Tony Marston's Blog About software development, PHP and OOP

Not the three greatest paragraphs ever written on encapsulation

Posted on 19th May 2013 by Tony Marston

Amended on 21st February 2015

Introduction
Chinese Whispers
Basic definitions
Encapsulation or Abstraction?
A proper definition of Encapsulation
Incorrect definitions
Separation of Concerns
Information Hiding
Ripple Effect
Correct definitions
Inheritance
Polymorphism
Conclusion
References
Amendment History
Comments

Introduction

This article is in response to The three greatest paragraphs ever written on encapsulation which I feel is a total failure. As far as I am concerned when you are supplying a description for one of the fundamental principles of Object Oriented Programming (OOP) you should do so in a way that can be used by a newcomer to OOP so that they may obtain a proper understanding of what it means and how it can be implemented. None of the paragraphs in that article meet those goals, which is why I regard them as "not fit for purpose".

I have been in software development for over 30 years, and in OOP for the last 10, and I have read volumes and volumes of articles which are supposed to be teaching aids written by gurus, but which, in my humble opinion, are nothing more than misguided ideas written by those who have lost the plot. I have debunked some of these erroneous definitions in What OOP is Not.


Chinese Whispers

The main problem, as I see it, is because of the richness of the English language:

The original idea is expressed using a particular combination of words, but a second person comes along and tries to express the same idea using different words, perhaps to prove how clever or intellectually superior he is. However, this second definition uses different words which themselves may have alternative meanings, and this causes the original idea to begin mutating into something else. Then a third person comes along and tries to express the second definition using different words which hang off one of these alternate meanings. And so it goes on. Different people try to express the same idea using different words, but as these words may have alternative meanings which are different from the one used by the preceding author, the original meaning can get corrupted or even lost altogether. This is exactly what happens in the children's game called Chinese Whispers where the original message "Send reinforcements, we're going to advance" gets changed into "Send three and four pence, we're going to a dance".


Basic definitions

First let me start with some basic definitions which I used as the basis for my migration from procedural to object oriented programming:

Object Oriented Programming Writing programs which are oriented around objects. Such programs can take advantage of Encapsulation, Polymorphism, and Inheritance to increase code reuse and decrease code maintenance.

Note that this excludes the use of static methods as they are accessed without the use of objects. If you are not using objects then how can it be called "object oriented"?

Note that the effectiveness of your implementation can be measured by the amount of reusable code that you produce. The more reusable code you have at your disposal then the less code you need to write to get the job done, the less time it will take and the more productive you will be.

Object An instance of a class. A class must be instantiated into an object before it can be used in the software. More than one instance of the same class can be in existence at any one time.
Class A class is a blueprint, or prototype, that defines the variables and the methods common to all objects (entities) of a certain kind.
Abstraction The process of separating the abstract from the concrete, the general from the specific, by examining a group of objects looking for both similarities and differences. The similarities can be shared by all members of that group while the differences are unique to individual members.

There are two flavours of abstraction:

Encapsulation The act of placing data and the operations that perform on that data in the same class. The class then becomes the 'capsule' or container for the data and operations. This binds together the data and the functions that manipulate the data.

More details can be found in Object-Oriented Programming for Heretics

Inheritance The reuse of base classes (superclasses) to form derived classes (subclasses). Methods and properties defined in the superclass are automatically shared by any subclass. A subclass may override any of the methods in the superclass, or may introduce new methods of its own.

More details can be found in Object-Oriented Programming for Heretics

Polymorphism Same interface, different implementation. The ability to substitute one class for another. This means that different classes may contain the same method signature, but the result which is returned by calling that method on a different object will be different as the code behind that method (the implementation) is different in each object.

More details can be found in Object-Oriented Programming for Heretics

A lot of my critics tell me that these definitions are too simple, that they have evolved into something more complicated, more "pure". Rather than my definitions being too simple I regard their definitions as being too complex.


Encapsulation or Abstraction?

Classes which can be instantiated into objects are the mainstay of Object Oriented Programming. The process by which external entities are converted into software classes is called encapsulation, but some people like to use the term abstraction instead. This is incorrect as they do not mean the same thing. Unfortunately the term "abstraction" identifies both a verb (process) and a noun (entity) in which you perform the process to produce a result which is a noun.

Abstraction is the process of separating the abstract from the concrete, the general from the specific, by examining a group of objects looking for both similarities and differences. The similarities can be shared by all members of that group while the differences are unique to individual members. The result of this process should then be an abstract superclass containing the shared characteristics and a separate concrete subclass to contain the differences for each unique instance. There are two flavours of abstration:


A proper definition of Encapsulation

The art of abstraction is supposed to reduce an idea or concept to its bare essentials and leave out any inessential details. As you can see from my Basic Definitions above I have reduced the concepts of OOP to a series of simple statements:

Encapsulation can therefore be defined as follows:

The act of placing an entity's data and the operations that perform on that data in the same class.

Note that this simple statement also implies the following:

In a typical software application not all classes may actually represent real-world objects or entities. In some cases they may be used to encapsulate or isolate certain functionality or processing which is performed on application data. In this article How to write testable code the author identifies three main object categories:

This is also discussed in When to inject: the distinction between newables and injectables.

The RADICORE framework is a combination of the Model-View-Controller design pattern and the 3-Tier Architecture. The MVC design pattern contains the following components:

The 3-Tier Architecture contains the following layers:

When combined these two produce the structure shown in Figure 1:

Figure 1 - MVC plus 3 Tier Architecture

model-view-controller-03a (5K)

The components in the RADICORE framework fall into the following categories:

Note that all domain knowledge is kept entirely with the Model classes for each domain/subsystem, which means that the Controllers, Views and DAOs are completely domain-agnostic. This means that they are not tied to any particular domain and can therefore be used with any domain.

With this structure it should be possible to have a large number of Model components, but a smaller collection of reusable Controllers, Views and DAOs which are capable of working with any Model in the application.

One thing you should not attempt to do is to define a single monolithic class to do everything within an entire application. This is sometimes referred to as a "God" class as it tries to be omnipotent, all-seeing and all-knowing. Just as you cannot put all the data you need in a single database table you cannot put all the code to process that data into a single software component. Just as data in a database is broken down into logical units according to the rules of normalisation, the program code should also be broken down into logical units or modules. In OOP these modules are usually constructed as classes. In this way you should be able to modify existing modules or add new modules without having a ripple effect on any existing modules. This also allows different programmers to work on different modules at the same time.

As well as trying not to create classes which are too big one should also avoid going too far in the opposite direction and create a huge number of small classes. A code base which has been abstracted into oblivion is as difficult to follow as a colony of ants. If you don't believe me then try working with a library that uses 100 classes to send a single email! That is what's known as Object Oriented Overkill.


Incorrect definitions

Now that I have provided you with the definition of encapsulation which I use, let me now go over the definitions in the article The three greatest paragraphs ever written on encapsulation which I consider to be incorrect. Firstly the author provides his own definition of encapsulation as:

Encapsulation: the property that the information contained in an object is accessible only through interactions at the interfaces supported by the object.

Encapsulation is not a property, it is a process which results in the creation of a class. For each real world object which needs to be modelled in your software you create a separate class. That class will contain all the object's data in the form of class properties, and all the operations which can be performed on that data in the form of class methods. These simple facts are missing from that definition, which makes that definition incomplete and therefore invalid.

That definition also tries to enforce an idea which was not covered by the original definition, and that is the concept of information hiding which is discussed below.

There now follows a critique of the three paragraphs:

Separation of Concerns

In his article the author does not actually supply a definition of Separation of Concerns (SoC) (also known as the Single Responsibility Principle or SRP). Instead he points to E. W. Dijkstra's article On the role of scientific thought:

We know that a program must be correct and we can study it from that viewpoint only; we also know that it should be efficient and we can study its efficiency on another day, so to speak. In another mood we may ask ourselves whether, and if so: why, the program is desirable. But nothing is gained - on the contrary! - by tackling these various aspects simultaneously. It is what I sometimes have called "the separation of concerns", which, even if not perfectly possible, is yet the only available technique for effective ordering of one's thoughts, that I know of.

This is not much of a definition as it only identifies the concept, but does not describe how to implement it. If you bother to read that article you will see that the author is stating that a program's correctness, efficiency and desirability should be examined separately and not together. This has absolutely nothing to do with encapsulation! Encapsulation is the act of creating a class while Separation of Concerns involves taking a monolithic "God" class (a single class that does everything) and splitting it into smaller yet coherent units. But how do you determine which parts of this "God" class can be split into smaller units? How do you now when to stop this splitting process? Robert C. Martin (Uncle Bob) provides this description in his article Test Induced Design Damage?:

How do you separate concerns? You separate behaviors that change at different times for different reasons. Things that change together you keep together. Things that change apart you keep apart.

GUIs change at a very different rate, and for very different reasons, than business rules. Database schemas change for very different reasons, and at very different rates than business rules. Keeping these concerns (GUI, business rules, database) separate is good design.

In a (slightly) later article called The Single Responsibility Principle he wrote the following:

This is the reason we do not put SQL in JSPs. This is the reason we do not generate HTML in the modules that compute results. This is the reason that business rules should not know the database schema. This is the reason we separate concerns.

It should be noted that both SoC and SRP are based on the principle of cohesion. Program code can either exhibit high cohesion or low cohesion where high cohesion is considered to be better. High cohesion can be said to have two faces:

This is why, for example, putting all the code which generates HTML into a single object is considered to be a good idea, as is putting all the code which generates SQL into a single object. This then leaves the remaining code, which deals with data validation and business rules, in its own object.

The idea of breaking an application down into three separate layers, where each layer contains objects which have a single responsibility or concern, has also been discussed by Martin Fowler in his article Presentation-Domain-Data Layering.

All of these description match the separation provided by the 3-Tier Architecture which has the following layers or tiers:

With this architecture it is possible to change the components in one layer without having to make any corresponding changes to the other layers.

In the RADICORE framework the 3-Tier Architecture has been merged with the Model-View-Controller design pattern to provide a structure such as that shown in Figure 1. This means that a program's logic is split into separate modules which exist as one of the following:

In the RADICORE framework each of these components is implemented as follows:

Only the Model classes need to generated by the application developer. All the others are supplied in the framework.

So now instead of a single object which tries to do everything we have a collection of objects, each of which is responsible for, or concerned with, a different part of the processing. The user sends in a request which is received by the Controller. The controller calls one or more methods on the Model(s) in order to process that request. The Models may or may not communicate with a database or some other data source. The Model's response to that request is then given to the View so that it can be converted to the desired format before being returned to the user.

When implemented correctly this "separation of concerns" or "separation of responsibilities" not only reduces the amount of duplicated code but also allows for new Controllers, Views and DAOs to be created with little or no impact on the Model components. It should also allow for new Model classes to be created without the need to amend any Controllers, Views or DAOs.

Too much or too little separation

The problem with this concept called "separation of concerns" is deciding how far you should go. If you don't go far enough you end with a compound object which deals with several entities, each of which should actually be handled by a separate class. Take the structure shown in Figure 2 which identifies the separate tables used to hold the data for a sales order:

Figure 2 - an "order" object

order-object (2K)

Far too many OO programmers are taught the idea that a sales order can be regarded as a single concept which may have a number of constituent parts, and that it should therefore have a single class/object in the software to deal with that single concept. This then becomes a compound object which handles multiple objects which exist in the outside world. However, when writing an application which has HTML forms at the front end and a relational database at the back end it is vitally important that the software be designed to handle those two different communication protocols. Anyone who knows anything at all about relational databases will tell you that compound tables, where you have a single "thing" through which you must navigate to get to an individual table, simply do not exist. Each table is a separate entity in its own right and can be accessed using the same CRUD operations as any other table without having to reference any other table. So, if each object in a database is a single table then my software should have its own class for each of those tables. The idea of compound classes, or composite objects as it is known, does not exist in my universe.

By creating a compound class, one which handles multiple database tables, the novice developer is actually breaking the rules and causing problems instead of solving them. Here are some of them:

On the other hand if you go too far with this splitting up of code according to perceived responsibilities instead of genuine responsibilities you end up with ravioli code, a mass of tiny classes which end up by being less readable, less usable, less efficient, less testable and less maintainable. This is like having an ant colony with a huge number of workers where each worker does something different. When you look at this mass of ants, how do you decide who does what? How do you identify the process flow? Where do you look to find the source of a bug, or where to make a change? A prime example of this is a certain open source email library which uses 100 classes, some of which contain single methods with a single line of code. 100 classes? For an email? WTF!!!

The only connection between "encapsulation" and "separation of concerns" is deciding which methods and properties go into which class for each of the separate areas of responsibility. Depending on the type of application you are writing you may identify different responsibilities which require different components, but deciding on this list of responsibilities is a totally separate exercise from encapsulating the methods and properties into suitable classes. The object of the exercise is to create classes which are neither too big (the "compound" class) nor too small (ravioli code).

Information Hiding

The second paragraph introduces us to a concept called information hiding which is taken from the paper On the Criteria To Be Used in Decomposing Systems into Modules written by D.L. Parnas:

The second decomposition was made using 'information hiding" as a criterion.
...
Every module in the second decomposition is characterized by its knowledge of a design decision which it hides from all others. Its interface or definition was chosen to reveal as little as possible about its inner workings.

Notice that I highlighted the words to reveal as little as possible about its inner workings. In my humble opinion the words "inner workings" do not refer to the data which is operated upon but the internal implementation of the module (the operations which are performed on that data). In my experience this has been the way that every subroutine/function in every programming language has ever been documented - it lists a series of APIs which identify nothing but the function name and its input and output arguments (the function signature), plus a brief description of what its does. This description does not include the code which actually implements what it does. This means that the function's internal implementation can be changed at any time without necessarily having to change the function's signature.

Each class has both properties and methods, but it was never intended that properties had to be hidden and could only be accessed via a method instead of directly. If you don't believe me then take a look at the following articles:

This observation is taken from the C2 wiki:

- Encapsulation is a programming language feature.
- Information Hiding is a design principle.

This observation is taken from Nat Pryce's article:

- Encapsulation ensures that the behaviour of an object can only be affected through the object's API.
- Information hiding conceals how an object implements its functionality behind the abstraction of the object's API.

The idea of data hiding has always seemed strange to me. Surely in an application whose sole purpose is to move data between a GUI and a database then hiding that data defeats that purpose?

Ripple Effect

The third paragraph comes from a paper entitled Structured design written for the IBM Systems Journal by W. Stevens, G. Myers and L. Constantine:

The fewer and simpler the connections between modules, the easier it is to understand each module without reference to other modules. Minimizing connections between modules also minimises the paths along which changes and errors can propagate into other parts of the system, thus eliminating disastrous 'Ripple effects' where changes in one part causes errors in another, necessitating additional changes elsewhere, giving rise to new errors, etc

What they are talking about here is called coupling, how modules interact, and the degree of mutual interdependence between modules. If two modules interact they can either be loosely coupled (which is supposed to be good) or tightly coupled (which is supposed to be bad). This is only relevant when you are assembling an application from several modules/classes, and has nothing to do with the construction of an individual module/class which is what encapsulation is all about.

Here is my explanation of "Coupling":

As shown previously in my implementation every Model class (which contains business logic) must have another entity (called a Controller in this document) which can instantiate it into an object so that a method can be called on that object. This is known as "coupling" as it defines how components interact with each other. "A" calls "B" therefore "A" and "B" are joined or coupled by that call. It is also referred to as "dependency" as it shows that the Controller ("A") is dependent on the Model ("B") in order to carry out its assigned task. "A" cannot work without "B", but "B" can be called by components other than "A". In this case "A" is dependent on "B", but "B" is not dependent on "A". How that coupling is implemented in the code decides whether it is "loose" or "tight". Looser coupling is better as it tends to create more reusable methods. Take a look at the following code samples which may be found in a typical controller:

Example 1a:
$object = new Foo;
$object->setField1($_POST['field1']);
$object->setField2($_POST['field2']);
$object->setField3($_POST['field3']);
$result = $object->insert();
Example 1b:
$object = new Foo;
$result = $object->insert($_POST['field1'], $_POST['field2'], $_POST['field3']);
Example 2:
$object = new $classname;
$object->insert($_POST);

Examples 1a and 1b both contain a hard-coded class name which means that it can only work with that particular class, which makes it non-reusable when it comes to other classes. It also has hard-coded property names, which means that if the list of class properties ever changes then both the Controller and the Model will have to be changed at the same time.

Example 2 does not contain a hard-coded class name, so provided that the method name is available in other objects then it can be used with any of those other objects. It also does not contain any hard-coded property names. The $_POST array can contain any number of fields, and the insert() method can accept that array as a single argument instead of a separate method/argument for each field. This means that the number of fields in that array can fluctuate without ever requiring a change to either the controller or the model.

Here is today's question: which of those two examples demonstrates lower coupling and higher reusability? Answers on a postcard to .....


Correct definitions

As well as giving what I consider to be incorrect definitions of encapsulation, the offending article also misses out on providing explanations for those other principles of OOP - inheritance and polymorphism.

Inheritance

Inheritance can be defined as follows:

The reuse of base classes (superclasses) to form derived classes (subclasses). Methods and properties defined in the superclass are automatically shared by any subclass.

What this means in real life is that you can create a "superclass" which contains sharable methods and properties, then create a "subclass" which inherits everything from the "superclass" by using the "extends" keyword, as in the following example:

include 'default.class.inc';
class entity1 extends Default_Table
{
    function __construct ()
    // constructor
    {
        $this->tablename   = 'entity1';
        $this->dbname      = 'foobar';
        
        $this->fieldlist   = array('column1', 'column2', 'column3', 'column4');
        $this->primary_key = array('column1');
        
    } // __construct
    
} // entity1

The real skill here is in identifying what can be a "superclass" and how many "subclasses" can actually share this code through inheritance. I have seen many suggestions and examples, mostly containing little sharable code from lots of small superclasses, but to my mind the most productive approach is as follows:

Here the term "abstract" means a class that does not contain enough detail to be instantiated into a viable object. A "concrete" class adds in those missing details. In my own development framework, which deals with database applications, I have a single abstract table class which contains everything which I may need for accessing an unspecified database table, and hundreds of concrete classes, one for each physical database table.

Polymorphism

Polymorphism can be defined as follows:

Same interface, different implementation. The ability to substitute one class for another. This means that different classes may contain the same method names, but the result which is returned by a particular method will be different as the code behind that method (the implementation) is different in each class.

What this means in real life is that you have the same method name appearing in more than one class, usually but not necessarily through inheritance, so that the piece of code which calls that method (a page controller, for example) can be used with any class that contains that method. By enabling a controller to be used on more than one class you effectively increase the reusability and sharability of that controller. When a Controller is not tightly bound to a single Model you achieve what is known as loose coupling, which is supposed to be a good thing.

In my own development framework I have a series of pre-written page controllers which use the generic methods defined in my abstract table class. This means that I do not have to have a separate set of controllers for each class. This in turn gives me the ability to use any page controller with any concrete class, which then maximises the level of polymorphism and makes the inter-module coupling as loose as it can possibly be. I can add a new table to my database and a new table class to my application without requiring a new controller to communicate with that table class, which also reduces the development times.

I also have a set of pre-written data access classes which handle all communication with a particular database engine (MySQL, PostgreSQL, Oracle and SQL Server). Although they all share the same method names there is no inheritance from an abstract superclass. When a Model wishes to communicate with the database it talks to whatever DAO has been assigned and gives it an instruction which is equivalent to "using these arguments please construct and execute the relevant SQL query and give me the result". The Model does not know which DBMS is being used, and does not contain any code which is specific to a particular DBMS. The underlying DBMS can therefore be switched without making any changes to any Model. A different DBMS can be introduced into the mix simply by creating a new class file for that DBMS, and provided that it uses the same method signatures its integration will be totally seamless and transparent.


Conclusion

All this confusion is the result of people rewording a perfectly good definition for no good reason other than to impress others with their lexicological skills. By using different words which may have different meanings the original definition can mutate into something else entirely. After multiple iterations of this rewording and mutating the eventual definition could reach a point where it bears no relation to or is a complete corruption of the original.

People keep losing sight of the simple truth that Object Oriented Programming is supposed to increase code reuse and decrease code maintenance through the mechanisms of encapsulation, inheritance and polymorphism, yet they consistently fail to provide either simple definitions or simple examples of these principles being put into practice. As far as I am concerned if you are unable to provide a description that a novice can understand and follow, then your own understanding is less than it should be. By attempting to pass this lack of knowledge onto others, especially when it is disguised with clever words and phrases, instead of adding to the pool of universal knowledge you are actually muddying the waters with pseudo-knowledge. This then makes it harder for novices to filter out the good from the bad, the wheat from the chaff, the excellent from the excrement, so instead of producing well-crafted OOP they end up producing steaming piles of POO.

References

The following articles describe aspects of my framework:

The following articles express my heretical views on the topic of OOP:

These are reasons why I consider some ideas to be complete rubbish:

Here are my views on changes to the PHP language and Backwards Compatibility:

The following are responses to criticisms of my methods:

Here are some miscellaneous articles:

Amendment History

21 Feb 2015 Modified Separation of Concerns to include a definition from Robert C. Martin.
20 Jul 2013 Added Correct definitions
22 Jun 2013 Added my explanation of "Separation of Concerns"
Added my explanation of "Coupling"

counter