do
      r.route "api"
    end
  end
end

run App.app

This allows you to set instance variables in the main route block and still have access to them inside the api route block.

Testing

It is very easy to test Roda with Rack::Test or Capybara. Roda’s own tests use RSpec. The default Rake task will run the specs for Roda, if RSpec is installed.

Settings

Each Roda app can store settings in the opts hash. The settings are inherited if you happen to subclass Roda.

Roda.opts[:layout] = "guest"

class Users < Roda; end
class Admin < Roda
  opts[:layout] = "admin"
end

Users.opts[:layout] # => 'guest'
Admin.opts[:layout] # => 'admin'

Feel free to store whatever you find convenient. Note that when subclassing, Roda only does a shallow clone of the settings.

If you store nested structures and plan to mutate them in subclasses, it is your responsibility to dup the nested structures inside Roda.inherited (making sure to call super). This should be is done so that that modifications to the parent class made after subclassing do not affect the subclass, and vice-versa.

The plugins that ship with Roda freeze their settings and only allow modification to their settings by reloading the plugin, and external plugins are encouraged to follow this approach.

The following options are respected by multiple plugins:

:add_script_name: Prepend the SCRIPT_NAME for the request to paths. This is useful if you mount the app as a path under another app.
:root: Set the root path for the app. This defaults to the current working directory of the process.

There may be other options supported by individual plugins, if so it will be mentioned in the documentation for the plugin.

Rendering

Roda ships with a render plugin that provides helpers for rendering templates. It uses Tilt, a gem that interfaces with many template engines. The erb engine is used by default.

Note that in order to use this plugin you need to have Tilt installed, along with the templating engines you want to use.

This plugin adds the render and view methods, for rendering templates. By default, view will render the template inside the default layout template; render will just render the template.

class App < Roda
  plugin :render

  route do |r|
    @var = '1'

    r.get "render" do
      # Renders the views/home.erb template, which will have access to
      # the instance variable @var, as well as local variable content.
      render("home", :locals=>{:content => "hello, world"})
    end

    r.get "view" do
      @var2 = '1'

      # Renders the views/home.erb template, which will have access to the
      # instance variables @var and @var2, and takes the output of that and
      # renders it inside views/layout.erb (which should yield where the
      # content should be inserted).
      view("home")
    end
  end
end

You can override the default rendering options by passing a hash to the plugin:

class App < Roda
  plugin :render,
    :escape => true,        # Automatically escape output in erb templates
    :views => 'admin_views', # Default views directory
    :layout_opts => {:template=>'admin_layout',
                     :ext=>'html.erb'},    # Default layout template options
    :template_opts => {:default_encoding=>'UTF-8'} # Default template options
end

Sessions

By default, Roda doesn’t turn on sessions, but most users are going to want to turn on session support. The simplest way to do this is to use the Rack::Session::Cookie middleware that comes with Rack:

require "roda"

class App < Roda
  use Rack::Session::Cookie, :secret => ENV['SECRET']
end

Security

Web application security is a very large topic, but here are some things you can do with Roda to prevent some common web application vulnerabilities.

Session Security

If you are using sessions, you should also always set a session secret, using the :secret option as shown above. Make sure that this secret is not disclosed, because if an attacker knows the :secret value, they can inject arbitrary session values. In the worst case scenario, this can lead to remote code execution.

Keep in mind that with Rack::Session::Cookie, the content in the session cookie is not encrypted, just signed to prevent tampering. This means you should not store any secret data in the session.

Cross Site Request Forgery (CSRF)

CSRF can be prevented by using the csrf plugin that ships with Roda, which uses the rack_csrf library. Just make sure that you include the CSRF token tags in your HTML, as appropriate.

It’s also possible to use the Rack::Csrf middleware directly; you don’t have to use the csrf plugin.

Cross Site Scripting (XSS)

The easiest way to prevent XSS with Roda is to use a template library that automatically escapes output by default. The :escape option to the render plugin sets the ERb template processor to escape by default, so that in your templates:

<%= '<>' %>  # outputs &lt;&gt; 
<%== '<>' %> # outputs <>

When using the +:escape_option, you will need to ensure that your layouts are not escaping the output of the content template:

<%== yield %> # not <%= yield %>

You can also provide a :escape_safe_classes option, which will make <%= %> not escape certain string subclasses, useful if you have helpers that already return escaped output using a string subclass instance.

This support requires Erubis.

You may want to look into setting the following HTTP headers, which can be done at the web server level, but can also be done at the application level using using the default_headers plugin:

Content-Security-Policy/X-Content-Security-Policy: Defines policy for how javascript and other types of content can be used on the page.
Frame-Options/X-Frame-Options: Provides click-jacking protection by not allowing usage inside a frame.
Strict-Transport-Security: Enforces SSL/TLS Connections to the application.
X-Content-Type-Options: Forces some browsers to respect a declared Content-Type header.
X-XSS-Protection: Enables an XSS mitigation filter in some browsers.

Example:

class App < Roda
  plugin :default_headers,
    'Content-Type'=>'text/html',
    'Content-Security-Policy'=>"default-src 'self'",
    'Strict-Transport-Security'=>'max-age=16070400;',
    'X-Frame-Options'=>'deny',
    'X-Content-Type-Options'=>'nosniff',
    'X-XSS-Protection'=>'1; mode=block'
end

Rendering Templates Derived From User Input

Roda’s rendering plugin assumes that template paths given to it are trusted by default. If you provide a path to the render/view methods that is derived from user input, you are opening yourself for people rendering arbitrary files on the system that that have a file name ending in the default template extension. For example, if you do:

class App < Roda
  plugin :render
  route do |r|
    view(r['page'])
  end
end

Then attackers can submit a page parameter such as '../../../../tmp/upload' to render the /tmp/upload.erb file. If you have another part of your system that allows users to create files with arbitrary extensions (even temporary files), then it may be possible to combine these two issues into a remote code execution exploit.

To mitigate against this issue, you can use the :check_paths => true render option, which will check that the full path of the template to be rendered begins with the :views directory, and raises an exception if not. You can also use the :allowed_paths render option to specify which paths are allowed. While :check_paths => true is not currently the default, it will become the default in Roda 3. Note that when specifying the :path option when rendering a template, Roda will not check paths, as it assumes that users and libraries that use this option will be checking such paths manually.

Reloading

Most rack-based reloaders will work with Roda, including:

Plugins

By design, Roda has a very small core, providing only the essentials. All nonessential features are added via plugins.

Roda’s plugins can override any Roda method and call super to get the default behavior, which makes Roda very extensible.

Roda ships with a large number of plugins, and some other libraries ship with support for Roda.

How to create plugins

Authoring your own plugins is pretty straightforward. Plugins are just modules, which may contain any of the following modules:

InstanceMethods: module included in the Roda class
ClassMethods: module that extends the Roda class
RequestMethods: module included in the class of the request
RequestClassMethods: module extending the class of the request
ResponseMethods: module included in the class of the response
ResponseClassMethods: module extending the class of the response

If the plugin responds to load_dependencies, it will be called first, and should be used if the plugin depends on another plugin.

If the plugin responds to configure, it will be called last, and should be used to configure the plugin.

Both load_dependencies and configure are called with the additional arguments and block that was given to the plugin call.

So, a simple plugin to add an instance method would be:

module MarkdownHelper
  module InstanceMethods
    def markdown(str)
      BlueCloth.new(str).to_html
    end
  end
end

Roda.plugin MarkdownHelper

Registering plugins

If you want to ship a Roda plugin in a gem, but still have Roda load it automatically via Roda.plugin :plugin_name, you should place it where it can be required via roda/plugins/plugin_name and then have the file register it as a plugin via Roda::RodaPlugins.register_plugin. It’s recommended, but not required, that you store your plugin module in the Roda::RodaPlugins namespace:

class Roda
  module RodaPlugins
    module Markdown
      module InstanceMethods
        def markdown(str)
          BlueCloth.new(str).to_html
        end
      end
    end

    register_plugin :markdown, Markdown
  end
end

To avoid namespace pollution, you should avoid creating your module directly in the Roda namespace. Additionally, any instance variables created inside InstanceMethods should be prefixed with an underscore (e.g., @_variable) to avoid polluting the scope. Finally, do not add any constants inside the InstanceMethods module, add constants to the plugin module itself (Markdown in the above example).

If you are planning on shipping your plugin in an external gem, it is recommended that you follow standard gem naming conventions for extensions. So if your plugin module is named FooBar, your gem name should be roda-foo_bar.

No Introspection

Because a routing tree does not store the routes in a data structure, but directly executes the routing tree block, you cannot introspect the routes when using a routing tree.

If you would like to introspect your routes when using Roda, there is an external plugin named roda-route_list, which allows you to add appropriate comments to your routing files, and has a parser that will parse those comments into routing metadata that you can then introspect.

Inspiration

Roda was inspired by Sinatra and Cuba. It started out as a fork of Cuba, from which it borrows the idea of using a routing tree (which Cuba in turn took from Rum). From Sinatra, it takes the ideas that route blocks should return the request bodies and that routes should be canonical. It pilfers the idea for an extensible plugin system from the Ruby database library Sequel.

License

MIT

Maintainer

Jeremy Evans <[email protected]>