This is the English version of my original post in Hungarian. Photo by Samuel Zeller on Unsplash.
The null reference is so ubiquitous, such an integral part of our experience as programmers, that it would be hard to imagine a world without it. Yet it is the source of so much frustration and actual financial loss that it is well worth the effort to think about some alternatives.
But what exactly is the problem?
Let's imagine we have a system where customers can place orders. The orders contain line items that have a reference to a product and they store the quantity ordered. They can also calculate the price based on the unit price and the quantity specified.
public class LineItem
{
public Product Product { get; }
public decimal Quantity { get; }
public decimal Price => Product.UnitPrice * Quantity;
public LineItem(Product product, decimal quantity)
{
if (quantity <= 0)
throw new ArgumentOutOfRangeException(nameof(quantity));
if (product == null)
throw new ArgumentNullException(nameof(product));
Product = product;
Quantity = quantity;
}
}
A single discount of various types can be applied to an order so we use the Strategy pattern to decouple orders from the calculation of discounts.
public interface IDiscount
{
decimal Calculate(decimal orderTotal);
}
public class CouponDiscount : IDiscount
{
public CouponDiscount(string couponCode, decimal rate) { /* ... */ }
public decimal Calculate(decimal orderTotal) => orderTotal * rate;
}
public class FrequentBuyerDiscount : IDiscount { /* ... */ }
Given the order total, a discount object can calculate the reduced price according to its own set of rules.
The order class provides methods to add line items and a discount, and a property that returns the total after applying the discount.
public class Order
{
public IEnumerable<LineItem> LineItems => lineItems.AsReadOnly();
public decimal Total => discount.Calculate(totalBeforeDiscount);
private decimal totalBeforeDiscount => lineItems.Sum(i => i.Price);
private IDiscount discount; // initialized to null
private List<LineItem> lineItems;
public Order()
{
lineItems = new List<LineItem>();
}
public void AddLineItem(LineItem lineItem)
{
if (lineItem == null) throw new ArgumentNullException(nameof(lineItem));
lineItems.Add(lineItem);
}
public void AddDiscount(IDiscount discount)
{
if (discount == null) throw new ArgumentNullException(nameof(discount));
this.discount = discount;
}
}
The class above seems completely innocent. What can go wrong? We even added null checks for all the arguments. But it is still very likely that it will throw a NullReferenceException. Can you spot the bug?
var order = new Order();
order.AddLineItem(new LineItem(product1, 2));
order.AddLineItem(new LineItem(product2, 1));
Console.WriteLine($"Total: {order.Total}"); // BOOM!!!
Oh, we forgot to add a discount!
But wait. Creating an order without a discount is a completely valid requirement. Let's fix the code to support that.
public decimal Total =>
discount == null ? totalBeforeDiscount : discount.Calculate(totalBeforeDiscount);
This code works, but it sure isn't beautiful. Even if we put aesthetic concerns aside, having to check for null every time we access the discount field is a recipe for disaster. Forget it even once, and a NullReferenceException appears. Also, we need to duplicate the default logic, returning the total before discount, every time. That logic isn't well encapsulated.
Fortunately, we can do better. Let's look at how we dealt with the exact same problem when adding the line items. We initialized the lineItems field to a safe default value, the empty list, in the constructor.
public Order()
{
lineItems = new List<LineItem>();
}
Having done that, there is no need to check whether lineItems is null in the AddLineItem method.
lineItems.Add(lineItem);
Can we imagine such a safe default value for discount too? Of course we can, that would be a NoDiscount class in which the Calculate method simply returns the orderTotal argument without modification.
public class NoDiscount : IDiscount
{
public decimal Calculate(decimal orderTotal) => orderTotal;
}
Such a default object is called a Null Object. Now we can use that as the initial value for discount in the constructor,
public Order()
{
lineItems = new List<LineItem>();
discount = new NoDiscount();
}
and get rid of the problematic null check in the Total property.
public decimal Total => discount.Calculate(totalBeforeDiscount);
As a final piece of advice, it is recommended to exercise restraint when applying the Null Object pattern. It is a very elegant solution when the default behavior is harmless and predictable. But if the behavior of the Null Object is not obvious to the consumer, getting a NullReferenceException is actually more useful than having a badly-designed Null Object behave unexpectedly.
Top comments (18)
The next version of C# will make huge strides toward making
null
a nonissue. Nullable reference types.AFAIK there are currently measures to deal with null issues in C#
They are:
Nullable types System.Nullable<
T
>Default literal
Null coalescing operator
Null conditional operator
Nullable value types introduced the concept of null where previously a value couldn’t be null. (It’s essentially an option type.) The rest of these were introduced as syntactic sugar to help us deal with the fact that null exists in the language. In upcoming versions of C# you will be able to opt into nullability. Soon, the default mode of operation in C# will be “things can’t be null”.
The benefit comes when you use Resharper because you can utilize code annotation attributes ([CanBeNull], [NotNull], [ItemCanBeNull], [ItemNotNull])
So that’s exactly what I think you’re missing here. In upcoming versions, you won’t need those annotations. Take some time to look into Nullable reference types.
I think it's far from being ready. Especailly if it introduces breaking changes. Switching midsets is difficult and if you hide a feature behind a compiler switch it's not going to work.
By the way John Skeet found 2 bugs straight away...
Awesome, I knew they were working on it but I didn't know they already have a working prototype I can download!
Very well written and straightforward - thanks!
Using the same pattern but in the context of collections, C# has ways to define "empty" collections. For example, a collection of
IEnumerable
can use the Null Object pattern this way:Looking forward to the next article!
Interesting pattern - for a more general purpose solution to the null-safety issue when no default can be defined, I'd recommend the Maybe / Option pattern, which is a universally equivalent (but type-safe) alternative to null references
Very good overview of the null object pattern.
Others already mentioned Maybe/Optional pattern. Here's my take on that pattern: github.com/j2jensen/callmemaybe
I'm definitely looking forward to nullable reference types in C# 8!
Nice post. I'm a huge fan of the "Null Object" and try to avoid any
null
in my code as good as possible. Of course, you have to considernull
as possible inputs from external sources (network, database, user input), but inside of my own code, I use "Null Object" most of the time.Well explained and easy to follow. Thank you!
You laid out the problem and then the solution very clearly. Thank you! Quite a satisfying read.
Thanks Joe!
Thank you Botond for providing a concrete example to make it easier to digest the Null Object pattern (NOP hereafter since it's too long to type 😉).
I've been using NOP happily and worked great when used in conjunction with Singleton Pattern (😱).
Shameless plug: I've touched on it briefly here.
A valid usage of Singleton Pattern (with Null object Pattern)
🕺🆂🆄🅽🅶⭐️🅺🅸🅼💃
This is awesome!
Well explained. I followed it !