This is the ‘eye’ of SOLID software design principles.

Definition:

Clients should not be forced to depend on methods that they do not use.

Alternate Definition (s):

The dependency of one class to another one should depend on the smallest possible interface

The idea

Consider two components- which interact with each other. The interaction between the two is defined by the API calls that one makes onto another. The component which makes calls to the other serves as the client.

The methods that the client can call upon the component serves as the contract between the two.This contract should be kept thin. It should have all that the client needs- and nothing more. Keep in mind- what the component exposes to the client is not limited only to the interface explicitly implemented by the component – but also included all the behavior inherited by that component from classes\interfaces up the hierarchy

Example of violation

Lets say I have  domain classes- PurchaseOrder, Appointment , Invoice etc. All domain classes  extend the same base class- DomainClass which contains common behavior.

All my domain class can be published to external entities. So we have a  method publish (). Now if we want to keep this behavior internal to these objects- i.e. only the domain classes will decide when its the time to be published- this method should be declared as protected. If we declare this method as public- all clients would have access to this method- if that was not the intention- we have violated the Interface Segregation Principle.

Another common violation is when we have an interface  say Animal with methods – run (); swim (); fly ()

And Classes – Fish , Horse etc.- all implementing the Animal interface

Now since flying is a behavior not exhibited by all animals- we provide a dummy implementation and throw and Error “NotSupported”. This is a voilation of the principle. You have exposed the method fly () to all clients of the class Horse when they didn’t even need it.

In fact – in a twisted sort of way – the Object class,the Parent of all classes in Java, is a violation of Interface Segregation Principle. All methods in Object are exposed to all clients.

The Benefits

This principle is important because it encourages two very important ingredients of a good software design

High cohesion – Keep all related methods together

Low coupling  – Keep dependence of one another to the bare minimum

Changes to fat interfaces tend to cause a ripple affect to classes who shouldn’t have been affected in the first place.

Credits

Read more about this from the Master , Robert Martin himself.

 

This is one of the most powerful principle of OOP.

The Principle

Simply stated- what this means is- once you are done coding an entity – e.g Class,function etc- you should have no reason to modify it again (…hence closed), but that entity should be available for modification via extensions e.g inheritance or any other smart mechanism you can think on (…hence open)

The only exception – when you would ‘open’ the class for changes is- to fix defects. That’s it

And why do we have this principle?

The rationale behind this s pretty simple. If you don’t touch it you can’t break it. It gives you a better and a solid design- which is resilient to changes and is yet adaptive to ever evolving nature of software

In extreme

The principle in extreme would mean- that once you are done with the class- you will never- ever again modify it (unless of course -you find a defect) – What that would mean is – any new method taht you would like to add should go to a different class- because any such change is not a bug-fix and hence is not a change to be made to this class

Hmmm…Sounds a bit extreme to me

I went and took a look at API for java.lang.String

So if Java team has been following this principle- they wouldn’t have added any new method to this class. Any new feature they would want to provide to this class- they would conjure some other mechanism .

Well- they didn’t do that . I found at least 4 new methods added in 1.5.

If you think about it- even though you have technically open the class for modification- you haven’t increased the chance of any new defects getting reported for the existing methods- though yes- as a whole- this class has become more prone to defects- just because it has newer methods added to it

So what really is a problem?

Software are ever evolving. New features are continuously being added, defects fixed, underlying technologies change.

The question is- how adaptive is your software to all this.

Consider a railway booking system. When it was first launched – the only way to book a ticket was via a terminal at the booking office. Later on – support for online booking is added. Now how does your booking engine adapt to this change? Do you go and start changing code all over to accommodate request coming in from a new medium or do you simply plug in new medium to the existing system ?

One good rule of thumb is

Whenever you see code using if- then-else statements or switch statements- you ‘could’ have stumbled upon a violation of this principle (This is not to mean that any usage of these constructs is a violation !!!)

Related Design patterns

Many design patterns help you confirm to this principle. Decorator, Strategy, Chain of Responsibility.

Caveat

Like I said before- all classes cannot be coded to confirm to the Open Closed principle. You have to pick and choose where would you like to apply these principles. Adding this principles adds complexity- You will want to add to those corner spots which are subject to extensions.

The credits

Dr Meyer- read more about him at wikipedia

http://en.wikipedia.org/wiki/Open/closed_principle

 

Visit to the bank

You walk in to the bank- to deposit cash.

You stand in the line- meet the Teller who does the transaction with you.

Do you really  care whether its Joe or Mary behind the counter?

No. You do not.

Because your contract with the Bank is via the teller (:read interface) not Joe who happens to be there – that day or maybe everyday (:read implementation)

It’s all about the contract

Interface is nothing but a formal contract . All classes which implement that interface- are basically declaring that we will honor this contract and will provide you the service as said by this interface.

The beauty of this paradigm is that – this keeps your two classes blissfully unaware of each other – allowing you to change one without drastically affecting another or at least controlling the affect on another.

Examples

Java API is actually full of very good examples of this.

e.g. JDBC Driver . All JDBC applications use the Driver interface – to get a connection etc.

Because we program against this Driver- we do not care what that implementation is.

The implementation could be an Oracle Driver or a MySQL Driver.

Your application would work the same- for both the databases- as long as you have programmed to the interface- without depending upon the actual implementation.

Another example: Java Collection API List,Set,Map Interfaces.

If your classes take as input one of these interfaces- the caller may send you a different class- ArrayList or LinkedList- and you wont notice a thing. That’s why its strongly encouraged that your input parameters or return types should be an interface rather than a class.

And Spring

The best thing I like about Spring is- it strongly encourages you to program to interfaces by using Config files- which bind implementing classes to each other using interfaces

You code uses only interfaces you refer to implementation only in the config file

How do I know if I did program to interfaces or not ?

Look at the input parameters and return type of your public API.

If you return a class rather than an interface- you are exposing your implementation detail.The caller is not programing to your interface -but to your classes.

If the input parameters of your public API is a class- you are dependent upon the internal implementation of the caller- and not on it’s interfaces.

© 2011 Technology Cafe Suffusion theme by Sayontan Sinha