Symfony Internals

by Geoffrey Bachelet

Have you ever wondered what happens to a HTTP request when it reaches a symfony application? If yes, then you are in the right place. This chapter will explain in depth how symfony processes each request in order to create and return the response. Of course, just describing the process would lack a bit of fun, so we'll also have a look at some interesting things you can do and where you can interact with this process.

The Bootstrap

It all begins in your application's controller. Say you have a frontend controller with a dev environment (a very classic start for any symfony project). In this case, you'll end up with a front controller located at web/frontend_dev.php. What exactly happens in this file? In just a few lines of code, symfony retrieves the application configuration and creates an instance of sfContext, which is responsible for dispatching the request. The application configuration is necessary when creating the sfContext object, which is the application-dependent engine behind symfony.

Retrieving the application's configuration is a very important step. First, sfProjectConfiguration is responsible for guessing the application's configuration class, usually ${application}Configuration, located in apps/${application}/config/${application}Configuration.class.php.

sfApplicationConfiguration actually extends ProjectConfiguration, meaning that any method in ProjectConfiguration can be shared between all applications. This also means that sfApplicationConfiguration shares its constructor with both ProjectConfiguration and sfProjectConfiguration. This is fortunate since much of the project is configured inside the sfProjectConfiguration constructor. First, several useful values are computed and stored, such as the project's root directory and the symfony library directory. sfProjectConfiguration also creates a new event dispatcher of type sfEventDispatcher, unless one was passed as the fifth argument of ProjectConfiguration::getApplicationConfiguration() in the front controller.

Just after that, you are given a chance to interact with the configuration process by overriding the setup() method of ProjectConfiguration. This is usually the best place to enable / disable plugins (using sfProjectConfiguration::setPlugins(), sfProjectConfiguration::enablePlugins(), sfProjectConfiguration::disablePlugins() or sfProjectConfiguration::enableAllPluginsExcept()).

Next the plugins are loaded by sfProjectConfiguration::loadPlugins() and the developer has a chance to interact with this process through the sfProjectConfiguration::setupPlugins() that can be overriden.

Plugin initialization is quite straight forward. For each plugin, symfony looks for a ${plugin}Configuration (e.g. sfGuardPluginConfiguration) class and instantiates it if found. Otherwise, sfPluginConfigurationGeneric is used. You can hook into a plugin's configuration through two methods:

• ${plugin}Configuration::configure(), before autoloading is done
• ${plugin}Configuration::initialize(), after autoloading

Next, sfApplicationConfiguration executes its configure() method, which can be used to customize each application's configuration before the bulk of the internal configuration initialization process begins in sfApplicationConfiguration::initConfiguration().

This part of symfony's configuration process is responsible for many things and there are several entry points if you want to hook into this process. For example, you can interact with the autoloader's configuration by connecting to the autoload.filter_config event. Next, several very important configuration files are loaded, including settings.yml and app.yml. Finally, a last bit of plugin configuration is available through each plugin's config/config.php file or configuration class's initialize() method.

If sf_check_lock is activated, symfony will now check for a lock file (the one created by the project:disable task, for example). If the lock is found, the following files are checked and the first available is included, followed immediately by termination of the script:

1. apps/${application}/config/unavailable.php,
2. config/unavailable.php,
3. web/errors/unavailable.php,
4. lib/vendor/symfony/lib/exception/data/unavailable.php,

Finally, the developer has one last chance to customize the application's initialization through the sfApplicationConfiguration::initialize() method.

Bootstrap and configuration summary

• Retrieval of the application's configuration
  • ProjectConfiguration::setup() (define your plugins here)
  • Plugins are loaded
  • ${plugin}Configuration::configure()
  • ${plugin}Configuration::initialize()
  • ProjectConfiguration::setupPlugins() (setup your plugins here)
  • ${application}Configuration::configure()
  • autoload.filter_config is notified
  • Loading of settings.yml and app.yml
  • ${application}Configuration::initialize()
• Creation of an sfContext instance

sfContext and Factories

Before diving into the dispatch process, let's talk about a vital part of the symfony workflow: the factories.

In symfony, factories are a set of components or classes that your application relies on. Examples of factories are logger, i18n, etc. Each factory is configured via factories.yml, which is compiled by a config handler (more on config handlers later) and converted into PHP code that actually instantiates the factory objects (you can view this code in your cache in the cache/frontend/dev/config/config_factories.yml.php file).

At this point, you can already customize a large part of symfony's behavior just by editing the factories.yml configuration. You can even replace symfony's built-in factory classes with your own!

If you looked at the generated config_factories.yml.php, you may have noticed that factories are instantiated in a certain order. That order is important since some factories are dependent on others (for example, the routing component obviously needs the request to retrieve the information it needs).

Let's talk in greater details about the request. By default, the sfWebRequest class represents the request. Upon instantiation, sfWebRequest::initialize() is called, which gathers relevant information such as the GET / POST parameters as well as the HTTP method. You're then given an opportunity to add your own request processing through the request.filter_parameters event.

Using the request.filter_parameter event

Let's say you're operating a website exposing a public API to your users. The API is available through HTTP, and each user wanting to use it must provide a valid API key through a request header (for example X_API_KEY) to be validated by your application. This can be easily achieved using the request.filter_parameter event:

class apiConfiguration extends sfApplicationConfiguration
{
  public function configure()
  {
    // ...
 
    $this->dispatcher->connect('request.filter_parameters', array(
      $this, 'requestFilterParameters'
    ));
  }
 
  public function requestFilterParameters(sfEvent $event, $parameters)
  {
    $request = $event->getSubject();
 
    $api_key = $request->getHttpHeader('X_API_KEY');
 
    if (null === $api_key || false === $api_user = Doctrine_Core::getTable('ApiUser')->findOneByToken($api_key))
    {
      throw new RuntimeException(sprintf('Invalid api key "%s"', $api_key));
    }
 
    $request->setParameter('api_user', $api_user);
 
    return $parameters;
  }
}

You will then be able to access your API user from the request:

public function executeFoobar(sfWebRequest $request)
{
  $api_user = $request->getParameter('api_user');
}

This technique can be used, for example, to validate webservice calls.

The next very important factory is the routing. Routing's initialization is fairly straightforward and consists mostly of gathering and setting specific options. You can, however, hook up to this process through the routing.load_configuration event.

routing.load_configuration event usage example

For example, you can easily add a route:

public function setup()
{
  // ...
 
  $this->dispatcher->connect('routing.load_configuration', array(
    $this, 'listenToRoutingLoadConfiguration'
  ));
}
 
public function listenToRoutingLoadConfiguration(sfEvent $event)
{
  $routing = $event->getSubject();
 
  if (!$routing->hasRouteName('my_route'))
  {
    $routing->prependRoute('my_route', new sfRoute(
      '/my_route', array('module' => 'default', 'action' => 'foo')
    ));
  }
}

URL parsing occurs right after initialization, via the sfPatternRouting::parse() method. There are quite a few methods involved, but it suffices to say that by the time we reach the end of the parse() method, the correct route has been found, instantiated and bound to relevant parameters.

Once all factories have been loaded and properly setup, the context.load_factories event is triggered. This event is important since it's the earliest event in the framework where the developer has access to all of symfony's core factory objects (request, response, user, logging, database, etc.).

This is also the time to connect to another very useful event: template.filter_parameters. This event occurs whenever a file is rendered by sfPHPView and allows the developer to control the parameters actually passed to the template. sfContext takes advantage of this event to add some useful parameters to each template (namely, $sf_context, $sf_request, $sf_params, $sf_response and $sf_user).

You can connect to the template.filter_parameters event in order to add additional custom global parameters to all templates.

Taking advantage of the template.filter_parameters event

Say you decide that every single template you use should have access to a particular object, say a helper object. You would then add the following code to ProjectConfiguration:

public function setup()
{
  // ...
 
  $this->dispatcher->connect('template.filter_parameters', array(
    $this, 'templateFilterParameters'
  ));
}
 
public function templateFilterParameters(sfEvent $event, $parameters)
{
  $parameters['my_helper_object'] = new MyHelperObject();
 
  return $parameters;
}

Now every template has access to an instance of MyHelperObject through $my_helper_object.

sfContext summary

1. Initialization of sfContext
2. Factory loading
3. Events notified:
   1. request.filter_parameters
   2. routing.load_configuration
   3. context.load_factories
4. Global templates parameters added

A Word on Config Handlers

Config handlers are at the heart of symfony's configuration system. A config handler is tasked with understanding the meaning behind a configuration file. Each config handler is simply a class that is used to translate a set of yaml configuration files into a block of PHP code that can be executed as needed. Each configuration file is assigned to one specific config handler in the config_handlers.yml file.

To be clear, the job of a config handler is not to actually parse the yaml files (this is handled by sfYaml). Instead each config handler creates a set of PHP directions based on the YAML information and saves those directions to a PHP file, which can be efficiently included later. The compiled version of each YAML configuration file can be found in the cache directory.

The most commonly used config handler is most certainly sfDefineEnvironmentConfigHandler, which allows for environment-specific configuration settings. This config handler takes care to fetch only the configuration settings of the current environment.

Still not convinced? Let's explore sfFactoryConfigHandler. This config handler is used to compile factories.yml, which is one of the most important configuration file in symfony. This config handler is very particular since it converts a YAML configuration file into the PHP code that ultimately instantiate the factories (the all-important components we talked about earlier). Not your average config handler, is it?

The Dispatching and Execution of the Request

Enough said about factories, let's get back on track with the dispatch process. Once sfContext is finished initializing, the final step is to call the controller's dispatch() method, sfFrontWebController::dispatch().

The dispatch process itself in symfony is very simple. In fact, sfFrontWebController::dispatch() simply pulls the module and action names from the request parameters and forwards the application via sfController::forward().

The forward method is responsible for a lot of pre-execution checks as well as preparing the configuration and data for the action to be executed.

First the controller checks for the presence of a generator.yml file for the current module. This check is performed first (after some basic module / action name cleanup) because the generator.yml config file (if it exists) is responsible for generating the base actions class for the module (through its config handler, sfGeneratorConfigHandler). This is needed for the next step, which checks if the module and action exists. This is delegated to the controller, through sfController::actionExists(), which in turn calls the sfController::controllerExists() method. Here again, if the actionExists() method fails, an sfError404Exception is raised.

There's not much you can do here besides overriding the sfApplicationConfiguration::getControllerDirs() method in the application's configuration class. This method returns an array of directories where the controller files live, with an additional parameter to tell symfony if it should check whether controllers in each directory are enabled via the sf_enabled_modules configuration option from settings.yml. For example, getControllerDirs() could look something like this:

/**
 * Controllers in /tmp/myControllers won't need to be enabled
 * to be detected
 */
public function getControllerDirs($moduleName)
{
  return array_merge(parent::getControllerDirs($moduleName), array(
    '/tmp/myControllers/'.$moduleName => false
  ));
}

The next step is to retrieve an instance of the controller containing the action. This is handled via the sfController::getAction() method, which, like actionExists() is a facade for the sfController::getController(), method. Finally, the controller instance is added to the action stack.

After a little more configuration loading, we'll be ready to execute our action. The module-specific configuration must still be loaded, which can be found in two distinct places. First symfony looks for the module.yml file (normally located in apps/frontend/modules/yourModule/config/module.yml) which, because it's a YAML config file, uses the config cache. Additionally, this configuration file can declare the module as internal, using the mod_yourModule_is_internal setting which will cause the request to fail at this point since an internal module cannot be called publicly.

Now that module.yml is loaded, it's time to check for a second time that the current module is enabled. Indeed, you can set the mod_$moduleName_enabled setting to false if you want to disable the module at this point.

The final opportunity to configure a module lies in the config.php file (apps/frontend/modules/yourModule/config/config.php) where you can place arbitrary PHP code to be run in the context of the sfController::forward() method (that is, you have access to the sfController instance via the $this variable, as the code is literally run inside the sfController class).

The Dispatching Process Summary

1. sfFrontWebController::dispatch() is called
2. sfController::forward() is called
3. Check for a generator.yml
4. Check if the module / action exists
5. Retrieve a list of controllers directories
6. Retrieve an instance of the action
7. Load module configuration through module.yml and/or config.php

The Filter Chain

Now that all the configuration has been done, it's time to start the real work. Real work, in this particular case, is the execution of the filter chain.

The configuration of the filter chain is pulled from the current module's filters.yml, which is why the action instance is needed. This is your chance to modify the set of filters executed by the chain. Just remember that the rendering filter should always be the first in the list (we will see why later). The default filters configuration is as follow:

• rendering
• security
• cache
• execution

The Security Filter

Since the rendering filter waits for everyone to be done before doing anything, the first filter that actually gets executed is the security filter. This filter ensures that everything is right according to the security.yml configuration file. Specifically, the filter forwards an unauthenticated user to the login module / action and a user with insufficient credentials to the secure module / action. Note that this filter is only executed if security is enabled for the given action.

The Cache Filter

Next comes the cache filter. This filter takes advantage of its ability to prevent further filters from being executed. Indeed, if the cache is activated, and if we have a hit, why even bother executing the action? Of course, this will work only for a fully cacheable page, which is not the case for the vast majority of pages.

But this filter has a second bit of logic that gets executed after the execution filter, and just before the rendering filter. This code is responsible for setting up the right HTTP cache headers, and placing the page into the cache if necessary, thanks to the sfViewCacheManager::setPageCache() method.

The Execution Filter

Last but not least, the execution filter will, finally, take care of executing your business logic and handling the associated view.

Everything starts when the filter checks the cache for the current action. Of course, if we have something in the cache, the actual action execution is skipped and the Success view is then executed.

If the action is not found in the cache, then it is time to execute the preExecute() logic of the controller, and finally to execute the action itself. This is accomplished by the action instance via a call to sfActions::execute(). This method doesn't do much: it simply checks that the action is callable, then calls it. Back in the filter, the postExecute() logic of the action is now executed.

One more step ahead, and it's view time. The filter checks for two special return values that your action may have returned, sfView::HEADER_ONLY and sfView::NONE. Each does exactly what their names say: sends HTTP headers only (internally handled via sfWebResponse::setHeaderOnly()) or skips rendering altogether.

Once we know that we do want to render something, we're ready to get into the final step of the filter: the actual view execution.

The first thing we do is retrieve an sfView object through the sfController::getView() method. This object can come from two different places. First you could have a custom view object for this specific action (assuming the current module/action is, let's keep it simple, module/action) actionSuccessView or module_actionSuccessView in a file called apps/frontend/modules/module/view/actionSuccessView.class.php. Otherwise, the class defined in the mod_module_view_class configuration entry will be used. This value defaults to sfPHPView.

There are three rendering modes possible for rendering the view:

1. sfView::RENDER_NONE": equivalent to sfView::NONE, this cancels any rendering from being actually, well, rendered.
2. sfView::RENDER_VAR: populates the action's presentation, which is then accessible through its stack entry's sfActionStackEntry::getPresentation() method.
3. sfView::RENDER_CLIENT, the default mode, will render the view and feed the response's content.

The Rendering Filter

We're almost done now, just one very last step. The filter chain has almost finished executing, but do you remember the rendering filter? It's been waiting since the beginning of the chain for everyone to complete their work so that it can do its own job. Namely, the rendering filter sends the response content to the browser, using sfWebResponse::send().

Summary of the filter chain execution

1. The filter chain is instantiated with configuration from the filters.yml file
2. The security filter checks for authorizations and credentials
3. The cache filter handles the cache for the current page
4. The execution filter actually executes the action
5. The rendering filter send the response through sfWebResponse

Global Summary

1. Retrieval of the application's configuration
2. Creation of an sfContext instance
3. Initialization of sfContext
4. Factories loading
5. Events notified:
   1. request.filter_parameters
   2. routing.load_configuration
   3. context.load_factories
6. Global templates parameters added
7. sfFrontWebController::dispatch() is called
8. sfController::forward() is called
9. Check for a generator.yml
10. Check if the module / action exists
11. Retrieve a list of controllers directories
12. Retrieve an instance of the action
13. Load module configuration through module.yml and/or config.php
14. The filter chain is instantiated with configuration from the filters.yml file
15. The security filter checks for authorizations and credentials
16. The cache filter handles the cache for the current page
17. The execution filter actually executes the action
18. The rendering filter send the response through sfWebResponse

Final Thoughts

That's it! The request has been handled and we're now ready for another one. Of course, we could write an entire book about symfony's internal processes, so this chapter serves only as an overview. You are more than welcome to explore the source by yourself - it is, and always will be, the best way to learn the true mechanics of any framework or library.