Kévin Dunglas

Symfony 3.0, the next major version of our preferred PHP framework, will be released in a few weeks. Basically, it shares the same code base as Symfony 2.8 but all deprecated features coming from older versions have been removed to simplify the framework and its maintenance:

Symfony 3 deprecation progress report. #symfony pic.twitter.com/Nz4pMZZwU4

— Symfony News (@symfony_en) October 6, 2015

Symfony 2.8 and 3.0 also come with a lot of new features including (but not limited to) the (awesome) Guard authentication system, LDAP support or a component to guess types of PHP properties. In this post we’ll discover another interesting feature proudly sponsored by Les-Tilleuls.coop I’ve added to the Dependency Injection Component: autowiring.

The #Symfony Dependency Injection Component now supports autowiring! https://t.co/6imXHSReGU #PHP

— Kévin Dunglas (@dunglas) October 3, 2015

Introduction

Autowiring allows to register services in the container with minimal configuration. It is practical in the field of rapid application development, when designing prototypes and in early stages of large projects. It makes it easy to bootstrap an app service graph and eases refactoring:

@dunglas right, that's an interesting aspect for cache invalidation as well: a change in the class itself should trigger a container build.

— Matthias Noback (@matthiasnoback) August 30, 2015

A demo containing all code snippets shown in this article is available in a dedicated GitHub repository.

Let’s see how it works. To do so we will build a fake API publishing statutes on a Twitter feed obfuscated with ROT13 (a special case of the Caesar cipher).

Start by creating a ROT13 transformer class:

And now a Twitter client using this transformer:

The Dependency Injection Component is now able to automatically register the dependencies of this  TwitterClient class. The twitter_client service definition just need to be marked as autowired:

The autowiring subsystem will parse the constructor of the TwitterClient class and detects its dependencies that way. Here it will find and fill the need for an instance of a  Rot13Transformer.

If an existing service definition (and only one – see below) is of the needed type, it will inject it. Here it’s not the case, but the subsystem is smart enough to automatically register a private service for the Rot13Transformer class and set it as first argument of the twitter_client  service. Again, it can work only if there is one class of the given type. If there are several classes of the same type, you must fallback to the explicit service definition or register a default implementation (I’ll present this feature in a few line).

As you can see, the autowiring feature drastically reduces the amount of configuration required to define a service. No more arguments section! It also makes it easy to change the dependencies of the  TwitterClient class: just add or remove typehinted arguments in the constructor and you’re done. There is no need anymore to search and edit related service definitions.

Here is a typical controller using the twitter_client services:

You can give a try to the API with  curl:

curl -d "user=kevin&key=ABCD&status=Salut" http://localhost:8000/tweet

It should return  OK.

Working with interfaces

This is nice but when the application grows, it’s recommended to code against abstractions instead of implementations: it allows to easily replace some dependencies without modifying the class depending of them.

To follow this best practice, constructor arguments must be typehinted with interfaces and not concrete classes. It allows to replace easily the current implementation if necessary.

Let’s introduce a Rot13TransformerInterface:

Then edit Rot13Transformer to make it implementing the new interface:

And update TwitterClient  to depend of this new interface:

Finally the service definition must be updated because, obviously, the autowiring subsystem isn’t able to find itself the interface implementation to register:

The autowiring subsystem detects that the rot13_transformer service implements the Rot13TransformerInterface and injects it automatically. Even when using interfaces (and you should), building the service graph and refactoring the project is easier than with standard definitions.

Dealing with multiple implementations of the same type

Last but not least, the autowiring feature allows to specify the default implementation of a given type. Let’s introduce a new implementation of the Rot13TransformerInterface returning the result of the ROT13 transformation uppercased:

This class is intended to decorate the standard ROT13 transformer (or any other implementation) and return it uppercased.

We can now refactor the controller to add another endpoint leveraging this new transformer:

The last step is to update service definitions to register this new implementation and a Twitter client using it:

It deserves some explanations. We now have 2 services implementing the  Rot13TransformerInterface. The autowiring subsystem cannot guess the which one to use, this leads to errors like:

Fortunately, the autowiring_types key is here to specify which implementation to use by default. This key can take a list of types if necessary (using a YAML array).

Thanks to this setting, the  rot13_transformer service is automatically injected as argument of the uppercase_rot13_transformer and twitter_client services. For the  uppercase_twitter_client, we use a standard service definition to inject the specific uppercase_rot13_transformer  service.

You now know everything you need to use the new autowiring feature! As this feature is directly available in the Dependency Injection Component, you can leverage it in any project using it, including Drupal 8, API Platform or BackBee once the component have been upgraded to 2.8+.

As for other RAD features such as the FrameworkBundle controller or annotations, keep in mind to not use autowiring in public bundles nor in large projects with complex maintenance needs.