FII Practic 2016

Programming principles

Chapter I - Object Oriented Programming

Encapsulation

Encapsulation facilitates bundling of data with the method(s) operating on that data.

To exemplify let's consider the following requirement: As an account holder I want to perform withdrawal and deposit of various amounts from and into my account.

To implement the requirement we consider the following:

  • An Account has a Balance
  • A Withdrawal is an operation which subtracts the amount from the Balance
  • A Deposit is an operation which adds the amount to the Balance

The code below should do it:

public class Account
{
    public Money Balance { get; set; }

}

public class Program
{
    static void Main(string[] args)
    {
	var account = new Account();
	// Make a deposit
	account.Balance += 50;
	// Withdraw some cash
	account.Balance -= 100;
    }
}

Houston, we have a problem. If we let the consumers of our code (in this case Program) to change the Balance at will, our system may end up in such funky situations where account holders can withdraw as much money as they want from the ATM without having enough funds.

Of course, we need to validate each withdrawal but we can't surround each account.Balance -= amount with an if(account.Balance > amount) because it's time consuming, inefficient and we simply may not have access to the code that uses our class. What we can do is to encapsulate the withdrawal and validation into a single operation like this:

public class Account
{
    public Money Balance {get; private set; } // The balance is now read-only

    public Money Withdraw(Money amount)
    {
	if( Balance <= amount)
	{
	    throw new InvalidOperationException("Insufficient funds.");
	}
	this.Balance -= amount;
	return amount;
    }

    public void Deposit(Money money)
    {
	this.Balance += money;
    }
}

public class Program
{
    static void Main(string[] args)
    {
	var account = new Account();
	account.Deposit(50);
	account.Withdraw(100); // => InvalidOperationException.
    }
}

Much better! Now we are in control and we can be sure that every withdrawal is validated against account balance. As a nice side effect our code gains readability.

Inheritance and Polymorphism

Inheritance is the ability of a new class to acquire the properties of another class.

Let's look at another requirement:

As an account holder I can have multiple accounts of various types such as savings account, debit account etc for which different withdrawal fees apply. For a debit account the withdrawal fee is zero; for a savings account the withdrawal fee is 0.5% of the amount withdrawn.

From the requirement we can deduct that an Account can be modeled as a class with two properties - a Balance and a Type which can take values from an enumeration.

Also, additional changes to the Withdraw method should be applied:

  • Each withdrawal calculates and subtracts a fee from the balance
  • The check for insufficient funds should also take into the consideration the fee alongside the amount

Given the above, our code would look like this:

public enum AccountType
{
    Debit,
    Savings
}

public class Account
{
    private readonly AccountType _accountType;

    public Account(AccountType accountType)
    {
	_accountType = accountType;
    }

    public AccountType AccountType { get { return _accountType; } }

    public Money Balance { get; private set; }

    public void Deposit(Money money)
    {
	this.Balance += money;
    }

    public Money Withdraw(Money amount)
    {
	var fee = CalculateWithdrawalFee(amount);
	amount = amount + fee;
	if( Balance <= amount)
	{
	    throw new InvalidOperationException("Insufficient funds.");
	}
	this.Balance -= amount;
    }

    private Money CalculateWithdrawalFee(Money amount)
    {
	var margin = _accountType == AccountType.Savings ? 0.5 : 0;
	return amount * margin / 100;
    }
}

Everything looks good except the magic numbers. Until we get a new requirement:

As an account holder I can have a credit account with a withdrawal fee of 0.7%

To accomodate this new requirement we just have to change the CalculateWithdrawalFee method and add a new value in our AccountType enum:

public enum AccountType
{
    Debit,
    Savings,
    Credit
}

...

private Money CalculateWithdrawalFee(Money amount)
{
    switch( _accountType )
	case AccountType.Savings:
	    return amount * 0.5 / 100;
	case AccountType.Credit:
	    return amount * 0.7 / 100;
	default:
	    return 0;
}

The problem with this approach is that we had to add changes in more than one place; and the hint that this way of implementing the new requirement is not ok was given by the and conjunction from the previous sentence. And if that wasn't enough then the switch should be the indicator of a possible code smell.

A better way to address these requirements is through inheritance by first determining which are the common traits of an Account and extracting them into a base class followed by modeling the specific behavior for each subsequent type of account into its separate class.

So, as before, the common traits of an account are:

  • All accounts have a Balance property
  • All accounts support Withdrawal and Deposit
  • All accounts validate available amount when a withdrawal is performed

The distinct traits of the accounts are:

  • Savings accounts have a withdrawal fee of 0.5% of the amount withdrawn
  • Credit accounts have a withdrawal fee of 0.7% of the amount withdrawn
  • Debit accounts don't have a withdrawal fee

Although we can see a clear difference between the common traits and specific features for each account type there is a certain amount of overlapping between them when performing a withdrawal. In other words and specifically for withdrawal operation we can say that:

  • All account types support withdrawal
  • Each account type applies a specific withdrawal fee
  • All account types validate the amount against available funds

In order to model such kind of mixture between common and specific behavior we can make use of polymorphism by defining an abstract method CalculateWithdrawalFee which will be overridden in the derived classes with specific values for each account type but still keeping the common behavior in a single place.

Polymorphisms means the ability to request that the same operations be performed by a wide range of different types of things.

Now, with this approach let's address each requirement in the order they arrived. First, let's create the base class:

public abstract class Account
{
    public Money Balance { get; private set; }

    public void Deposit(Money money)
    {
	this.Balance += money;
    }

    public Money Withdraw(Money amount)
    {
	var fee = CalculateWithdrawalFee(amount);
	amount = amount + fee;
	if( Balance <= amount)
	{
	    throw new InvalidOperationException("Insufficient funds.");
	}
	this.Balance -= amount;
    }

    protected abstract Money CalculateWithdrawalFee(Money amount);
}

With the base class in place, let's model the first two account types - debit and savings:

public class DebitAccount : Account
{
    protected override Money CalculateWithdrawalFee(Money amount)
    {
	return 0;
    }
}

public class SavingsAccount : Account
{
    protected override Money CalculateWithdrawalFee(Money amount)
    {
	return amount * 0.5 / 100;
    }
}

With this structure in place adding a new account type is as simple as adding a new derived class and all edits are in a single place - the newly added file:

public class CreditAccount : Account
{
    protected override Money CalculateWithdrawalFee(Money amount)
    {
	return amount * 0.7 / 100;
    }
}

Chapter II - Structure

How you structure your code is very important in software development. A good code structure promotes low coupling and high cohesion which is a must for all good software.

Let's take a look at what coupling and cohesion mean.

Low coupling - The case of the Singleton

Coupling is the manner and degree of interdependence between software modules; a measure of how closely connected two routines or modules are; the strength of the relationships between modules.

To exemplify what low coupling is and its importance let's take a look at the well-kown Singleton pattern:

public class FileSystem
{
    private static readonly FileSystem _root = new FileSystem();

    private FileSystem()
    {
    }

    public static FileSystem Root
    {
	get { return _root; }
    }

    public string[] List(string path)
    {
	return Directory.EnumerateDirectories(path)
	    .Union(Directory.EnumerateFiles(path))
	    .ToArray();
    }
}

public class Explorer
{
    public string CurrentDirectory { get; set; }

    public IEnumerable<string> Contents { get; private set;}

    public void NavigateTo(string path)
    {
	var contents = FileSystem.Root.List(path);
	CurrentDirectory = path;
	Contents = contents;

	foreach(var item in contents)
	{
	    Display(item);
	}
    }
}

public class Program
{
    static void Main(string[] args)
    {
	var explorer = new Explorer();
	explorer.NavigateTo(@"c:\users\petru\Documents\");
    }
}

What is wrong in the code above? The Explorer class is tightly coupled with FileSystem singleton, i.e. Explorer class is highly denendent on the FileSystem class and cannot be used without the FileSystem.

To demonstrate this tight coupling let's create a test for Explorer class to verify that it returns proper values from a known directory:

public class ExplorerTests
{
    protected string Directory { get; set; }

    protected Explorer Explorer { get; private set;}

    protected void Initialize()
    {
	Explorer = new Explorer();
    }

    [TestClass]
    public class WhenNavigatingToTempDirectory : ExplorerTests
    {
	[TestInitialize]
	public void TestInitialize()
	{
	    Initialize();
	    Directory = @"C:\Temp\";
	}

	[TestMethod]
	public void ShouldDisplayDirectoryContents()
	{
	    Explorer.NavigateTo(Directory);
	    Assert.AreEqual(Directory, Explorer.CurrentDirectory);
	    Assert.AreEqual("C:\Temp\test.txt", Explorer.Contents.Single());
	}
    }
}

This test may succeed on our machine because we know that there's a file named text.txt in our C:\Temp\ directory because we've put it there. But what if we execute that test on another machine? Will that file be there?

Having such code means that not only our Explorer class is tightly coupled with the FileSystem class but also that this chain of dependency can grow to depend on the contents of the machine it's running on. We'll see how to break this dependecies in Chapter III - SOLID code.

High cohesion - Why can't I have all code in Program.cs?

Cohesion refers to the degree to which the elements of a module belong together.

The main reason for properly structuring your application is that similar entities belong toghether.

Let's consider the previous example in a less cohesive form:

public class Program
{
    static void Main(string[] args)
    {
	NavigateTo(@"C:\users\petru\Documents\");
    }

    static void NavigateTo(string path)
    {
	var contents = Directory.EnumerateDirectories(path)
	    .Union(Directory.EnumerateFiles(path));
	foreach(var item in contents)
	{
	    Display(item);
	}
    }
}

In this case our Program is nothing more than a God object - it does everything and this is a problem. It may not be a problem for such a trivial application but when it comes to real applications things get more complicated.

Even small production applications have a lot of objects and that number keeps growing alongside the growth of the application and at some point having everything in one place is simply unbearable. Furthermore, chances are that more than one developer will be working on the same codebase and you can't just have everything in a single file.

Chapter III - SOLID code

Single Responsibility Principle

Every module or class should have responsibility over a single part of the functionality provided by the software, and that responsibility should be entirely encapsulated by the class.

Let's look again at our Account example:

public abstract class Account
{
    public Money Balance { get; private set; }

    public void Deposit(Money money)
    {
	this.Balance += money;
    }

    public Money Withdraw(Money amount)
    {
	var fee = CalculateWithdrawalFee(amount);
	amount = amount + fee;
	if( Balance <= amount)
	{
	    throw new InvalidOperationException("Insufficient funds.");
	}
	this.Balance -= amount;
    }

    protected abstract Money CalculateWithdrawalFee(Money amount);
}

As you can see from the code, the Withdraw method does more than one thing, namely:

  • It calculates the withdrawal fee
  • It subtracts the amount and withdrawal fee from the balance

The purpose of this method is to withdraw the amount from the available balance; calculating the withdrawal fee should be someone else's responsibility. And that someone else should be a WithdrawalAmountCalculator:

public class WithdrawalAmountCalculator
{
    public Money CalculateWithdrawalAmount(Money amount, decimal percentage)
    {
	return amount * percentage / 100;
    }
}

Now, let's put it all together and remove the extra responsibilities from Account class:

public class WithdrawalAmountCalculator
{
    public Money CalculateWithdrawalAmount(Money amount, decimal feePercentage)
    {
	return amount * feePercentage / 100;
    }
}

public abstract class Account
{
    private readonly WithdrawalAmountCalculator _amountCalculator;

    protected Account(WithdrawalFeeCalculator amountCalculator)
    {
	_amountCalculator = amountCalculator;
    }

    public Money Balance { get; private set; }

    public abstract decimal WithdrawalInterestRate { get; }

    public void Deposit(Money money)
    {
	this.Balance += money;
    }

    public Money Withdraw(Money amount)
    {
	var amountToWithdraw = _amountCalculator.CalculateWithdrawalAmount(amount, WithdrawalInterestRate);
	if( Balance <= amountToWithdraw)
	{
	    throw new InvalidOperationException("Insufficient funds.");
	}
	this.Balance -= amountToWithdraw;
	return amount;
    }
}

public class DebitAccount : Account
{
    public DebitAccount(WithdrawalAmountCalculator amountCalculator) : base(amountCalculator)
    {
    }

    public override decimal WithdrawalInterestRate
    {
	{ get { return 0m; }}
    }
}

public class Program
{
    static void Main(string[] args)
    {
	var calculator = new WithdrawalAmountCalculator();
	var account = new DebitAccount(calculator);
	account.Deposit(new Money(100));

	var cash = account.Withdraw(50);
    }
}

Much better! Each class does only one thing, namely:

  • WithdrawalAmountCalculator calculates the withdrawal amount and
  • Account handles the changes to balance.

Open/Closed principle

Software entities (classes, modules, functions, etc.) should be open for extension, but closed for modification

Let's go back again to a previous example, the one with multiple account types:

public enum AccountType
{
    Debit,
    Savings,
    Credit
}

...

private Money CalculateWithdrawalFee(Money amount)
{
    switch( _accountType )
	case AccountType.Savings:
	    return amount * 0.5 / 100;
	case AccountType.Credit:
	    return amount * 0.7 / 100;
	default:
	    return 0;
}

The code from the example above is the exact opposite of Open/Closed principle:

  • It is open for modification because every time we'll need to create a new account type we'll have to add another case to the switch statement within CalculateWithdrawalAmount method
  • It is closed for extension in the sense that even if we do create a derived type of Account that type will have to always take into consideration the AccountType property when implementing its custom behavior.

Liskov substitution principle

Objects in a program should be replaceable with instances of their subtypes without altering the correctness of that program.

To exemplify let's consider the overused Rectangle and Square example where one would model the relationship between them using the following statement: a Square is a Rectangle having Width == Height as behavioral description. In other words, a Square behaves exactly like a Rectangle but the Square has Width == Heigth.

As an object hierarchy this would look like the following:

public class Rectangle
{
    protected int _width;

    protected int _height;

    public virtual int Width
    {
	get { return _width; }
	set { _width = value; }
    }

    public virtual int Height
    {
	get { return _height; }
	set { _height = value; }
    }
}

public class Square : Rectangle
{
    public override int Width
    {
	get { return _width; }
	set
	{
	    _width = value;
	    _height = value;
	}
    }

    public override int Height
    {
	get { return _height; }
	set
	{
	    _height = value;
	    _width = value;
	}
    }
}

Now let' think of consumer of this code: it knows it's working with a Rectangle and sets all the values accordingly but it is given a Square to work with. Let's see that in code:

public class Program
{
    private const int ImageWidth = 800;
    private const int ImageHeight = 600;

    static void Main(string[] args)
    {
	var rectangle = new Square();
	InitializeRectangle(rectangle);
	var area = rectangle.Width * rectangle.Height;
	Debug.Assert(area == ImageWidth * ImageHeight);
    }

    void InitializeRectangle(Rectangle rectangle)
    {
	rectangle.Width = ImageWidth;
	rectangle.Height = ImageHeight;
    }
}

The method InitializeRectangle will perform it's work on a Rectangle thus will set its Width and Height properties and the Main method will be expecting to have an rectangle with the area of ImageWidth * ImageHeight but since the Square overrides the Width when Height is set (and viceversa) the actual area of the rectangle will be ImageHeight * ImageHeight.

This shows that although a Square may have a similar structure with an Rectangle it cannot be used as a substitute for a Rectangle because it doesn't have the same behavior.

Interface Segregation principle

Many client-specific interfaces are better than one general-purpose interface.

To exemplify this let's consider some existing types within .NET Framework: MembershipProvider from System.Web.Security namespace and IUserStore from Microsoft.AspNet.Identity namespace:

public abstract class MembershipProvider : ProviderBase
{
    public abstract bool EnablePasswordRetrieval { get; }

    public abstract bool EnablePasswordReset { get; }

    public abstract bool RequiresQuestionAndAnswer { get; }

    public abstract string ApplicationName { get; set; }

    public abstract int MaxInvalidPasswordAttempts { get; }

    public abstract int PasswordAttemptWindow { get; }

    public abstract bool RequiresUniqueEmail { get; }

    public abstract MembershipPasswordFormat PasswordFormat { get; }

    public abstract int MinRequiredPasswordLength { get; }

    public abstract int MinRequiredNonAlphanumericCharacters { get; }

    public abstract string PasswordStrengthRegularExpression { get; }

    public event MembershipValidatePasswordEventHandler ValidatingPassword

    public abstract MembershipUser CreateUser(string username, string password, string email, string passwordQuestion, string passwordAnswer, bool isApproved, object providerUserKey, out MembershipCreateStatus status);

    public abstract bool ChangePasswordQuestionAndAnswer(string username, string password, string newPasswordQuestion, string newPasswordAnswer);

    public abstract string GetPassword(string username, string answer);

    public abstract bool ChangePassword(string username, string oldPassword, string newPassword);

    public abstract string ResetPassword(string username, string answer);

    public abstract void UpdateUser(MembershipUser user);

    public abstract bool ValidateUser(string username, string password);

    public abstract MembershipUser GetUser(object providerUserKey, bool userIsOnline);

    public abstract MembershipUser GetUser(string username, bool userIsOnline);

    internal MembershipUser GetUser(string username, bool userIsOnline, bool throwOnError)

    public abstract string GetUserNameByEmail(string email);

    public abstract bool DeleteUser(string username, bool deleteAllRelatedData);

    public abstract MembershipUserCollection GetAllUsers(int pageIndex, int pageSize, out int totalRecords);

    public abstract int GetNumberOfUsersOnline();

    public abstract MembershipUserCollection FindUsersByName(string usernameToMatch, int pageIndex, int pageSize, out int totalRecords);

    public abstract MembershipUserCollection FindUsersByEmail(string emailToMatch, int pageIndex, int pageSize, out int totalRecords);

    protected virtual byte[] EncryptPassword(byte[] password)

    protected virtual byte[] EncryptPassword(byte[] password, MembershipPasswordCompatibilityMode legacyPasswordCompatibilityMode)

    protected virtual byte[] DecryptPassword(byte[] encodedPassword)

    protected virtual void OnValidatingPassword(ValidatePasswordEventArgs e)

}


public interface IUserStore<TUser, in TKey> : IDisposable where TUser : class, IUser<TKey>
{
    Task CreateAsync(TUser user);

    Task UpdateAsync(TUser user);

    Task DeleteAsync(TUser user);

    Task<TUser> FindByIdAsync(TKey userId);

    Task<TUser> FindByNameAsync(string userName);
}

 public interface IUserPasswordStore<TUser, in TKey> : IUserStore<TUser, TKey>, IDisposable where TUser : class, IUser<TKey>
  {
    Task SetPasswordHashAsync(TUser user, string passwordHash);

    Task<string> GetPasswordHashAsync(TUser user);

    Task<bool> HasPasswordAsync(TUser user);
  }

If we would have to implement a simple authentication mechanism the functionality of IUserStore and IUserPasswordStore would be sufficient for our needs. Compare implementing only the functionality required for these two interfaces with the effort required to derive from the MembershipProvider class. Even worse, I can say from my experience that with the MembershipProvider you'll be writing code which will never be used.

Dependency Inversion principle

Depend upon Abstractions. Do not depend upon concretions.

Let's go back to our Singleton example. In the previous module we stated that the Explorer class is tightly coupled with the FileSystem class thus depending on it and also depending on underlying file system for tests to succed.

To break this dependency we need to hide the functionality of our FileSystem behind an abstraction and we do that using an interface IFileSystem which will be implemented by FileSystem class:

public interface IFileSystem
{
    string[] List(string path);
}

public class FileSystem : IFileSystem
{
    public string[] List(string path)
    {
	return Directory.EnumerateDirectories(path)
	    .Union(Directory.EnumerateFiles(path))
	    .ToArray();
    }
}

Now we make the Explorer class depend upon the IFileSystem interface:

public class Explorer
{
    private readonly IFileSystem _fileSystem;

    public Explorer(IFileSystem fileSystem)
    {
	_fileSystem = fileSystem;
    }

    public string CurrentDirectory { get; set; }

    public IEnumerable<string> Contents { get; private set;}

    public void NavigateTo(string path)
    {
	var contents = _fileSystem.Root.List(path);
	CurrentDirectory = path;
	Contents = contents;

	foreach(var item in contents)
	{
	    Display(item);
	}
    }
}

Now it's up to the Program class to glue them together:

public class Program
{
    static void Main(string[] args)
    {
	var fs = new FileSystem();
	var explorer = new Explorer(fs);
	explorer.NavigateTo(@"c:\users\petru\Documents\");
    }
}

This way we can test the Explorer class separately from the FileSystem class by passing it an instance of a class that implements IFileSystem and returns some stub data.

As an additional bonus, the semantics of the Singleton is still in place - there is only one instance of the FileSystem class for the whole lifetime of the application.