Design Patterns, pt. 3

This week we’ll focus on three structural patterns that are frequent in application development: Proxy, Adapter and Flyweight.

Proxy

Consider the following example. You design a campus app that provides information such as timetables, room plans, cafeteria meal plan, etc. The class responsible for retrieving the meal plan might look like this:

class Meal {
	String name;
	List<String> notes;
}
class MensaService {
	interface OpenMensaApi {
		@GET("canteens/229/days/{date}/meals")
		Call<List<Meal>> listMeals(@Path("date") String date);
	}

	OpenMensaApi api;

	MensaService() {
		Retrofit retrofit = new Retrofit.Builder()
			.baseUrl("https://openmensa.org/api/v2/")
			.addConverterFactory(GsonConverterFactory.create())
			.build();

		api = retrofit.create(OpenMensaApi.class);
	}

	List<Meal> getMeals(String date) throws IOException {
		Call<List<Meal>> call = api.listMeals(date);
		Response<List<Meal>> resp = call.execute();
		return resp.body();
	}
}

Later in your app, you might use this class as follows:

class SomeApp {
	public static void main(String... args) {
		MensaService ms = new MensaService();

		List<Meal> meals = ms.getMeals("20170612");
	}
}

You test your product and observe that students keep looking at the app every 5 minutes in the morning. Clearly, every request to get the meals of a certain date will result in a subsequent (network) call to the OpenMensa API. This is unfortunate: first, the remote server may become unreachable if wifi drops, or slow during “rush hour”; second (and more importantly), the information is quite static – it usually doesn’t change!

This is where the proxy pattern comes in. We create a subclass that satisfies the same interface as the base class, but adds caching functionality:

class SomeApp {
	public static void main(String... args) {
		// anonymous derived class for brevity
		MensaService proxy = new MensaService() {
			Map<String, List<Meal>> cache = new HashMap<>();
			List<Meal> getMeals(String date) throws IOException {
				if (cache.containsKey(date))
					return cache.get(date);

				List<Meal> meals = super.getMeals(date);
				cache.put(date, meals);
				return meals;
			}
		};

		List<Meal> meals = proxy.getMeals("20170612");
	}
}

This way, the request for the meal plan of a certain date will only be executed once; for subsequent calls, the proxy returns the cached responses. The fact that the proxy has the same interface allows the client to dynamically select to use the proxy or not.

Note: Subclassing is one option; more frequently, both the “real” service and the proxy would implement the same interface, and the proxy would maintain a reference to the real service.

Structure

dp-proxy dp-proxy-process

Proxies come in different flavors:

  • Remote proxy (aka. Ambassador): Provides local proxy to remote object (different process or physical location)
  • Virtual proxy: Creates expensive objects on demand; not to be confused with singleton (unique instance)
  • Protection Proxy: controls access to the original object, e.g. read-only access that simulates write.

Note: A smart reference is a proxy, too: it behaves just like the underlying object, but manages the state of the instance.

Examples

  • Caching for network requests
  • Log output routing
  • Lazy initialization of expensive objects
  • Related: security facade; behaves like proxy, but hides error handling or authentication

Proxy, Decorator and Composite

Proxy, Decorator and Composite pattern have a similar structure using recursive composition. However, the point of the

  • decorator is to add functionality without subclassing: one enclosed instance plus extra logic;
  • composite is to model a recursive structure, such as user interface widgets: arbitrary number of enclosed instances, logic typically restricted to traversing the structure or specific to leaf classes;
  • proxy is to mimic the original object (!) while adding access control or caching.

Note: In edge cases, a proxy actually behaves like a decorator (or vice versa). However, decorators can typically be stacked, often in arbitrary order; Proxy hierarchies are typically very flat: either there is a proxy, or there is none.


Adapter

Let’s stick with the example above, where you implemented a MensaService class that allows you to get the list of meals via getMeals(String date). Now you meet with your friend who is writing the front-end part of the app. Well, they were expecting you to provide the meals in form of an Iterable where you can set the date:

interface MealProvider extends Iterable<Meal> {
	void setDate(String date);
	// Iterator<Meal> iterator();  <-- from Iterable!
}

Doh. This is quite a different interface, but there is no way that either of you changes their code – think of all the refactoring of the unit tests etc.!

This is where the adapter pattern comes in. Just like you use adapters for tools if they don’t fit, you can create an adapter that fits both ends:

class MealAdapter extends MensaService implements MealProvider {
	private String date;
	
	@Override
	public void setDate(String date) { this.date = date; }
	
	@Override
	public Iterator<Meal> iterator() {
		try {
			// optionally: use today if date == null?
			return super.getMeals(date).iterator();
		} catch (IOException e) {
			return Collections.emptyIterator();
		}
	}
}

Voila, this is your class adapter:

MealProvider mp = new MealAdapter();
mp.setDate("20171206");
for (Meal m : mp)
	System.out.println(m);

Alternatively, you could write an object adapter, that implements the target interface, but maintains a reference to an instance of the class to be adapted (see structure below).

Structure

Class adapter:

dp-adapter-class

Object adapter:

dp-adapter-object

Note: The Adapter is not to be confused with the Facade, in which a whole subsystem is abstracted into a new class, typically implementing a new interface. An example for a Facade would be to couple the classes Engine, Transmission and Starter into the class Auto, which adds the logic on how to start, drive and stop.

Examples

  • ArrayAdapter in Android to render data arrays in views
  • Wrappers for third-party libraries
  • Object adapter often best choice if implementation of Adaptee unknown

Flyweight

Consider the following example: you want to build a “text based” web browser (e.g. for visually impaired). Here is a simple page that contains a list of a few images.

<ul>
	<li><img alt="Exhibit 1" src="picasso.png"></li>
	<li><img alt="Exhibit 2" src="vangogh.png"></li>
	<li><img alt="Exhibit 3" src="munch.png"></li>
	<li><img alt="Exhibit 4" src="monet.png"></li>
</ul>

In Java, we could use an Img class to represent each image:

class Img {
	final Image image;
	final String caption;

	Img(String caption, String path) throws IOException, URISyntaxException {
		this.caption = caption;

		// get resource file uri
		File file = new File(getClass().getClassLoader()
				.getResource(path).toURI());

		// load image; use helper function from JavaX package
		// https://docs.oracle.com/javase/8/docs/api/javax/imageio/ImageIO.html
		this.image = ImageIO.read(file);
	}

	void describe(PrintStream ps) {
		ps.println(String.format("%s: %d x %d",
				caption,
				image.getHeight(null),
				image.getWidth(null)
		));
	}
}

You could now instantiate the list with a few image tags and print them to make it a text based browser.

List<Img> items = new LinkedList<>();

// allocate items
items.add(new Img("Exhibit 1", "picasso.png"));
items.add(new Img("Exhibit 2", "vangogh.png"));
items.add(new Img("Exhibit 3", "munch.png"));
items.add(new Img("Exhibit 4", "monet.png"));

// print them out
for (Img e : items)
	e.describe(System.out);

This works alright as long as every image is different, but is fairly inefficient if images are displayed multiple times:

<ul>
	<li><img alt="Exhibit 1" src="picasso.png"></li>
	<li><img alt="Also Picasso" src="picasso.png"></li>
	<li><img alt="Picasso, too" src="picasso.png"></li>
	<li><img alt="Oh look, Picasso" src="picasso.png"></li>
</ul>

This may sound hypothetical, but think of recurring images in an endless scroll page such as the “Like” button on Facebook.

Clearly, re-loading the picasso.png is not only inefficient in terms of load times and network traffic, it also has bad effect on memory.

This is where the Flyweight pattern comes into play. The general idea is to separate intrinsic (static, unchanged; here: picasso.png) information from extrinsic (variable; here: alt caption) information.

The intrinsic share becomes the flyweight, and it will be shared among all different img that have the same src.

class Flyweight {
	// intrinsic state
	private final Image image;

	Flyweight(String path) throws URISyntaxException, IOException {
		// get resource file uri
		File file = new File(getClass().getClassLoader()
				.getResource(path).toURI());

		// load image (the intrinsic state)
		this.image = ImageIO.read(file);
	}

	void describe(PrintStream ps, Img es) {
		ps.println(String.format("%s: %d x %d",
				es.caption,
				image.getHeight(null),
				image.getWidth(null)
		));
	}
}

These flyweights are managed by a factory; that is, the user never allocates a flyweight manually, but retrieves instances from the factory, which facilitates the sharing.

class FlyweightFactory {
	private Map<String, Flyweight> flyweights = new HashMap<>();

	Flyweight getFlyweight(String path) throws URISyntaxException, IOException {
		if (flyweights.containsKey(path))
			return flyweights.get(path);

		// allocate new flyweight
		Flyweight fw = new Flyweight(path);
		flyweights.put(path, fw);

		return fw;
	}
}

The extrinsic share becomes the new Img class; it will have individual alt captions, but maintain references to the shared flyweight.

class Img {
	final String caption;
	final Flyweight flyweight;  // reference!

	Img(String caption, Flyweight flyweight) {
		this.caption = caption;
		this.flyweight = flyweight;
	}

	void describe(PrintStream ps) {
		// inject extrinsic state to flyweight
		flyweight.describe(ps, this);
	}
}

Back to the original example, our text browser. Instead of allocating the Img tags

List<Img> items = new LinkedList<>();
FlyweightFactory factory = new FlyweightFactory();

// allocate items
items.add(new Img("Exhibit 1", factory.getFlyweight("picasso.png")));
items.add(new Img("Also Picasso", factory.getFlyweight("picasso.png")));
items.add(new Img("Picasso, too", factory.getFlyweight("picasso.png")));
items.add(new Img("Oh look, Picasso", factory.getFlyweight("picasso.png")));

// print them out
for (Img e : items)
	e.describe(System.out);

This way, the picasso.png is only loaded once and then shared among all the other Img instances. As a result: the application is faster (single loading) and needs less memory (all static data just once). You can easily try it by loading a few hundreds of images: you will see how much faster (and less memory) the flyweight uses.

Structure

dp-flyweight

Notes:

  • The term flyweight is misleading: it is light in a sense of less and static parts, but often contains the “memory heavy” objects.
  • In a variant, there is no operation() but just a shared reference to the flyweights, which act as a pool of “heavy” objects.

Examples

  • Glyph (letter) rendering for text fields; intrinsic state: true-type fonts (often several MB), extrinsic state: position on screen, scale (size).
  • Browser rendering the same media multiple times; intrinsic state: actual media (image, video, audio), extrinsic state: location on screen
  • Android RecyclerView; intrinsic state: inflated layout of RecycleView, extrinsic state: actual contents to be displayed (often nested with further Flyweight)
  • Video games rendering/tiling engines; intrinsic state: actual texture or tile, extrinsic state: 3D location and orientation
  • Boost.Flyweight, a C++ implementation that abstracts the caching of heavy, potentially duplicated, objects. (This may be similar to the Java-internal management of String literals/instances.)

Design Patterns Summary

There is a total of 23 design patterns described by Gamma et al. Throughout this course, we already discussed quite a few of those:

Creational Patterns

  • Factory and factory method: Provide an interface for creating families of related or dependent objects without specifying their concrete class.
  • Singleton: Guarantee unique instance of class, and provide global access.

Structural Patterns

  • Adapter: Make a piece of software fit your needs.
  • Composite: Recursive data structure with containers and leaves, to represent part-whole hierarchies; composite lets client treeat objects and compositions uniformly.
  • Decorator: Add functionality to objects without changing their basic interface.
  • Flyweight: Share common data to support large numbers of similar objects.
  • Proxy: Provide a surrogate to allow caching or access control; indistinguishable to the client (same interface).

Behavioral Patterns

  • Command: create objects that can do or undo certain actions; use it to realize undo, macros and transactions.
  • Iterator: Provide access to elements of aggregate without exposing the underlying structure/representation.
  • Observer: Subscribe to an object to get notified on state change.
  • State: Allow an object to alter its behavior when its internal state changes; objects will appear to change their class.
  • Strategy: Define family of algorithms and make them interchangeable.
  • Template method: Define skeleton of algorithm/functionality in an operation, deferring certain steps/parts to subclasses.