AMP Toolbox for PHP

A collection of AMP tools making it easier to publish and host AMP pages with PHP.

The following tools are part of this library:

AMP Optimizer for PHP

AMP Optimizer for PHP

AMP Optimizer is a library for doing server-side optimization to AMP markup by automatically applying AMP performance best practices and enabling AMP server-side-rendering.

Table of Contents (click to expand)

Conceptual Overview
Installation
Basic Usage
Usage with a DOM Representation
Working with Errors
Included Transformers
Configuring the Transformers
Creating a Custom Transformer
Making a Custom Transformer Configurable
Transformers Requesting External Data
Adapting the Handling of Remote Requests

Conceptual Overview

The AMP Optimizer is a AmpProject\Optimizer\TransformationEngine object that sets up a pipeline of consecutive AmpProject\Optimizer\Transformer objects. The engine takes unoptimized input in the form of either a HTML markup string or an AmpProject\Dom\Document object and turns it into an optimized HTML markup string.

During the process, errors might occur that make parts of the optimization impossible. These are collected within an AmpProject\Optimizer\ErrorCollection object that you can then iterate over to find out more and provide feedback as needed.

Installation

The AMP Optimizer is part of the AMP Toolbox for PHP library that you can pull into your project via the Composer PHP package manager:

composer require ampproject/amp-toolbox

Basic Usage

The following code snippet shows the most basic way of using the AMP Optimizer:

use AmpProject\Optimizer\ErrorCollection;
use AmpProject\Optimizer\TransformationEngine;

$transformationEngine = new TransformationEngine();    // 1.
$errorCollection      = new ErrorCollection;           // 2.

$optimizedHtml = $transformationEngine->optimizeHtml(  // 3.
    $unoptimizedHtml,                                  // 4.
    $errorCollection                                   // 5.
);

First we instantiate the transformation engine itself.
Then we instantiate an AmpProject\Optimizer\ErrorCollection object as we need a "bag" to collect the errors in and pass them around.
As a final step, we store the result of calling the transformation engine's optimizeHtml() method, which requires...
... the unoptimized input HTML markup as a string and ...
... the empty error collection we've already instantiated. After the transformation engine ran, this collection will contain all errors that were encountered during the transformation.

Usage with a DOM Representation

If you already have a DOM representation, there's no need to save it as HTML first to use it with the transformation engine. The transformation engine accepts an AmpProject\Dom\Document object[^1] directly via its optimizeDom() method.

If you have a regular PHP built-in DOMDocument instead, you can turn it into an AmpProject\Dom\Document using AmpProject\Dom\Document::fromNode().

use AmpProject\Dom\Document;
use AmpProject\Optimizer\ErrorCollection;
use AmpProject\Optimizer\TransformationEngine;

if (! $dom instanceof Document) {
    $dom = Document::fromNode($dom);
}

$transformationEngine = new TransformationEngine();
$errorCollection      = new ErrorCollection;
$transformationEngine->optimizeDom($dom, $errorCollection);

Do note that the optimizeDom() doesn't have a return value, as it changes the provided AmpProject\Dom\Document in-place.

Working with Errors

The AmpProject\Optimizer\ErrorCollection that you pass into the transformation engine's optimizeHtml() or optimizeDom() method should ideally stay empty after the optimization pass.

To check whether errors were found, you can iterate over the collection, which will provide you with 0 or more AmpProject\Optimizer\Error objects.

$errorCollection = new ErrorCollection;

// Do the transformation here, while passing in the $errorCollection object.

foreach ($errorCollection as $error) {
	printf(
	    "Error code: %s\nError Message: %s\n",
	    $error->getCode(),
	    $error->getMessage()
	);
}

A quick count of the errors can be done for early returns as needed:

if ($errorCollection->count() > 0) {
	$this->log('The AMP serverside optimization process produced one or more errors.');
}

You can check whether the collection of errors contains an error with a specific code as well. The current convention is that all errors have their class shortname (the class name without the namespace) as the error code.

if ($errorCollection->has('CannotRemoveBoilerplate')) {
	$this->log('The boilerplate was not removed by the Optimizer.');
}

Note that this only lets you check whether an error "category" popped up. It can be one or more errors with that same code. If you need a more detailed check, you should iterate over the collection instead.

Included Transformers

Class (short name)	Description
`AmpBoilerplate`	Transformer that removes AMP boilerplate `<style>` and `<noscript>` tags in `<head>`, keeping only the `amp-custom` style tag. It then (re-)inserts the `amp-boilerplate` unless the document is marked with the `i-amphtml-no-boilerplate` attribute.
`AmpRuntimeCss`	Transformer adding `https://cdn.ampproject.org/v0.css` if server-side-rendering is applied (known by the presence of the `<style amp-runtime>` tag). AMP runtime css (`v0.css`) will always be inlined as it'll get automatically updated to the latest version once the AMP runtime has loaded.
`PreloadHeroImage`	Transformer that optimizes image rendering times for hero images by adding preload and serverside-rendered `<img>` tags when possible. Viable hero images are `<amp-img>` tags, `<amp-video>` tags with a `poster` attribute as well as `<amp-iframe>` and `<amnp-video-iframe>` tags with a `placeholder` attribute. The first viable image that is encountered is used by default, but this behavior can be overridden by adding the `data-hero` attribute to a maximum of two images. The preloads only work work images that don't use `srcset`, as that is not supported as a preload in most browsers. The serverside-rendered image will not be created for `<amp-video>` tags.
`ReorderHead`	Transformer applying the head reordering transformations to the HTML input. `ReorderHead` reorders the children of `<head>`. Specifically, it orders the `<head>` like so: (0) `<meta charset>` tag (1) `<style amp-runtime>` (inserted by `AmpRuntimeCss`) (2) remaining `<meta>` tags (those other than `<meta charset>`) (3) AMP runtime `.js` `<script>` tag (4) AMP viewer runtime `.js` `<script>` (5) `<script>` tags that are render delaying (6) `<script>` tags for remaining extensions (7) `<link>` tag for favicons (8) `<link>` tag for resource hints (9) `<link rel=stylesheet>` tags before `<style amp-custom>` (10) `<style amp-custom>` (11) any other tags allowed in `<head>` (12) AMP boilerplate (first `<style>` boilerplate, then `<noscript>`)
`ServerSideRendering`	Transformer applying the server-side rendering transformations to the HTML input. This does immediately on the server what would normally be done on the client after the runtime was downloaded and executed to process the DOM. As such, it allows for the removal of the boilerplate CSS that hides the page while it has not yet been processed on the client, drastically improving time it takes for the First Contentful Paint (FCP).
`TransformedIdentifier`	Transformer applying the transformed identifier transformations to the HTML input. This is what marks an AMP document as "already optimized", so that the AMP runtime does not need to process it anymore.

Configuring the Transformers

You can inject a configuration object into the AmpProject\Optimizer\TransformationEngine to override the default configuration.

The main AmpProject\Optimizer\Configuration object will provide the list of transformers to use, as well as give access to child objects it stores that are Transformer-specific configuration objects.

To override the list of transformers to use, you can provide an array containing the AmpProject\Optimizer\Configuration::KEY_TRANSFORMERS key.

use AmpProject\Optimizer\Configuration;
use AmpProject\Optimizer\TransformationEngine;
use AmpProject\Optimizer\Transformer;

$configurationData = [
	Configuration::KEY_TRANSFORMERS => [
		Transformer\ServerSideRendering::class,
		Transformer\AmpRuntimeCss::class,
		Transformer\TransformedIdentifier::class,
	],
];

$transformationEngine = new TransformationEngine(
	new Configuration($configurationData)
);

Configuration values for the transformers can be stored under the fully qualified class name of these transformers. This can be easily done by using their ::class constant.

They will also usually provide publicly accessible constants for the known configuration keys as well.

In the following example, we configure the AmpProject\Optimizer\Transformer\AmpRuntimeCss transformer by setting its 'canary' option to true (which would default to false).

use AmpProject\Optimizer\Configuration;
use AmpProject\Optimizer\TransformationEngine;
use AmpProject\Optimizer\Transformer;

$configurationData = [
	Transformer\AmpRuntimeCss::class => [
		Configuration\AmpRuntimeCssConfiguration::CANARY => true,
	],
];

$transformationEngine = new TransformationEngine(
	new Configuration($configurationData)
);

Creating a Custom Transformer

A custom transformer is at the most basic level an object that implements the AmpProject\Optimizer\Transformer interface.

This means it needs to have at the very least the following method:

public function transform(Document $document, ErrorCollection $errors)
{
	// Apply transformations to the provided $document and ...

	// ... add any encountered errors to the $errors collection.
}

To make this transformer then known to the transformation engine, you add it to the AmpProject\Optimizer\Configuration::KEY_TRANSFORMERS key of the AmpProject\Optimizer\Configuration object you pass into it.

use AmpProject\Optimizer\Configuration;
use AmpProject\Optimizer\TransformationEngine;
use MyProject\MyCustomTransformer;

$configurationData = [
	Configuration::KEY_TRANSFORMERS => array_merge(
		Configuration::DEFAULT_TRANSFORMERS,
		[
			MyCustomTransformer::class
		],
	),
];

$transformationEngine = new TransformationEngine(
	new Configuration($configurationData)
);

Making a Custom Transformer Configurable

Configuration objects for the individual transformers need to be registered with the main AmpProject\Optimizer\Configuration object using its registerConfigurationClass() method, which takes a fully qualified class name of the transformer as well as a fully qualified class name of the corresponding configuration object as its two arguments.

The configuration objects for the transformers that ship with this library are already registered by default. But if you add third-party or custom transformers, you'll need to register whatever configuration objects they might need with the main AmpProject\Optimizer\Configuration object first.

In the following example, we add a new MyProject\MyCustomTransformer transformer in addition to the default set and configure it with a default value, and then we register its corresponding configuration object to make sure the configuration can be properly validated and passed around.

use AmpProject\Optimizer\Configuration;
use AmpProject\Optimizer\TransformationEngine;
use MyProject\MyCustomTransformer;
use MyProject\MyCustomTransformerConfiguration;

$configurationData = [
	Configuration::KEY_TRANSFORMERS => array_merge(
		Configuration::DEFAULT_TRANSFORMERS,
		[
			MyCustomTransformer::class
		],
	),
	MyCustomTransformer::class => [
		MyCustomTransformerConfiguration::SOME_CONFIG_KEY => 'some value',
	],
];

$configuration = new Configuration($configurationData);

$configuration->registerConfigurationClass(
	MyCustomTransformer::class,
	MyCustomTransformerConfiguration::class
);

$transformationEngine = new TransformationEngine($configuration);

For the wiring to work correctly, the MyProject\MyCustomTransformer class needs to accept within its constructor an object implementing the AmpProject\Optimizer\TransformerConfiguration interface. The transformation engine will then inject the appropriate implementation at runtime when the transformer is being instantiated.

The MyProject\MyCustomTransformerConfiguration class should then implement that same AmpProject\Optimizer\TransformerConfiguration interface. For convenience, it can do so easily by extending the abstract AmpProject\Optimizer\Configuration\BaseTransformerConfiguration base class.

The configuration object will then be automatically injected into the transformer's constructor as needed.

Here's an example configuration class for our custom MyProject\MyCustomTransformer transformer:

namespace MyProject;

use AmpProject\Optimizer\Configuration\BaseTransformerConfiguration;

final class MyCustomTransformerConfiguration extends BaseTransformerConfiguration
{
	const SOME_CONFIG_KEY = 'some_config_key';

	protected function getAllowedKeys()
	{
		return [
			self::SOME_CONFIG_KEY => 'default value',
		];
	}

	protected function validate($key, $value)
	{
		switch ($key) {
			case self::SOME_CONFIG_KEY:
				// Validate configuration value here.
		}

		return $value;
	}
}

Here's how the transformer itself can accept and make use of the configuration object:

namespace MyProject;

use AmpProject\Dom\Document;
use AmpProject\Optimizer\Configurable;
use AmpProject\Optimizer\ErrorCollection;
use AmpProject\Optimizer\TransformerConfiguration;
use AmpProject\Optimizer\Transformer;

final class MyCustomTransformer implements Transformer
{
	private $configuration;

	public function __construct(TransformerConfiguration $configuration)
	{
		$this->configuration = $configuration;
	}

	public function transform(Document $document, ErrorCollection $errors)
	{
		// Bogus transformer logic that adds the configuration value as a body attribute.
		$document->body->setAttribute(
			'data-my-custom-transformer-body-attribute,
			$this->configuration->get(
				MyCustomTransformerConfiguration::SOME_CONFIG_KEY
			)
		);
	}
}

Transformers Requesting External Data

In case your transformer needs to make remote requests to fetch external data (like the AmpProject\Optimizer\Transformer\AmpRuntimeCss does for fetching the latest version of the CSS to inline), you need to accept an AmpProject\RemoteGetRequest object as an argument in your constructor. The transformation engine will then inject the appropriate implementation at runtime when the transformer is being instantiated.

This layer of abstraction allows code outside of the transformation engine to control the specific conditions and limits that govern these remote request, like for example throttling them or integrating them with the caching subsystem of the framework in use.

namespace MyProject;

use AmpProject\Dom\Document;
use AmpProject\RemoteGetRequest;
use AmpProject\Optimizer\ErrorCollection;
use AmpProject\Optimizer\Transformer;
use Throwable;

final class MyCustomTransformer implements Transformer
{
	const END_POINT = 'https://example.com/some_endpoint/';

	private $remoteRequest;

	public function __construct(RemoteGetRequest $remoteRequest)
	{
		$this->remoteRequest = $remoteRequest;
	}

	public function transform(Document $document, ErrorCollection $errors)
	{
		try {
			$response = $this->remoteRequest->get(self::END_POINT);
		} catch (Throwable $exception) {
			// Add error handling here.
		}

		$statusCode = $response->getStatusCode();

		if (200 < $statusCode || $statusCode >= 300) {
			// Add error handling here.
		}

		$content = $response->getBody();
		
		// Make use of the $content you've just retrieved from an external source.
	}
}

Adapting the Handling of Remote Requests

The implementation to use for fulfilling requests made via the AmpProject\RemoteGetRequest interface can be injected into the AmpProject\Optimizer\TransformationEngine via its second, optional argument:

use AmpProject\Optimizer\Configuration;
use AmpProject\Optimizer\TransformationEngine;

$transformationEngine = new TransformationEngine(
	new Configuration(),

	// A custom implementation that lets you control how remote requests are handled.
	new MyCustomRemoteGetRequestImplementation()
);

If this optional second argument is not provided when instancing the transformation engine, the default AmpProject\RemoteRequest\CurlRemoteGetRequest implementation is used.

There are other implementations already provided that can be useful:

Class (short name)	Description
`CurlRemoteGetRequest`	Remote request transport using cURL. This is the default implementation that will be used if you don't provide one explicitly. It has the following configuration settings as constructor arguments: $sslVerify - Whether to verify SSL certificates. Defaults to true. $timeout - Timeout value to use in seconds. Defaults to 10. $retries - Number of retry attempts to do if an error code was thrown that is worth retrying. Defaults to 2.
`FallbackRemoteGetRequest`	Fallback pipeline implementation to go through a series of fallback requests until a request succeeds. The request will be tried with the first instance provided, and follow the instance series from one to the next until a successful response was returned. It has the following configuration settings as constructor arguments: ...$pipeline - Variadic array of RemoteGetRequest instances to use as consecutive fallbacks.
`FilesystemRemoteGetRequest`	Fetch the response for a remote request from the local filesystem instead. This can be used to provide offline fallbacks. It has the following configuration settings as constructor arguments: $argumentMap - Associative array of data for mapping between provided URLs and the filepaths they should map to.
`StubbedRemoteGetRequest`	Stub for simulating remote requests. This is mainly used for writing tests. It has the following configuration settings as constructor arguments: $argumentMap - Associative array of data for mapping between provided URLs and the body content they should return.

The following code shows an example of how to use a remote request via cURL while falling back to files stored on the disk when an external request fails (probably due to network issues).

use AmpProject\Optimizer\Configuration;
use AmpProject\Optimizer\TransformationEngine;
use AmpProject\RemoteRequest\CurlRemoteGetRequest;
use AmpProject\RemoteRequest\FallbackRemoteGetRequest;
use AmpProject\RemoteRequest\FilesystemRemoteGetRequest;

const FALLBACK_MAPPING = [
	'https://example.com/some_endpoint/' => __DIR__ . '/../fallback_files/some_endpoint.json',
];

$remoteRequest = new FallbackRemoteGetRequest(
	new CurlRemoteGetRequest(true, 5, 0),                  // 5 second timeout with no retries, and ...
	new FilesystemRemoteGetRequest(self::FALLBACK_MAPPING) // ... fall back to shipped files.
);

$transformationEngine = new TransformationEngine(new Configuration(), $remoteRequest);

To build your own transport, you'll need to implement the AmpProject\RemoteGetRequest interface. For a more involved example of a custom transport or for integrating with your stack of choice, see the two implementations provided by the Amp for WordPress WordPress plugin:

jeffochoa/amp-toolbox-php