Learning JavaScript Design Patterns
Volume 1.5.2
(original online at: http://addyosmani.com/resources/essentialjsdesignpatterns/book/)
Copyright © Addy Osmani 2012.
Learning JavaScript Design Patterns is released under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 unported license. It is available for purchase via O'Reilly
Media but will remain available for both free online and as a physical
(or eBook) purchase for readers wishing to support the project.
Preface
Design patterns are reusable solutions to commonly occurring problems
in software design. They are both exciting and a fascinating topic to
explore in any programming language.
One reason for this is that they help us build upon the combined
experience of many developers that came before us and ensure we
structure our code in an optimized way, meeting the needs of problems
we're attempting to solve.
Design patterns also provide us a common vocabulary to describe
solutions. This can be significantly simpler than describing syntax and
semantics when we're attempting to convey a way of structuring a
solution in code form to others.
In this book we will explore applying both classical and modern design patterns to the JavaScript programming language.
Target Audience
This book is targeted at professional developers wishing to improve
their knowledge of design patterns and how they can be applied to the
JavaScript programming language.
Some of the concepts covered (closures, prototypal inheritance) will
assume a level of basic prior knowledge and understanding. If you find
yourself needing to read further about these topics, a list of suggested
titles is provided for convenience.
If you would like to learn how to write beautiful, structured and organized code, I believe this is the book for you.
Acknowledgments
I will always be grateful for the talented technical reviewers who
helped review and improve this book, including those from the community
at large. The knowledge and enthusiasm they brought to the project was
simply amazing. The official technical reviewers tweets and blogs are
also a regular source of both ideas and inspiration and I wholeheartedly
recommend checking them out.
I would also like to thank Rebecca Murphey (http://rebeccamurphey.com, @rmurphey)
for providing the inspiration to write this book and more importantly,
continue to make it both available on GitHub and via O'Reilly.
Finally, I would like to thank my wonderful wife Ellie, for all of her support while I was putting together this publication.
Credits
Whilst some of the patterns covered in this book were implemented
based on personal experience, many of them have been previously
identified by the JavaScript community. This work is as such the
production of the combined experience of a number of developers. Similar
to Stoyan Stefanov's logical approach to preventing interruption of the
narrative with credits (in JavaScript Patterns), I have listed credits and suggested reading for any content covered in the references section.
If any articles or links have been missed in the list of references,
please accept my heartfelt apologies. If you contact me I'll be sure to
update them to include you on the list.
Reading
Whilst this book is targeted at both beginners and intermediate
developers, a basic understanding of JavaScript fundamentals is assumed.
Should you wish to learn more about the language, I am happy to
recommend the following titles:
- JavaScript: The Definitive Guide by David Flanagan
- Eloquent JavaScript by Marijn Haverbeke
- JavaScript Patterns by Stoyan Stefanov
- Writing Maintainable JavaScript by Nicholas Zakas
- JavaScript: The Good Parts by Douglas Crockford
Table Of Contents
#
Introduction
One of the most important aspects of writing maintainable code is
being able to notice the recurring themes in that code and optimize
them. This is an area where knowledge of design patterns can prove
invaluable.
In the first part of this book, we will explore the history and
importance of design patterns which can really be applied to any
programming language. If you're already sold on or are familiar with
this history, feel free to skip to the chapter "What is a Pattern?" to continue reading.
Design patterns can be traced back to the early work of an architect named Christopher Alexander.
He would often write publications about his experience in solving
design issues and how they related to buildings and towns. One day, it
occurred to Alexander that when used time and time again, certain design
constructs lead to a desired optimal effect.
In collaboration with Sara Ishikawa and Murray Silverstein,
Alexander produced a pattern language that would help empower anyone
wishing to design and build at any scale. This was published back in
1977 in a paper titled "A Pattern Language", which was later released as
a complete hardcover book.
Some 30 years ago, software engineers began to incorporate the
principles Alexander had written about into the first documentation
about design patterns, which was to be a guide for novice developers
looking to improve their coding skills. It's important to note that the
concepts behind design patterns have actually been around in the
programming industry since its inception, albeit in a less formalized
form.
One of the first and arguably most iconic formal works published
on design patterns in software engineering was a book in 1995 called Design Patterns: Elements Of Reusable Object-Oriented Software. This was written by Erich Gamma, Richard Helm, Ralph Johnson and John Vlissides - a group that became known as the Gang of Four (or GoF for short).
The GoF's publication is considered quite instrumental to pushing
the concept of design patterns further in our field as it describes a
number of development techniques and pitfalls as well as providing
twenty-three core Object-Oriented design patterns frequently used around
the world today. We will be covering these patterns in more detail in
the section "Categories of Design Patterns".
In this book, we will take a look at a number of popular
JavaScript design patterns and explore why certain patterns may be more
suitable for your projects than others. Remember that patterns can be
applied not just to vanilla JavaScript (i.e standard JavaScript code),
but also to abstracted libraries such as jQuery or dojo as well. Before we begin, let’s look at the exact definition of a "pattern" in software design.
What is a Pattern?
A pattern is a reusable solution that can be applied to commonly
occurring problems in software design - in our case - in writing
JavaScript web applications. Another way of looking at patterns are as
templates for how we solve problems - ones which can be used in quite a
few different situations.
So, why is it important to understand patterns and be familiar with them? Design patterns have three main benefits:
- Patterns are proven solutions: They
provide solid approaches to solving issues in software development
using proven techniques that reflect the experience and insights
the developers that helped define them bring to the pattern.
- Patterns can be easily reused: A
pattern usually reflects an out of the box solution that can be
adapted to suit our own needs. This feature makes them quite
robust.
- Patterns can be expressive: When we look at a pattern there’s generally a set structure and vocabulary to the solution presented that can help express rather large solutions quite elegantly.
Patterns are not an exact solution. It’s
important that we remember the role of a pattern is merely to provide us
with a solution scheme. Patterns don’t solve all design problems nor do
they replace good software designers, however, they do support them. Next we’ll take a look at some of the other advantages patterns have to offer.
- Reusing patterns assists in preventing minor issues that can cause major problems in the application development process. What
this means is when code is built on proven patterns, we can afford to
spend less time worrying about the structure of our code and more time
focusing on the quality of our overall solution. This is because
patterns can encourage us to code in a more structured and organized
fashion avoiding the need to refactor it for cleanliness purposes in the
future.
- Patterns can provide generalized solutions which are
documented in a fashion that doesn't require them to be tied to a
specific problem. This generalized approach means that
regardless of the application (and in many cases the programming
language) we are working with, design patterns can be applied to improve
the structure of our code.
- Certain patterns can actually decrease the overall file-size footprint of our code by avoiding repetition. By encouraging developers to look more closely at their solutions for areas where instant reductions in repetition
can be made, e.g. reducing the number of functions performing similar
processes in favor of a single generalized function, the overall size of
our codebase can be decreased. This is also known as making code more DRY.
- Patterns add to a developers vocabulary, which makes communication faster.
- Patterns that are frequently used can be improved over
time by harnessing the collective experiences other developers using
those patterns contribute back to the design pattern community.
In some cases this leads to the creation of entirely new design
patterns whilst in others it can lead to the provision of improved
guidelines on how specific patterns can be best used. This can ensure
that pattern-based solutions continue to become more robust than ad-hoc
solutions may be.
We already use patterns everyday
To understand how useful patterns can be, let's review a very simple
element selection problem that the jQuery library solves for us.
Imagine that we have a script where for each DOM element found on a page
with class "foo" we wish to increment a counter. What's the most
efficient way to query for this collection of elements? Well, there are a
few different ways this problem could be tackled:
- Select all of the elements in the page and then store references
to them. Next, filter this collection and use regular expressions (or
another means) to only store those with the class "foo".
-
Use a modern native browser feature such as
querySelectorAll()
to select all of the elements with the class "foo".
- Use a native feature such as
getElementsByClassName()
to similarly get back the desired collection.
So, which of these options is the fastest? It's actually option 3. by a factor of 8-10 times the alternatives.
In a real-world application however, 3. will not work in versions of
Internet Explorer below 9 and thus it's necessary to use 1. where both
2. and 3. aren't supported.
Developers using jQuery don't have to worry about this problem however, as it's luckily abstracted away for us using the Facade pattern. As we'll review in more detail later, this pattern provides a simple set of abstracted interfaces (e.g $el.css()
, $el.animate()
)
to several more complex underlying bodies of code. As we've seen, this
means less time having to be concerned about implementation level
details.
Behind the scenes, the library simply opts for the most optimal
approach to selecting elements depending on what our current browser
supports and we just consume the abstraction layer.
We're probably all also familiar with jQuery's $("selector")
. This is significantly more easy to use for selecting HTML elements on a page versus having to manually handle opt for getElementById()
, getElementsByClassName()
, getElementByTagName
and so on.
Although we know that querySelectorAll()
attempts to
solve this problem, compare the effort involved in using jQuery's Facade
interfaces vs. selecting the most optimal selection paths ourselves.
There's no contest! Abstractions using patterns can offer real-world
value.
We'll be looking at this and more design patterns later on in the book.
"Pattern"-ity Testing, Proto-Patterns & The Rule Of Three
Remember that not every algorithm, best practice or solution represents
what might be considered a complete pattern. There may be a few key
ingredients here that are missing and the pattern community is generally
wary of something claiming to be one unless it has been heavily vetted.
Even if something is presented to us which appears to
meet the criteria for a pattern, it should not be considered one until
it has undergone suitable periods of scrutiny and testing by others.
Looking back upon the work by Alexander once more, he claims that a
pattern should both be a process and a "thing". This definition is
obtuse on purpose as he follows by saying that it is the process should
create the "thing". This is a reason why patterns generally focus on
addressing a visually identifiable structure i.e we should be able to
visually depict (or draw) a picture representing the structure that
placing the pattern into practice results in.
In studying design patterns, it's not irregular to come across the term
"proto-pattern". What is this? Well, a pattern that has not yet been
known to pass the "pattern"-ity tests is usually referred to as a
proto-pattern. Proto-patterns may result from the work of someone that
has established a particular solution that is worthy of sharing with the
community, but may not have yet had the opportunity to have been vetted
heavily due to its very young age.
Alternatively, the individual(s) sharing the pattern may not have the
time or interest of going through the "pattern"-ity process and might
release a short description of their proto-pattern instead. Brief
descriptions or snippets of this type of pattern are known as patlets.
The work involved in fully documenting a qualified pattern can be quite
daunting. Looking back at some of the earliest work in the field of
design patterns, a pattern may be considered "good" if it does the
following:
- Solves a particular problem: Patterns are not
supposed to just capture principles or strategies. They need to
capture solutions. This is one of the most essential ingredients
for a good pattern.
- The solution to this problem cannot be obvious:
We can find that problem-solving techniques often attempt to derive
from well-known first principles. The best design patterns usually
provide solutions to problems indirectly - this is considered a
necessary approach for the most challenging problems related to
design.
- The concept described must have been proven:
Design patterns require proof that they function as described and
without this proof the design cannot be seriously considered. If a
pattern is highly speculative in nature, only the brave may attempt
to use it.
- It must describe a relationship: In some
cases it may appear that a pattern describes a type of module.
Although an implementation may appear this way, the official
description of the pattern must describe much deeper system
structures and mechanisms that explain its relationship to code.
We would be forgiven for thinking that a proto-pattern which fails to
meet guidelines isn't worth learning from, however, this is far from the
truth. Many proto-patterns are actually quite good. I’m not saying that
all proto-patterns are worth looking at, but there are quite a few
useful ones in the wild that could assist us with future projects. Use
best judgment with the above list in mind and you’ll be fine in your
selection process.
One of the additional requirements for a pattern to be valid is that
they display some recurring phenomenon. This is often something that can
be qualified in at least three key areas, referred to as the rule of three. To show recurrence using this rule, one must demonstrate:
- Fitness of purpose - how is the pattern considered successful?
- Usefulness - why is the pattern considered successful?
- Applicability - is the design worthy of being a
pattern because it has wider applicability? If so, this needs to be
explained. When reviewing or defining a pattern, it is important to
keep the above in mind.
#
The Structure Of A Design Pattern
You may be curious about how a pattern author might approach
outlining structure, implementation and purpose of a new pattern. A
pattern is initially presented in the form of a rule that establishes a relationship between:
- A context
- A system of forces that arises in that context and
- A configuration that allows these forces to resolve themselves in context
With this in mind, let’s now take a look at a summary of the
component elements for a design pattern. A design pattern should have a:
- Pattern name and a description
- Context outline – the contexts in which the pattern is effective in responding to the users needs.
- Problem statement – a statement of the problem being addressed so we can understand the intent of the pattern.
- Solution – a description of how the user’s problem is being solved in an understandable list of steps and perceptions.
- Design – a description of the pattern’s design and in particular, the user’s behavior in interacting with it
- Implementation – a guide to how the pattern would be implemented
- Illustrations – a visual representation of classes in the pattern (e.g. a diagram))
- Examples – an implementation of the pattern in a minimal form
- Co-requisites – what other patterns may be needed to support use of the pattern being described?
- Relations – what patterns does this pattern resemble? does it closely mimic any others?
- Known usage – is the pattern being used in the wild? If so, where and how?
- Discussions – the team or author’s thoughts on the exciting benefits of the pattern
Design patterns are quite a powerful approach to getting all of the
developers in an organization or team on the same page when creating or
maintaining solutions. If considering working on a pattern of your own,
remember that although they may have a heavy initial cost in the
planning and write-up phases, the value returned from that investment
can be quite worth it. Always research thoroughly before working on new
patterns however, as you may find it more beneficial to use or build on
top of existing proven patterns than starting afresh.
#
Writing Design Patterns
Although this book is aimed at those new to design patterns, a
fundamental understanding of how a design pattern is written can offer a
number of useful benefits. For starters, we can gain a deeper
appreciation for the reasoning behind why a pattern is needed. We can
also learn how to tell if a pattern (or proto-pattern) is up to scratch
when reviewing it for our own needs.
Writing good patterns is a challenging task. Patterns not only need to
(ideally) provide a substantial quantity of reference material for
end-users, but they also need to be able to defend why they are
necessary.
Having read the previous section on what a pattern is, we
may think that this in itself is enough to help us identify patterns we
see in the wild. This is actually not completely true. It's not always
clear if a piece of code we're looking at is following a set pattern or
just accidentally happens to appear like it does.
When we're looking at a body of code we think may be using a pattern, we
should consider writing down some of the aspects of the code that we
believe falls under a particular existing pattern or set of patterns.
In many cases of pattern-analysis we can find that we're just
looking at code that follows good principles and design practices that
could happen to overlap with the rules for a pattern by accident.
Remember - solutions in which neither interactions nor defined rules
appear are not patterns.
If interested in venturing down the path of writing your own design
patterns I recommend learning from others who have already been through
the process and done it well. Spend time absorbing the information from a
number of different design pattern descriptions and take in what’s
meaningful to you.
Explore structure and semantics - this can be done by examining the
interactions and context of the patterns you are interested in so you
can identify the principles that assist in organizing those patterns
together in useful configurations.
Once we've exposed ourselves to a wealth of information on pattern
literature, we may wish to begin writing our pattern using an existing format and see if we can brainstorm new ideas for improving it or integrating our ideas in there.
An example of a developer that did this is in recent years is Christian Heilmann, who took the existing Module pattern and made some fundamentally useful changes to it to create the Revealing Module pattern (this is one of the patterns covered later in this book).
The following are tips I would suggest if interesting in creating a new design pattern:
- How practical is the pattern?: Ensure the pattern
describes proven solutions to recurring problems rather than just
speculative solutions which haven’t been qualified.
- Keep best practices in mind: The design decisions we make should be based on principles we derive from an understanding of best practices.
- Our design patterns should be transparent to the user:
Design patterns should be entirely transparent to any type of
user-experience. They are primarily there to serve the developers using
them and should not force changes to behavior in the user-experience
that would not be incurred without the use of a pattern.
- Remember that originality is not key in pattern design:
When writing a pattern, we do not need to be the original discoverer of
the solutions being documented nor do you have to worry about our
design overlapping with minor pieces of other patterns. If the approach
is strong enough to have broad useful applicability, it has a chance of
being recognized as a valid pattern.
- Pattern need a strong set of examples: A good
pattern description needs to be followed by an equally strong set
of examples demonstrating the successful application of our
pattern. To show broad usage, examples that exhibit good design
principles are ideal.
Pattern writing is a careful balance between creating a design that is
general, specific and above all, useful. Try to ensure that if writing a
pattern you cover the widest possible areas of application and you
should be fine. I hope that this brief introduction to writing
patterns has given you some insights that will assist your learning
process for the next sections of this book.
Anti-Patterns
If we consider that a pattern represents a best practice, an
anti-pattern represents a lesson that has been learned. The term
anti-patterns was coined in 1995 by Andrew Koenig in the November C++
Report that year, inspired by the GoF's book Design Patterns. In Koenig’s report, there are two notions of anti-patterns that are presented. Anti-Patterns:
- Describe a bad solution to a particular problem which resulted in a bad situation occurring
- Describe how to get out of said situation and how to go from there to a good solution
On this topic, Alexander writes about the difficulties in
achieving a good balance between good design structure and good context:
“These notes are about the process of design; the process of
inventing physical things which display a new physical order,
organization, form, in response to function.…every design problem begins
with an effort to achieve fitness between two entities: the form in
question and its context. The form is the solution to the problem; the
context defines the problem”.
While it’s quite important to be aware of design patterns, it can be
equally important to understand anti-patterns. Let us qualify the reason
behind this. When creating an application, a project’s life-cycle
begins with construction however once you’ve got the initial release
done, it needs to be maintained. The quality of a final solution will
either be good or bad, depending on the level of skill and time the team have invested in it. Here good and bad are considered in context - a ‘perfect’ design may qualify as an anti-pattern if applied in the wrong context.
The bigger challenges happen after an application has hit production
and is ready to go into maintenance mode. A developer working on such a
system who hasn’t worked on the application before may introduce a bad design into the project by accident. If said bad
practices are created as anti-patterns, they allow developers a means
to recognize these in advance so that they can avoid common mistakes
that can occur - this is parallel to the way in which design patterns
provide us with a way to recognize common techniques that are useful.
To summarize, an anti-pattern is a bad design that is worthy of
documenting. Examples of anti-patterns in JavaScript are the following:
- Polluting the global namespace by defining a large number of variables in the global context
- Passing strings rather than functions to either setTimeout or setInterval as this triggers the use of
eval()
internally.
- Modifying the
Object
class prototype (this is a particularly bad anti-pattern)
- Using JavaScript in an inline form as this is inflexible
- The use of document.write where native DOM alternatives such
as document.createElement are more appropriate. document.write has
been grossly misused over the years and has quite a few disadvantages
including that if it's executed after the page has been loaded it can
actually overwrite the page we're on, whilst document.createElement does
not. We can see here
for a live example of this in action. It also doesn't work with XHTML
which is another reason opting for more DOM-friendly methods such as
document.createElement is favorable.
Knowledge of anti-patterns is critical for success. Once we are able
to recognize such anti-patterns, we're able to refactor our code to
negate them so that the overall quality of our solutions improves
instantly.
#
Categories Of Design Pattern
A glossary from the well-known design book, Domain-Driven Terms, rightly states that:
“A design pattern names, abstracts, and identifies the key aspects
of a common design structure that make it useful for creating a
reusable object-oriented design. The design pattern identifies the
participating classes and their instances, their roles and
collaborations, and the distribution of responsibilities.
Each design pattern focuses on a particular object-oriented design
problem or issue. It describes when it applies, whether or not it can
be applied in view of other design constraints, and the consequences and
trade-offs of its use. Since we must eventually implement our designs, a
design pattern also provides sample ... code to illustrate an
implementation.
Although design patterns describe object-oriented designs, they
are based on practical solutions that have been implemented in
mainstream object-oriented programming languages ....”
Design patterns can be broken down into a number of different
categories. In this section we’ll review three of these categories and
briefly mention a few examples of the patterns that fall into these
categories before exploring specific ones in more detail.
Creational Design Patterns
Creational design patterns focus on handling object creation
mechanisms where objects are created in a manner suitable for the
situation we're are working in. The basic approach to object creation
might otherwise lead to added complexity in a project whilst these
patterns aim to solve this problem by controlling the creation process.
Some of the patterns that fall under this category are: Constructor, Factory, Abstract, Prototype, Singleton and Builder.
Structural Design Patterns
Structural patterns are concerned with object composition and
typically identify simple ways to realize relationships between
different objects. They help ensure that when one part of a system
changes, the entire structure of the system doesn't need to do the same.
They also assist in recasting parts of the system which don't fit a
particular purpose into those that do.
Patterns that fall under this category include: Decorator, Facade, Flyweight, Adapter and Proxy.
Behavioral Design Patterns
Behavioral patterns focus on improving or streamlining the communication between disparate objects in a system.
Some behavioral patterns include: Iterator, Mediator, Observer and Visitor.
#
Design Pattern Categorization
In my early experiences of learning about design patterns, I
personally found the following table a very useful reminder of what a
number of patterns has to offer - it covers the 23 Design Patterns
mentioned by the GoF. The original table was summarized by Elyse Nielsen
back in 2004
and I've modified it where necessary to suit our discussion in this
section of the book.
I recommend using this table as reference, but do remember that
there are a number of additional patterns that are not mentioned here
but will be discussed later in the book.
A brief note on classes
Keep in mind that there will be patterns in this table that reference
the concept of "classes". JavaScript is a class-less language, however
classes can be simulated using functions.
The most common approach to achieving this is by defining a JavaScript function where we then create an object using the new
keyword. this
can be used to help define new properties and methods for the object as follows:
02 | function Car( model ) { |
05 | this .color = "silver" ; |
08 | this .getInfo = function () { |
09 | return this .model + " " + this .year; |
We can then instantiate the object using the Car constructor we defined above like this:
1 | var myCar = new Car( "ford" ); |
5 | console.log( myCar.getInfo() ); |
For more ways to define "classes" using JavaScript, see Stoyan Stefanov's useful post on them:
Introduction
JavaScript is a very flexible object-oriented language when it comes to syntax. In this article you can find three ways of defining and instantiating an object. Even if you have already picked your favorite way of doing it, it helps to know some alternatives in order to read other people's code.
It's important to note that there are no classes in JavaScript. Functions can be used to somewhat simulate classes, but in general JavaScript is a class-less language. Everything is an object. And when it comes to inheritance, objects inherit from objects, not classes from classes as in the "class"-ical languages.
1. Using a function
This is probably one of the most common ways. You define a normal JavaScript function and then create an object by using the new
keyword. To define properties and methods for an object created using function()
, you use the this
keyword, as seen in the following example.
function Apple (type) {
this.type = type;
this.color = "red";
this.getInfo = getAppleInfo;
}
function getAppleInfo() {
return this.color + ' ' + this.type + ' apple';
}
To instantiate an object using the Apple constructor function, set some properties and call methods you can do the following:
var apple = new Apple('macintosh');
apple.color = "reddish";
alert(apple.getInfo());
1.1. Methods defined internally
In the example above you see that the method getInfo() of the Apple "class" was defined in a separate function getAppleInfo(). While this works fine, it has one drawback – you may end up defining a lot of these functions and they are all in the "global namespece". This means you may have naming conflicts if you (or another library you are using) decide to create another function with the same name. The way to prevent pollution of the global namespace, you can define your methods within the constructor function, like this:
function Apple (type) {
this.type = type;
this.color = "red";
this.getInfo = function() {
return this.color + ' ' + this.type + ' apple';
};
}
Using this syntax changes nothing in the way you instantiate the object and use its properties and methods.
1.2. Methods added to the prototype
A drawback of 1.1. is that the method getInfo() is recreated every time you create a new object. Sometimes that may be what you want, but it's rare. A more inexpensive way is to add getInfo() to the prototype of the constructor function.
function Apple (type) {
this.type = type;
this.color = "red";
}
Apple.prototype.getInfo = function() {
return this.color + ' ' + this.type + ' apple';
};
Again, you can use the new objects exactly the same way as in 1. and 1.1.
2. Using object literals
Literals are shorter way to define objects and arrays in JavaScript. To create an empty object using you can do:
var o = {};
instead of the "normal" way:
var o = new Object();
For arrays you can do:
var a = [];
instead of:
var a = new Array();
So you can skip the class-like stuff and create an instance (object) immediately. Here's the same functionality as described in the previous examples, but using object literal syntax this time:
var apple = {
type: "macintosh",
color: "red",
getInfo: function () {
return this.color + ' ' + this.type + ' apple';
}
}
In this case you don't need to (and cannot) create an instance of the class, it already exists. So you simply start using this instance.
apple.color = "reddish";
alert(apple.getInfo());
Such an object is also sometimes called singleton. It "classical" languages such as Java, singleton means that you can have only one single instance of this class at any time, you cannot create more objects of the same class. In JavaScript (no classes, remember?) this concept makes no sense anymore since all objects are singletons to begin with.
3. Singleton using a function
Again with the singleton, eh?
The third way presented in this article is a combination of the other two you already saw. You can use a function to define a singleton object. Here's the syntax:
var apple = new function() {
this.type = "macintosh";
this.color = "red";
this.getInfo = function () {
return this.color + ' ' + this.type + ' apple';
};
}
So you see that this is very similar to 1.1. discussed above, but the way to use the object is exactly like in 2.
apple.color = "reddish";
alert(apple.getInfo());
new function(){...}
does two things at the same time: define a function (an anonymous constructor function) and invoke it with new
. It might look a bit confusing if you're not used to it and it's not too common, but hey, it's an option, when you really want a constructor function that you'll use only once and there's no sense of giving it a name.
Summary
You saw three (plus one) ways of creating objects in JavaScript. Remember that (despite the article's title) there's no such thing as a class in JavaScript. Looking forward to start coding using the new knowledge? Happy JavaScript-ing!
Tell your friends about this post:
Facebook,
Twitter,
Google+
Let us now proceed to review the table.
Creational |
Based on the concept of creating an object. |
Class |
Factory Method |
This makes an instance of several derived classes based on interfaced data or events. |
Object |
Abstract Factory |
Creates an instance of several families of classes without detailing concrete classes. |
Builder |
Separates object construction from its representation, always creates the same type of object. |
Prototype |
A fully initialized instance used for copying or cloning. |
Singleton |
A class with only a single instance with global access points. |
|
|
|
|
|
|
|
|
Structural |
Based on the idea of building blocks of objects |
Class |
Adapter |
Match interfaces of different classes therefore classes can work together despite incompatible interfaces |
Object |
Adapter |
Match interfaces of different classes therefore classes can work together despite incompatible interfaces |
Bridge |
Separates an object's interface from its implementation so the two can vary independently |
Composite |
A structure of simple and composite objects which makes the total object more than just the sum of its parts. |
Decorator |
Dynamically add alternate processing to objects. |
Facade |
A single class that hides the complexity of an entire subsystem. |
Flyweight |
A fine-grained instance used for efficient sharing of information that is contained elsewhere. |
Proxy |
A place holder object representing the true object |
|
Behavioral |
Based on the way objects play and work together. |
Class |
Interpreter |
A way to include language elements in an application to match the grammar of the intended language. |
Template Method |
Creates the shell of an algorithm in a method, then defer the exact steps to a subclass. |
Object |
Chain of Responsibility |
A way of passing a request between a chain of objects to find the object that can handle the request. |
Command |
Encapsulate a command request as an object to
enable, logging and/or queuing of requests, and provides error-handling
for unhandled requests. |
Iterator |
Sequentially access the elements of a collection without knowing the inner workings of the collection. |
Mediator |
Defines simplified communication between classes to prevent a group of classes from referring explicitly to each other. |
Memento |
Capture an object's internal state to be able to restore it later. |
Observer |
A way of notifying change to a number of classes to ensure consistency between the classes. |
State |
Alter an object's behavior when its state changes |
Strategy |
Encapsulates an algorithm inside a class separating the selection from the implementation |
Visitor |
Adds a new operation to a class without changing the class |
#
JavaScript Design Patterns
In this section, we will explore JavaScript implementations of a number of both classic and modern design patterns.
Developers commonly wonder whether there is an ideal pattern
or set of patterns they should be using in their workflow. There isn't a
true single answer to this question; each script and web application we
work on is likely to have its own individual needs and we need to think
about where we feel a pattern can offer real value to an
implementation.
For example, some projects may benefit from the decoupling benefits
offered by the Observer pattern (which reduces how dependent parts of an
application are on one another) whilst others may simply be too small
for decoupling to be a concern at all.
That said, once we have a firm grasp of design patterns and the
specific problems they are best suited to, it becomes much easier to
integrate them into our application architectures.
The patterns we will be exploring in this section are the:
#
The Constructor Pattern
In classical object-oriented programming languages, a constructor is a
special method used to initialize a newly created object once memory
has been allocated for it. In JavaScript, as almost everything is an
object, we're most often interested in object constructors.
Object constructors are used to create specific types of objects -
both preparing the object for use and accepting arguments which a
constructor can use to set the values of member properties and methods
when the object is first created.
Object Creation
The three common ways to create new objects in JavaScript are as follows:
6 | var newObject = Object.create( null ); |
9 | var newObject = new Object(); |
Where the "Object" constructor in the final example creates an object
wrapper for a specific value, or where no value is passed, it will
create an empty object and return it.
There are then four ways in which keys and values can then be assigned to an object:
06 | newObject.someKey = "Hello World" ; |
09 | var key = newObject.someKey; |
16 | newObject[ "someKey" ] = "Hello World" ; |
19 | var key = newObject[ "someKey" ]; |
29 | Object.defineProperty( newObject, "someKey" , { |
30 | value: "for more control of the property's behavior" , |
39 | var defineProp = function ( obj, key, value ){ |
41 | Object.defineProperty( obj, key, config ); |
45 | var person = Object.create( null ); |
48 | defineProp( person, "car" , "Delorean" ); |
49 | defineProp( person, "dateOfBirth" , "1981" ); |
50 | defineProp( person, "hasBeard" , false ); |
56 | Object.defineProperties( newObject, { |
As we will see a little later in the book, these methods can even be used for inheritance, as follows:
04 | var driver = Object.create( person ); |
07 | defineProp(driver, "topSpeed" , "100mph" ); |
10 | console.log( driver.dateOfBirth ); |
13 | console.log( driver.topSpeed ); |
Basic Constructors
As we saw earlier, JavaScript didn't support the concept of classes until the incorporation of ECMAScript 6 recommendations
but it has supported special constructor functions that work with
objects. By simply prefixing a call to a constructor function with the
keyword "new", we can tell JavaScript we would like the function to
behave like a constructor and instantiate a new object with the members
defined by that function.
Inside a constructor, the keyword this references the new object that's being created. Revisiting object creation, a basic constructor may look as follows:
01 | function Car( model, year, miles ) { |
07 | this .toString = function () { |
08 | return this .model + " has done " + this .miles + " miles" ; |
15 | var civic = new Car( "Honda Civic" , 2009, 20000 ); |
16 | var mondeo = new Car( "Ford Mondeo" , 2010, 5000 ); |
21 | console.log( civic.toString() ); |
22 | console.log( mondeo.toString() ); |
The above is a simple version of the constructor pattern but it does
suffer from some problems. One is that it makes inheritance difficult
and the other is that functions such as toString()
are
redefined for each of the new objects created using the Car constructor.
This isn't very optimal as the function should ideally be shared
between all of the instances of the Car type.
Thankfully as there are a number of both ES3 and ES5-compatible
alternatives to constructing objects, it's trivial to work around this
limitation.
Constructors With Prototypes
Functions in JavaScript have a property called a prototype. When we
call a JavaScript constructor to create an object, all the properties of
the constructor's prototype are then made available to the new object.
In this fashion, multiple Car objects can be created which access the
same prototype. We can thus extend the original example as follows:
01 | function Car( model, year, miles ) { |
12 | Car.prototype.toString = function () { |
13 | return this .model + " has done " + this .miles + " miles" ; |
18 | var civic = new Car( "Honda Civic" , 2009, 20000 ); |
19 | var mondeo = new Car( "Ford Mondeo" , 2010, 5000 ); |
21 | console.log( civic.toString() ); |
22 | console.log( mondeo.toString() ); |
Above, a single instance of toString() will now be shared between all of the Car objects.
#
The Module Pattern
Modules
Modules are an integral piece of any robust application's architecture
and typically help in keeping the units of code for a project both
cleanly separated and organized.
In JavaScript, there are several options for implementing modules. These include:
- The Module pattern
- Object literal notation
- AMD modules
- CommonJS modules
- ECMAScript Harmony modules
We will be exploring the latter three of these options later on in the book in the section Modern Modular JavaScript Design Patterns.
The Module pattern is based in part on object literals and so it makes sense to refresh our knowledge of them first.
Object Literals
In object literal notation, an object is described as a set of comma-separated name/value pairs enclosed in curly braces ({}
).
Names inside the object may be either strings or identifiers that are
followed by a colon. There should be no comma used after the final
name/value pair in the object as this may result in errors.
3 | variableKey: variableValue, |
5 | functionKey: function () { |
Object literals don't require instantiation using the new
operator but shouldn't be used at the start of a statement as the opening {
may be interpreted as the beginning of a block. Outside of an object,
new members may be added to it using assignment as follows myModule.property = "someValue";
Below we can see a more complete example of a module defined using object literal notation:
03 | myProperty: "someValue" , |
13 | myMethod: function () { |
14 | console.log( "Where in the world is Paul Irish today?" ); |
18 | myMethod2: function () { |
19 | console.log( "Caching is:" + ( this .myConfig.useCaching ) ? "enabled" : "disabled" ); |
23 | myMethod3: function ( newConfig ) { |
25 | if ( typeof newConfig === "object" ) { |
26 | this .myConfig = newConfig; |
27 | console.log( this .myConfig.language ); |
Using object literals can assist in encapsulating and organizing your
code and Rebecca Murphey has previously written about this topic in depth should you wish to read into object literals further.
That said, if we're opting for this technique, we may be equally as
interested in the Module pattern. It still uses object literals but only
as the return value from a scoping function.
The Module Pattern
The Module pattern was originally defined as a way to provide both
private and public encapsulation for classes in conventional software
engineering.
In JavaScript, the Module pattern is used to further emulate
the concept of classes in such a way that we're able to include both
public/private methods and variables inside a single object, thus
shielding particular parts from the global scope. What this results in
is a reduction in the likelihood of our function names conflicting with
other functions defined in additional scripts on the page.
Privacy
The Module pattern encapsulates "privacy", state and organization using
closures. It provides a way of wrapping a mix of public and private
methods and variables, protecting pieces from leaking into the global
scope and accidentally colliding with another developer's interface.
With this pattern, only a public API is returned, keeping everything
else within the closure private.
This gives us a clean solution for shielding logic doing the heavy
lifting whilst only exposing an interface we wish other parts of our
application to use. The pattern is quite similar to an
immediately-invoked functional expression (IIFE - see the section on namespacing patterns for more on this) except that an object is returned rather than a function.
It should be noted that there isn't really an explicitly true sense of
"privacy" inside JavaScript because unlike some traditional languages,
it doesn't have access modifiers. Variables can't technically be
declared as being public nor private and so we use function scope to
simulate this concept. Within the Module pattern, variables or methods
declared are only available inside the module itself thanks to closure.
Variables or methods defined within the returning object however are
available to everyone.
History
From a historical perspective, the Module pattern was originally developed by a number of people including Richard Cornford
in 2003. It was later popularized by Douglas Crockford in his lectures.
Another piece of trivia is that if you've ever played with Yahoo's YUI
library, some of its features may appear quite familiar and the reason
for this is that the Module pattern was a strong influence for YUI when
creating their components.
Examples
Let's begin looking at an implementation of the Module pattern by creating a module which is self-contained.
01 | var testModule = ( function () { |
07 | incrementCounter: function () { |
11 | resetCounter: function () { |
12 | console.log( "counter value prior to reset: " + counter ); |
22 | testModule.incrementCounter(); |
26 | testModule.resetCounter(); |
Here, other parts of the code are unable to directly read the value of our incrementCounter()
or resetCounter()
.
The counter variable is actually fully shielded from our global scope
so it acts just like a private variable would - its existence is limited
to within the module's closure so that the only code able to access its
scope are our two functions. Our methods are effectively namespaced so
in the test section of our code, we need to prefix any calls with the
name of the module (e.g. "testModule").
When working with the Module pattern, we may find it useful to define
a simple template that we use for getting started with it. Here's one
that covers namespacing, public and private variables:
01 | var myNamespace = ( function () { |
03 | var myPrivateVar, myPrivateMethod; |
09 | myPrivateMethod = function ( foo ) { |
19 | myPublicFunction: function ( bar ) { |
25 | myPrivateMethod( bar ); |
Looking at another example, below we can see a shopping basket
implemented using the this pattern. The module itself is completely
self-contained in a global variable called basketModule
. The basket
array in the module is kept private and so other parts of our
application are unable to directly read it. It only exists with the
module's closure and so the only methods able to access it are those
with access to its scope (ie. addItem()
, getItem()
etc).
01 | var basketModule = ( function () { |
07 | function doSomethingPrivate() { |
11 | function doSomethingElsePrivate() { |
19 | addItem: function ( values ) { |
24 | getItemCount: function () { |
29 | doSomething: doSomethingPrivate, |
32 | getTotal: function () { |
34 | var q = this .getItemCount(), |
Inside the module, you may have noticed that we return an object
. This gets automatically assigned to basketModule
so that we can interact with it as follows:
14 | console.log( basketModule.getItemCount() ); |
17 | console.log( basketModule.getTotal() ); |
24 | console.log( basketModule.basket ); |
The methods above are effectively namespaced inside basketModule
.
Notice how the scoping function in the above basket module is wrapped
around all of our functions, which we then call and immediately store
the return value of. This has a number of advantages including:
- The freedom to have private functions which can only be consumed
by our module. As they aren't exposed to the rest of the page (only our
exported API is), they're considered truly private.
- Given that functions are declared normally and are named, it can
be easier to show call stacks in a debugger when we're attempting to
discover what function(s) threw an exception.
- As T.J Crowder has pointed out in the past, it also enables us to
return different functions depending on the environment. In the past,
I've seen developers use this to perform UA testing in order to provide a
code-path in their module specific to IE, but we can easily opt for
feature detection these days to achieve a similar goal.
Module Pattern Variations
Import mixins
This variation of the pattern demonstrates how globals (e.g jQuery,
Underscore) can be passed in as arguments to our module's anonymous
function. This effectively allows us to import them and locally alias them as we wish.
02 | var myModule = ( function ( jQ, _ ) { |
04 | function privateMethod1(){ |
05 | jQ( ".container" ).html( "test" ); |
08 | function privateMethod2(){ |
09 | console.log( _.min([10, 5, 100, 2, 1000]) ); |
13 | publicMethod: function (){ |
21 | myModule.publicMethod(); |
Exports
This next variation allows us to declare globals without consuming
them and could similarly support the concept of global imports seen in
the last example.
02 | var myModule = ( function () { |
06 | privateVariable = "Hello World" ; |
08 | function privateMethod() { |
12 | module.publicProperty = "Foobar" ; |
13 | module.publicMethod = function () { |
14 | console.log( privateVariable ); |
Toolkit And Framework-specific Module Pattern Implementations
Dojo
Dojo provides a convenience method for working with objects called dojo.setObject()
. This takes as its first argument a dot-separated string such as myObj.parent.child
which refers to a property called "child" within an object "parent" defined inside "myObj". Using setObject()
allows us to set the value of children, creating any of the
intermediate objects in the rest of the path passed if they don't
already exist.
For example, if we wanted to declare basket.core
as an object of the store
namespace, this could be achieved as follows using the traditional way:
01 | var store = window.store || {}; |
03 | if ( !store[ "basket" ] ) { |
07 | if ( !store.basket[ "core" ] ) { |
08 | store.basket.core = {}; |
Or as follows using Dojo 1.7 (AMD-compatible version) and above:
01 | require([ "dojo/_base/customStore" ], function ( store ){ |
04 | store.setObject( "basket.core" , ( function () { |
08 | function privateMethod() { |
13 | publicMethod: function (){ |
For more information on dojo.setObject()
, see the official documentation.
ExtJS
For those using Sencha's ExtJS, you're in for some luck as the official documentation incorporates examples that do demonstrate how to correctly use the Module pattern with the framework.
Below we can see an example of how to define a namespace which can
then be populated with a module containing both a private and public
API. With the exception of some semantic differences, it's quite close
to how the Module pattern is implemented in vanilla JavaScript:
02 | Ext.namespace( "myNameSpace" ); |
05 | myNameSpace.app = function () { |
14 | var btn1Handler = function ( button, event ) { |
15 | console.log( "privVar1=" + privVar1 ); |
16 | console.log( "this.btn1Text=" + this .btn1Text ); |
29 | btn1 = new Ext.Button({ |
37 | btn1 = new Ext.Button( "btn1-ct" , { |
YUI
Similarly, we can also implement the Module pattern when building
applications using YUI3. The following example is heavily based on the
original YUI Module pattern implementation by Eric Miraglia, but again,
isn't vastly different from the vanilla JavaScript version:
01 | Y.namespace( "store.basket" ) = ( function () { |
03 | var myPrivateVar, myPrivateMethod; |
06 | myPrivateVar = "I can be accessed only within Y.store.basket." ; |
09 | myPrivateMethod = function () { |
10 | Y.log( "I can be accessed only from within YAHOO.store.basket" ); |
14 | myPublicProperty: "I'm a public property." , |
16 | myPublicMethod: function () { |
17 | Y.log( "I'm a public method." ); |
20 | Y.log( myPrivateVar ); |
21 | Y.log( myPrivateMethod() ); |
25 | Y.log( this .myPublicProperty ); |
jQuery
There are a number of ways in which jQuery code unspecific to plugins
can be wrapped inside the Module pattern. Ben Cherry previously
suggested an implementation where a function wrapper is used around
module definitions in the event of there being a number of commonalities
between modules.
In the following example, a library
function is defined which declares a new library and automatically binds up the init
function to document.ready
when new libraries (ie. modules) are created.
01 | function library( module ) { |
12 | var myLibrary = library( function () { |
Advantages
We've seen why the Singleton pattern can be useful, but why is the
Module pattern a good choice? For starters, it's a lot cleaner for
developers coming from an object-oriented background than the idea of
true encapsulation, at least from a JavaScript perspective.
Secondly, it supports private data - so, in the Module pattern,
public parts of our code are able to touch the private parts, however
the outside world is unable to touch the class's private parts (no
laughing! Oh, and thanks to David Engfer for the joke).
Disadvantages
The disadvantages of the Module pattern are that as we access both
public and private members differently, when we wish to change
visibility, we actually have to make changes to each place the member
was used.
We also can't access private members in methods that are added to the
object at a later point. That said, in many cases the Module pattern is
still quite useful and when used correctly, certainly has the potential
to improve the structure of our application.
Other disadvantages include the inability to create automated unit
tests for private members and additional complexity when bugs require
hot fixes. It's simply not possible to patch privates. Instead, one must
override all public methods which interact with the buggy privates.
Developers can't easily extend privates either, so it's worth
remembering privates are not as flexible as they may initially appear.
For further reading on the Module pattern, see Ben Cherry's excellent in-depth article on it.
#
The Revealing Module Pattern
Now that we're a little more familiar with the module pattern, let’s
take a look at a slightly improved version - Christian Heilmann’s
Revealing Module pattern.
The Revealing Module pattern came about as Heilmann was frustrated
with the fact that he had to repeat the name of the main object when we
wanted to call one public method from another or access public
variables. He also disliked the Module pattern’s requirement for
having to switch to object literal notation for the things he wished to
make public.
The result of his efforts was an updated pattern where we would
simply define all of our functions and variables in the private scope
and return an anonymous object with pointers to the private
functionality we wished to reveal as public.
An example of how to use the Revealing Module pattern can be found below:
01 | var myRevealingModule = function () { |
03 | var privateVar = "Ben Cherry" , |
04 | publicVar = "Hey there!" ; |
06 | function privateFunction() { |
07 | console.log( "Name:" + privateVar ); |
10 | function publicSetName( strName ) { |
14 | function publicGetName() { |
23 | setName: publicSetName, |
25 | getName: publicGetName |
30 | myRevealingModule.setName( "Paul Kinlan" ); |
The pattern can also be used to reveal private functions and properties with a more specific naming scheme if we would prefer:
01 | var myRevealingModule = function () { |
03 | var privateCounter = 0; |
05 | function privateFunction() { |
09 | function publicFunction() { |
13 | function publicIncrement() { |
17 | function publicGetCount(){ |
18 | return privateCounter; |
25 | start: publicFunction, |
26 | increment: publicIncrement, |
32 | myRevealingModule.start(); |
Advantages
This pattern allows the syntax of our scripts to be more consistent.
It also makes it more clear at the end of the module which of our
functions and variables may be accessed publicly which eases
readability.
Disadvantages
A disadvantage of this pattern is that if a private function refers
to a public function, that public function can't be overridden if a
patch is necessary. This is because the private function will continue
to refer to the private implementation and the pattern doesn't apply to
public members, only to functions.
Public object members which refer to private variables are also subject to the no-patch rule notes above.
As a result of this, modules created with the Revealing Module
pattern may be more fragile than those created with the original Module
pattern, so care should be taken during usage.
#
The Singleton Pattern
The Singleton pattern is thus known because it restricts
instantiation of a class to a single object. Classically, the Singleton
pattern can be implemented by creating a class with a method that
creates a new instance of the class if one doesn't exist. In the event
of an instance already existing, it simply returns a reference to that
object.
Singletons differ from static classes (or objects) as we can
delay their initialization, generally because they require some
information that may not be available during initialization time. They
don't provide a way for code that is unaware of a previous reference to
them to easily retrieve them. This is because it is neither the object
or "class" that's returned by a Singleton, it's a structure. Think of
how closured variables aren't actually closures - the function scope
that provides the closure is the closure.
In JavaScript, Singletons serve as a shared resource namespace which
isolate implementation code from the global namespace so as to provide a
single point of access for functions.
We can implement a Singleton as follows:
01 | var mySingleton = ( function () { |
11 | function privateMethod(){ |
12 | console.log( "I am private" ); |
15 | var privateVariable = "Im also private" ; |
17 | var privateRandomNumber = Math.random(); |
22 | publicMethod: function () { |
23 | console.log( "The public can see me!" ); |
26 | publicProperty: "I am also public" , |
28 | getRandomNumber: function () { |
29 | return privateRandomNumber; |
40 | getInstance: function () { |
53 | var myBadSingleton = ( function () { |
62 | var privateRandomNumber = Math.random(); |
66 | getRandomNumber: function () { |
67 | return privateRandomNumber; |
77 | getInstance: function () { |
91 | var singleA = mySingleton.getInstance(); |
92 | var singleB = mySingleton.getInstance(); |
93 | console.log( singleA.getRandomNumber() === singleB.getRandomNumber() ); |
95 | var badSingleA = myBadSingleton.getInstance(); |
96 | var badSingleB = myBadSingleton.getInstance(); |
97 | console.log( badSingleA.getRandomNumber() !== badSingleB.getRandomNumber() ); |
What makes the Singleton is the global access to the instance (generally through MySingleton.getInstance()
) as we don't (at least in static languages) call new MySingleton()
directly. This is however possible in JavaScript.
In the GoF book, the applicability of the Singleton pattern is described as follows:
- There must be exactly one instance of a class, and it must be accessible to clients from a well-known access point.
- When the sole instance should be extensible by subclassing, and
clients should be able to use an extended instance without modifying
their code.
The second of these points refers to a case where we might need code such as:
01 | mySingleton.getInstance = function (){ |
02 | if ( this ._instance == null ) { |
04 | this ._instance = new FooSingleton(); |
06 | this ._instance = new BasicSingleton(); |
09 | return this ._instance; |
Here, getInstance
becomes a little like a Factory method and we don't need to update each point in our code accessing it. FooSingleton
above would be a subclass of BasicSingleton
and implement the same interface.
Why is deferring execution considered important for a Singleton?:
In C++ it serves to isolate from the unpredictability of the
order of dynamic initialization, returning control to the programmer.
It is important to note the difference between a static instance of a
class (object) and a Singleton: whilst a Singleton can be implemented
as a static instance, it can also be constructed lazily, without the
need for resources nor memory until this is actually needed.
If we have a static object that can be initialized directly, we need
to ensure the code is always executed in the same order (e.g in case objCar
needs objWheel
during its initialization) and this doesn't scale when you have a large number of source files.
Both Singletons and static objects are useful but they shouldn't be
overused - the same way in which we shouldn't overuse other patterns.
In practice, the Singleton pattern is useful when exactly one object
is needed to coordinate others across a system. Here is one example with
the pattern being used in this context:
01 | var SingletonTester = ( function () { |
05 | function Singleton( options ) { |
09 | options = options || {}; |
12 | this .name = "SingletonTester" ; |
14 | this .pointX = options.pointX || 6; |
16 | this .pointY = options.pointY || 10; |
26 | name: "SingletonTester" , |
30 | getInstance: function ( options ) { |
31 | if ( instance === undefined ) { |
32 | instance = new Singleton( options ); |
44 | var singletonTest = SingletonTester.getInstance({ |
50 | console.log( singletonTest.pointX ); |
Whilst the Singleton has valid uses, often when we find ourselves
needing it in JavaScript it's a sign that we may need to re-evaluate our
design.
They're often an indication that modules in a system are either
tightly coupled or that logic is overly spread across multiple parts of a
codebase. Singletons can be more difficult to test due to issues
ranging from hidden dependencies, the difficulty in creating multiple
instances, difficulty in stubbing dependencies and so on.
Miller Medeiros has previously recommended this excellent article on the Singleton and its various issues for further reading as well as the comments to this
article, discussing how Singletons can increase tight coupling. I'm
happy to second these recommendations as both pieces raise many
important points about this pattern that are also worth noting.
#
The Observer Pattern
The Observer is a design pattern where an an object (known as a
subject) maintains a list of objects depending on it (observers),
automatically notifying them of any changes to state.
When a subject needs to notify observers about something interesting
happening, it broadcasts a notification to the observers (which can
include specific data related to the topic of the notification).
When we no longer wish for a particular observer to be notified of
changes by the subject they are registered with, the subject can remove
them from the list of observers.
It's often useful to refer back to published definitions of design
patterns that are language agnostic to get a broader sense of their
usage and advantages over time. The definition of the Observer pattern
provided in the GoF book, Design Patterns: Elements of Reusable Object-Oriented Software, is:
"One or more observers are interested in the state of a subject and
register their interest with the subject by attaching themselves. When
something changes in our subject that the observer may be interested in,
a notify message is sent which calls the update method in each
observer. When the observer is no longer interested in the subject's
state, they can simply detach themselves."
We can now expand on what we've learned to implement the Observer pattern with the following components:
- Subject: maintains a list of observers, facilitates adding or removing observers
- Observer: provides a update interface for objects that need to be notified of a Subject's changes of state
- ConcreteSubject: broadcasts notifications to observers on changes of state, stores the state of ConcreteObservers
- ConcreteObserver: stores a reference to the
ConcreteSubject, implements an update interface for the Observer to
ensure state is consistent with the Subject's
First, let's model the list of dependent Observers a subject may have:
01 | function ObserverList(){ |
02 | this .observerList = []; |
05 | ObserverList.prototype.Add = function ( obj ){ |
06 | return this .observerList.push( obj ); |
09 | ObserverList.prototype.Empty = function (){ |
10 | this .observerList = []; |
13 | ObserverList.prototype.Count = function (){ |
14 | return this .observerList.length; |
18 | ObserverList.prototype.Get = function ( index ){ |
19 | if ( index > -1 && index < this .observerList.length ){ |
20 | return this .observerList[ index ]; |
24 | ObserverList.prototype.Insert = function ( obj, index ){ |
28 | this .observerList.unshift( obj ); |
30 | } else if ( index === this .observerList.length ){ |
31 | this .observerList.push( obj ); |
38 | ObserverList.prototype.IndexOf = function ( obj, startIndex ){ |
39 | var i = startIndex, pointer = -1; |
41 | while ( i < this .observerList.length ){ |
42 | if ( this .observerList[i] === obj ){ |
52 | ObserverList.prototype.RemoveAt = function ( index ){ |
54 | this .observerList.shift(); |
55 | } else if ( index === this .observerList.length -1 ){ |
56 | this .observerList.pop(); |
62 | function extend( extension, obj ){ |
63 | for ( var key in extension ){ |
64 | obj[key] = extension[key]; |
Next, let's model the Subject and the ability to add, remove or notify observers on the observer list.
02 | this .observers = new ObserverList(); |
05 | Subject.prototype.AddObserver = function ( observer ){ |
06 | this .observers.Add( observer ); |
09 | Subject.prototype.RemoveObserver = function ( observer ){ |
10 | this .observers.RemoveAt( this .observers.IndexOf( observer, 0 ) ); |
13 | Subject.prototype.Notify = function ( context ){ |
14 | var observerCount = this .observers.Count(); |
15 | for ( var i=0; i < observerCount; i++){ |
16 | this .observers.Get(i).Update( context ); |
We then define a skeleton for creating new Observers. The Update
functionality here will be overwritten later with custom behaviour.
3 | this .Update = function (){ |
In our sample application using the above Observer components, we now define:
- A button for adding new observable checkboxes to the page
- A control checkbox which will act as a subject, notifying other checkboxes they should be checked
- A container for the new checkboxes being added
We then define ConcreteSubject and ConcreteObserver handlers for both
adding new observers to the page and implementing the updating
interface. See below for inline comments on what these components do in
the context of our example.
HTML:
1 | <button id= "addNewObserver" >Add New Observer checkbox</button> |
2 | <input id= "mainCheckbox" type= "checkbox" /> |
3 | <div id= "observersContainer" ></div> |
Sample script:
03 | var controlCheckbox = document.getElementById( "mainCheckbox" ), |
04 | addBtn = document.getElementById( "addNewObserver" ), |
05 | container = document.getElementById( "observersContainer" ); |
11 | extend( new Subject(), controlCheckbox ); |
14 | controlCheckbox[ "onclick" ] = new Function( "controlCheckbox.Notify(controlCheckbox.checked)" ); |
17 | addBtn[ "onclick" ] = AddNewObserver; |
21 | function AddNewObserver(){ |
24 | var check = document.createElement( "input" ); |
25 | check.type = "checkbox" ; |
28 | extend( new Observer(), check ); |
31 | check.Update = function ( value ){ |
37 | controlCheckbox.AddObserver( check ); |
40 | container.appendChild( check ); |
In this example, we looked at how to implement and utilize the
Observer pattern, covering the concepts of a Subject, Observer,
ConcreteSubject and ConcreteObserver.
Differences Between The Observer And Publish/Subscribe Pattern
Whilst the Observer pattern is useful to be aware of, quite often in
the JavaScript world, we'll find it commonly implemented using a
variation known as the Publish/Subscribe pattern. Whilst very similar,
there are differences between these patterns worth noting.
The Observer pattern requires that the observer (or object) wishing
to receive topic notifications must subscribe this interest to the
object firing the event (the subject).
The Publish/Subscribe pattern however uses a topic/event channel
which sits between the objects wishing to receive notifications
(subscribers) and the object firing the event (the publisher). This
event system allows code to define application specific events which can
pass custom arguments containing values needed by the subscriber. The
idea here is to avoid dependencies between the subscriber and publisher.
This differs from the Observer pattern as it allows any subscriber
implementing an appropriate event handler to register for and receive
topic notifications broadcast by the publisher.
Here is an example of how one might use the Publish/Subscribe if provided with a functional implementation powering publish()
,subscribe()
and unsubscribe()
behind the scenes:
10 | var subscriber1 = subscribe( "inbox/newMessage" , function ( topic, data ) { |
13 | console.log( "A new message was received: " , topic ); |
17 | $( ".messageSender" ).html( data.sender ); |
18 | $( ".messagePreview" ).html( data.body ); |
28 | var subscriber2 = subscribe( "inbox/newMessage" , function ( topic, data ) { |
30 | $( '.newMessageCounter' ).html( mailCounter++ ); |
34 | publish( "inbox/newMessage" , [{ |
35 | sender: "hello@google.com" , |
36 | body: "Hey there! How are you doing today?" |
The general idea here is the promotion of loose coupling. Rather than
single objects calling on the methods of other objects directly, they
instead subscribe to a specific task or activity of another object and
are notified when it occurs.
Advantages
The Observer and Publish/Subscribe patterns encourage us to think hard
about the relationships between different parts of our application. They
also help us identify what layers containing direct relationships which
could instead be replaced with sets of subjects and observers. This
effectively could be used to break down an application into smaller,
more loosely coupled blocks to improve code management and potentials
for re-use.
Further motivation behind using the Observer pattern is where we need
to maintain consistency between related objects without making classes
tightly coupled. For example, when an object needs to be able to notify
other objects without making assumptions regarding those objects.
Dynamic relationships can exist between observers and subjects when
using either pattern. This provides a great deal of flexibility which
may not be as easy to implement when disparate parts of our application
are tightly coupled.
Whilst it may not always be the best solution to every problem, these
patterns remain one of the best tools for designing decoupled systems
and should be considered an important tool in any JavaScript developer's
utility belt.
Disadvantages
Consequently, some of the issues with these patterns actually stem from
their main benefits. In Publish/Subscribe, by decoupling publishers from
subscribers, it can sometimes become difficult to obtain guarantees
that particular parts of our applications are functioning as we may
expect.
For example, publishers may make an assumption that one or more
subscribers are listening to them. Say that we're using such an
assumption to log or output errors regarding some application process.
If the subscriber performing the logging crashes (or for some reason
fails to function), the publisher won't have a way of seeing this due to
the decoupled nature of the system.
Another draw-back of the pattern is that subscribers are quite
ignorant to the existence of each other and are blind to the cost of
switching publishers. Due to the dynamic relationship between
subscribers and publishers, the update dependency can be difficult to
track.
Publish/Subscribe Implementations
Publish/Subscribe fits in very well in JavaScript ecosystems, largely
because at the core, ECMAScript implementations are event driven. This
is particularly true in browser environments as the DOM uses events as
its main interaction API for scripting.
That said, neither ECMAScript nor DOM provide core objects or methods
for creating custom events systems in implementation code (with the
exception of perhaps the DOM3 CustomEvent, which is bound to the DOM and
is thus not generically useful).
Luckily, popular JavaScript libraries such as dojo, jQuery (custom
events) and YUI already have utilities that can assist in easily
implementing a Publish/Subscribe system with very little effort. Below
we can see some examples of this:
04 | $( el ).trigger( "/login" , [{username: "test" , userData: "test" }] ); |
07 | dojo.publish( "/login" , [{username: "test" , userData: "test" }] ); |
10 | el.publish( "/login" , {username: "test" , userData: "test" } ); |
16 | $( el ).on( "/login" , function ( event ){...} ); |
19 | var handle = dojo.subscribe( "/login" , function (data){..} ); |
22 | el.on( "/login" , function ( data ){...} ); |
28 | $( el ).off( "/login" ); |
31 | dojo.unsubscribe( handle ); |
For those wishing to use the Publish/Subscribe pattern with vanilla JavaScript (or another library) AmplifyJS includes a clean, library-agnostic implementation that can be used with any library or toolkit. Radio.js (http://radio.uxder.com/), PubSubJS (https://github.com/mroderick/PubSubJS) or Pure JS PubSub by Peter Higgins (https://github.com/phiggins42/bloody-jquery-plugins/blob/55e41df9bf08f42378bb08b93efcb28555b61aeb/pubsub.js) are also similar alternatives worth checking out.
jQuery developers in particular have quite a few other options and can
opt to use one of the many well-developed implementations ranging from
Peter Higgins's jQuery plugin to Ben Alman's (optimized) Pub/Sub jQuery
gist on GitHub. Links to just a few of these can be found below.
So that we are able to get an appreciation for how many of the vanilla
JavaScript implementations of the Observer pattern might work, let's
take a walk through of a minimalist version of Publish/Subscribe I
released on GitHub under a project called pubsubz. This demonstrates the core concepts of subscribe, publish as well as the concept of unsubscribing.
I've opted to base our examples on this code as it sticks closely to
both the method signatures and approach of implementation I would expect
to see in a JavaScript version of the classic Observer pattern.
A Publish/Subscribe Implementation
11 | q.publish = function ( topic, args ) { |
13 | if ( !topics[topic] ) { |
17 | var subscribers = topics[topic], |
18 | len = subscribers ? subscribers.length : 0; |
21 | subscribers[len].func( topic, args ); |
31 | q.subscribe = function ( topic, func ) { |
37 | var token = ( ++subUid ).toString(); |
48 | q.unsubscribe = function ( token ) { |
49 | for ( var m in topics ) { |
51 | for ( var i = 0, j = topics[m].length; i < j; i++ ) { |
52 | if ( topics[m][i].token === token) { |
53 | topics[m].splice( i, 1 ); |
Example: Using Our Implementation
We can now use the implementation to publish and subscribe to events of interest as follows:
05 | var messageLogger = function ( topics, data ) { |
06 | console.log( "Logging: " + topics + ": " + data ); |
12 | var subscription = pubsub.subscribe( "inbox/newMessage" , messageLogger ); |
17 | pubsub.publish( "inbox/newMessage" , "hello world!" ); |
20 | pubsub.publish( "inbox/newMessage" , [ "test" , "a" , "b" , "c" ] ); |
23 | pubsub.publish( "inbox/newMessage" , { |
24 | sender: "hello@google.com" , |
35 | pubsub.publish( "inbox/newMessage" , "Hello! are you still there?" ); |
Example: User-Interface Notifications
Next, let's imagine we have a web application responsible for displaying real-time stock information.
The application might have a grid for displaying the stock stats and a
counter for displaying the last point of update. When the data model
changes, the application will need to update the grid and counter. In
this scenario, our subject (which will be publishing
topics/notifications) is the data model and our subscribers are the grid
and counter.
When our subscribers receive notification that the model itself has changed, they can update themselves accordingly.
In our implementation, our subscriber will listen to the topic
"newDataAvailable" to find out if new stock information is available. If
a new notification is published to this topic, it will trigger gridUpdate
to add a new row to our grid containing this information. It will also update a last updated counter to log the last time data was added
02 | getCurrentTime = function (){ |
04 | var date = new Date(), |
05 | m = date.getMonth() + 1, |
07 | y = date.getFullYear(), |
08 | t = date.toLocaleTimeString().toLowerCase(); |
10 | return (m + "/" + d + "/" + y + " " + t); |
14 | function addGridRow( data ) { |
17 | console.log( "updated grid component with:" + data ); |
23 | function updateCounter( data ) { |
26 | console.log( "data last updated at: " + getCurrentTime() + " with " + data); |
31 | gridUpdate = function ( topic, data ){ |
33 | if ( data !== "undefined" ) { |
35 | updateCounter( data ); |
41 | var subscriber = pubsub.subscribe( "newDataAvailable" , gridUpdate ); |
48 | pubsub.publish( "newDataAvailable" , { |
49 | summary: "Apple made $5 billion" , |
54 | pubsub.publish( "newDataAvailable" , { |
55 | summary: "Microsoft made $20 million" , |
Example: Decoupling applications using Ben Alman's Pub/Sub implementation
In the following movie ratings example, we'll be using Ben Alman's
jQuery implementation of Publish/Subscribe to demonstrate how we can
decouple a user interface. Notice how submitting a rating only has the
effect of publishing the fact that new user and rating data is
available.
It's left up to the subscribers to those topics to then delegate what
happens with that data. In our case we're pushing that new data into
existing arrays and then rendering them using the Underscore library's .template()
method for templating.
HTML/Templates
01 | <script id= "userTemplate" type= "text/html" > |
06 | <script id= "ratingsTemplate" type= "text/html" > |
07 | <li><strong><%= title %></strong> was rated <%= rating %>/5</li> |
13 | <div class= "sampleForm" > |
15 | <label for = "twitter_handle" >Twitter handle:</label> |
16 | <input type= "text" id= "twitter_handle" /> |
19 | <label for = "movie_seen" >Name a movie you've seen this year:</label> |
20 | <input type= "text" id= "movie_seen" /> |
24 | <label for = "movie_rating" >Rate the movie you saw:</label> |
25 | <select id= "movie_rating" > |
26 | <option value= "1" >1</option> |
27 | <option value= "2" >2</option> |
28 | <option value= "3" >3</option> |
29 | <option value= "4" >4</option> |
30 | <option value= "5" selected>5</option> |
36 | <button id= "add" >Submit rating</button> |
42 | <div class= "summaryTable" > |
43 | <div id= "users" ><h3>Recent users</h3></div> |
44 | <div id= "ratings" ><h3>Recent movies rated</h3></div> |
JavaScript
05 | userTemplate = _.template($( "#userTemplate" ).html()), |
06 | ratingsTemplate = _.template($( "#ratingsTemplate" ).html()); |
10 | $.subscribe( "/new/user" , function ( e, data ){ |
14 | $( '#users' ).append( userTemplate( data )); |
23 | $.subscribe( "/new/rating" , function ( e, data ){ |
29 | $( "#ratings" ).append( ratingsTemplate( data ); |
36 | $( "#add" ).on( "click" , function ( e ) { |
40 | var strUser = $( "#twitter_handle" ).val(), |
41 | strMovie = $( "#movie_seen" ).val(), |
42 | strRating = $( "#movie_rating" ).val(); |
45 | $.publish( "/new/user" , { name: strUser } ); |
48 | $.publish( "/new/rating" , { title: strMovie, rating: strRating} ); |
Example: Decoupling an Ajax-based jQuery application
In our final example, we're going to take a practical look at how
decoupling our code using Pub/Sub early on in the development process
can save us some potentially painful refactoring later on.
Quite often in Ajax-heavy applications, once we've received a response
to a request we want to achieve more than just one unique action. One
could simply add all of their post-request logic into a success
callback, but there are drawbacks to this approach.
Highly coupled applications sometimes increase the effort required to
reuse functionality due to the increased inter-function/code dependency.
What this means is that although keeping our post-request logic
hardcoded in a callback might be fine if we're just trying to grab a
result set once, it's not as appropriate when we want to make further
Ajax-calls to the same data source (and different end-behavior) without
rewriting parts of the code multiple times. Rather than having to go
back through each layer that calls the same data-source and generalizing
them later on, we can use pub/sub from the start and save time.
Using Observers, we can also easily separate application-wide
notifications regarding different events down to whatever level of
granularity we're comfortable with - something which can be less
elegantly done using other patterns.
Notice how in our sample below, one topic notification is made when a
user indicates they want to make a search query and another is made when
the request returns and actual data is available for consumption. It's
left up to the subscribers to then decide how to use knowledge of these
events (or the data returned). The benefits of this are that, if we
wanted, we could have 10 different subscribers utilizing the data
returned in different ways but as far as the Ajax-layer is concerned, it
doesn't care. Its sole duty is to request and return data then pass it
on to whoever wants to use it. This separation of concerns can make the
overall design of our code a little cleaner.
HTML/Templates:
01 | <form id= "flickrSearch" > |
03 | <input type= "text" name= "tag" id= "query" /> |
05 | <input type= "submit" name= "submit" value= "submit" /> |
11 | <div id= "lastQuery" ></div> |
13 | <div id= "searchResults" ></div> |
17 | <script id= "resultTemplate" type= "text/html" > |
18 | <% _.each(items, function ( item ){ %> |
19 | <li><p><img src= "<%= item.media.m %>" /></p></li> |
JavaScript:
04 | var resultTemplate = _.template($( "#resultTemplate" ).html()); |
07 | $.subscribe( "/search/tags" , function ( tags ) { |
09 | .html( "<p>Searched for:<strong>" + tags + "</strong></p>" ); |
13 | $.subscribe( "/search/resultSet" , function ( results ){ |
15 | $( "#searchResults" ).append(resultTemplate( results )); |
20 | $( "#flickrSearch" ).submit( function ( e ) { |
23 | var tags = $( this ).find( "#query" ).val(); |
29 | $.publish( "/search/tags" , [ $.trim(tags) ]); |
39 | $.subscribe( "/search/tags" , function ( tags ) { |
49 | if ( !data.items.length ) { |
53 | $.publish( "/search/resultSet" , data.items ); |
The Observer pattern is useful for decoupling a number of different
scenarios in application design and if you haven't been using it, I
recommend picking up one of the pre-written implementations mentioned
today and just giving it a try out. It's one of the easier design
patterns to get started with but also one of the most powerful.
The dictionary refers to a mediator as a neutral party that assists in negotiations and conflict resolution.
In our world, a mediator is a behavioral design pattern that allows us
to expose a unified interface through which the different parts of a
system may communicate.
If it appears a system has too many direct relationships between
components, it may be time to have a central point of control that
components communicate through instead. The Mediator promotes loose
coupling by ensuring that instead of components referring to each other
explicitly, their interaction is handled through this central point.
This can help us decouple systems and improve the potential for
component reusability.
A real-world analogy could be a typical airport traffic control system. A
tower (Mediator) handles what planes can take off and land because all
communications (notifications being listened out for or broadcast) are
done from the planes to the control tower, rather than from
plane-to-plane. A centralized controller is key to the success of this
system and that's really the role a Mediator plays in software design.
In implementation terms, the Mediator pattern is essentially a shared
subject in the Observer pattern. This might assume that a direction
Publish/Subscribe relationship between objects or modules in such
systems is sacrificed in order to maintain a central point of contact.
It may also be considered supplemental - perhaps used for
application-level notifications such as a communication between
different subsystems that are themselves complex and may desire internal
component decoupling through Publish/Subscribe relationships.
Another analogy would be DOM event bubbling and event delegation. If
all subscriptions in a system are made against the document rather than
individual nodes, the document effectively serves as a Mediator. Instead
of binding to the events of the individual nodes, a higher level object
is given the responsibility of notifying subscribers about interaction
events.
Basic Implementation
A simple implementation of the Mediator pattern can be found below, exposing both publish()
and subscribe()
methods for use:
01 | var mediator = ( function (){ |
08 | var subscribe = function ( topic, fn ){ |
10 | if ( !topics[topic] ){ |
14 | topics[topic].push( { context: this , callback: fn } ); |
20 | var publish = function ( topic ){ |
24 | if ( !topics[topic] ){ |
28 | args = Array.prototype.slice.call( arguments, 1 ); |
29 | for ( var i = 0, l = topics[topic].length; i < l; i++ ) { |
31 | var subscription = topics[topic][i]; |
32 | subscription.callback.apply( subscription.context, args ); |
40 | installTo: function ( obj ){ |
41 | obj.subscribe = subscribe; |
42 | obj.publish = publish; |
Advanced Implementation
For those interested in a more advanced implementation, read on for a
walk-through of my trimmed down version of Jack Lawson's excellent Mediator.js.
Amongst other improvements, this version supports topic namespaces,
subscriber removal and a much more robust Publish/Subscribe system for
our Mediator. If however, you wish to skip this walk-through, you can go
directly to the next example to continue reading.
Thanks to Jack for his excellent code comments which assisted with this section.
To start, let's implement the notion of Subscriber, which we can consider an instance of a Mediators topic registration.
By generating object instances, we can easily update Subscribers
later without the need to unregister and re-register them. Subscribers
can be written as constructors that take a function fn
to be called, an options
object and a context
.
05 | function guidGenerator() { } |
08 | function Subscriber( fn, options, context ){ |
10 | if ( !( this instanceof Subscriber) ) { |
12 | return new Subscriber( fn, context, options ); |
21 | this .id = guidGenerator(); |
23 | this .options = options; |
24 | this .context = context; |
Topics in our Mediator hold a list of callbacks and sub-topics that are fired when Mediator.Publish
is called on our Mediator instance. It also contains methods for manipulating lists of data.
05 | function Topic( namespace ){ |
07 | if ( !( this instanceof Topic) ) { |
08 | return new Topic( namespace ); |
11 | this .namespace = namespace || "" ; |
25 | AddSubscriber: function ( fn, options, context ){ |
27 | var callback = new Subscriber( fn, options, context ); |
29 | this ._callbacks.push( callback ); |
31 | callback.topic = this ; |
Our topic instance is passed along as an argument to the Mediator
callback. Further callback propagation can then be called using a handy
utility method called StopPropagation()
:
1 | StopPropagation: function (){ |
We can also make it easy to retrieve existing Subscribers when given a GUID identifier:
01 | GetSubscriber: function ( identifier ){ |
03 | for ( var x = 0, y = this ._callbacks.length; x < y; x++ ){ |
04 | if ( this ._callbacks[x].id == identifier || this ._callbacks[x].fn == identifier ){ |
05 | return this ._callbacks[x]; |
09 | for ( var z in this ._topics ){ |
10 | if ( this ._topics.hasOwnProperty( z ) ){ |
11 | var sub = this ._topics[z].GetSubscriber( identifier ); |
12 | if ( sub !== undefined ){ |
Next, in case we need them, we can offer easy ways to add new topics, check for existing topics or retrieve topics as well:
01 | AddTopic: function ( topic ){ |
02 | this ._topics[topic] = new Topic( ( this .namespace ? this .namespace + ":" : "" ) + topic ); |
05 | HasTopic: function ( topic ){ |
06 | return this ._topics.hasOwnProperty( topic ); |
09 | ReturnTopic: function ( topic ){ |
10 | return this ._topics[topic]; |
We can also explicitly remove Subscribers if we feel they are no
longer necessary. The following will recursively remove a Subscriber
through its sub-topics:
01 | RemoveSubscriber: function ( identifier ){ |
06 | for ( var z in this ._topics ){ |
07 | if ( this ._topics.hasOwnProperty(z) ){ |
08 | this ._topics[z].RemoveSubscriber( identifier ); |
13 | for ( var y = 0, x = this ._callbacks.length; y < x; y++ ) { |
14 | if ( this ._callbacks[y].fn == identifier || this ._callbacks[y].id == identifier ){ |
15 | this ._callbacks[y].topic = null ; |
16 | this ._callbacks.splice( y,1 ); |
Next we include the ability to Publish
arbitrary arguments to Subscribers recursively through sub-topics.
01 | Publish: function ( data ){ |
03 | for ( var y = 0, x = this ._callbacks.length; y < x; y++ ) { |
05 | var callback = this ._callbacks[y], l; |
06 | callback.fn.apply( callback.context, data ); |
08 | l = this ._callbacks.length; |
16 | for ( var x in this ._topics ){ |
18 | if ( this ._topics.hasOwnProperty( x ) ){ |
19 | this ._topics[x].Publish( data ); |
Here we expose the Mediator
instance we will primarily be interacting with. It is through here that events are registered and removed from topics.
3 | if ( !( this instanceof Mediator) ) { |
6 | this ._topics = new Topic( "" ); |
For more advanced use-cases, we can get our Mediator supporting namespaces for topics such as inbox:messages:new:read
.GetTopic
below returns topic instances based on a namespace.
03 | GetTopic: function ( namespace ){ |
04 | var topic = this ._topics, |
05 | namespaceHierarchy = namespace.split( ":" ); |
07 | if ( namespace === "" ){ |
11 | if ( namespaceHierarchy.length > 0 ){ |
12 | for ( var i = 0, j = namespaceHierarchy.length; i < j; i++ ){ |
14 | if ( !topic.HasTopic( namespaceHierarchy[i]) ){ |
15 | topic.AddTopic( namespaceHierarchy[i] ); |
18 | topic = topic.ReturnTopic( namespaceHierarchy[i] ); |
In this section we define a Mediator.Subscribe
method, which accepts a topic namespace, a function fn
to be executed, options
and once again a context
to call the function in to Subscribe. This creates a topic if one doesn't exist.
1 | Subscribe: function ( topiclName, fn, options, context ){ |
2 | var options = options || {}, |
3 | context = context || {}, |
4 | topic = this .GetTopic( topicName ), |
5 | sub = topic.AddSubscriber( fn, options, context ); |
Following on from this, we can also define further utilities for
accessing specific subscribers or removing them from topics recursively.
03 | GetSubscriber: function ( identifier, topic ){ |
04 | return this .GetTopic( topic || "" ).GetSubscriber( identifier ); |
10 | Remove: function ( topicName, identifier ){ |
11 | this .GetTopic( topicName ).RemoveSubscriber( identifier ); |
Our primary Publish
method allows us to arbitrarily publish data to a chosen topic namespace and can be seen below.
Topics are called recursively downwards. For example, a post to inbox:messages
will post to inbox:messages:new
and inbox:messages:new:read
. It is used as follows: Mediator.Publish( "inbox:messages:new", [args] );
1 | Publish: function ( topicName ){ |
2 | var args = Array.prototype.slice.call( arguments, 1), |
3 | topic = this .GetTopic( topicName ); |
7 | this .GetTopic( topicName ).Publish( args ); |
Finally we can easily expose our Mediator for attachment to the object passed in to root
:
1 | root.Mediator = Mediator; |
2 | Mediator.Topic = Topic; |
3 | Mediator.Subscriber = Subscriber; |
Example
Using either of the implementations from above (both the simple
option and the more advanced one), we can then put together a simple
chat logging system as follows:
HTML
03 | <label for = "fromBox" >Your Name:</label> |
04 | <input id= "fromBox" type= "text" /> |
06 | <label for = "toBox" >Send to:</label> |
07 | <input id= "toBox" type= "text" /> |
09 | <label for = "chatBox" >Message:</label> |
10 | <input id= "chatBox" type= "text" /> |
11 | <button type= "submit" >Chat</button> |
14 | <div id= "chatResult" ></div> |
JavaScript
01 | $( "#chatForm" ).on( "submit" , function (e) { |
05 | var text = $( "#chatBox" ).val(), |
06 | from = $( "#fromBox" ).val(), |
07 | to = $( "#toBox" ).val(); |
10 | mediator.publish( "newMessage" , { message: text, from: from, to: to } ); |
14 | function displayChat( data ) { |
15 | var date = new Date(), |
16 | msg = data.from + " said \"" + data.message + "\" to " + data.to; |
19 | .prepend( "<p>" + msg + " (" + date.toLocaleTimeString() + ")</p>" ); |
23 | function logChat( data ) { |
24 | if ( window.console ) { |
33 | mediator.subscribe( "newMessage" , displayChat ); |
34 | mediator.subscribe( "newMessage" , logChat ); |
39 | function amITalkingToMyself( data ) { |
40 | return data.from === data.to; |
43 | function iAmClearlyCrazy( data ) { |
44 | $( "#chatResult" ).prepend( "<p>" + data.from + " is talking to himself.</p>" ); |
47 | mediator.Subscribe( amITalkingToMyself, iAmClearlyCrazy ); |
#
Advantages & Disadvantages
The largest benefit of the Mediator pattern is that it reduces the
communication channels needed between objects or components in a system
from many to many to just many to one. Adding new publishers and
subscribers is relatively easy due to the level of decoupling present.
Perhaps the biggest downside of using the pattern is that it can
introduce a single point of failure. Placing a Mediator between modules
can also cause a performance hit as they are always communicating
indirectly. Because of the nature of loose coupling, it's difficult to
establish how a system might react by only looking at the broadcasts.
That said, it's useful to remind ourselves that decoupled systems
have a number of other benefits - if our modules communicated with each
other directly, changes to modules (e.g another module throwing an
exception) could easily have a domino effect on the rest of our
application. This problem is less of a concern with decoupled systems.
At the end of the day, tight coupling causes all kinds of headaches
and this is just another alternative solution, but one which can work
very well if implemented correctly.
Developers often wonder what the differences are between the Mediator
pattern and the Observer pattern. Admittedly, there is a bit of overlap,
but let's refer back to the GoF for an explanation:
"In the Observer pattern, there is no single object that
encapsulates a constraint. Instead, the Observer and the Subject must
cooperate to maintain the constraint. Communication patterns are
determined by the way observers and subjects are interconnected: a
single subject usually has many observers, and sometimes the observer of
one subject is a subject of another observer."
Both Mediators and Observers promote loose coupling, however, the
Mediator achieves this by having objects communicate strictly through
the Mediator. The Observer pattern creates observable objects which
publish events of interest occur to objects which are subscribed to
them.
We will be covering the Facade pattern shortly, but for reference
purposes some developers may also wonder whether there are similarities
between the Mediator and Facade patterns. They do both abstract the
functionality of existing modules, but there are some subtle
differences.
The Mediator centralizes communication between modules where it's
explicitly referenced by these modules. In a sense this is
multidirectional. The Facade however just defines a simpler interface to
a module or system but doesn't add any additional functionality. Other
modules in the system aren't directly aware of the concept of a facade
and could be considered unidirectional.
#
The Prototype Pattern
The GoF refer to the prototype pattern as one which creates objects based on a template of an existing object through cloning.
We can think of the prototype pattern as being based on prototypal
inheritance where we create objects which act as prototypes for other
objects. The prototype object itself is effectively used as a blueprint
for each object the constructor creates. If the prototype of the
constructor function used contains a property called name
for example (as per the code sample lower down), then each object
created by that same constructor will also have this same property.
Reviewing the definitions for this pattern in existing (non-JavaScript) literature, we may
find references to classes once again. The reality is that prototypal
inheritance avoids using classes altogether. There isn't a "definition"
object nor a core object in theory. we're simply creating copies of
existing functional objects.
One of the benefits of using the prototype pattern is that we're
working with the prototypal strengths JavaScript has to offer natively
rather than attempting to imitate features of other languages. With
other design patterns, this isn't always the case.
Not only is the pattern an easy way to implement inheritance, but it
can also come with a performance boost as well: when defining a function
in an object, they're all created by reference (so all child objects
point to the same function) instead of creating their own individual
copies.
For those interested, real prototypal inheritance, as defined in the ECMAScript 5 standard, requires the use of Object.create
(which we previously looked at earlier in this section). To remind ourselves, Object.create
creates an object which has a specified prototype and optionally contains specified properties as well (e.g Object.create( prototype, optionalDescriptorObjects )
).
We can see this demonstrated in the example below:
06 | console.log( "Weeee. I'm driving!" ); |
10 | console.log( "Wait. How do you stop this thing?" ); |
16 | var yourCar = Object.create( myCar ); |
19 | console.log( yourCar.name ); |
Object.create
also allows us to easily implement advanced
concepts such as differential inheritance where objects are able to
directly inherit from other objects. We saw earlier that Object.create
allows us to initialise object properties using the second supplied argument. For example:
02 | getModel: function () { |
03 | console.log( "The model of this vehicle is.." + this .model ); |
07 | var car = Object.create(vehicle, { |
10 | value: MY_GLOBAL.nextId(), |
Here the properties can be initialized on the second argument of Object.create
using an object literal with a syntax similar to that used by the Object.defineProperties
and Object.defineProperty
methods that we looked at previously.
It is worth noting that prototypal relationships can cause trouble
when enumerating properties of objects and (as Crockford recommends)
wrapping the contents of the loop in a hasOwnProperty()
check.
If we wish to implement the prototype pattern without directly using Object.create
, we can simulate the pattern as per the above example as follows:
01 | var vehiclePrototype = { |
03 | init: function ( carModel ) { |
04 | this .model = carModel; |
07 | getModel: function () { |
08 | console.log( "The model of this vehicle is.." + this .model); |
13 | function vehicle( model ) { |
16 | F.prototype = vehiclePrototype; |
25 | var car = vehicle( "Ford Escort" ); |
Note: This alternative does not allow the user to
define read-only properties in the same manner (as the vehiclePrototype
may be altered if not careful).
A final alternative implementation of the Prototype pattern could be the following:
1 | var beget = ( function () { |
5 | return function ( proto ) { |
One could reference this method from the vehicle
function. Note, however that vehicle
here is emulating a constructor, since the prototype pattern does not
include any notion of initialization beyond linking an object to a
prototype.
#
The Command Pattern
The Command pattern aims to encapsulate method invocation, requests
or operations into a single object and gives us the ability to both
parameterize and pass method calls around that can be executed at our
discretion. In addition, it enables us to decouple objects invoking the
action from the objects which implement them, giving us a greater degree
of overall flexibility in swapping out concrete classes (objects).
Concrete classes are best explained in terms of class-based programming languages and are related to the idea of abstract classes. An abstract
class defines an interface, but doesn't necessarily provide
implementations for all of its member functions. It acts as a base class
from which others are derived. A derived class which implements the
missing functionality is called a concrete class.
The general idea behind the Command pattern is that it provides us a
means to separate the responsibilities of issuing commands from anything
executing commands, delegating this responsibility to different objects
instead.
Implementation wise, simple command objects bind together both an
action and the object wishing to invoke the action. They consistently
include an execution operation (such as run()
or execute()
).
All Command objects with the same interface can easily be swapped as
needed and this is considered one of the larger benefits of the pattern.
To demonstrate the Command pattern we're going to create a simple car purchasing service.
06 | requestInfo: function ( model, id ){ |
07 | return "The information for " + model + " with ID " + id + " is foobar" ; |
11 | buyVehicle: function ( model, id ){ |
12 | return "You have successfully purchased Item " + id + ", a " + model; |
16 | arrangeViewing: function ( model, id ){ |
17 | return "You have successfully booked a viewing of " + model + " ( " + id + " ) " ; |
Taking a look at the above code, it would be trivial to invoke our CarManager
methods by directly accessing the object. We would all be forgiven for
thinking there is nothing wrong with this - technically, it's completely
valid JavaScript. There are however scenarios where this may be
disadvantageous.
For example, imagine if the core API behind the CarManager
changed. This would require all objects directly accessing these
methods within our application to also be modified. This could be viewed
as a layer of coupling which effectively goes against the OOP
methodology of loosely coupling objects as much as possible. Instead, we
could solve this problem by abstracting the API away further.
Let's now expand on our CarManager
so that our application of the Command pattern results in the following: accept any named methods that can be performed on the CarManager
object, passing along any data that might be used such as the Car model and ID.
Here is what we would like to be able to achieve:
1 | CarManager.execute( "buyVehicle" , "Ford Escort" , "453543" ); |
As per this structure we should now add a definition for the "CarManager.execute" method as follows:
1 | CarManager.execute = function ( name ) { |
2 | return CarManager[name] && CarManager[name].apply( CarManager, [].slice.call(arguments, 1) ); |
Our final sample calls would thus look as follows:
1 | CarManager.execute( "arrangeViewing" , "Ferrari" , "14523" ); |
2 | CarManager.execute( "requestInfo" , "Ford Mondeo" , "54323" ); |
3 | CarManager.execute( "requestInfo" , "Ford Escort" , "34232" ); |
4 | CarManager.execute( "buyVehicle" , "Ford Escort" , "34232" ); |
#
The Facade Pattern
When we put up a facade, we present an outward appearance to the
world which may conceal a very different reality. This was the
inspiration for the name behind the next pattern we're going to review -
the Facade pattern. This pattern provides a convenient higher-level
interface to a larger body of code, hiding its true underlying
complexity. Think of it as simplifying the API being presented to other
developers, something which almost always improves usability.
Facades are a structural pattern which can often be seen in JavaScript
libraries like jQuery where, although an implementation may support
methods with a wide range of behaviors, only a "facade" or limited
abstraction of these methods is presented to the public for use.
This allows us to interact with the Facade directly rather than the subsystem behind the scenes. Whenever we use jQuery's $(el).css()
or $(el).animate()
methods, we're actually using a Facade - the simpler public interface
that avoid us having to manually call the many internal methods in
jQuery core required to get some behavior working. This also avoids the
need to manually interact with DOM APIs and maintain state variables.
The jQuery core methods should be considered intermediate
abstractions. The more immediate burden to developers is the DOM API and
facades are what make the jQuery library so easy to use.
To build on what we've learned, the Facade pattern both simplifies the
interface of a class and it also decouples the class from the code that
utilizes it. This gives us the ability to indirectly interact with
subsystems in a way that can sometimes be less prone to error than
accessing the subsystem directly. A Facade's advantages include ease of
use and often a small size-footprint in implementing the pattern.
Let’s take a look at the pattern in action. This is an unoptimized code
example, but here we're utilizing a Facade to simplify an interface for
listening to events cross-browser. We do this by creating a common
method that can be used in one’s code which does the task of checking
for the existence of features so that it can provide a safe and
cross-browser compatible solution.
01 | var addMyEvent = function ( el,ev,fn ){ |
03 | if ( el.addEventListener ){ |
04 | el.addEventListener( ev,fn, false ); |
05 | } else if (el.attachEvent){ |
06 | el.attachEvent( "on" + ev, fn ); |
In a similar manner, we're all familiar with jQuery's $(document).ready(..)
. Internally, this is actually being powered by a method called bindReady()
, which is doing this:
01 | bindReady: function () { |
03 | if ( document.addEventListener ) { |
05 | document.addEventListener( "DOMContentLoaded" , DOMContentLoaded, false ); |
08 | window.addEventListener( "load" , jQuery.ready, false ); |
11 | } else if ( document.attachEvent ) { |
13 | document.attachEvent( "onreadystatechange" , DOMContentLoaded ); |
16 | window.attachEvent( "onload" , jQuery.ready ); |
This is another example of a Facade, where the rest of the world simply uses the limited interface exposed by $(document).ready(..)
and the more complex implementation powering it is kept hidden from sight.
Facades don't just have to be used on their own, however. They can also
be integrated with other patterns such as the Module pattern. As we can
see below, our instance of the module patterns contains a number of
methods which have been privately defined. A Facade is then used to
supply a much simpler API to accessing these methods:
01 | var module = ( function () { |
06 | console.log( "current value:" + this .i); |
08 | set : function ( val ) { |
12 | console.log( "running" ); |
15 | console.log( "jumping" ); |
21 | facade : function ( args ) { |
22 | _private.set(args.val); |
33 | module.facade( {run: true , val:10} ); |
In this example, calling module.facade()
will actually
trigger a set of private behavior within the module, but again, the user
isn't concerned with this. we've made it much easier for them to
consume a feature without needing to worry about implementation-level
details.
Notes on abstraction
Facades generally have few disadvantages, but one concern worth
noting is performance. Namely, one must determine whether there is an
implicit cost to the abstraction a Facade offers to our implementation
and if so, whether this cost is justifiable. Going back to the jQuery
library, most of us are aware that both getElementById("identifier")
and $("#identifier")
can be used to query an element on a page by its ID.
Did you know however that getElementById()
on its own is
significantly faster by a high order of magnitude? Take a look at this
jsPerf test to see results on a per-browser level: http://jsperf.com/getelementbyid-vs-jquery-id.
Now of course, we have to keep in mind that jQuery (and Sizzle - its
selector engine) are doing a lot more behind the scenes to optimize our
query (and that a jQuery object, not just a DOM node is returned).
The challenge with this particular Facade is that in order to provide
an elegant selector function capable of accepting and parsing multiple
types of queries, there is an implicit cost of abstraction. The user
isn't required to access jQuery.getById("identifier")
or jQuery.getbyClass("identifier")
and so on. That said, the trade-off in performance has been tested in
practice over the years and given the success of jQuery, a simple Facade
actually worked out very well for the team.
When using the pattern, try to be aware of any performance costs
involved and make a call on whether they are worth the level of
abstraction offered.
#
The Factory Pattern
The Factory pattern is another creational pattern concerned with the
notion of creating objects. Where it differs from the other patterns in
its category is that it doesn't explicitly require us use a constructor.
Instead, a Factory can provide a generic interface for creating
objects, where we can specify the type of factory object we wish to be
created.
Imagine that we have a UI factory where we are asked to create a type
of UI component. Rather than creating this component directly using the
new
operator or via another creational constructor, we ask
a Factory object for a new component instead. We inform the Factory
what type of object is required (e.g "Button", "Panel") and it
instantiates this, returning it to us for use.
This is particularly useful if the object creation process is
relatively complex, e.g. if it strongly depends on dynamic factors or
application configuration.
Examples of this pattern can be found in UI libraries such as ExtJS
where the methods for creating objects or components may be further
subclassed.
The following is an example that builds upon our previous snippets
using the Constructor pattern logic to define cars. It demonstrates how a
Vehicle Factory may be implemented using the Factory pattern:
04 | function Car( options ) { |
07 | this .doors = options.doors || 4; |
08 | this .state = options.state || "brand new" ; |
09 | this .color = options.color || "silver" ; |
14 | function Truck( options){ |
16 | this .state = options.state || "used" ; |
17 | this .wheelSize = options.wheelSize || "large" ; |
18 | this .color = options.color || "blue" ; |
25 | function VehicleFactory() {} |
30 | VehicleFactory.prototype.vehicleClass = Car; |
33 | VehicleFactory.prototype.createVehicle = function ( options ) { |
35 | if ( options.vehicleType === "car" ){ |
36 | this .vehicleClass = Car; |
38 | this .vehicleClass = Truck; |
41 | return new this .vehicleClass( options ); |
46 | var carFactory = new VehicleFactory(); |
47 | var car = carFactory.createVehicle( { |
55 | console.log( car instanceof Car ); |
Approach #1: Modify a VehicleFactory instance to use the Truck class
01 | var movingTruck = carFactory.createVehicle( { |
05 | wheelSize: "small" } ); |
10 | console.log( movingTruck instanceof Truck ); |
14 | console.log( movingTruck ); |
Approach #2: Subclass VehicleFactory to create a factory class that builds Trucks
01 | function TruckFactory () {} |
02 | TruckFactory.prototype = new VehicleFactory(); |
03 | TruckFactory.prototype.vehicleClass = Truck; |
05 | var truckFactory = new TruckFactory(); |
06 | var myBigTruck = truckFactory.createVehicle( { |
07 | state: "omg..so bad." , |
09 | wheelSize: "so big" } ); |
13 | console.log( myBigTruck instanceof Truck ); |
17 | console.log( myBigTruck ); |
When To Use The Factory Pattern
The Factory pattern can be especially useful when applied to the following situations:
- When our object or component setup involves a high level of complexity
- When we need to easily generate different instances of objects depending on the environment we are in
- When we're working with many small objects or components that share the same properties
- When composing objects with instances of other objects that need
only satisfy an API contract (aka, duck typing) to work. This is useful
for decoupling.
When Not To Use The Factory Pattern
When applied to the wrong type of problem, this pattern can introduce an
unnecessarily great deal of complexity to an application. Unless
providing an interface for object creation is a design goal for the
library or framework we are writing, I would suggest sticking to
explicit constructors to avoid the unnecessary overhead.
Due to the fact that the process of object creation is effectively
abstracted behind an interface, this can also introduce problems with
unit testing depending on just how complex this process might be.
Abstract Factories
It is also useful to be aware of the Abstract Factory
pattern, which aims to encapsulate a group of individual factories with
a common goal. It separates the details of implementation of a set of
objects from their general usage.
An Abstract Factory should be used where a system must be independent
from the way the objects it creates are generated or it needs to work
with multiple types of objects.
An example which is both simple and easier to understand is a vehicle
factory, which defines ways to get or register vehicles types. The
abstract factory can be named AbstractVehicleFactory. The Abstract
factory will allow the definition of types of vehicle like "car" or
"truck" and concrete factories will implement only classes that fulfill
the vehicle contract (e.g Vehicle.prototype.drive
and Vehicle.prototype.breakDown
).
01 | var AbstractVehicleFactory = ( function () { |
07 | getVehicle: function ( type, customizations ) { |
08 | var Vehicle = types[type]; |
10 | return (Vehicle ? new Vehicle(customizations) : null ); |
13 | registerVehicle: function ( type, Vehicle ) { |
14 | var proto = Vehicle.prototype; |
17 | if ( proto.drive && proto.breakDown ) { |
18 | types[type] = Vehicle; |
21 | return AbstractVehicleFactory; |
29 | AbstractVehicleFactory.registerVehicle( "car" , Car ); |
30 | AbstractVehicleFactory.registerVehicle( "truck" , Truck ); |
33 | var car = AbstractVehicleFactory.getVehicle( "car" , { |
35 | state: "like new" } ); |
38 | var truck = AbstractVehicleFactory.getVehicle( "truck" , { |
40 | color: "neon yellow" } ); |
#
The Mixin Pattern
In traditional programming languages such as C++ and Lisp, Mixins are
classes which offer functionality that can be easily inherited by a
sub-class or group of sub-classes for the purpose of function re-use.
Sub-classing
For developers unfamiliar with sub-classing, we will go through a
brief beginners primer on them before diving into Mixins and Decorators
further.
Sub-classing is a term that refers to inheriting properties for a new object from a base or superclass object. In traditional object-oriented programming, a class B
is able to extend another class A
. Here we consider A
a superclass and B
a subclass of A
. As such, all instances of B
inherit the methods from A
. B
is however still able to define its own methods, including those that override methods originally defined by A
.
Should B
need to invoke a method in A
that has been overridden, we refer to this as method chaining. Should B
need to invoke the constructor A
(the superclass), we call this constructor chaining.
In order to demonstrate sub-classing, we first need a base object
that can have new instances of itself created. let's model this around
the concept of a person.
1 | var Person = function ( firstName , lastName ){ |
3 | this .firstName = firstName; |
4 | this .lastName = lastName; |
Next, we'll want to specify a new class (object) that's a subclass of the existing Person
object. Let us imagine we want to add distinct properties to distinguish a Person
from a Superhero
whilst inheriting the properties of the Person
"superclass". As superheroes share many common traits with normal
people (e.g. name, gender), this should hopefully illustrate how
sub-classing works adequately.
02 | var clark = new Person( "Clark" , "Kent" ); |
05 | var Superhero = function ( firstName, lastName , powers ){ |
11 | Person.call( this , firstName, lastName ); |
17 | SuperHero.prototype = Object.create( Person.prototype ); |
18 | var superman = new Superhero( "Clark" , "Kent" , [ "flight" , "heat-vision" ] ); |
19 | console.log( superman ); |
The Superhero
constructor creates an object which descends from Person
. Objects of this type have attributes of the objects that are above it in the chain and if we had set default values in the Person
object, Superhero
is capable of overriding any inherited values with values specific to it's object.
Mixins
In JavaScript, we can look at inheriting from Mixins as a means of
collecting functionality through extension. Each new object we define
has a prototype from which it can inherit further properties. Prototypes
can inherit from other object prototypes but, even more importantly,
can define properties for any number of object instances. We can
leverage this fact to promote function re-use.
Mixins allow objects to borrow (or inherit) functionality from them
with a minimal amount of complexity. As the pattern works well with
JavaScripts object prototypes, it gives us a fairly flexible way to
share functionality from not just one Mixin, but effectively many
through multiple inheritance.
They can be viewed as objects with attributes and methods that can be
easily shared across a number of other object prototypes. Imagine that
we define a Mixin containing utility functions in a standard object
literal as follows:
04 | console.log( "move up" ); |
08 | console.log( "move down" ); |
12 | console.log( "stop! in the name of love!" ); |
We can then easily extend the prototype of existing constructor
functions to include this behavior using a helper such as the
Underscore.js _.extend()
method:
02 | function carAnimator(){ |
03 | this .moveLeft = function (){ |
04 | console.log( "move left" ); |
09 | function personAnimator(){ |
10 | this .moveRandomly = function (){ }; |
14 | _.extend( carAnimator.prototype, myMixins ); |
15 | _.extend( personAnimator.prototype, myMixins ); |
18 | var myAnimator = new carAnimator(); |
As we can see, this allows us to easily "mix" in common behaviour into object constructors fairly trivially.
In the next example, we have two constructors: a Car and a Mixin.
What we're going to do is augment (another way of saying extend) the Car
so that it can inherit specific methods defined in the Mixin, namely driveForward()
and driveBackward()
. This time we won't be using Underscore.js.
Instead, this example will demonstrate how to augment a constructor
to include functionality without the need to duplicate this process for
every constructor function we may have.
02 | var Car = function ( settings ) { |
04 | this .model = settings.model || "no model provided" ; |
05 | this .color = settings.color || "no colour provided" ; |
10 | var Mixin = function () {}; |
14 | driveForward: function () { |
15 | console.log( "drive forward" ); |
18 | driveBackward: function () { |
19 | console.log( "drive backward" ); |
22 | driveSideways: function () { |
23 | console.log( "drive sideways" ); |
30 | function augment( receivingClass, givingClass ) { |
34 | for ( var i = 2, len = arguments.length; i < len; i++ ) { |
35 | receivingClass.prototype[arguments[i]] = givingClass.prototype[arguments[i]]; |
40 | for ( var methodName in givingClass.prototype ) { |
45 | if ( !Object.hasOwnProperty(receivingClass.prototype, methodName) ) { |
46 | receivingClass.prototype[methodName] = givingClass.prototype[methodName]; |
59 | augment( Car, Mixin, "driveForward" , "driveBackward" ); |
79 | var mySportsCar = new Car({ |
84 | mySportsCar.driveSideways(); |
Advantages & Disadvantages
Mixins assist in decreasing functional repetition and increasing
function re-use in a system. Where an application is likely to require
shared behaviour across object instances, we can easily avoid any
duplication by maintaining this shared functionality in a Mixin and thus
focusing on implementing only the functionality in our system which is
truly distinct.
That said, the downsides to Mixins are a little more debatable. Some
developers feel that injecting functionality into an object prototype is
a bad idea as it leads to both prototype pollution and a level of
uncertainly regarding the origin of our functions. In large systems this
may well be the case.
I would argue that strong documentation can assist in minimizing the
amount of confusion regarding the source of mixed in functions, but as
with every pattern, if care is taken during implementation we should be
okay.
The Decorator Pattern
Decorators are a structural design pattern that aim to promote code
re-use. Similar to Mixins, they can be considered another viable
alternative to object sub-classing.
Classically, Decorators offered the ability to add behaviour to existing classes in a system dynamically. The idea was that the decoration itself wasn't essential to the base functionality of the class, otherwise it would be baked into the superclass itself.
They can be used to modify existing systems where we wish to add
additional features to objects without the need to heavily modify the
underlying code using them. A common reason why developers use them is
their applications may contain features requiring a large quantity of
distinct types of object. Imagine having to define hundreds of different
object constructors for say, a JavaScript game.
The object constructors could represent distinct player types, each with differing capabilities. A Lord of the Rings game could require constructors for Hobbit
, Elf
, Orc
, Wizard
, Mountain Giant
, Stone Giant
and so on, but there could easily be hundreds of these. If we then
factored in capabilities, imagine having to create sub-classes for each
combination of capability type e.g HobbitWithRing
,HobbitWithSword
, HobbitWithRingAndSword
and so on.This isn't very practical and certainly isn't manageable when we factor in a growing number of different abilities.
The Decorator pattern isn't heavily tied to how objects are created
but instead focuses on the problem of extending their functionality.
Rather than just relying on prototypal inheritance, we work with a
single base object and progressively add decorator objects which provide
the additional capabilities. The idea is that rather than sub-classing,
we add (decorate) properties or methods to a base object so it's a
little more streamlined.
Adding new attributes to objects in JavaScript is a very
straight-forward process so with this in mind, a very simplistic
decorator may be implemented as follows:
Example 1: Decorating Constructors With New Functionality
02 | function vehicle( vehicleType ){ |
05 | this .vehicleType = vehicleType || "car" ; |
06 | this .model = "default" ; |
07 | this .license = "00000-000" ; |
12 | var testInstance = new vehicle( "car" ); |
13 | console.log( testInstance ); |
19 | var truck = new vehicle( "truck" ); |
22 | truck.setModel = function ( modelName ){ |
23 | this .model = modelName; |
26 | truck.setColor = function ( color ){ |
31 | truck.setModel( "CAT" ); |
32 | truck.setColor( "blue" ); |
40 | var secondInstance = new vehicle( "car" ); |
41 | console.log( secondInstance ); |
This type of simplistic implementation is functional, but it doesn't
really demonstrate all of the strengths Decorators have to offer. For
this, we're first going to go through my variation of the Coffee example
from an excellent book called Head First Design Patterns by Freeman, Sierra and Bates, which is modeled around a Macbook purchase.
Example 2: Decorating Objects With Multiple Decorators
04 | this .cost = function () { return 997; }; |
05 | this .screenSize = function () { return 11.6; }; |
10 | function Memory( macbook ) { |
12 | var v = macbook.cost(); |
13 | macbook.cost = function () { |
20 | function Engraving( macbook ){ |
22 | var v = macbook.cost(); |
23 | macbook.cost = function (){ |
30 | function Insurance( macbook ){ |
32 | var v = macbook.cost(); |
33 | macbook.cost = function (){ |
39 | var mb = new MacBook(); |
45 | console.log( mb.cost() ); |
48 | console.log( mb.screenSize() ); |
In the above example, our Decorators are overriding the MacBook()
super-class objects .cost()
function to return the current price of the Macbook
plus the cost of the upgrade being specified.
It's considered a decoration as the original Macbook
objects constructor methods which are not overridden (e.g. screenSize()
) as well as any other properties which we may define as a part of the Macbook
remain unchanged and intact.
There isn't really a defined interface in the above example
and we're shifting away the responsibility of ensuring an object meets
an interface when moving from the creator to the receiver.
Pseudo-classical Decorators
We're now going to examine a variation of the Decorator first presented in a JavaScript form in Pro JavaScript Design Patterns (PJDP) by Dustin Diaz and Ross Harmes.
Unlike some of the examples from earlier, Diaz and Harmes stick more
closely to how decorators are implemented in other programming languages
(such as Java or C++) using the concept of an "interface", which we
will define in more detail shortly.
Note: This particular variation of the Decorator
pattern is provided for reference purposes. If finding it overly
complex, I recommend opting for one of the simpler implementations
covered earlier.
Interfaces
PJDP describes the Decorator as a pattern that is used to
transparently wrap objects inside other objects of the same interface.
An interface is a way of defining the methods an object should have, however, it doesn't actually directly specify how those methods should be implemented.
They can also indicate what parameters the methods take, but this is considered optional.
So, why would we use an interface in JavaScript? The idea is that
they're self-documenting and promote reusability. In theory, interfaces
also make code more stable by ensuring changes to them must also be made
to the objects implementing them.
Below is an example of an implementation of interfaces in JavaScript
using duck-typing - an approach that helps determine whether an object
is an instance of constructor/object based on the methods it implements.
08 | var reminder = new Interface( "List" , [ "summary" , "placeOrder" ] ); |
11 | name: "Remember to buy the milk" , |
15 | return "Remember to buy the milk, we are almost out!" ; |
17 | placeOrder: function (){ |
18 | return "Ordering milk from your local grocery store" ; |
26 | function Todo( config ){ |
32 | Interface.ensureImplements( config.actions, reminder ); |
34 | this .name = config.name; |
35 | this .methods = config.actions; |
41 | var todoItem = Todo( properties ); |
45 | console.log( todoItem.methods.summary() ); |
46 | console.log( todoItem.methods.placeOrder() ); |
In the above, Interface.ensureImplements
provides strict functionality checking and code for both this and the Interface
constructor can be found here.
The biggest problem with interfaces is that, as there isn't built-in
support for them in JavaScript, there is a danger of us attempting to
emulate a feature of another language that may not be an ideal fit.
Lightweight interfaces can be used without a great performance cost
however and we will next look at Abstract Decorators using this same concept.
Abstract Decorators
To demonstrate the structure of this version of the Decorator
pattern, we're going to imagine we have a superclass that models a Macbook
once again and a store that allows us to "decorate" our Macbook with a number of enhancements for an additional fee.
Enhancements can include upgrades to 4GB or 8GB Ram, engraving,
Parallels or a case. Now if we were to model this using an individual
sub-class for each combination of enhancement options, it might look
something like this:
01 | var Macbook = function (){ |
05 | var MacbookWith4GBRam = function (){}, |
06 | MacbookWith8GBRam = function (){}, |
07 | MacbookWith4GBRamAndEngraving = function (){}, |
08 | MacbookWith8GBRamAndEngraving = function (){}, |
09 | MacbookWith8GBRamAndParallels = function (){}, |
10 | MacbookWith4GBRamAndParallels = function (){}, |
11 | MacbookWith8GBRamAndParallelsAndCase = function (){}, |
12 | MacbookWith4GBRamAndParallelsAndCase = function (){}, |
13 | MacbookWith8GBRamAndParallelsAndCaseAndInsurance = function (){}, |
14 | MacbookWith4GBRamAndParallelsAndCaseAndInsurance = function (){}; |
and so on.
This would be an impractical solution as a new subclass would be
required for every possible combination of enhancements that are
available. As we would prefer to keep things simple without maintaining a
large set of subclasses, let's look at how decorators may be used to
solve this problem better.
Rather than requiring all of the combinations we saw earlier, we
should simply have to create five new decorator classes. Methods that
are called on these enhancement classes would be passed on to our Macbook
class.
In our next example, decorators transparently wrap around their
components and can interestingly be interchanged astray use the same
interface.
Here's the interface we're going to define for the Macbook:
01 | var Macbook = new Interface( "Macbook" , |
09 | var MacbookPro = function (){ |
13 | MacbookPro.prototype = { |
14 | addEngraving: function (){ |
16 | addParallels: function (){ |
18 | add4GBRam: function (){ |
To make it easier for us to add as many more options as needed later
on, an Abstract Decorator class is defined with default methods required
to implement the Macbook
interface, which the rest of the
options will sub-class. Abstract Decorators ensure that we can decorate a
base class independently with as many decorators as needed in different
combinations (remember the example earlier?) without needing to derive a
class for every possible combination.
03 | var MacbookDecorator = function ( macbook ){ |
05 | Interface.ensureImplements( macbook, Macbook ); |
06 | this .macbook = macbook; |
10 | MacbookDecorator.prototype = { |
11 | addEngraving: function (){ |
12 | return this .macbook.addEngraving(); |
14 | addParallels: function (){ |
15 | return this .macbook.addParallels(); |
17 | add4GBRam: function (){ |
18 | return this .macbook.add4GBRam(); |
21 | return this .macbook.add8GBRam(); |
24 | return this .macbook.addCase(); |
27 | return this .macbook.getPrice(); |
What's happening in the above sample is that the Macbook
Decorator is taking an object to use as the component. It's using the Macbook
interface we defined earlier and for each method is just calling the
same method on the component. We can now create our option classes just
by using the Macbook
Decorator - simply call the superclass constructor and any methods can be overridden as per necessary.
01 | var CaseDecorator = function ( macbook ){ |
04 | this .superclass.constructor( macbook ); |
09 | extend( CaseDecorator, MacbookDecorator ); |
11 | CaseDecorator.prototype.addCase = function (){ |
12 | return this .macbook.addCase() + "Adding case to macbook" ; |
15 | CaseDecorator.prototype.getPrice = function (){ |
16 | return this .macbook.getPrice() + 45.00; |
As we can see, most of this is relatively straight-forward to implement. What we're doing is overriding the addCase()
and getPrice()
methods that need to be decorated and we're achieving this by first executing the component's method and then adding to it.
As there's been quite a lot of information presented in this section
so far, let's try to bring it all together in a single example that will
hopefully highlight what we have learned.
02 | var myMacbookPro = new MacbookPro(); |
05 | console.log( myMacbookPro.getPrice() ); |
08 | myMacbookPro = new CaseDecorator( myMacbookPro ); |
11 | console.log( myMacbookPro.getPrice() ); |
As decorators are able to modify objects dynamically, they're a
perfect pattern for changing existing systems. Occasionally, it's just
simpler to create decorators around an object versus the trouble of
maintaining individual sub-classes for each object type. This makes
maintaining applications that may require a large number of sub-classed
objects significantly more straight-forward.
Decorators With jQuery
As with other patterns we've covered, there are also examples of the Decorator pattern that can be implemented with jQuery. jQuery.extend()
allows us to extend (or merge) two or more objects (and their
properties) together into a single object either at run-time or
dynamically at a later point.
In this scenario, a target object can be decorated with new
functionality without necessarily breaking or overriding existing
methods in the source/superclass object (although this can be done).
In the following example, we define three objects: defaults, options and settings. The aim of the task is to decorate the defaults
object with additional functionality found in optionssettings
. We must:
(a) Leave "defaults" in an untouched state where we don't lose the
ability to access the properties or functions found in it a later point
(b) Gain the ability to use the decorated properties and functions found
in "options"
01 | var decoratorApp = decoratorApp || {}; |
10 | welcome: function () { |
11 | console.log( "welcome!" ); |
18 | helloWorld: function () { |
19 | console.log( "hello world" ); |
25 | printObj: function ( obj ) { |
28 | $.each( obj, function ( key, val ) { |
30 | next += $.isPlainObject(val) ? printObj( val ) : val; |
34 | return "{ " + arr.join( ", " ) + " }" ; |
40 | decoratorApp.settings = $.extend({}, decoratorApp.defaults, decoratorApp.options); |
47 | .append( decoratorApp.printObj(decoratorApp.settings) + |
48 | + decoratorApp.printObj(decoratorApp.options) + |
49 | + decoratorApp.printObj(decoratorApp.defaults)); |
Advantages & Disadvantages
Developers enjoy using this pattern as it can be used transparently
and is also fairly flexible - as we've seen, objects can be wrapped or
"decorated" with new behavior and then continue to be used without
needing to worry about the base object being modified. In a broader
context, this pattern also avoids us needing to rely on large numbers of
subclasses to get the same benefits.
There are however drawbacks that we should be aware of when
implementing the pattern. If poorly managed, it can significantly
complicate our application architecture as it introduces many small, but
similar objects into our namespace. The concern here is that in
addition to becoming hard to manage, other developers unfamiliar with
the pattern may have a hard time grasping why it's being used.
Sufficient commenting or pattern research should assist with the
latter, however as long as we keep a handle on how widespread we use the
decorator in our applications we should be fine on both counts.
#
Flyweight
The Flyweight pattern is a classical structural solution for
optimizing code that is repetitive, slow and inefficiently shares data.
It aims to minimize the use of memory in an application by sharing as
much data as possible with related objects (e.g application
configuration, state and so on).
The pattern was first conceived by Paul Calder and Mark Linton in
1990 and was named after the boxing weight class that includes fighters
weighing less than 112lb. The name Flyweight itself is derived from this
weight classification as it refers to the small weight (memory
footprint) the pattern aims to help us achieve.
In practice, Flyweight data sharing can involve taking several
similar objects or data constructs used by a number of objects and
placing this data into a single external object. We can pass through
this object through to those depending on this data, rather than storing
identical data across each one.
Using Flyweights
There are two ways in which the Flyweight pattern can be applied. The
first is at the data-layer, where we deal with the concept of sharing
data between large quantities of similar objects stored in memory.
The second is at the DOM-layer where the Flyweight can be used as a
central event-manager to avoid attaching event handlers to every child
element in a parent container we wish to have some similar behavior.
As the data-layer is where the flyweight pattern is most used traditionally, we'll take a look at this first.
Flyweights and sharing data
For this application, there are a few more concepts around the
classical Flyweight pattern that we need to be aware of. In the
Flyweight pattern there's a concept of two states - intrinsic and
extrinsic. Intrinsic information may be required by internal methods in
our objects which they absolutely cannot function without. Extrinsic
information can however be removed and stored externally.
Objects with the same intrinsic data can be replaced with a single
shared object, created by a factory method. This allows us to reduce the
overall quantity of implicit data being stored quite significantly.
The benefit of this is that we're able to keep an eye on objects that
have already been instantiated so that new copies are only ever created
should the intrinsic state differ from the object we already have.
We use a manager to handle the extrinsic states. How this is
implemented can vary, but one approach to this to have the manager
object contain a central database of the extrinsic states and the
flyweight objects which they belong to.
Implementing Classical Flyweights
As the Flyweight pattern hasn't been heavily used in JavaScript in
recent years, many of the implementations we might use for inspiration
come form the Java and C++ worlds.
Our first look at Flyweights in code is my JavaScript implementation
of the Java sample of the Flyweight pattern from Wikipedia (http://en.wikipedia.org/wiki/Flyweight_pattern).
We will be making use of three types of Flyweight components in this implementation, which are listed below:
- Flyweight corresponds to an interface through which flyweights are able to receive and act on extrinsic states
- Concrete Flyweight actually implements the
Flyweight interface and stores intrinsic state. Concrete Flyweights need
to be sharable and capable of manipulating state that is extrinsic
- Flyweight Factory manages flyweight objects and
creates them too. It makes sure that our flyweights are shared and
manages them as a group of objects which can be queried if we require
individual instances. If an object has been already created in the group
it returns it, otherwise it adds a new object to the pool and returns
it.
These correspond to the following definitions in our implementation:
- CoffeeOrder: Flyweight
- CoffeeFlavor: Concrete Flyweight
- CoffeeOrderContext: Helper
- CoffeeFlavorFactory: Flyweight Factory
- testFlyweight: Utilization of our Flyweights
Duck punching "implements"
Duck punching allows us to extend the capabilities of a language or
solution without necessarily needing to modify the runtime source. As
this next solution requires the use of a Java keyword (implements
) for implementing interfaces and isn't found in JavaScript natively, let's first duck punch it.
Function.prototype.implementsFor
works on an object
constructor and will accept a parent class (function) or object and
either inherit from this using normal inheritance (for functions) or
virtual inheritance (for objects).
02 | Function.prototype.implementsFor = function ( parentClassOrObject ){ |
03 | if ( parentClassOrObject.constructor === Function ) |
06 | this .prototype = new parentClassOrObject(); |
07 | this .prototype.constructor = this ; |
08 | this .prototype.parent = parentClassOrObject.prototype; |
13 | this .prototype = parentClassOrObject; |
14 | this .prototype.constructor = this ; |
15 | this .prototype.parent = parentClassOrObject; |
We can use this to patch the lack of an implements
keyword by having a function inherit an interface explicitly. Below, CoffeeFlavor
implements the CoffeeOrder
interface and must contain its interface methods in order for us to
assign the functionality powering these implementations to an object.
005 | serveCoffee: function (context){}, |
006 | getFlavor: function (){} |
013 | function CoffeeFlavor( newFlavor ){ |
015 | var flavor = newFlavor; |
019 | if ( typeof this .getFlavor === "function" ){ |
020 | this .getFlavor = function () { |
025 | if ( typeof this .serveCoffee === "function" ){ |
026 | this .serveCoffee = function ( context ) { |
027 | console.log( "Serving Coffee flavor " |
029 | + " to table number " |
030 | + context.getTable()); |
038 | CoffeeFlavor.implementsFor( CoffeeOrder ); |
042 | function CoffeeOrderContext( tableNumber ) { |
044 | getTable: function () { |
051 | function CoffeeFlavorFactory() { |
056 | getCoffeeFlavor: function (flavorName) { |
058 | var flavor = flavors[flavorName]; |
059 | if (flavor === undefined) { |
060 | flavor = new CoffeeFlavor(flavorName); |
061 | flavors[flavorName] = flavor; |
067 | getTotalCoffeeFlavorsMade: function () { |
076 | function testFlyweight(){ |
080 | var flavors = new CoffeeFlavor(), |
083 | tables = new CoffeeOrderContext(), |
091 | function takeOrders( flavorIn, table) { |
092 | flavors[ordersMade] = flavorFactory.getCoffeeFlavor( flavorIn ); |
093 | tables[ordersMade++] = new CoffeeOrderContext( table ); |
096 | flavorFactory = new CoffeeFlavorFactory(); |
098 | takeOrders( "Cappuccino" , 2); |
099 | takeOrders( "Cappuccino" , 2); |
100 | takeOrders( "Frappe" , 1); |
101 | takeOrders( "Frappe" , 1); |
102 | takeOrders( "Xpresso" , 1); |
103 | takeOrders( "Frappe" , 897); |
104 | takeOrders( "Cappuccino" , 97); |
105 | takeOrders( "Cappuccino" , 97); |
106 | takeOrders( "Frappe" , 3); |
107 | takeOrders( "Xpresso" , 3); |
108 | takeOrders( "Cappuccino" , 3); |
109 | takeOrders( "Xpresso" , 96); |
110 | takeOrders( "Frappe" , 552); |
111 | takeOrders( "Cappuccino" , 121); |
112 | takeOrders( "Xpresso" , 121); |
114 | for ( var i = 0; i < ordersMade; ++i) { |
115 | flavors[i].serveCoffee(tables[i]); |
118 | console.log( "total CoffeeFlavor objects made: " + flavorFactory.getTotalCoffeeFlavorsMade()); |
Converting code to use the Flyweight pattern
Next, let's continue our look at Flyweights by implementing a system to
manage all of the books in a library. The important meta-data for each
book could probably be broken down as follows:
- ID
- Title
- Author
- Genre
- Page count
- Publisher ID
- ISBN
We'll also require the following properties to keep track of which
member has checked out a particular book, the date they've checked it
out on as well as the expected date of return.
- checkoutDate
- checkoutMember
- dueReturnDate
- availability
Each book would thus be represented as follows, prior to any optimization using the Flyweight pattern:
01 | var Book = function ( id, title, author, genre, pageCount,publisherID, ISBN, checkoutDate, checkoutMember, dueReturnDate,availability ){ |
07 | this .pageCount = pageCount; |
08 | this .publisherID = publisherID; |
10 | this .checkoutDate = checkoutDate; |
11 | this .checkoutMember = checkoutMember; |
12 | this .dueReturnDate = dueReturnDate; |
13 | this .availability = availability; |
19 | getTitle: function () { |
23 | getAuthor: function () { |
32 | updateCheckoutStatus: function ( bookID, newStatus, checkoutDate , checkoutMember, newReturnDate ){ |
35 | this .availability = newStatus; |
36 | this .checkoutDate = checkoutDate; |
37 | this .checkoutMember = checkoutMember; |
38 | this .dueReturnDate = newReturnDate; |
42 | extendCheckoutPeriod: function ( bookID, newReturnDate ){ |
45 | this .dueReturnDate = newReturnDate; |
49 | isPastDue: function (bookID){ |
51 | var currentDate = new Date(); |
52 | return currentDate.getTime() > Date.parse( this .dueReturnDate ); |
This probably works fine initially for small collections of books,
however as the library expands to include a larger inventory with
multiple versions and copies of each book available, we may find the
management system running slower and slower over time. Using thousands
of book objects may overwhelm the available memory, but we can optimize
our system using the Flyweight pattern to improve this.
We can now separate our data into intrinsic and extrinsic states as follows: data relevant to the book object (title
, author
etc) is intrinsic whilst the checkout data (checkoutMember
, dueReturnDate
etc) is considered extrinsic. Effectively this means that only one Book
object is required for each combination of book properties. it's still a
considerable quantity of objects, but significantly fewer than we had
previously.
The following single instance of our book meta-data combinations will be
shared among all of the copies of a book with a particular title.
02 | var Book = function ( title, author, genre, pageCount, publisherID, ISBN ) { |
07 | this .pageCount = pageCount; |
08 | this .publisherID = publisherID; |
As we can see, the extrinsic states have been removed. Everything to do
with library check-outs will be moved to a manager and as the object
data is now segmented, a factory can be used for instantiation.
A Basic Factory
Let's now define a very basic factory. What we're going to have it do is
perform a check to see if a book with a particular title has been
previously created inside the system; if it has, we'll return it - if
not, a new book will be created and stored so that it can be accessed
later. This makes sure that we only create a single copy of each unique
intrinsic piece of data:
02 | var BookFactory = ( function () { |
03 | var existingBooks = {}, existingBook; |
06 | createBook: function ( title, author, genre, pageCount, publisherID, ISBN ) { |
10 | existingBook = existingBooks[ISBN]; |
11 | if ( !!existingBook ) { |
16 | var book = new Book( title, author, genre, pageCount, publisherID, ISBN ); |
17 | existingBooks[ISBN] = book; |
Managing the extrinsic states
Next, we need to store the states that were removed from the Book
objects somewhere - luckily a manager (which we'll be defining as a
Singleton) can be used to encapsulate them. Combinations of a Book
object and the library member that's checked them out will be called
Book records. Our manager will be storing both and will also include
checkout related logic we stripped out during our flyweight optimization
of the Book class.
02 | var BookRecordManager = ( function () { |
04 | var bookRecordDatabase = {}; |
08 | addBookRecord: function ( id, title, author, genre, pageCount, publisherID, ISBN, checkoutDate, checkoutMember, dueReturnDate, availability ) { |
10 | var book = bookFactory.createBook( title, author, genre, pageCount, publisherID, ISBN ); |
12 | bookRecordDatabase[id] = { |
13 | checkoutMember: checkoutMember, |
14 | checkoutDate: checkoutDate, |
15 | dueReturnDate: dueReturnDate, |
16 | availability: availability, |
20 | updateCheckoutStatus: function ( bookID, newStatus, checkoutDate, checkoutMember, newReturnDate ) { |
22 | var record = bookRecordDatabase[bookID]; |
23 | record.availability = newStatus; |
24 | record.checkoutDate = checkoutDate; |
25 | record.checkoutMember = checkoutMember; |
26 | record.dueReturnDate = newReturnDate; |
29 | extendCheckoutPeriod: function ( bookID, newReturnDate ) { |
30 | bookRecordDatabase[bookID].dueReturnDate = newReturnDate; |
33 | isPastDue: function ( bookID ) { |
34 | var currentDate = new Date(); |
35 | return currentDate.getTime() > Date.parse( bookRecordDatabase[bookID].dueReturnDate ); |
The result of these changes is that all of the data that's been extracted from the Book class
is now being stored in an attribute of the BookManager singleton
(BookDatabase) - something considerably more efficient than the large
number of objects we were previously using. Methods related to book
checkouts are also now based here as they deal with data that's
extrinsic rather than intrinsic.
This process does add a little complexity to our final solution, however
it's a small concern when compared to the performance issues that have
been tackled.Data wise, if we have 30 copies of the same book, we are
now only storing it once. Also, every function takes up memory. With the
flyweight pattern these functions exist in one place (on the manager)
and not on every object, thus saving on memory use.
The Flyweight pattern and the DOM
The DOM (Document Object Model) supports two approaches that allow
objects to detect events - either top down (event capture) or bottom up
(event bubbling).
In event capture, the event is first captured by the outer-most
element and propagated to the inner-most element. In event bubbling, the
event is captured and given to the inner-most element and then
propagated to the outer-elements.
One of the best metaphors for describing Flyweights in this context was written by Gary Chisholm and it goes a little like this:
Try to think of the flyweight in terms of a pond. A
fish opens its mouth (the event), bubbles rise to the surface (the
bubbling) a fly sitting on the top flies away when the bubble reaches
the surface (the action). In this example we can easily transpose the
fish opening its mouth to a button being clicked, the bubbles as the
bubbling effect and the fly flying away to some function being run
Bubbling was introduced to handle situations where a single event
(e.g a click) may be handled by multiple event handlers defined at
different levels of the DOM hierarchy. Where this happens, event
bubbling executes event handlers defined for specific elements at the
lowest level possible. From there on, the event bubbles up to containing
elements before going to those even higher up.
Flyweights can be used to tweak the event bubbling process further, as we will see shortly.
Example 1: Centralized event handling
For our first practical example, imagine we have a number of similar
elements in a document with similar behavior executed when a user-action
(e.g click, mouse-over) is performed against them.
Normally what we do when constructing our own accordion component, menu
or other list-based widget is bind a click event to each link element in
the parent container (e.g $('ul li a').on(..)
. Instead of
binding the click to multiple elements, we can easily attach a Flyweight
to the top of our container which can listen for events coming from
below. These can then be handled using logic that is as simple or
complex as required.
As the types of components mentioned often have the same repeating
markup for each section (e.g. each section of an accordion), there's a
good chance the behavior of each element that may be clicked is going to
be quite similar and relative to similar classes nearby. We'll use this
information to construct a very basic accordion using the Flyweight
below.
A stateManager namespace is used here to encapsulate our flyweight logic
whilst jQuery is used to bind the initial click to a container div. In
order to ensure that no other logic on the page is attaching similar
handles to the container, an unbind event is first applied.
Now to establish exactly what child element in the container is clicked, we make use of a target
check which provides a reference to the element that was clicked,
regardless of its parent. We then use this information to handle the
click event without actually needing to bind the event to specific
children when our page loads.
HTML
02 | <div class= "toggle" href= "#" >More Info (Address) |
04 | This is more information |
06 | <div class= "toggle" href= "#" >Even More Info (Map) |
JavaScript
07 | $( "#container" ).unbind().on( "click" , function ( e ) { |
08 | var target = $( e.originalTarget || e.srcElement ); |
09 | if ( target.is( "div.toggle" ) ) { |
10 | self.handleClick( target ); |
15 | handleClick: function ( elem ) { |
16 | elem.find( "span" ).toggle( "slow" ); |
The benefit here is that we're converting many independent actions into a shared ones (potentially saving on memory).
Example 2: Using the Flyweight for performance optimization
In our second example, we'll reference some further performance gains that can be achieved using Flyweights with jQuery.
James Padolsey previously wrote an article called 76 bytes for faster jQuery
where he reminded us that each time jQuery fires off a callback,
regardless of type (filter, each, event handler), we're able to access
the function's context (the DOM element related to it) via the this
keyword.
Unfortunately, many of us have become used to the idea of wrapping this
in $()
or jQuery()
, which means that a new instance of jQuery is unnecessarily constructed every time, rather than simply doing this:
01 | $( "div" ).on( "click" , function () { |
02 | console.log( "You clicked: " + $( this ).attr( "id" )); |
09 | $( "div" ).on( "click" , function () { |
10 | console.log( "You clicked:" + this .id ); |
James had wanted to use jQuery's jQuery.text
in the following context, however he disagreed with the notion that a new jQuery object had to be created on each iteration:
$( "a" ).map( function () {
return $( this ).text();
});
Now with respect to redundant wrapping, where possible with jQuery's utility methods, it's better to use jQuery.methodName
(e.g jQuery.text
) as opposed to jQuery.fn.methodName
(e.g jQuery.fn.text
) where methodName represents a utility such as each()
or text
.
This avoids the need to call a further level of abstraction or
construct a new jQuery object each time our function is called as as jQuery.methodName
is what the library itself uses at a lower-level to power jQuery.fn.methodName
.
Because however not all of jQuery's methods have corresponding
single-node functions, Padolsey devised the idea of a jQuery.single
utility.
The idea here is that a single jQuery object is created and used for
each call to jQuery.single (effectively meaning only one jQuery object
is ever created). The implementation for this can be found below and as
we're consolidating data for multiple possible objects into a more
central singular structure, it is technically also a Flyweight.
01 | jQuery.single = ( function ( o ){ |
03 | var collection = jQuery([1]); |
04 | return function ( element ) { |
07 | collection[0] = element; |
An example of this in action with chaining is:
1 | $( "div" ).on( "click" , function () { |
3 | var html = jQuery.single( this ).next().html(); |
Note: Although we may believe that simply caching our jQuery code may
offer just as equivalent performance gains, Padolsey claims that
$.single() is still worth using and can perform better. That's not to
say don't apply any caching at all, just be mindful that this approach
can assist. For further details about $.single, I recommend reading
Padolsey's full post:
#
JavaScript MV* Patterns
In this section, we're going to review three very important
architectural patterns - MVC (Model-View-Controller), MVP
(Model-View-Presenter) and MVVM (Model-View-ViewModel). In the past,
these patterns have been heavily used for structuring desktop and
server-side applications but it's only been in recent years that come to
being applied to JavaScript.
As the majority of JavaScript developers currently using these
patterns opt to utilize libraries such as Backbone.js for implementing
an MVC/MV*-like structure, we will compare how modern solutions such as
it differ in their interpretation of MVC compared to classical takes on
these patterns.
Let us first now cover the basics.
MVC
MVC is an architectural design pattern that encourages improved
application organization through a separation of concerns. It enforces
the isolation of business data (Models) from user interfaces (Views),
with a third component (Controllers) traditionally managing logic and
user-input. The pattern was originally designed by Trygve Reenskaug
during his time working on Smalltalk-80 (1979) where it was initially
called Model-View-Controller-Editor. MVC went on to be described in
depth in 1995's “Design Patterns: Elements of Reusable Object-Oriented Software” (The "GoF" book), which played a role in popularizing its use.
Smalltalk-80 MVC
It's important to understand what the original MVC pattern was aiming
to solve as it's mutated quite heavily since the days of its origin.
Back in the 70's, graphical user-interfaces were few and far between and
a concept known as Separated Presentation
began to be used as a means to make a clear division between domain
objects which modeled concepts in the real world (e.g a photo, a person)
and the presentation objects which were rendered to the user's screen.
The Smalltalk-80 implementation of MVC took this concept further and
had an objective of separating out the application logic from the user
interface. The idea was that decoupling these parts of the application
would also allow the reuse of models for other interfaces in the
application. There are some interesting points worth noting about
Smalltalk-80's MVC architecture:
- A Model represented domain-specific data and was ignorant of the
user-interface (Views and Controllers). When a model changed, it would
inform its observers.
- A View represented the current state of a Model. The Observer
pattern was used for letting the View know whenever the Model was
updated or modified.
- Presentation was taken care of by the View, but there wasn't just a
single View and Controller - a View-Controller pair was required for
each section or element being displayed on the screen.
- The Controllers role in this pair was handling user interaction
(such as key-presses and actions e.g clicks), making decisions for the
View.
Developers are sometimes surprised when they learn that the Observer
pattern (nowadays commonly implemented as the Publish/Subscribe
variation) was included as a part of MVC's architecture many decades
ago. In Smalltalk-80's MVC, the View observes the Model. As mentioned in
the bullet point above, anytime the Model changes, the Views react. A
simple example of this is an application backed by stock market data -
in order for the application to be useful, any change to the data in our
Models should result in the View being refreshed instantly.
Martin Fowler has done an excellent job of writing about the origins
of MVC over the years and if interested in some further historical
information about Smalltalk-80's MVC, I recommend reading his work.
MVC For JavaScript Developers
We've reviewed the 70's, but let us now return to the here and now.
In modern times, the MVC pattern has been applied to a diverse range of
programming languages including of most relevance to us: JavaScript.
JavaScript now has a number of frameworks boasting support for MVC (or
variations on it, which we refer to as the MV* family), allowing
developers to easily add structure to their applications without great
effort.
These frameworks include the likes of Backbone, Ember.js and
AngularJS. Given the importance of avoiding "spaghetti" code, a term
which describes code that is very difficult to read or maintain due to
its lack of structure, it's imperative that the modern JavaScript
developer understand what this pattern provides. This allows us to
effectively appreciate what these frameworks enable us to do
differently.
We know that MVC is composed of three core components:
Models
Models manage the data for an application. They are concerned with
neither the user-interface nor presentation layers but instead represent
unique forms of data that an application may require. When a model
changes (e.g when it is updated), it will typically notify its observers
(e.g views, a concept we will cover shortly) that a change has occurred
so that they may react accordingly.
To understand models further, let us imagine we have a JavaScript
photo gallery application. In a photo gallery, the concept of a photo
would merit its own model as it represents a unique kind of
domain-specific data. Such a model may contain related attributes such
as a caption, image source and additional meta-data. A specific photo
would be stored in an instance of a model and a model may also be
reusable. Below we can see an example of a very simplistic model
implemented using Backbone.
01 | var Photo = Backbone.Model.extend({ |
05 | src: "placeholder.jpg" , |
06 | caption: "A default image" , |
11 | initialize: function () { |
12 | this .set( { "src" : this .defaults.src} ); |
The built-in capabilities of models vary across frameworks, however
it is quite common for them to support validation of attributes, where
attributes represent the properties of the model, such as a model
identifier. When using models in real-world applications we generally
also desire model persistence. Persistence allows us to edit and update
models with the knowledge that its most recent state will be saved in
either: memory, in a user's localStorage data-store or synchronized with
a database.
In addition, a model may also have multiple views observing it. If
say, our photo model contained meta-data such as its location (longitude
and latitude), friends that were present in the photo (a list of
identifiers) and a list of tags, a developer may decide to provide a
single view to display each of these three facets.
It is not uncommon for modern MVC/MV* frameworks to provide a means
to group models together (e.g. in Backbone, these groups are referred to
as "collections"). Managing models in groups allows us to write
application logic based on notifications from the group should any model
it contains be changed. This avoids the need to manually observe
individual model instances.
A sample grouping of models into a simplified Backbone collection can be seen below.
01 | var PhotoGallery = Backbone.Collection.extend({ |
09 | return this .filter( function ( photo ){ |
10 | return photo.get( "viewed" ); |
16 | unviewed: function () { |
17 | return this .without.apply( this , this .viewed() ); |
Older texts on MVC may also contain reference to a notion of models managing application state.In JavaScript applications state
has a different connotation, typically referring to the current "state"
i.e view or sub-view (with specific data) on a users screen at a fixed
point. State is a topic which is regularly discussed when looking at
Single-page applications, where the concept of state needs to be
simulated.
So to summarize, models are primarily concerned with business data.
Views
Views are a visual representation of models that present a filtered
view of their current state. Whilst Smalltalk views are about painting
and maintaining a bitmap, JavaScript views are about building and
maintaining a DOM element.
A view typically observes a model and is notified when the model
changes, allowing the view to update itself accordingly. Design pattern
literature commonly refers to views as "dumb" given that their knowledge
of models and controllers in an application is limited.
Users are able to interact with views and this includes the ability
to read and edit (i.e get or set the attribute values in) models. As the
view is the presentation layer, we generally present the ability to
edit and update in a user-friendly fashion. For example, in the former
photo gallery application we discussed earlier, model editing could be
facilitated through an "edit' view where a user who has selected a
specific photo could edit its meta-data.
The actual task of updating the model falls to controllers (which we will be covering shortly).
Let's explore views a little further using a vanilla JavaScript
sample implementation. Below we can see a function that creates a single
Photo view, consuming both a model instance and a controller instance.
We define a render()
utility within our view which is responsible for rendering the contents of the photoModel
using a JavaScript templating engine (Underscore templating) and updating the contents of our view, referenced by photoEl
.
The photoModel
then adds our render()
callback as one of its subscribers so that through the Observer pattern
we can trigger the view to update when the model changes.
One may wonder where user-interaction comes into play here. When
users click on any elements within the view, it's not the view's
responsibility to know what to do next. It relies on a controller to
make this decision for it. In our sample implementation, this is
achieved by adding an event listener to photoEl
which will
delegate handling the click behavior back to the controller, passing the
model information along with it in case it's needed.
The benefit of this architecture is that each component plays its own
separate role in making the application function as needed.
01 | var buildPhotoView = function ( photoModel, photoController ) { |
03 | var base = document.createElement( "div" ), |
04 | photoEl = document.createElement( "div" ); |
06 | base.appendChild(photoEl); |
08 | var render = function () { |
12 | photoEl.innerHTML = _.template( "#photoTemplate" , { |
13 | src: photoModel.getSrc() |
17 | photoModel.addSubscriber( render ); |
19 | photoEl.addEventListener( "click" , function () { |
20 | photoController.handleEvent( "click" , photoModel ); |
23 | var show = function () { |
24 | photoEl.style.display = "" ; |
27 | var hide = function () { |
28 | photoEl.style.display = "none" ; |
Templating
In the context of JavaScript frameworks that support MVC/MV*, it is
worth briefly discussing JavaScript templating and its relationship to
views as we briefly touched upon it in the last section.
It has long been considered (and proven) a performance bad practice
to manually create large blocks of HTML markup in-memory through string
concatenation. Developers doing so have fallen prey to inperformantly
iterating through their data, wrapping it in nested divs and using
outdated techniques such as document.write
to inject the
"template" into the DOM. As this typically means keeping scripted markup
inline with our standard markup, it can quickly become both difficult
to read and more importantly, maintain such disasters, especially when
building non-trivially sized applications.
JavaScript templating solutions (such as Handlebars.js and Mustache)
are often used to define templates for views as markup (either stored
externally or within script tags with a custom type - e.g text/template)
containing template variables. Variables may be delimitated using a
variable syntax (e.g {{name}}) and frameworks are typically smart enough
to accept data in a JSON form (of which model instances can be
converted to) such that we only need be concerned with maintaining clean
models and clean templates. Most of the grunt work to do with
population is taken care of by the framework itself. This has a large
number of benefits, particularly when opting to store templates
externally as this can give way to templates being dynamically loaded on
an as-needed basis when it comes to building larger applications.
Below we can see two examples of HTML templates. One implemented
using the popular Handlebars.js framework and another using Underscore's
templates.
Handlebars.js:
3 | <img class= "source" src= "{{src}}" /> |
4 | <div class= "meta-data" > |
Underscore.js Microtemplates:
2 | <h2><%= caption %></h2> |
3 | <img class= "source" src= "<%= src %>" /> |
4 | <div class= "meta-data" > |
Note that templates are not themselves views. Developers coming from a
Struts Model 2 architecture may feel like a template *is a view, but it
isn't. A view is an object which observes a model and keeps the visual
representation up-to-date. A template *might* be a declarative way to
specify part or even all of a view object so that it can be generated
from the template specification.
It is also worth noting that in classical web development, navigating
between independent views required the use of a page refresh. In
Single-page JavaScript applications however, once data is fetched from a
server via Ajax, it can simply be dynamically rendered in a new view
within the same page without any such refresh being necessary.
The role of navigation thus falls to a "router", which assists in
managing application state (e.g allowing users to bookmark a particular
view they have navigated to). As routers are, however, neither a part of
MVC nor present in every MVC-like framework, I will not be going into
them in greater detail in this section.
To summarize, views are a visual representation of our application data.
Controllers
Controllers are an intermediary between models and views which are
classically responsible for updating the model when the user manipulates
the view.
In our photo gallery application, a controller would be responsible
for handling changes the user made to the edit view for a particular
photo, updating a specific photo model when a user has finished editing.
Remember that the controllers fulfill one role in MVC: the
facilitation of the Strategy pattern for the view. In the Strategy
pattern regard, the view delegates to the controller at the view's
discretion. So, that's how the strategy pattern works. The view could
delegate handling user events to the controller when the view sees fit.
The view *could* delegate handling model change events to the controller
if the view sees fit, but this is not the traditional role of the
controller.
In terms of where most JavaScript MVC frameworks detract from what is
conventionally considered "MVC" however, it is with controllers. The
reasons for this vary, but in my honest opinion, it is that framework
authors initially look at the server-side interpretation of MVC, realize
that it doesn't translate 1:1 on the client-side and re-interpret the C
in MVC to mean something they feel makes more sense. The issue with
this however is that it is subjective, increases the complexity in both
understanding the classical MVC pattern and of course the role of
controllers in modern frameworks.
As an example, let's briefly review the architecture of the popular
architectural framework Backbone.js. Backbone contains models and views
(somewhat similar to what we reviewed earlier), however it doesn't
actually have true controllers. Its views and routers act a little
similar to a controller, but neither are actually controllers on their
own.
In this respect, contrary to what might be mentioned in the official
documentation or in blog posts, Backbone is neither a truly MVC/MVP nor
MVVM framework. It's in fact better to consider it a member of the MV*
family which approaches architecture in its own way. There is of course
nothing wrong with this, but it is important to distinguish between
classical MVC and MV* should we begin relying on advice from classical
literature on the former to help with the latter.
Controllers in another library (Spine.js) vs Backbone.js
Spine.js
We now know that controllers are traditionally responsible for
updating the model when the user updates the view. It's interesting to
note that the most popular JavaScript MVC/MV* framework at the time of
writing (Backbone) does not have it's own explicit concept of controllers.
It can thus be useful for us to review the controller from another
MVC framework to appreciate the difference in implementations and
further demonstrate how nontraditionally frameworks approach the role of
the controller. For this, let's take a look at a sample controller from
Spine.js:
In this example, we're going to have a controller called PhotosController
which will be in charge of individual photos in the application. It
will ensure that when the view updates (e.g a user edited the photo
meta-data) the corresponding model does too.
Note: We won't be delving heavily into Spine.js at all, but will just take a ten-foot view of what its controllers can do:
03 | var PhotosController = Spine.Controller.sub({ |
06 | this .item.bind( "update" , this .proxy( this .render )); |
07 | this .item.bind( "destroy" , this .proxy( this .remove )); |
12 | this .replace( $( "#photoTemplate" ).tmpl( this .item ) ); |
In Spine, controllers are considered the glue for an application,
adding and responding to DOM events, rendering templates and ensuring
that views and models are kept in sync (which makes sense in the context
of what we know to be a controller).
What we're doing in the above example is setting up listeners in the update
and destroy
events using render()
and remove()
.
When a photo entry gets updated, we re-render the view to reflect the
changes to the meta-data. Similarly, if the photo gets deleted from the
gallery, we remove it from the view. In the render()
function, we're using Underscore micro-templating (via _.template()
)
to render a JavaScript template with the ID #photoTemplate. This simply
returns a compiled HTML string used to populate the contents of photoEl
.
What this provides us with is a very lightweight, simple way to manage changes between the model and the view.
Backbone.js
Later on in this section we're going to revisit the differences
between Backbone and traditional MVC, but for now let's focus on
controllers.
In Backbone, one shares the responsibility of a controller with both the Backbone.View
and Backbone.Router
. Some time ago Backbone did once come with its own Backbone.Controller
,
but as the naming for this component didn't make sense for the context
in which it was being used, it was later renamed to Router.
Routers handle a little more of the controller responsibility as it's
possible to bind the events there for models and have our view respond
to DOM events and rendering. As Tim Branyen (another Bocoup-based
Backbone contributor) has also previously pointed out, it's possible to
get away with not needing Backbone.Router
at all for this, so a way to think about it using the Router paradigm is probably:
01 | var PhotoRouter = Backbone.Router.extend({ |
02 | routes: { "photos/:id" : "route" }, |
04 | route: function ( id ) { |
05 | var item = photoCollection.get( id ); |
06 | var view = new PhotoView( { model: item } ); |
08 | $( '.content' ).html( view.render().el ); |
To summarize, the takeaway from this section is that controllers
manage the logic and coordination between models and views in an
application.
What does MVC give us?
This separation of concerns in MVC facilitates simpler modularization of an application's functionality and enables:
- Easier overall maintenance. When updates need to be made to the
application it is very clear whether the changes are data-centric,
meaning changes to models and possibly controllers, or merely visual,
meaning changes to views.
- Decoupling models and views means that it is significantly more straight-forward to write unit tests for business logic
- Duplication of low-level model and controller code (i.e what we
may have been using instead) is eliminated across the application
- Depending on the size of the application and separation of roles,
this modularity allows developers responsible for core logic and
developers working on the user-interfaces to work simultaneously
Smalltalk-80 MVC In JavaScript
Although the majority of modern-day JavaScript frameworks attempt to
evolve the MVC paradigm to better fit the differing needs of web
application development, there is one framework which attempts to adhere
to the pure form of the pattern found in Smalltalk-80. Maria.js (https://github.com/petermichaux/maria)
by Peter Michaux offers an implementation which is faithful to MVCs
origins - Models are models, Views are views and Controllers are nothing
but controllers. Whilst some developers might feel an MV* framework
should address more concerns, this is a useful reference to be aware of
in case you would like a JavaScript implementation of the original MVC.
Delving deeper
At this point in the book, we should have a basic understanding of
what the MVC pattern provides, but there's still some fascinating
information about it worth noting.
The GoF do not refer to MVC as a design pattern, but rather consider it a set of classes to build a user interface.
In their view, it's actually a variation of three classical design
patterns: the Observer, Strategy and Composite patterns. Depending on
how MVC has been implemented in a framework, it may also use the Factory
and Template patterns. The GoF book mentions these patterns as useful
extras when working with MVC.
As we have discussed, models represent application data whilst views
are what the user is presented on screen. As such, MVC relies on the
Observer pattern for some of its core communication (something that
surprisingly isn't covered in many articles about the MVC pattern). When
a model is changed it notifies its observers (Views) that something has
been updated - this is perhaps the most important relationship in MVC.
The observer nature of this relationship is also what facilitates
multiple views being attached to the same model.
For developers interested in knowing more about the decoupled nature
of MVC (once again, depending on the implementation), one of the goals
of the pattern is to help define one-to-many relationships between a
topic (data object) and its observers. When a topic changes, its
observers are updated. Views and controllers have a slightly different
relationship. Controllers facilitate views to respond to different user
input and are an example of the Strategy pattern.
Summary
Having reviewed the classical MVC pattern, we should now understand
how it allows us to cleanly separate concerns in an application. We
should also now appreciate how JavaScript MVC frameworks may differ in
their interpretation of the MVC pattern, which although quite open to
variation, still shares some of the fundamental concepts the original
pattern has to offer.
When reviewing a new JavaScript MVC/MV* framework, remember - it can
be useful to step back and review how it's opted to approach
architecture (specifically, how it supports implementing models, views,
controllers or other alternatives) as this can better help us grok how
the framework expects to be used.
MVP
Model-view-presenter (MVP) is a derivative of the MVC design pattern
which focuses on improving presentation logic. It originated at a
company named Taligent
in the early 1990s while they were working on a model for a C++
CommonPoint environment. Whilst both MVC and MVP target the separation
of concerns across multiple components, there are some fundamental
differences between them.
For the purposes of this summary we will focus on the version of MVP most suitable for web-based architectures.
Models, Views & Presenters
The P in MVP stands for presenter. It's a component which contains
the user-interface business logic for the view. Unlike MVC, invocations
from the view are delegated to the presenter, which are decoupled from
the view and instead talk to it through an interface. This allows for
all kinds of useful things such as being able to mock views in unit
tests.
The most common implementation of MVP is one which uses a Passive
View (a view which is for all intents and purposes "dumb"), containing
little to no logic. If MVC and MVP are different it is because the C
and P do different things. In MVP, the P observes models and updates
views when models change. The P effectively binds models to views, a
responsibility which was previously held by controllers in MVC.
Solicited by a view, presenters perform any work to do with user
requests and pass data back to them. In this respect, they retrieve
data, manipulate it and determine how the data should be displayed in
the view. In some implementations, the presenter also interacts with a
service layer to persist data (models). Models may trigger events but
it's the presenters role to subscribe to them so that it can update the
view. In this passive architecture, we have no concept of direct data
binding. Views expose setters which presenters can use to set data.
The benefit of this change from MVC is that it increases the
testability of our application and provides a more clean separation
between the view and the model. This isn't however without its costs as
the lack of data binding support in the pattern can often mean having to
take care of this task separately.
Although a common implementation of a Passive View
is for the view to implement an interface, there are variations on it,
including the use of events which can decouple the View from the
Presenter a little more. As we don't have the interface construct in
JavaScript, we're using more a protocol than an explicit interface here.
It's technically still an API and it's probably fair for us to refer to
it as an interface from that perspective.
There is also a Supervising Controller variation of MVP, which is closer to the MVC and MVVM
patterns as it provides data-binding from the Model directly from the
View. Key-value observing (KVO) plugins (such as Derick Bailey's
Backbone.ModelBinding plugin) tend to bring Backbone out of the Passive
View and more into the Supervising Controller or MVVM variations.
MVP or MVC?
MVP is generally used most often in enterprise-level applications
where it's necessary to reuse as much presentation logic as possible.
Applications with very complex views and a great deal of user
interaction may find that MVC doesn't quite fit the bill here as solving
this problem may mean heavily relying on multiple controllers. In MVP,
all of this complex logic can be encapsulated in a presenter, which can
simplify maintenance greatly.
As MVP views are defined through an interface and the interface is
technically the only point of contact between the system and the view
(other than a presenter), this pattern also allows developers to write
presentation logic without needing to wait for designers to produce
layouts and graphics for the application.
Depending on the implementation, MVP may be easier to automatically
unit test than MVC. The reason often cited for this is that the
presenter can be used as a complete mock of the user-interface and so it
can be unit tested independent of other components. In my experience
this really depends on the languages we are implementing MVP in (there's
quite a difference between opting for MVP for a JavaScript project over
one for say, ASP.net).
At the end of the day, the underlying concerns we may have with MVC
will likely hold true for MVP given that the differences between them
are mainly semantic. As long as we are cleanly separating concerns into
models, views and controllers (or presenters) we should be achieving
most of the same benefits regardless of the variation we opt for.
MVC, MVP and Backbone.js
There are very few, if any architectural JavaScript frameworks that
claim to implement the MVC or MVP patterns in their classical form as
many JavaScript developers don't view MVC and MVP as being mutually
exclusive (we are actually more likely to see MVP strictly implemented
when looking at web frameworks such as ASP.net or GWT). This is because
it's possible to have additional presenter/view logic in our application
and yet still consider it a flavor of MVC.
Backbone contributor Irene Ros (of
Boston-based Bocoup) subscribes to this way of thinking as when she
separates views out into their own distinct components, she needs
something to actually assemble them for her. This could either be a
controller route (such as a Backbone.Router
, covered later in the book) or a callback in response to data being fetched.
That said, some developers do however feel that Backbone.js better
fits the description of MVP than it does MVC. Their view is that:
- The presenter in MVP better describes the
Backbone.View
(the layer between View templates and the data bound to it) than a controller does
- The model fits
Backbone.Model
(it isn't greatly different to the models in MVC at all)
- The views best represent templates (e.g Handlebars/Mustache markup templates)
A response to this could be that the view can also just be a View (as
per MVC) because Backbone is flexible enough to let it be used for
multiple purposes. The V in MVC and the P in MVP can both be
accomplished by Backbone.View
because they're able to
achieve two purposes: both rendering atomic components and assembling
those components rendered by other views.
We've also seen that in Backbone the responsibility of a controller
is shared with both the Backbone.View and Backbone.Router and in the
following example we can actually see that aspects of that are certainly
true.
Our Backbone PhotoView
uses the Observer pattern to "subscribe" to changes to a View's model in the line this.model.bind("change",...)
. It also handles templating in the render()
method, but unlike some other implementations, user interaction is also handled in the View (see events
).
01 | var PhotoView = Backbone.View.extend({ |
08 | template: _.template( $( "#photo-template" ).html() ), |
12 | "click img" : "toggleViewed" |
21 | initialize: function () { |
22 | this .model.on( "change" , this .render, this ); |
23 | this .model.on( "destroy" , this .remove, this ); |
28 | $( this .el ).html( this .template( this .model.toJSON() )); |
33 | toggleViewed: function () { |
Another (quite different) opinion is that Backbone more closely resembles Smalltalk-80 MVC, which we went through earlier.
As regular Backbone blogger Derick Bailey has previously
put it, it's ultimately best not to force Backbone to fit any specific
design patterns. Design patterns should be considered flexible guides to
how applications may be structured and in this respect, Backbone fits
neither MVC nor MVP. Instead, it borrows some of the best concepts from
multiple architectural patterns and creates a flexible framework that
just works well.
It is however worth understanding where and why these
concepts originated, so I hope that my explanations of MVC and MVP have
been of help. Call it the Backbone way, MV* or whatever
helps reference its flavor of application architecture. Most structural
JavaScript frameworks will adopt their own take on classical patterns,
either intentionally or by accident, but the important thing is that
they help us develop applications which are organized, clean and can be
easily maintained.
#
MVVM
MVVM (Model View ViewModel) is an architectural pattern based on MVC
and MVP, which attempts to more clearly separate the development of
user-interfaces (UI) from that of the business logic and behavior in an
application. To this end, many implementations of this pattern make use
of declarative data bindings to allow a separation of work on Views from
other layers.
This facilitates UI and development work occurring almost
simultaneously within the same codebase. UI developers write bindings to
the ViewModel within their document markup (HTML), where the Model and
ViewModel are maintained by developers working on the logic for the
application.
History
MVVM (by name) was originally defined by Microsoft for use with Windows Presentation Foundation (WPF) and Silverlight, having been officially announced in 2005 by John Grossman
in a blog post about Avalon (the codename for WPF). It also found some
popularity in the Adobe Flex community as an alternative to simply using
MVC.
Prior to Microsoft adopting the MVVM name, there was however a movement in the community to go from MVP to MVPM: Model View PresentationModel. Martin Fowler wrote an article on PresentationModels back in 2004 for those interested in reading more about it. The idea of a PresentationModel
had been around much longer than this article, however it was
considered the big break in the idea and greatly helped popularize it.
There was quite a lot of uproar in the "alt.net" circles after
Microsoft announced MVVM as an alternative to MVPM. Many claimed the
company's dominance in the GUI world was giving them the opportunity to
take over the community as a whole, renaming existing concepts as they
pleased for marketing purposes. A progressive crowd recognized that
whilst MVVM and MVPM were effectively the same idea, they came in
slightly different packages.
In recent years, MVVM has been implemented in JavaScript in the form of structural frameworks such as KnockoutJS, Kendo MVVM and Knockback.js, with an overall positive response from the community.
Let’s now review the three components that compose MVVM.
Model
As with other members of the MV* family, the Model in MVVM represents
domain-specific data or information that our application will be
working with. A typical example of domain-specific data might be a user
account (e.g name, avatar, e-mail) or a music track (e.g title, year,
album).
Models hold information, but typically don’t handle behavior. They
don’t format information or influence how data appears in the browser as
this isn’t their responsibility. Instead, formatting of data is handled
by the View, whilst behavior is considered business logic that should
be encapsulated in another layer that interacts with the Model - the
ViewModel.
The only exception to this rule tends to be validation and it’s
considered acceptable for Models to validate data being used to define
or update existing models (e.g does an e-mail address being input meet
the requirements of a particular regular expression?).
In KnockoutJS, Models fall under the above definition, but often make
Ajax calls to a server-side service to both read and write Model data.
If we were constructing a simple Todo application, a KnockoutJS Model representing a single Todo item could look as follows:
1 | var Todo = function ( content, done ) { |
2 | this .content = ko.observable(content); |
3 | this .done = ko.observable(done); |
4 | this .editing = ko.observable( false ); |
Note: One may notice in the above snippet that we are calling the method observable()
on the KnockoutJS namespace ko
.
In KnockoutJS, observables are special JavaScript objects that can
notify subscribers about changes and automatically detect dependencies.
This allows us to synchronize Models and ViewModels when the value of a
Model attribute is modified.
View
As with MVC, the View is the only part of the application that users
actually interact with. They are an interactive UI that represent the
state of a ViewModel. In this sense, the view is considered active
rather than passive, but this is also true for views in MVC and MVP. In
MVC, MVP and MVVM a view can also be passive, but what does this mean?
A passive View only outputs a display and does not accept any user input.
Such a view may also have no real knowledge of the models in our
application and could be manipulated by a presenter. MVVM’s active View
contains the data-bindings, events and behaviors which requires an
understanding of the ViewModel. Although these behaviors can be mapped
to properties, the View is still responsible for handling events from
the ViewModel.
It’s important to remember the View isn’t responsible here for handling state - it keeps this in sync with the ViewModel.
A KnockoutJS View is simply a HTML document with declarative bindings
to link it to the ViewModel. KnockoutJS Views display information from
the ViewModel, pass commands to it (e.g a user clicking on an element)
and update as the state of the ViewModel changes. Templates generating
markup using data from the ViewModel can however also be used for this
purpose.
To give a brief initial example, we can look to the JavaScript MVVM
framework KnockoutJS for how it allows the definition of a ViewModel and
its related bindings in markup:
ViewModel:
2 | contactName: ko.observable( "John" ) |
4 | ko.applyBindings(aViewModel); |
View:
1 | <p><input id= "source" data-bind= "value: contactName, valueUpdate: 'keyup'" /></p> |
2 | <div data-bind= "visible: contactName().length > 10" > |
3 | You have a really long name! |
5 | <p>Contact name: <strong data-bind= "text: contactName" ></strong></p> |
Our input text-box (source) obtains it's initial value from contactName
,
automatically updating this value whenever contactName changes. As the
data binding is two-way, typing into the text-box will update contactName
accordingly so the values are always in sync.
Although implementation specific to KnockoutJS, the <div>
containing the "You have a really long name!" text also contains simple
validation (once again in the form of data bindings). If the input
exceeds 10 characters, it will display, otherwise it will remain hidden.
Moving on to a more advanced example, we can return to our Todo
application. A trimmed down KnockoutJS View for this, including all the
necessary data-bindings may look as follows.
04 | <input id= "new-todo" type= "text" data-bind= "value: current, valueUpdate: 'afterkeydown', enterKey: add" |
05 | placeholder= "What needs to be done?" /> |
07 | <section id= "main" data-bind= "block: todos().length" > |
09 | <input id= "toggle-all" type= "checkbox" data-bind= "checked: allCompleted" > |
10 | <label for = "toggle-all" >Mark all as complete</label> |
12 | <ul id= "todo-list" data-bind= "foreach: todos" > |
15 | <li data-bind= "css: { done: done, editing: editing }" > |
16 | <div class= "view" data-bind= "event: { dblclick: $root.editItem }" > |
17 | <input class= "toggle" type= "checkbox" data-bind= "checked: done" > |
18 | <label data-bind= "text: content" ></label> |
19 | <a class= "destroy" href= "#" data-bind= "click: $root.remove" ></a> |
21 | <input class= "edit' type=" text" |
22 | data-bind= "value:
content, valueUpdate: 'afterkeydown', enterKey: $root.stopEditing,
selectAndFocus: editing, event: { blur: $root.stopEditing }" /> |
Note that the basic layout of the mark-up is relatively straight-forward, containing an input textbox (new-todo
) for adding new items, togglers for marking items as complete and a list (todo-list
) with a template for a Todo item in the form of an li
.
The data bindings in the above markup can be broken down as follows:
- The input textbox
new-todo
has a data-binding for the current
property, which is where the value of the current item being added is stored. Our ViewModel (shown shortly) observes the current
property and also has a binding against the add
event. When the enter key is pressed, the add
event is triggered and our ViewModel can then trim the value of current
and add it to the Todo list as needed
- The input checkbox
toggle-all
can mark all of the current items as completed if clicked. If checked, it triggers the allCompleted
event, which can be seen in our ViewModel
- The item
li
has the class done
. When a task is marked as done, the CSS class editing
is marked accordingly. If double-clicking on the item, the $root.editItem
callback will be executed
- The checkbox with the class
toggle
shows the state of the done
property
- A label contains the text value of the Todo item (
content
)
- There is also a remove button that will call the
$root.remove
callback when clicked.
- An input textbox used for editing mode also holds the value of the Todo item
content
. The enterKey
event will set the editing
property to true or false
ViewModel
The ViewModel can be considered a specialized Controller that acts as
a data converter. It changes Model information into View information,
passing commands from the View to the Model.
For example, let us imagine that we have a model containing a date
attribute in unix format (e.g 1333832407). Rather than our models being
aware of a user's view of the date (e.g 04/07/2012 @ 5:00pm), where it
would be necessary to convert the attribute to its display format, our
model simply holds the raw format of the data. Our View contains the
formatted date and our ViewModel acts as a middle-man between the two.
In this sense, the ViewModel might be looked upon as more of a Model
than a View but it does handle most of the View's display logic. The
ViewModel may also expose methods for helping to maintain the View's
state, update the model based on the action's on a View and trigger
events on the View.
In summary, the ViewModel sits behind our UI layer. It exposes data
needed by a View (from a Model) and can be viewed as the source our
Views go to for both data and actions.
KnockoutJS interprets the ViewModel as the representation of data and
operations that can be performed on a UI. This isn't the UI itself nor
the data model that persists, but rather a layer that can also hold the
yet to be saved data a user is working with. Knockout's ViewModels are
implemented JavaScript objects with no knowledge of HTML markup. This
abstract approach to their implementation allows them to stay simple,
meaning more complex behavior can be more easily managed on-top as
needed.
A partial KnockoutJS ViewModel for our Todo application could thus look as follows:
02 | var ViewModel = function ( todos ) { |
06 | self.todos = ko.observableArray( |
07 | ko.utils.arrayMap( todos, function ( todo ) { |
08 | return new Todo( todo.content, todo.done ); |
12 | self.current = ko.observable(); |
15 | self.add = function ( data, event ) { |
16 | var newTodo, current = self.current().trim(); |
18 | newTodo = new Todo( current ); |
19 | self.todos.push( newTodo ); |
25 | self.remove = function ( todo ) { |
26 | self.todos.remove( todo ); |
30 | self.removeCompleted = function () { |
31 | self.todos.remove( function (todo) { |
37 | self.allCompleted = ko.computed({ |
41 | return !self.remainingCount(); |
45 | write: function ( newValue ) { |
46 | ko.utils.arrayForEach( self.todos(), function ( todo ) { |
48 | todo.done( newValue ); |
54 | self.editItem = function ( item ) { |
Above we are basically providing the methods needed to add, edit or
remove items as well as the logic to mark all remaining items as having
been completed Note: The only real difference worth noting from previous
examples in our ViewModel are observable arrays. In KnockoutJS, if we
wish to detect and respond to changes on a single object, we would use observables
. If however we wish to detect and respond to changes of a collection of things, we can use an observableArray
instead. A simpler example of how to use observables arrays may look as follows:
2 | var myObservableArray = ko.observableArray(); |
5 | myObservableArray.push( 'A new todo item' ); |
Note: The complete Knockout.js Todo application we reviewed above can be grabbed from TodoMVC if interested.
Recap: The View and the ViewModel
Views and ViewModels communicate using data-bindings and events. As
we saw in our initial ViewModel example, the ViewModel doesn’t just
expose Model attributes but also access to other methods and features
such as validation.
Our Views handle their own user-interface events, mapping them to the
ViewModel as necessary. Models and attributes on the ViewModel are
synchronized and updated via two-way data-binding.
Triggers (data-triggers) also allow us to further react to changes in the state of our Model attributes.
Recap: The ViewModel and the Model
Whilst it may appear the ViewModel is completely responsible for the
Model in MVVM, there are some subtleties with this relationship worth
noting. The ViewModel can expose a Model or Model attributes for the
purposes of data-binding and can also contain interfaces for fetching
and manipulating properties exposed in the view.
Pros and Cons
We now hopefully have a better appreciation for what MVVM is and how
it works. Let’s now review the advantages and disadvantages of employing
the pattern:
Advantages
- MVVM Facilitates easier parallel development of a UI and the building blocks that power it
- Abstracts the View and thus reduces the quantity of business logic (or glue) required in the code behind it
- The ViewModel can be easier to unit test than event-driven code
- The ViewModel (being more Model than View) can be tested without concerns of UI automation and interaction
Disadvantages
- For simpler UIs, MVVM can be overkill
- Whilst data-bindings can be declarative and nice to work with,
they can be harder to debug than imperative code where we simply set
breakpoints
- Data-bindings in non-trivial applications can create a lot of
book-keeping. We also don’t want to end up in a situation where bindings
are heavier than the objects being bound to
- In larger applications, it can be more difficult to design the ViewModel up front to get the necessary amount of generalization
MVVM With Looser Data-Bindings
It’s not uncommon for JavaScript developers from an MVC or MVP
background to review MVVM and complain about its true separation of
concerns. Namely, the quantity of inline data-bindings maintained in the
HTML markup of a View.
I must admit that when I first reviewed implementations of MVVM (e.g
KnockoutJS, Knockback), I was surprised that any developer would want to
return to the days of old where we mixed logic (JavaScript) with our
markup and found it quickly unmaintainable. The reality however is that
MVVM does this for a number of good reasons (which we’ve covered),
including facilitating designers to more easily bind to logic from their
markup.
For the purists among us, you’ll be happy to know that we can now
also greatly reduce how reliant we are on data-bindings thanks to a
feature known as custom binding providers, introduced in KnockoutJS 1.3
and available in all versions since.
KnockoutJS by default has a data-binding provider which searches for any elements with data-bind
attributes on them such as in the below example.
1 | <input id= "new-todo" type= "text" data-bind= "value: current, valueUpdate: 'afterkeydown', enterKey: add" placeholder= "What needs to be done?" /> |
When the provider locates an element with this attribute, it parses
it and turns it into a binding object using the current data context.
This is the way KnockoutJS has always worked, allowing us to
declaratively add bindings to elements which KnockoutJS binds to the
data at that layer.
Once we start building Views that are no longer trivial, we may end
up with a large number of elements and attributes whose bindings in
markup can become difficult to manage. With custom binding providers
however, this is no longer a problem.
A binding provider is primarily interested in two things:
- When given a DOM node, does it contain any data-bindings?
- If the node passed this first question, what does the binding object look like in the current data context?
Binding providers implement two functions:
nodeHasBindings
: this takes in a DOM node which doesn’t necessarily have to be an element
getBindings
: returns an object representing the bindings as applied to the current data context
A skeleton binding provider might thus look as follows:
1 | var ourBindingProvider = { |
2 | nodeHasBindings: function ( node ) { |
6 | getBindings: function ( node, bindingContext ) { |
Before we get to fleshing out this provider, let’s briefly discuss logic in data-bind attributes.
If when using Knockout’s MVVM we find yourself dissatisfied with the
idea of application logic being overly tied into your View, we can
change this. We could implement something a little like CSS classes to
assign bindings by name to elements. Ryan Niemeyer (of knockmeout.net)
has previously suggested using data-class
for this to avoid confusing presentation classes with data classes, so let’s get our nodeHasBindings
function supporting this:
2 | function nodeHasBindings( node ) { |
3 | return node.getAttribute ? node.getAttribute( "data-class" ) : false ; |
Next, we need a sensible getBindings()
function. As
we’re sticking with the idea of CSS classes, why not also consider
supporting space-separated classes to allow us to share binding specs
between different elements?
Let’s first review what our bindings will look like. We create an
object to hold them where our property names need to match the keys we
wish to use in our data-classes.
Note: There isn’t a great deal of work required to convert a
KnockoutJS application from using traditional data-bindings over to
unobtrusive bindings with custom binding providers. We simply pull our
all of our data-bind attributes, replace them with data-class attributes
and place our bindings in a binding object as per below:
01 | var viewModel = new ViewModel( todos || [] ), |
05 | value: viewModel.current, |
06 | valueUpdate: "afterkeydown" , |
07 | enterKey: viewModel.add |
11 | visible: viewModel.showTooltip |
14 | visible: viewModel.todos().length |
17 | checked: viewModel.allCompleted |
21 | foreach: viewModel.todos |
23 | todoListItem: function () { |
30 | todoListItemWrapper: function () { |
37 | todoCheckBox: function () { |
42 | todoContent: function () { |
50 | todoDestroy: function () { |
52 | click: viewModel.remove |
56 | todoEdit: function () { |
59 | valueUpdate: "afterkeydown" , |
60 | enterKey: this .stopEditing, |
62 | blur: this .stopEditing |
68 | visible: viewModel.remainingCount |
71 | text: viewModel.remainingCount |
73 | remainingCountWord: function () { |
75 | text: viewModel.getLabel(viewModel.remainingCount) |
79 | visible: viewModel.completedCount |
82 | click: viewModel.removeCompleted |
85 | text: viewModel.completedCount |
87 | completedCountWord: function () { |
89 | text: viewModel.getLabel(viewModel.completedCount) |
93 | visible: viewModel.todos().length |
There are however two lines missing from the above snippet - we still need our getBindings
function, which will loop through each of the keys in our data-class
attributes and build up the resulting object from each of them. If we
detect that the binding object is a function, we call it with our
current data using the context this
. Our complete custom binding provider would look as follows:
03 | ko.customBindingProvider = function ( bindingObject ) { |
04 | this .bindingObject = bindingObject; |
07 | this .nodeHasBindings = function ( node ) { |
08 | return node.getAttribute ? node.getAttribute( "data-class" ) : false ; |
13 | this .getBindings = function ( node, bindingContext ) { |
16 | classes = node.getAttribute( "data-class" ); |
19 | classes = classes.split( "" ); |
22 | for ( var i = 0, j = classes.length; i < j; i++ ) { |
24 | var bindingAccessor = this .bindingObject[classes[i]]; |
25 | if ( bindingAccessor ) { |
26 | var binding = typeof bindingAccessor === "function" ? bindingAccessor.call(bindingContext.$data) : bindingAccessor; |
27 | ko.utils.extend(result, binding); |
Thus, the final few lines of our bindings
object can be defined as follows:
2 | ko.bindingProvider.instance = new ko.customBindingProvider( bindings ); |
5 | ko.applyBindings( viewModel ); |
What we’re doing here is effectively defining constructor for our
binding handler which accepts an object (bindings) which we use to
lookup our bindings. We could then re-write the markup for our
application View using data-classes as follows:
02 | <input id= "new-todo" data-class= "newTodo" placeholder= "What needs to be done?" /> |
03 | <span class= "ui-tooltip-top" data-class= "taskTooltip" style= "display: none;" >Press Enter to save this task</span> |
06 | <div data-class= "checkAllContainer" > |
07 | <input id= "check-all" class= "check" type= "checkbox" data-class= "checkAll" /> |
08 | <label for = "check-all" >Mark all as complete</label> |
10 | <ul id= "todo-list" data-class= "todos" > |
11 | <li data-class= "todoListItem" > |
12 | <div class= "todo" data-class= "todoListItemWrapper" > |
14 | <input class= "check" type= "checkbox" data-class= "todoCheckBox" /> |
15 | <div class= "todo-content" data-class= "todoContent" style= "cursor: pointer;" ></div> |
16 | <span class= "todo-destroy" data-class= "todoDestroy" ></span> |
19 | <input class= "todo-input" data-class="todoEdit'/> |
Neil Kerkin has put together a complete TodoMVC demo app using the above, which can be accessed and played around with here.
Whilst it may look like quite a lot of work in the explanation above, now that we have a generic getBindings
method written, it’s a lot more trivial to simply re-use it and use
data-classes rather than strict data-bindings for writing our KnockoutJS
applications instead. The net result is hopefully cleaner markup with
our data bindings being shifted from the View to a bindings object
instead.
MVC Vs. MVP Vs. MVVM
Both MVP and MVVM are derivatives of MVC. The key difference between
it and its derivatives is the dependency each layer has on other layers
as well as how tightly bound they are to each other.
In MVC, the View sits on top of our architecture with the controller
beside it. Models sit below the controller and so our Views know about
our controllers and controllers know about Models. Here, our Views have
direct access to Models. Exposing the complete Model to the View however
may have security and performance costs, depending on the complexity of
our application. MVVM attempts to avoid these issues.
In MVP, the role of the controller is replaced with a Presenter.
Presenters sit at the same level as views, listening to events from both
the View and model and mediating the actions between them. Unlike MVVM,
there isn’t a mechanism for binding Views to ViewModels, so we instead
rely on each View implementing an interface allowing the Presenter to
interact with the View.
MVVM consequently allows us to create View-specific subsets of a
Model which can contain state and logic information, avoiding the need
to expose the entire Model to a View. Unlike MVP’s Presenter, a
ViewModel is not required to reference a View. The View can bind to
properties on the ViewModel which in turn expose data contained in
Models to the View. As we’ve mentioned, the abstraction of the View
means there is less logic required in the code behind it.
One of the downsides to this however is that a level of
interpretation is needed between the ViewModel and the View and this can
have performance costs. The complexity of this interpretation can also
vary - it can be as simple as copying data or as complex as manipulating
them to a form we would like the View to see. MVC doesn’t have this
problem as the whole Model is readily available and such manipulation
can be avoided.
Backbone.js Vs. KnockoutJS
Understanding the subtle differences between MVC, MVP and MVVM are
important but developers ultimately will ask whether they should
consider using KnockoutJS over Backbone based in what we’ve learned. The
following notes may be of help here:
-
Both libraries are designed with different goals in mind and its often not as simple as just choosing MVC or MVVM
-
If data-binding and two-way communication are your main concerns,
KnockoutJS is definitely the way to go.Practically any attribute or
value stored in DOM nodes can be mapped to JavaScript objects with this
approach.
-
Backbone excels with its ease of integration with RESTful
services, whilst KnockoutJS Models are simply JavaScript objects and
code needed for updating the Model must be written by the developer.
-
KnockoutJS has a focus on automating UI bindings, which requires
significantly more verbose custom code if attempting to do this with
Backbone. This isn't a problem with Backbone itself per se as it
purposefully attempts to stay out of the UI. Knockback does however
attempt to assist with this problem.
-
With KnockoutJS, we can bind our own functions to ViewModel
observables, which are executed anytime the observable changes. This
allows us the same level of flexibility as can be found in Backbone
-
Backbone has a solid routing solution built-in, whilst KnockoutJS
offers no routing options out of the box. One can however easily fill
this behavior in if needed using Ben Alman’s BBQ plugin or a standalone routing system like Miller Medeiros’s excellent Crossroads.
To conclude, I personally find KnockoutJS more suitable for smaller
applications whilst Backbone’s feature set really shines when building
anything non-trivial. That said, many developers have used both
frameworks to write applications of varying complexity and I recommend
trying out both at a smaller scale before making a decision on which
might work best for your project.
For further reading about MVVM or Knockout, I recommend the following articles:
#
Modern Modular JavaScript Design Patterns
The Importance Of Decoupling Applications
In the world of scalable JavaScript, when we say an application is modular, we often mean it's composed of a set of highly decoupled, distinct pieces of functionality stored in modules. Loose coupling facilitates easier maintainability of apps by removing dependencies
where possible. When this is implemented efficiently, it's quite easy
to see how changes to one part of a system may affect another.
Unlike some more traditional programming languages however, the current iteration of JavaScript (ECMA-262)
doesn't provide developers with the means to import such modules of
code in a clean, organized manner. It's one of the concerns with
specifications that haven't required great thought until more recent
years where the need for more organized JavaScript applications became
apparent.
Instead, developers at present are left to fall back on variations of the module or object literal
patterns, which we covered earlier in the book. With many of these,
module scripts are strung together in the DOM with namespaces being
described by a single global object where it's still possible to incur
naming collisions in our architecture. There's also no clean way to
handle dependency management without some manual effort or third party
tools.
Whilst native solutions to these problems will be arriving in ES Harmony
(likely to be the next version of JavaScript), the good news is that
writing modular JavaScript has never been easier and we can start doing
it today.
In this section, we're going to look at three formats for writing modular JavaScript: AMD, CommonJS and proposals for the next version of JavaScript, Harmony.
A Note On Script Loaders
It's difficult to discuss AMD and CommonJS modules without talking about the elephant in the room - script loaders.
At the time of writing this book, script loading is a means to a goal,
that goal being modular JavaScript that can be used in applications
today - for this, use of a compatible script loader is unfortunately
necessary. In order to get the most out of this section, I recommend
gaining a basic understanding of how popular script loading tools work so the explanations of module formats make sense in context.
There are a number of great loaders for handling module loading in the
AMD and CommonJS formats, but my personal preferences are RequireJS and curl.js. Complete tutorials on these tools are outside the scope of this book, but I can recommend reading John Hann's article about curl.js and James Burke's RequireJS API documentation for more.
From a production perspective, the use of optimization tools (like
the RequireJS optimizer) to concatenate scripts is recommended for
deployment when working with such modules. Interestingly, with the Almond
AMD shim, RequireJS doesn't need to be rolled in the deployed site and
what one might consider a script loader can be easily shifted outside of
development.
That said, James Burke would probably say that being able to
dynamically load scripts after page load still has its use cases and
RequireJS can assist with this too. With these notes in mind, let's get
started.
AMD
A Format For Writing Modular JavaScript In The Browser
The overall goal for the AMD (Asynchronous Module Definition) format
is to provide a solution for modular JavaScript that developers can use
today. It was born out of Dojo's real world experience using XHR+eval
and proponents of this format wanted to avoid any future solutions
suffering from the weaknesses of those in the past.
The AMD module format itself is a proposal for defining modules where both the module and dependencies can be asynchronously
loaded. It has a number of distinct advantages including being both
asynchronous and highly flexible by nature which removes the tight
coupling one might commonly find between code and module identity. Many
developers enjoy using it and one could consider it a reliable stepping
stone towards the module system proposed for ES Harmony.
AMD began as a draft specification for a module format on the
CommonJS list but as it wasn't able to reach full consensus, further
development of the format moved to the amdjs group.
Today it's embraced by projects including Dojo, MooTools, Firebug and even jQuery . Although the term CommonJS AMD format
has been seen in the wild on occasion, it's best to refer to it as just
AMD or Async Module support as not all participants on the CommonJS
list wished to pursue it.
Note: There was a
time when the proposal was referred to as Modules Transport/C, however
as the spec wasn't geared towards transporting existing CommonJS
modules, but rather - for defining modules - it made more sense to opt
for the AMD naming convention.
Getting Started With Modules
The first two concepts worth noting about AMD are the idea of a define
method for facilitating module definition and a require
method for handling dependency loading. define is used to define named or unnamed modules based using the following signature:
As we can tell by the inline comments, the module_id
is
an optional argument which is typically only required when non-AMD
concatenation tools are being used (there may be some other edge cases
where it's useful too). When this argument is left out, we refer to the
module as anonymous.
When working with anonymous modules, the idea of a module's identity
is DRY, making it trivial to avoid duplication of filenames and code.
Because the code is more portable, it can be easily moved to other
locations (or around the file-system) without needing to alter the code
itself or change its module ID. Consider the module_id
similar to the concept of folder paths.
Note: Developers can run this same code on multiple environments just
by using an AMD optimizer that works with a CommonJS environment such
as r.js.
Back to the define
signature, the dependencies argument
represents an array of dependencies which are required by the module we
are defining and the third argument ("definition function" or "factory
function") is a function that's executed to instantiate our module. A
bare bone module could be defined as follows:
Understanding AMD: define()
08 | function ( foo, bar ) { |
15 | console.log( "Yay! Stuff" ); |
27 | function ( math, graph ) { |
34 | plot: function ( x, y ){ |
35 | return graph.drawPie( math.randomGrid( x, y ) ); |
require on the other hand is typically used to load code in a
top-level JavaScript file or within a module should we wish to
dynamically fetch dependencies. An example of its usage is:
Understanding AMD: require()
6 | require([ "foo" , "bar" ], function ( foo, bar ) { |
Dynamically-loaded Dependencies
01 | define( function ( require ) { |
02 | var isReady = false , foobar; |
05 | require([ "foo" , "bar" ], function ( foo, bar ) { |
07 | foobar = foo() + bar(); |
Understanding AMD: plugins
The following is an example of defining an AMD-compatible plugin:
06 | define( [ "./templates" , "text!./template.md" , "css!./template.css" ], |
08 | function ( templates, template ){ |
09 | console.log( templates ); |
Note: Although css!
is included for loading CSS dependencies in the above example, it's
important to remember that this approach has some caveats such as it not
being fully possible to establish when the CSS is fully loaded.
Depending on how we approach our build process, it may also result in
CSS being included as a dependency in the optimized file, so use CSS as a
loaded dependency in such cases with caution. If interested in doing
the above, we can also explore @VIISON's RequireJS CSS plugin further
here: https://github.com/VIISON/RequireCSS
Loading AMD Modules Using RequireJS
1 | require([ "app/myModule" ], |
6 | var module = new myModule(); |
This example could simply be looked at as requirejs(["app/myModule"], function(){})
which indicates the loader's top level globals are being used. This is
how to kick off top-level loading of modules with different AMD loaders
however with a define()
function, if it's passed a local require all require([])
examples apply to both types of loader (curl.js and RequireJS).
Loading AMD Modules Using curl.js
1 | curl([ "app/myModule.js" ], |
6 | var module = new myModule(); |
Modules With Deferred Dependencies
05 | define([ "lib/Deferred" ], function ( Deferred ){ |
06 | var defer = new Deferred(); |
08 | require([ "lib/templates/?index.html" , "lib/data/?stats" ], |
09 | function ( template, data ){ |
10 | defer.resolve( { template: template, data:data } ); |
13 | return defer.promise(); |
AMD Modules With Dojo
Defining AMD-compatible modules using Dojo is fairly
straight-forward. As per above, define any module dependencies in an
array as the first argument and provide a callback (factory) which will
execute the module once the dependencies have been loaded. e.g:
1 | define([ "dijit/Tooltip" ], function ( Tooltip ){ |
Note the anonymous nature of the module, which can now be both
consumed by a Dojo asynchronous loader, RequireJS or the standard dojo.require() module loader.
There are some interesting gotchas with module referencing that are
useful to know here. Although the AMD-advocated way of referencing
modules declares them in the dependency list with a set of matching
arguments, this isn't supported by the older Dojo 1.6 build system - it
really only works for AMD-compliant loaders. e.g:
1 | define([ "dojo/cookie" , "dijit/Tooltip" ], function ( cookie, Tooltip ){ |
3 | var cookieValue = cookie( "cookieName" ); |
This has many advantages over nested namespacing as modules no longer
need to directly reference complete namespaces every time - all we
require is the "dojo/cookie" path in dependencies, which once aliased to
an argument, can be referenced by that variable. This removes the need
to repeatedly type out "dojo." in our applications.
The final gotcha to be aware of is that if we wish to continue using
the older Dojo build system or wish to migrate older modules to this
newer AMD-style, the following more verbose version enables easier
migration. Notice that dojo and dijit and referenced as dependencies
too:
1 | define([ "dojo" , "dijit', " dojo/cookie ", " dijit/Tooltip"], function ( dojo, dijit ){ |
2 | var cookieValue = dojo.cookie( "cookieName" ); |
3 | new dijit.Tooltip(...); |
AMD Module Design Patterns (Dojo)
As we've seen in previous sections, design patterns can be highly
effective in improving how we approach structuring solutions to common
development problems. John Hann
has given some excellent presentations about AMD module design patterns
covering the Singleton, Decorator, Mediator and others and I highly
recommend checking out his slides if we get a chance.
A selection of AMD design patterns can be found below.
Decorator pattern:
02 | define([ "dojo" , "dojo/store/Observable" ], function ( dojo, Observable ) { |
03 | return function UpdatableObservable ( store ) { |
05 | var observable = dojo.isFunction( store.notify ) ? store : |
06 | new Observable(store); |
08 | observable.updated = function ( object ) { |
09 | dojo.when( object, function ( itemOrArray) { |
10 | dojo.forEach( [].concat(itemOrArray), this .notify, this ); |
22 | define([ "mylib/UpdatableObservable" ], function ( makeUpdatable ) { |
28 | updatable = makeUpdatable( observable ); |
Adapter pattern
02 | define([ "dojo/_base/lang" , "dojo/_base/array" ], function ( lang, array ) { |
03 | return lang.delegate( array, { |
04 | each: function ( arr, lambda ) { |
05 | array.forEach( arr, function ( item, i ) { |
06 | lambda.call( item, i, item ); |
14 | define([ "mylib/Array" ], function ( array ) { |
15 | array.each( [ "uno" , "dos" , "tres" ], function ( i, esp ) { |
AMD Modules With jQuery
Unlike Dojo, jQuery really only comes with one file, however given
the plugin-based nature of the library, we can demonstrate how
straight-forward it is to define an AMD module that uses it below.
01 | define([ "js/jquery.js" , "js/jquery.color.js" , "js/underscore.js" ], |
03 | function ( $, colorPlugin, _ ){ |
10 | var shuffleColor = _.first( _.shuffle( "#666" , "#333" , "#111" ] ) ); |