easejs/doc/impl-details.texi

@c  This document is part of the ease.js manual
@c  Copyright (c) 2011 Mike Gerwitz
@c    Permission is granted to copy, distribute and/or modify this document
@c    under the terms of the GNU Free Documentation License, Version 1.3
@c    or any later version published by the Free Software Foundation;
@c    with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
@c    Texts.  A copy of the license is included in the section entitled ``GNU
@c    Free Documentation License''.

@node Implementation Details
@appendix Implementation Details / Rationale
The majority of the development time spent on ease.js was not hacking away at
the source code. Rather, it was spent with pen and paper. Every aspect of
ease.js was heavily planned from the start. Every detail was important to ensure
a consistent implementation that worked, was fast and that developers would
enjoy working with. Failures upfront or alterations to the design in later
versions would break backwards compatibility unnecessarily and damage the
reputation of the project.

When using ease.js, developers may wonder why things were implemented in the
manner that they were. Perhaps they have a problem with the implementation, or
just want to learn how the project works. This project was an excellent learning
experience that deals very closely with the power and flexibility of prototypal
programming. In an attempt to appease both parties, this appendix is provided to
provide some details and rationale behind ease.js.

@menu
* Class Module Design::
* Visibility Implementation::
* Internal Methods/Objects::
@end menu


@node Class Module Design
@section Class Module Design
The @var{Class} module, which is accessible via @samp{require( 'easejs'
).Class}, is the backbone of the entire project. In a class-based
Object-Oriented model, as one could guess by the name, the class is the star
player. When the project began, this was the only initial implementation detail.
Everything else was later layered atop of it.

As such, developing the Class module took the most thought and presented the
largest challenge throughout the project. Every detail of its implementation
exists for a reason. Nothing was put in place because the author simply ``felt
like it''. The project aims to exist as a strong, reliable standard for the
development of JavaScript-based applications. If such a goal is to be attained,
the feature set and implementation details would have to be strongly functional,
easy to use and make sense to the Object-Oriented developer community.

The design also requires a strong understanding of Object-Oriented development.
Attention was paid to the nuances that could otherwise introduce bugs or an
inconsistent implementation.

@menu
* Class Declaration Syntax::
* Class Storage::
* Constructor Implementation::
* Static Implementation::
@end menu

@node Class Declaration Syntax
@subsection Class Declaration Syntax
Much thought was put into how a class should be declared. The chosen style
serves as syntatic sugar, making the declarations appear very similar to classes
in other Object-Oriented languages.

The original style was based on John Resig's blog post about a basic means of
extending class-like objects (@pxref{About}). That style was
@samp{Class.extend()} to declare a new class and @samp{Foo.extend()} to extend
an existing class. This implementation is still supported for creating anonymous
classes. However, a means needed to be provided to create named classes. In
addition, invoking @code{extend()} on an empty class seemed unnecessary.

The next incarnation made the @var{Class} module invokable. Anonymous classes
could be defined using @samp{Class( @{@} )} and named classes could be defined
by passing in a string as the first argument: @samp{Class( 'Foo', @{@} )}.
Classes could still be extended using the previously mentioned syntax, but that
did no justice if we need to provide a class name. Therefore, the @samp{Class(
'SubFoo' ).extend( Supertype, @{@} )} syntax was also adopted.

JavaScript's use of curly braces to represent objects provides a very convenient
means of making class definitions look like actual class definitions. By
convention, the opening brace for the declaration object is on its own line, to
make it look like an opening block.

@float Figure, f:class-def-syntax
@verbatim
    Class( 'Foo' )
        .implement( Bar )
        .extend(
    {
        'public foo': function()
        {
        }
    } );
@end verbatim
@caption{Syntax and style of class definition}
@end float

Syntax for implementing interfaces and extending classes was another
consideration. The implementation shown above was chosen for a couple of
reasons. Firstly, verbs were chosen in order to (a) prevent the use of reserved
words and (b) to represent that the process was taking place at @emph{runtime},
@emph{as} the code was being executed. Unlike a language like C++ or Java, the
classes are not prepared at compile-time.

@node Class Storage
@subsection Class Storage
One of the more powerful features of ease.js is how classes (and other objects,
such as Interfaces) are stored. Rather than adopting its own model, the decision
was instead to blend into how JavaScript already structures its data. Everything
in JavaScript can be assigned to a variable, including functions. Classes are no
different.

One decision was whether or not to store classes internally by name, then permit
accessing it globally (wherever ease.js is available). This is how most
Object-Oriented languages work. If the file in which the class is defined is
available, the class can generally be referenced by name. This may seem natural
to developers coming from other Object-Oriented languages. The decision was to
@emph{not} adopt this model.

By storing classes @emph{only} in variables, we have fine control over the
scope and permit the developer to adopt their own mechanism for organizing their
classes. For example, if the developer wishes to use namespacing, then he/she is
free to assign the class to a namespace (e.g. @samp{org.foo.my.ns.Foo =
Class( @{@} )}). More importantly, we can take advantage of the CommonJS format
that ease.js was initially built for by assigning the class to
@code{module.exports}. This permits @samp{require( 'filename' )} to return the
class.

This method also permits defining anonymous classes (while not necessarily
recommended, they have their uses just as anonymous functions do), mimic the
concept of Java's inner classes and create temporary classes (@pxref{Temporary
Classes}). Indeed, we can do whatever scoping that JavaScript permits.

@subsubsection Memory Management
Memory management is perhaps one of the most important considerations.
Initially, ease.js encapsulated class metadata and visibility structures
(@pxref{Hacking Around the Issue of Encapsulation}).  However, it quickly became
apparent that this method of storing data, although excellent for protecting it
from being manipulated, caused what appeared to be memory leaks in long-running
software. These were in fact not memory leaks, but ease.js keeping references to
class data with no idea when to free them.

To solve this issue, all class data is stored within the class itself (that is,
the constructor in JavaScript terms). They are stored in obscure variables that
are non-enumerable and subject to change in future releases. This ensures that
developers cannot rely on using them for reflection purposes or for manipulating
class data during runtime. This is important, since looking at such members can
give access to protected and private instance data. In the future, the names may
be randomly chosen at runtime to further mitigate exploits. Until that time,
developers should be aware of potential security issues.

If the globally accessible model would have been adopted (storing classes
internally by class name rather than in variables), classes would not have been
freed from memory when they went out of scope. This raises the memory footprint
unnecessarily, especially for temporary classes. It would make sense that, after
a temporary class is done being used, that the class be freed from memory.

Given this fact alone, the author firmly believes that the model that was chosen
was the best choice.

@node Constructor Implementation
@subsection Constructor Implementation
ease.js uses a PHP-style constructor. Rather than using the class name as the
constructor, a @code{__construct()} method is used. This was chosen primarily
because ease.js does not always know the name of the class. In fact, in the
early stages of development, named classes were unsupported. With the PHP-style
constructor, the class name does not need to be known, allowing constructors to
be written for anonymous and named classes alike.

In addition, the PHP-style constructor is consistent between class definitions.
To look up a constructor, one need only search for ``__construct'', rather than
the class name. This makes certain operations, such as global searching (using
@command{grep} or any other utility), much simpler.

One difference from PHP is the means of preventing instantiation. In PHP, if the
constructor is declared as non-public, then an error will be raised when the
developer attempts to instantiate the class. ease.js did not go this route, as
the method seems cryptic. Instead, an exception should be thrown in the
constructor if the developer doesn't wish the class to be instantiated. In the
future, a common method may be added for consistency/convenience.

The constructor is optional. If one is not provided, nothing is done after the
class is instantiated (aside from the internal ease.js initialization tasks).

The constructor is called after all initialization tasks have been completed.

@node Static Implementation
@subsection Static Implementation
The decisions behind ease.js's static implementation were very difficult. More
thought and time was spent on paper designing how the static implementation
should be represented than most other features in the project. The reason for
this is not because the concept of static members is complicated. Rather, it is
due to limitations of pre-ECMAScript 5 engines.

@subsubsection How Static Members Are Supposed To Work
The first insight into the problems a static implementation would present was
the concept itself. Take any common Object-Oriented language such as C++, Java,
or even PHP. Static members are inherited by subtypes @emph{by reference}. What
does this mean? Consider two classes: @var{Foo} and @var{SubFoo}, the latter of
which inherits from the former. @var{Foo} defines a static property @var{count}
to be incremented each time the class is instantiated. The subtype @var{SubFoo},
when instantiated (assuming the constructor is not overridden), would increment
that very same count. Therefore, we can represent this by stating that
@samp{Foo.count === SubFoo.count}. In the example below, we demonstrate this
concept in pseudocode:

@float Figure, f:static-ref-pseudocode
@verbatim
    let Foo = Class
        public static count = 0
    let SubFoo extend from Foo

    Foo.count = 5
    SubFoo.count === 5 // true
    SubFoo.count = 6
    Foo.count === 6 // true
@end verbatim
@caption{Representing static properties in pseudocode}
@end float

As you may imagine, this is a problem. The above example does not look very
JS-like. That is because it isn't. JS does not provide a means for variables to
share references to the same primitive. In fact, even Objects are passed by
value in the sense that, if the variable is reassigned, the other variable
remains unaffected. The concept we are looking to support is similar to a
pointer in C/C++, or a reference in PHP.

We have no such luxury.

@subsubsection Emulating References
Fortunately, ECMAScript 5 provides a means to @emph{emulate} references --
getters and setters. Taking a look at @ref{f:static-ref-pseudocode}, we can
clearly see that @var{Foo} and @var{SubFoo} are completely separate objects.
They do not share any values by references. We shouldn't share primitives by
reference even if we wanted to. This issue can be resolved by using
getters/setters on @var{SubFoo} and @emph{forwarding} gets/sets to the
supertype:

@float Figure, f:static-ref-forward
@verbatim
    var obj1 = { val: 1 },
        obj2 = {
            get val()
            {
                return obj1.val;
            },

            set val( value )
            {
                obj1.val = value;
            },
        }
    ;

    obj2.val; // 1
    obj2.val = 5;
    obj1.val; // 5

    obj1.val = 6;
    obj2.val // 6
@end verbatim
@caption{Emulating references with getters/setters (proxy)}
@end float

This comes with considerable overhead when compared to accessing the properties
directly (in fact, at the time of writing this, V8 doesn't even attempt to
optimize calls to getters/setters, so it is even slower than invoking accessor
methods). That point aside, it works well and accomplishes what we need it to.

There's just one problem. @emph{This does not work in pre-ES5 environments!}
ease.js needs to support older environments, falling back to ensure that
everything operates the same (even though features such as visibility aren't
present).

This means that we cannot use this proxy implementation. It is used for
visibility in class instances, but that is because a fallback is possible. It is
not possible to provide a fallback that works with two separate objects. If
there were, we wouldn't have this problem in the first place.

@subsubsection Deciding On a Compromise
A number of options were available regarding how static properties should be
implemented. Methods are not a problem -- they are only accessed by reference,
never written to. Therefore, they can keep their convenient @samp{Foo.method()}
syntax. Unfortunately, that cannot be the case for properties without the
ability to implement a proxy through the use of getters/setters (which, as
aforementioned, requires the services of ECMAScript 5, which is not available in
older environments).

The choices were has follows:

@enumerate
@item
Add another object to be shared between classes (e.g. @samp{Foo.$}).

@item
Do not inherit by reference. Each subtype would have their own distinct value.

@item
Access properties via an accessor method (e.g. @samp{Foo.$('var')}), allowing us
to properly proxy much like a getter/setter.
@end enumerate

There are problems with all of the above options. The first option, which
involves sharing an object, would cause awkward inheritance in the case of a
fallback. Subtypes would set their static properties on the object, which would
make that property available to the @emph{supertype}! That is tolerable in the
case of a fallback. However, the real problem lies in two other concepts: when a
class has two subtypes that attempt to define a property with the same name, or
when a subtype attempts to override a property. The former would cause both
subtypes (which are entirely separate from one-another, with the exception of
sharing the same parent) to share the same values, which is unacceptable. The
latter case can be circumvented by simply preventing overriding of static
properties, but the former just blows this idea out of the water entirely.

The second option is to @emph{not} inherit by reference. This was the initial
implementation (due to JavaScript limitations) until it was realized that this
caused far too many inconsistencies between other Object-Oriented languages.
There is no use in introducing a different implementation when we are attempting
to mirror classic Object-Oriented principals to present a familiar paradigm to
developers. Given this inconsistency alone, this option simply will not work.

The final option is to provide an accessor method, much like the style of
jQuery. This would serve as an ugly alternative for getters/setters. It would
operate as follows:

@float Figure, f:static-accessor-impl
@verbatim
    // external
    Foo.$('var'); // getter
    Foo.$( 'var, 'foo' ); // setter

    // internal
    this.__self.$('var'); // getter
    this.__self.$( 'var', 'foo' ); // setter
@end verbatim
@caption{Accessor implementation for static properties}
@end float

Obviously, this is highly inconsistent with the rest of the framework, which
permits accessing properties in the conventional manner. However, this
implementation does provide a number key benefits:

@itemize
@item
It provides an implementation that is @emph{consistent with other
Object-Oriented languages}. This is the most important point.

@item
The accessor method parameter style is common in other frameworks like jQuery.

@item
The method name (@var{$}) is commonly used to denote a variable in scripting
languages (such as PHP and shells, or to denote a scalar in Perl).

@item
It works consistently in ES5 and pre-ES5 environments alike.
@end itemize

So, although the syntax is inconsistent with the rest of the framework, it does
address all of our key requirements. This makes it a viable option for our
implementation.

@subsubsection Appeasing ES5-Only Developers
There is another argument to be had. ease.js is designed to operate across all
major browsers for all major versions, no matter how ridiculous (e.g. Internet
Explorer 5.5), so long as it does not require unreasonable development effort.
That is great and all, but what about those developers who are developing
@emph{only} for an ECMAScript 5 environment? This includes developers leveraging
modern HTML 5 features and those using Node.js who do not intend to share code
with pre-ES5 clients. Why should they suffer from an ugly, unnecessary syntax
when a beautiful, natural [and elegant] implementation is available using
proxies via getters/setters?

There are certainly two sides to this argument. On one hand, it is perfectly
acceptable to request a natural syntax if it is supported. On the other hand,
this introduces a number of problems:

@itemize
@item
This may make libraries written using ease.js unportable (to older
environments). If written using an ES5-only syntax, they would have no way to
fall back for static properties.

@item
The syntax differences could be very confusing, especially to those beginning to
learn ease.js. They may not clearly understand the differences, or may go to use
a library in their own code, and find that things do not work as intended.
Code examples would also have to make clear note of what static syntax they
decided to use. It adds a layer of complexity.
@end itemize

Now, those arguing for the cleaner syntax can also argue that all newer
environments moving forward will support the clean, ES5-only syntax, therefore
it would be beneficial to have. Especially when used for web applications that
can fall back to an entirely different implementation or refuse service entirely
to older browsers. Why hold ease.js back for those stragglers if there's no
intent on ever supporting them?

Both arguments are solid. Ultimately, ease.js will likely favor the argument of
implementing the cleaner syntax by providing a runtime flag. If enabled, static
members will be set using proxies. If not, it will fall back to the uglier
implementation using the accessor method. If the environment doesn't support the
flag when set, ease.js will throw an error and refuse to run, or will invoke a
fallback specified by the developer to run an alternative code base that uses
the portable, pre-ES5 syntax.

This decision will ultimately be made in the future. For the time being, ease.js
will support and encourage use of the portable static property syntax.


@node Visibility Implementation
@section Visibility Implementation
One of the major distinguishing factors of ease.js is its full visibility
support (@pxref{Access Modifiers}). This feature was the main motivator behind
the project. Before we can understand the use of this feature, we have to
understand certain limitations of JavaScript and how we may be able to work
around them.

@menu
* Encapsulation In JavaScript::
* Hacking Around the Issue of Encapsulation::
* The Visibility Object::
* Method Wrapping::
@end menu

@node Encapsulation In JavaScript
@subsection Encapsulation In JavaScript
Encapsulation is a cornerstone of many strong software development paradigms
(@pxref{Encapsulation}). This concept is relatively simply to achieve using
closures in JavaScript, as shown in the following example stack implementation:

@float Figure, f:js-encapsulation-ex
@verbatim
var stack = {};

( function( exports )
{
    var data = [];

    exports.push = function( data )
    {
        data.push( data );
    };

    exports.pop = function()
    {
        return data.pop();
    };
} )( stack );

stack.push( 'foo' );
stack.pop(); // foo
@end verbatim
@caption{Encapsulation example using closures in JavaScript}
@end float

Because functions introduce scope in JavaScript, data can be hidden within them.
In @ref{f:js-encapsulation-ex} above, a self-executing function is used to
encapsulate the actual data in the stack (@var{data}). The function accepts a
single argument, which will hold the functions used to push and pop values
to/from the stack respectively. These functions are closures that have access to
the @var{data} variable, allowing them to alter its data. However, nothing
outside of the self-executing function has access to the data. Therefore, we
present the user with an API that allows them to push/pop from the stack, but
never allows them to see what data is actually @emph{in} the stack@footnote{The
pattern used in the stack implementation is commonly referred to as the
@dfn{module} pattern and is the same concept used by CommonJS. Another common
implementation is to return an object containing the functions from the
self-executing function, rather than accepting an object to store the values in.
We used the former implementation here for the sake of clarity and because it
more closely represents the syntax used by CommonJS.}.

Let's translate some of the above into Object-Oriented terms:

@itemize
@item @var{push} and @var{pop} are public members of @var{stack}.
@item @var{data} is a private member of @var{stack}.
@item @var{stack} is a Singleton.
@end itemize

We can take this a bit further by defining a @code{Stack} prototype so that we
can create multiple instances of our stack implementation. A single instance
hardly seems useful for reuse. However, in attempting to do so, we run into a
bit of a problem:

@float Figure, f:js-proto-inst-noencapsulate
@verbatim
var Stack = function()
{
    this._data = [];
};

Stack.prototype = {
    push: function( val )
    {
        this._data.push( val );
    },

    pop: function()
    {
        return this._data.pop();
    },
};

// create a new instance of our Stack object
var inst = new Stack();

// what's this?
inst.push( 'foo' );
console.log( inst._data ); // [ 'foo' ]

// uh oh.
inst.pop(); // foo
console.log( inst._data ); // []
@end verbatim
@caption{Working easily with instance members in JavaScript breaks
encapsulation}
@end float

By defining our methods on the prototype and our data in the constructor, we
have created a bit of a problem. Although the data is easy to work with,
@emph{it is no longer encapsulated}. The @var{_data} property is now public,
accessible for the entire work to inspect and modify. As such, a common practice
in JavaScript is to simply declare members that are "supposed to be" private
with an underscore prefix, as we have done above, and then trust that nobody
will make use of them. Not a great solution.

Another solution is to use a concept called @dfn{privileged members}, which uses
closures defined in the constructor rather than functions defined in the
prototype:

@float Figure, f:js-privileged-members
@verbatim
var Stack = function()
{
    var data = [];

    this.push = function( data )
    {
        data.push( data );
    };

    this.pop = function()
    {
        return data.pop();
    };
};

// create a new instance of our Stack object
var inst = new Stack();

// can no longer access "privileged" member _data
inst.push( 'foo' );
console.log( inst._data ); // undefined
@end verbatim
@caption{Privileged members in JavaScript}
@end float

You may notice a strong similarity between @ref{f:js-encapsulation-ex} and
@ref{f:js-privileged-members}. They are doing essentially the same thing, the
only difference being that @ref{f:js-encapsulation-ex} is returning a single
object and @ref{f:js-privileged-members} represents a constructor that may be
instantiated.

When using privileged members, one would define all members that need access to
such members in the constructor and define all remaining members in the
prototype. However, this introduces a rather large problem that makes this
design decision a poor one in practice: @emph{Each time @var{Stack} is
instantiated, @var{push} and @var{pop} have to be redefined, taking up
additional memory and CPU cycles}. Those methods will be kept in memory until
the instance of @var{Stack} is garbage collected.

In @ref{f:js-privileged-members}, these considerations may not seem like much of
an issue. However, consider a constructor that defines tens of methods and could
potentially have hundreds of instances. For this reason, you will often see the
concepts demonstrated in @ref{f:js-proto-inst-noencapsulate} used more
frequently in libraries that have even modest performance requirements.

@node Hacking Around the Issue of Encapsulation
@subsection Hacking Around the Issue of Encapsulation
Since neither @ref{f:js-encapsulation-ex} nor @ref{f:js-privileged-members} are
acceptable implementations for strong Classical Object-Oriented code, another
solution is needed. Based on what we have seen thus far, let's consider our
requirements:

@itemize
@item
Our implementation must not break encapsulation. That is - we should be
enforcing encapsulation, not simply trusting our users not to touch.
@item
We must be gentle with our memory allocations and processing. This means placing
@emph{all} methods within the prototype.
@item
We should not require any changes to how the developer uses the
constructor/object. It should operate just like any other construct in
JavaScript.
@end itemize

We can accomplish the above by using the encapsulation concepts from
@ref{f:js-encapsulation-ex} and the same prototype model demonstrated in
@ref{f:js-proto-inst-noencapsulate}. The problem with
@ref{f:js-encapsulation-ex}, which provided proper encapsulation, was that it
acted as a Singleton. We could not create multiple instances of it and, even if
we could, they would end up sharing the same data. To solve this problem, we
need a means of distinguishing between each of the instances so that we can
access the data of each separately:

@float Figure, f:js-encapsulate-instance
@verbatim
var Stack = ( function()
{
    var idata = [],
        iid   = 0;

    var S = function()
    {
        // set the instance id of this instance, then increment it to ensure the
        // value is unique for the next instance
        this.__iid = iid++;

        // initialize our data for this instance
        idata[ this.__iid ] = {
            stack: [],
        };
    }:

    S.prototype = {
        push: function( val )
        {
            idata[ this.__iid ].stack.push( val );
        },

        pop: function()
        {
            return idata[ this.__iid ].stack.pop();
        }
    };

    return S;
} )();

var stack1 = new Stack();
var stack2 = new Stack();

stack1.push( 'foo' );
stack2.push( 'bar' );

stack1.pop(); // foo
stack2.pop(); // bar
@end verbatim
@caption{Encapsulating data per instance}
@end float

This would seem to accomplish each of our above goals. Our implementation does
not break encapsulation, as nobody can get at the data. Our methods are part of
the @var{Stack} prototype, so we are not redefining it with each instance,
eliminating our memory and processing issues. Finally, @var{Stack} instances can
be instantiated and used just like any other object in JavaScript; the developer
needn't adhere to any obscure standards in order to emulate encapsulation.

Excellent! However, our implementation does introduce a number of issues that we
hadn't previously considered:

@itemize
@item
Our implementation is hardly concise. Working with our ``private'' properties
requires that we add ugly instance lookup code@footnote{We could encapsulate
this lookup code, but we would then have the overhead of an additional method
call with very little benefit; we cannot do something like: @samp{this.stack}.},
obscuring the actual domain logic.
@item
Most importantly: @emph{this implementation introduces memory leaks}.
@end itemize

What do we mean by ``memory leaks''? Consider the usage example in
@ref{f:js-encapsulate-instance}. What happens when were are done using
@var{stack1} and @var{stack2} and they fall out of scope? They will be GC'd.
However, take a look at our @var{idata} variable. The garbage collector will not
know to free up the data for our particular instance. Indeed, it cannot, because
we are still holding a reference to that data as a member of the @var{idata}
array.

Now imagine that we have a long-running piece of software that makes heavy use
of @var{Stack}. This software will use thousands of instances throughout its
life, but they are used only briefly and then discarded. Let us also imagine
that the stacks are very large, perhaps holding hundreds of elements, and that
we do not necessarily @code{pop()} every element off of the stack before we
discard it.

Imagine that we examine the memory usage throughout the life of this software.
Each time a stack is used, additional memory will be allocated. Each time we
@code{push()} an element onto the stack, additional memory is allocated for that
element. Because our @var{idata} structure is not freed when the @var{Stack}
instance goes out of scope, we will see the memory continue to rise. The memory
would not drop until @var{Stack} itself falls out of scope, which may not be
until the user navigates away from the page.

From our perspective, this is not a memory leak. Our implementation is working
exactly as it was developer. However, to the user of our stack implementation,
this memory management is out of their control. From their perspective, this is
indeed a memory leak that could have terrible consequences on their software.

This method of storing instance data was ease.js's initial ``proof-of-concept''
implementation (@pxref{Class Storage}). Clearly, this was not going to work;
some changes to this implementation were needed.

@anchor{Instance Memory Considerations}
@subsubsection Instance Memory Considerations
JavaScript does not provide destructors to let us know when an instance is about
to be GC'd, so we unfortunately cannot know when to free instance data from
memory in @ref{f:js-encapsulate-instance}. We are also not provided with an API
that can return the reference count for a given object. We could provide a
method that the user could call when they were done with the object, but that is
not natural to a JavaScript developer and they could easily forget to call the
method.

As such, it seems that the only solution for this rather large issue is to store
instance data on the instance itself so that it will be freed with the instance
when it is garbage collected (remember, we decided that privileged members were
not an option in the discussion of @ref{f:js-privileged-members}). Hold on - we
already did that in @ref{f:js-proto-inst-noencapsulate}; that caused our data to
be available publicly. How do we approach this situation?

If we are adding data to an instance itself, there is no way to prevent it from
being accessed in some manner, making true encapsulation impossible. The only
options are to obscure it as best as possible, to make it too difficult to
access in any sane implementation. For example:

@itemize
@item
The property storing the private data could be made non-enumerable, requiring
the use of a debugger or looking at the source code to determine the object
name.
  @itemize
  @item This would work only with ECMAScript 5 and later environments.
  @end itemize
@item
We could store all private data in an obscure property name, such as
@var{___$$priv$$___}, which would make it clear that it should not be accessed.
  @itemize
  @item
  We could take that a step further and randomize the name, making it very
  difficult to discover at runtime, especially if it were
  non-enumerable@footnote{Note that ease.js does not currently randomize its
  visibility object name.}.
  @end itemize
@end itemize

Regardless, it is clear that our data will only be ``encapsulated'' in the sense
that it will not be available conveniently via a public API. Let's take a look
at how something like that may work:

@float Figure, f:js-obscure-private
@verbatim
var Stack = ( function()
{
    // implementation of getSomeRandomName() is left up to the reader
    var _privname = getSomeRandomName();

    var S = function()
    {
        // define a non-enumerable property to store our private data (will only
        // work in ES5+ environments)
        Object.defineProperty( this, _privname, {
            enumerable:   false,
            writable:     false,
            configurable: false,

            value: {
                stack: []
            }
        } );
    };

    S.prototype = {
        push: function( val )
        {
            this[ _privname ].stack.push( val );
        },

        pop: function()
        {
            return this[ _privname ].stack.pop();
        },
    };

    return S;
} );

var inst = new Stack();
inst.push( 'foo' );
inst.pop(); // foo
@end verbatim
@caption{Using a random, non-enumerable property name to store private members}
@end float

Now we are really starting to hack around what JavaScript provides for us. We
seem to be combining the encapsulation issues presented in
@ref{f:js-proto-inst-noencapsulate} and the obscurity demonstrated in
@ref{f:js-encapsulate-instance}. In addition, we our implementation depends on
ECMAScript 5 (ideally, we would detect that and fall back to normal, enumerable
properties in pre-ES5 environments, which ease.js does indeed do). This seems to
be a case of encapsulation through obscurity@footnote{A play on ``security
through obscurity''.}. While our implementation certainly makes it difficult to
get at the private member data, it is also very obscure and inconvenient to work
with. Who wants to write Object-Oriented code like that?

@subsubsection Other Considerations
We have conveniently omitted a number of other important factors in our
discussion thus far. Before continuing, they deserve some mention and careful
consideration.

How would we implement private methods? We could add them to our private member
object, just as we defined @var{stack} in @ref{f:js-obscure-private}, but that
would cause it to be redefined with each instance, raising the same issues that
were discussed with @ref{f:js-privileged-members}. Therefore, we would have to
define them in a separate ``prototype'', if you will, that only we have access
to:

@float Figure, f:js-obscure-private-methods
@verbatim
var Stack = ( function()
{
    // implementation of getSomeRandomName() is left up to the reader
    var _privname = getSomeRandomName();

    var S = function()
    {
        // define a non-enumerable property to store our private data (will only
        // work in ES5+ environments)
        Object.defineProperty( this, _privname, {
            // ... (see previous example)
        } );
    };

    // private methods that only we will have access to
    var priv_methods = {
        getStack: function()
        {
            // note that, in order for 'this' to be bound to our instance, it
            // must be passed as first argument to call() or apply()
            return this[ _privname ].stack;
        },
    };

    // public methods
    S.prototype = {
        push: function( val )
        {
            var stack = priv_methods.getStack.call( this );
            stack.push( val );
        },

        pop: function()
        {
            var stack = priv_methods.getStack.call( this );
            return stack.pop();
        },
    };

    return S;
} );

var inst = new Stack();
inst.push( 'foo' );
inst.pop(); // foo
@end verbatim
@caption{A possible private method implementation}
@end float

While this does solve our problem, it further reduces code clarity. The
implementation in @ref{f:js-obscure-private-methods} is certainly a far cry from
something like @samp{this._getStack()}, which is all you would need to do in
ease.js.

Another consideration is a protected (@pxref{Access Modifiers}) member
implementation, the idea being that subtypes should inherit both public and
protected members. Inheritance is not something that we had to worry about with
private members, so this adds an entirely new layer of complexity to the
implementation. This would mean somehow making a protected prototype available
to subtypes through the public prototype. Given our implementation in the
previous figures, this would likely mean an awkward call that somewhat
resembles: @samp{this[ _protname ].name}.

Although the implementations show in @ref{f:js-obscure-private} and
@ref{f:js-obscure-private-methods} represent a creative hack, this is precisely
one of the reasons ease.js was created - to encapsulate such atrocities that
would make code that is difficult to use, hard to maintain and easy to introduce
bugs. One shouldn't have to have a deep understanding of JavaScript's prototype
model in order to write the most elementary of Classical Object-Oriented code.
For example, the constructors in the aforementioned figures directly set up an
object in which to store private members. ease.js will do this for you before
calling the @code{__construct()} method. Furthermore, ease.js does not require
referencing that object directly, like we must do in our methods in
@ref{f:js-obscure-private}. Nor does ease.js have an awkward syntax for invoking
private methods. We will explore how this is handled in the following section.


@node The Visibility Object
@subsection The Visibility Object
Let's consider how we may rewrite @var{Stack} in
@ref{f:js-obscure-private-methods} using ease.js:

@float Figure, f:stack-using-easejs
@verbatim
var Stack = Class( 'Stack',
{
    'private _data': [],

    'public push': function( val )
    {
        this._data.push( val );
    },

    'public pop': function()
    {
        return this._data.pop();
    }
} );

var inst = Stack();
inst.push( 'foo' );
inst.pop(); // foo
@end verbatim
@caption{Stack implementation using ease.js}
@end float

The above implementation is much less impressive looking than our prior
examples. What we have done is encapsulate the excess logic needed to emulate a
class and got right down to business. ease.js will take the class definition
above and generate an object much like we had done in the prior examples, with a
few improvements.

If you have not read over the previous sections, you are recommended to do so
before continuing in order to better understand the rationale and finer
implementation details.

The secret behind ease.js's visibility implementation (@pxref{Access
Modifiers}) is referred to internally as the @dfn{visibility object} (or, in
older commits and some notes, the @dfn{property object}). Consider the problem
regarding the verbosity of our private property accessors and method calls in
@ref{f:js-obscure-private-methods}. It would be much more convenient if the
properties and methods were bound to @var{this} so that they can be accessed
more naturally, as would be expected by a programmer familiar with classes in
other Classical Object-Oriented languages (@pxref{f:stack-using-easejs}). This
can be done using @code{call()} or @code{apply()}:

@float Figure, f:function-context
@verbatim
function getName()
{
    return this.name;
}

var obj = { name: "foo" };
getName.call( obj ); // "foo"
@end verbatim
@caption{Calling a function within the context of a given object}
@end float

@ref{f:function-context} demonstrates the concept we are referring to. Given an
arbitrary object @var{obj}, we can call any given method (in this case,
@code{getName()}, binding @var{this} to that object. This is precisely what
ease.js does with each method call. To understand this process, we have to
explore two concepts: the visibility object itself and method wrapping. We will
start by discussing the visibility object in more detail and cover method
wrapping later on (@pxref{Method Wrapping}).

@menu
* Visibility Object Implementation::  Design of the visibility object
* Property Proxies::                  Overcoming prototype limitations
@end menu

@node Visibility Object Implementation
@subsubsection Visibility Object Implementation
The visibility object is mostly simply represented in the following diagram:

@float Figure, f:visobj
@image{img/visobj}
@caption{Structure of the visibility object}
@end float

Specifically, the visibility object is a prototype chain containing the private
members of the class associated with the method currently being invoked on the
current instance, its protected members (including those inherited from its
supertype) and the public members (also including those inherited from the
supertype). To accomplish this, the visibility object has the following
properties:

@itemize
@item
The private object is @dfn{swappable} - that is, it is the only portion of the
prototype chain that is replaced between calls to various methods.
  @itemize
  @item
  It is for this reason that the private object is placed atop the prototype
  chain. This allows it to be swapped very cheaply by simply passing different
  objects to be bound to @code{this}.
  @end itemize
@item
Both the private and protected objects are initialized during instantiation by
the @code{__initProps()} method attached by @code{ClassBuilder} to each class
during definition.
  @itemize
  @item
  Properties are cloned to ensure references are not shared between instances.
  @item
  Methods are copied by reference, since their implementations are immutable.
  @item
  This must be done because neither protected nor private members may be added
  to the prototype chain of a class.
    @itemize
    @item
    Doing so would effectively make them public.
    @item
    Doing so would also cause private members to be inherited by subtypes.
    @end itemize
  @end itemize
@item
Public members are a part of the class prototype chain as you would expect in
any conventional prototype.
  @itemize
  @item
  Public @emph{properties} only are initialized by @code{__initProps()}, just as
  private and protected properties, to ensure that no references are shared
  between instances.
  @end itemize
@end itemize

As a consequence of the above, it is then clear that there must be a separate
visibility object (prototype chain) @emph{for each supertype of each instance},
because there must be a separate private object for each subtype of each
instance. Let us consider for a moment why this is necessary with the following
sample of code:

@float Figure, f:priv-swap-rationale
@verbatim
var C1 = Class(
    {
        'private _name': 'Foo',

        'public getName': function()
        {
            return this._name;
        },

        // ...
    } ),

    // note the naming convention using descending ids for the discussion
    // following this example
    C0 = C1.extend(
    {
        // ...
    } );

C1().getName(); // "Foo"
C0().getName(); // "Foo"
@end verbatim
@caption{Why private member swapping is necessary}
@end float

@ref{f:priv-swap-rationale} demonstrates why the private object
swapping@footnote{The term ``swapping'' can be a bit deceptive. While we are
swapping in the sense that we are passing an entirely new private object as the
context to a method, we are @emph{not} removing an object from the prototype chain and
adding another in place of it. They @emph{do}, however, share the same prototype
chain.} is indeed necessary. If a subtype does @emph{not} override a super
method that uses a private member, it is important that the private member be
accessible to the method when it is called. In @ref{f:priv-swap-rationale}, if
we did not swap out the object, @var{_name} would be undefined when invoked on
@var{C2}.

Given this new information, the implementation would more formally be
represented as a collection of objects @var{V} for class @var{C} and each of its
supertypes as denoted by @var{C\_n}, with @var{C\_0} representing the class
having been instantiated and any integer @var{n} >= 0 representing the closest
supertype, such that each @var{V\_n} is associated with @var{C\_n},
@var{V\_n\^x} is the visibility object bound to any method associated with class
@var{C\_x} and each @var{V} shares the same prototype chain @var{P\_n} for any
given instance of @var{C\_n}:

@float Figure, f:visobj-collection
@image{img/visobj-collection-wide}
@caption{Collection of visibility objects @var{V} for each class @var{C}}
@end float

Fortunately, as shown in @ref{f:visobj-collection}, the majority of the
prototype chain can be reused under various circumstances:

@itemize
@item
For each instance of class @var{C\_n}, @var{P\_n} is re-used as the
prototype of every @var{V\_n}.
@item
@var{C\_n} is re-used as the prototype for each @var{P\_n}.
@end itemize

Consequently, on instantiation of class @var{C\_n}, we incur a performance hit
from @code{__initProps()} for the initialization of each member of @var{V\_x}
and @var{P\_x}, as well as each property of @var{C\_x}, recursively for each
value of @var{m} >= @var{x} >= @var{n} (that is, supertypes are initialized
first), where @var{m} is equal to the number of supertypes of class @var{C\_n} +
@var{n}.@footnote{There is room for optimization in this implementation, which
will be left for future versions of ease.js.}

The instance stores a reference to each of the visibility objects @var{V},
indexed by an internal class identifier (which is simply incremented for each
new class definition, much like we did with the instance id in
@ref{f:js-encapsulate-instance}. When a method is called, the visibility object
that matches the class identifier associated with the invoked method is then
passed as the context (bound to @var{this}) for that method (@pxref{Method
Wrapping}).

@node Property Proxies
@subsubsection Property Proxies
Astute readers may notice that the visibility implementation described in the
previous section (@pxref{Visibility Object Implementation}) has one critical
flaw stemming from how prototypes in JavaScript are implemented: setting a
property on the visibility object bound to the method will set the property on
that object, but @emph{not necessarily on its correct object}. The following
example will demonstrate this issue:

@float Figure, f:proto-set-issue
@verbatim
var pub = {
        foo:    'bar',
        method: function()
        {
            return 'baz';
        },
    },

    // what will become our visibility object
    priv = function() {}
;

// set up our visibility object's prototype chain (we're leaving the protected
// layer out of the equation)
priv.prototype = pub;

// create our visibility object
var vis = new priv();

// retrieving properties works fine, as do method invocations
vis.foo;       // "bar"
vis.method();  // "baz"

// but when it comes to setting values...
vis.foo = 'newval';

// ...we stop short
vis.foo;  // "newval"
pub.foo;  // "bar"

vis.foo = undefined;
vis.foo;  // undefined
delete vis.foo;
vis.foo;  // "bar"
pub.foo;  // "bar"

pub.foo = 'moo';
vis.foo; // "moo"
@end verbatim
@caption{Demonstrating property set issues with prototype chains}
@end float

Retrieving property values and invoking methods are not a problem. This is
because values further down the prototype chain peek through ``holes'' in
objects further up the chain. Since @var{vis} in @ref{f:proto-set-issue} has no
value for property @var{foo} (note that a value of @code{undefined} is still a
value), it looks at its prototype, @var{pub}, and finds the value there.

However, the story changes a bit when we try to set a value. When we assign a
value to member @var{foo} of @var{vis}, we are in fact setting the property on
@var{vis} itself, @emph{not} @var{pub}. This fills that aforementioned ``hole'',
masking the value further down the prototype chain (our value in @var{pub}).
This has the terrible consequence that if we were to set a public/protected
property value from within a method, it would only be accessible from within
that instance, for @emph{only that visibility object}.

To summarize:

@itemize
@item
Methods are never an issue, as they are immutable (in the sense of a class).
@item
Reading properties are never an issue; they properly ``peek'' through holes in
the prototype chain.
@item
Writing private values are never an issue, as they will be properly set on that
visibility object. The value needn't be set on any other visibility objects,
since private values are to remain exclusive to that instance within the context
of that class only (it should not be available to methods of supertypes).
@item
We run into issues when @emph{setting} public or protected values, as they are
not set on their appropriate object.
@end itemize

This issue is huge. Before ECMAScript 5, it may have been a show-stopper,
preventing us from using a familiar @code{this.prop} syntax within classes and
making the framework more of a mess than an elegant implementation. It is also
likely that this is the reason that frameworks like ease.js did not yet exist;
ECMAScript 5 and browsers that actually implement it are still relatively new.

Fortunately, ECMAScript 5 provides support for getters and setters. Using these,
we can create a proxy from our visibility object to the appropriate members of
the other layers (protected, public). Let us demonstrate this by building off of
@ref{f:proto-set-issue}:

@float Figure, f:proto-getset
@verbatim
// proxy vis.foo to pub.foo using getters/setters
Object.defineProperty( vis, 'foo', {
    set: function( val )
    {
        pub.foo = val;
    },

    get: function()
    {
        return pub.foo;
    },
} );

vis.foo; // "moo"
pub.foo; // "moo"

vis.foo = "bar";
vis.foo; // "bar"
pub.foo; // "bar"

pub.foo = "changed";
vis.foo; // "changed"
@end verbatim
@caption{Using getters/setters to proxy values to the appropriate object}
@end float

The implementation in @ref{f:proto-getset} is precisely how ease.js implements
and @emph{enforces} the various levels of visibility.@footnote{One may wonder
why we implemented a getter in @ref{f:proto-getset} when we had no trouble
retrieving the value to begin with. In defining a @emph{setter} for @var{foo} on
object @var{vis}, we filled that ``hole'', preventing us from ``seeing through''
into the prototype (@var{pub}). Unfortunately, that means that we must use a
getter in order to provide the illusion of the ``hole''.} This is both fortunate
and unfortunate; the project had been saved by getters/setters, but with a
slight performance penalty. In order to implement this proxy, the following must
be done:

@itemize
@item
For each public property, proxy from the protected object to the public.
@item
For each protected property, proxy from the private object to the
protected.@footnote{One may also notice that we are not proxying public
properties from the private member object to the public object. The reason for
this is that getters/setters, being functions, @emph{are} properly invoked when
nestled within the prototype chain. The reader may then question why ease.js did
not simply convert each property to a getter/setter, which would prevent the
need for proxying. The reason for this was performance - with the current
implementation, there is only a penalty for accessing public members from within
an instance, for example. However, accessing public members outside of the class
is as fast as normal property access. By converting all properties to
getters/setters, we would cause a performance hit across the board, which is
unnecessary.}
@end itemize

Consequently, this means that accessing public properties from within the class
will be slower than accessing the property outside of the class. Furthermore,
accessing a protected property will @emph{always} incur a performance
hit@footnote{How much of a performance hit are we talking? This will depend on
environment. In the case of v8 (Node.js is used to run the performance tests
currently), getters/setters are not yet optimized (converted to machine code),
so they are considerably more slow than direct property access.

For example: on one system using v8, reading public properties externally took
only 0.0000000060s (direct access), whereas accessing the same property
internally took 0.0000001120s (through the proxy), which is a significant
(18.6x) slow-down. Run that test 500,000 times, as the performance test does,
and we're looking at 0.005s for direct access vs 0.056s for proxy access.},
because it is always hidden behind the provide object and it cannot be accessed
from outside of the class.  On the upside, accessing private members is fast (as
in - ``normal'' speed). This has the benefit of encouraging proper OO practices
by discouraging the use of public and protected properties. Note that methods,
as they are not proxied, do not incur the same performance hit.

Given the above implementation details, it is clear that ease.js has been
optimized for the most common use case, indicative of proper OO development -
the access of private properties from within classes, for which there will be no
performance penalty.

@node Method Wrapping
@subsection Method Wrapping
The visibility object (@pxref{The Visibility Object}) is a useful tool for
organizing the various members, but we still need some means of binding it to a
method call. This is accomplished by wrapping each method in a closure that,
among other things@footnote{The closure also sets the @code{__super()} method
reference, if a super method exists, and returns the instance if @var{this} is
returned from the method.}, uses @code{apply()} to forward the arguments to the
method, binding @var{this} to the appropriate visibility object. This is very
similar to the ES5 @code{Function.bind()} call.

The following example demonstrates in an overly-simplistic way how ease.js
handles class definitions and method wrapping.@footnote{ease.js, of course,
generates its own visibility objects internally. However, for the sake of
brevity, we simply provide one in our example.}

@float Figure, f:method-wrapping
@verbatim
/**
 * Simple function that returns a prototype ("class"), generated from the given
 * definition and all methods bound to the provided visibility object
 */
function createClass( vis, dfn )
{
    var C      = function() {},
        hasOwn = Object.hasOwnProperty;

    for ( name in dfn )
    {
        // ignore any members that are not part of our object (further down the
        // chain)
        if ( hasOwn.call( dfn, name ) === false )
        {
            continue;
        }

        // simply property impl (WARNING: copies by ref)
        if ( typeof dfn[ name ] !== 'function' )
        {
            C.prototype[ name ] = dfn[ name ];
            continue;
        }

        // enclose name in a closure to preserve it (otherwise it'll contain the
        // name of the last member in the loop)
        C.prototype[ name ] = ( function( mname )
        {
            return function()
            {
                // call method with the given argments, bound to the given
                // visibility object
                dfn[ mname ].apply( vis, arguments );
            };
        } )( name );
    }

    return C;
};

var vis = { _data: "foo" },

    Foo = createClass( vis,
    {
        getData: function()
        {
            return this._data;
        },
    } );

var inst = new Foo();

// getData() will be bound to vis and should return its _data property
inst.getData(); // "foo"
@end verbatim
@caption{Basic "class" implementation with method binding}
@end float

There are some important considerations with the implementation in
@ref{f:method-wrapping}, as well as ease.js's implementation:

@itemize
@item
Each method call, unless optimized away by the engine, is equivalent to two
function invocations, which cuts down on the available stack space.
  @itemize
  @item
  The method wrapping may complicate tail call optimization, depending on the
  JavaScript engine's implementation and whether or not it will optimize across
  the stack, rather than just a single-depth recursive call.
  @item
  As such, for operations that are highly dependent on stack space, one may wish
  to avoid method calls and call functions directly.
  @end itemize
@item
There is a very slight performance hit (though worrying about this is likely to
be a micro-optimization in the majority of circumstances).
@end itemize

As mentioned previously, each visibility object is indexed by class identifier
(@pxref{Visibility Object Implementation}). The appropriate visibility object is
bound dynamically on method invocation based on the matching class identifier.
Previously in this discussion, it was not clear how this identifier was
determined at runtime. Since methods are shared by reference between subtypes,
we cannot store a class identifier on the function itself.

The closure that wraps the actual method references the arguments that were
passed to the function that created it when the class was defined. Among these
arguments are the class identifier and a lookup method used to determine the
appropriate visibility object to use for binding.@footnote{See
@file{lib/MethodWrappers.js} for the method wrappers and
@code{ClassBuilder.getMethodInstance()} for the lookup function.} Therefore, the
wrapper closure will always know the appropriate class identifier. The lookup
method is also passed @var{this}, which is bound to the instance automatically
by JavaScript for the method call. It is on this object that the visibility
objects are stored (non-enumerable; @pxref{Instance Memory Considerations}),
indexed by class identifier. The appropriate is simply returned.

If no visibility object is found, @code{null} is returned by the lookup
function, which causes the wrapper function to default to @var{this} as
determined by JavaScript, which will be the instance that the method was invoked
on, or whatever was bound to the function via a call to @code{call()} or
@code{apply()}. This means that, currently, a visibility object can be
explicitly specified for any method by invoking the method in the form of:
@samp{inst.methodName.apply( visobj, arguments )}, which is consistent with how
JavaScript is commonly used with other prototypes. However, it should be noted
that this behavior is undocumented and subject to change in future releases
unless it is decided that this implementation is ideal. It is therefore
recommended to avoid using this functionality for the time being.@footnote{One
one hand, keeping this feature is excellent in the sense that it is predictable.
If all other prototypes work this way, why not ``classes'' as created through
ease.js? At the same time, this is not very class-like. It permits manipulating
the internal state of the class, which is supposed to be encapsulated. It also
allows bypassing constructor logic and replacing methods at runtime. This is
useful for mocking, but a complete anti-pattern in terms of Classical
Object-Oriented development.}


@node Internal Methods/Objects
@section Internal Methods/Objects
There are a number of internal methods/objects that may be useful to developers
who are looking to use some features of ease.js without using the full class
system. An API will be provided to many of these in the future, once refactoring
is complete. Until that time, it is not recommended that you rely on any of the
functionality that is not provided via the public API (@code{index.js} or the
global @var{easejs} object).