This is an excerpt from my book Refactoring TypeScript: Keeping Your Code Healthy.
Identifying The Problem
Billion Dollar Mistake
Did you know that the inventor of the concept of "null" has called this his "Billion Dollar Mistake!"
As simple as it seems, once you get into larger projects and codebases you'll inevitably find some code that goes "off the deep end" in its use of nulls.
Sometimes, we desire to make a property of an object optional:
class Product{
public id: number;
public title: string;
public description: string;
}
In TypeScript, a string
property can be assigned the value null
.
But... so can a number
property!
const chocolate: Product = new Product();
chocolate.id = null;
chocolate.description = null;
Hmmm....
Another Example
That doesn't look so bad at first glance.
But, it can lead to the possibility of doing something like this:
const chocolate: Product = new Product(null, null, null);
What's wrong with that? Well, it allows your code (in this case, the Product
class) to get into an inconsistent state.
Does it ever make sense to have a Product
in your system that has no id
? Probably not.
Ideally, as soon as you create your Product
it should have an id
.
So... what happens in other places that have to deal with logic around dealing with Products?
Here's the sad truth:
let title: string;
if(product != null) {
if(product.id != null) {
if(product.title != null) {
title = product.title;
} else {
title = "N/A";
}
} else {
title = "N/A"
}
} else {
title = "N/A"
}
Is that even real code someone would write?
Yes.
Let's look at why this code is unhealthy and considered a "code smell" before we look at some techniques to fix it.
Is It That Bad?
This code is hard to read and understand. Therefore, it's very prone to bugs when changed.
I think we can agree that having code like this scattered in your app is not ideal. Especially when this kind of code is inside the important and critical parts of your application!
A Side-Note About Non-Nullable Types In TypeScript
As a relevant side note, someone might raise the fact that TypeScript supports non-nullable types.
This allows you to add a special flag to your compilation options and will prevent, by default, any variables to allow null
as a value.
A few points about this argument:
Most of us are dealing with existing codebases that would take tons of work and time to fix these compilation errors.
Without testing the code well, and carefully avoiding assumptions, we could still potentially cause run-time errors by these changes.
This article (taken from my book) teaches you about solutions that can be applied to other languages - which may not have this option available.
Either way, it's always safer to apply smaller more targeted improvements to our code. Again, this allows us to make sure the system still behaves the same and avoids introducing a large amount of risk when making these improvements.
One Solution: Null Object Pattern
Empty Collections
Imagine you work for a company that writes software for dealing with legal cases.
As you are working on a feature, you discover some code:
const legalCases: LegalCase[] = await fetchCasesFromAPI();
for (const legalCase of legalCases) {
if(legalCase.documents != null) {
uploadDocuments(legalCase.documents);
}
}
Remember that we should be wary of null checks? What if some other part of the code forgot to check for a null
array?
The Null Object Pattern can help: you create an object that represents an "empty" or null
object.
Fixing It Up
Let's look at the fetchCasesFromAPI()
method. We'll apply a version of this pattern that's a very common practice in JavaScript and TypeScript when dealing with arrays:
const fetchCasesFromAPI = async function() {
const legalCases: LegalCase[] = await $http.get('legal-cases/');
for (const legalCase of legalCases) {
// Null Object Pattern
legalCase.documents = legalCase.documents || [];
}
return legalCases;
}
Instead of leaving empty arrays/collections as null
, we are assigning it an actual empty array.
Now, no one else will need to make a null check!
But... what about the entire legal case collection itself? What if the API returns null
?
const fetchCasesFromAPI = async function() {
const legalCasesFromAPI: LegalCase[] = await $http.get('legal-cases/');
// Null Object Pattern
const legalCases = legalCasesFromAPI || [];
for (const case of legalCases) {
// Null Object Pattern
case.documents = case.documents || [];
}
return legalCases;
}
Cool!
Now we've made sure that everyone who uses this method does not need to be worried about checking for nulls.
Take 2
Other languages like C#, Java, etc. won't allow you to assign a mere empty array to a collection due to rules around strong typing (i.e. []
).
In those cases, you can use something like this version of the Null Object Pattern:
class EmptyArray<T> {
static create<T>() {
return new Array<T>()
}
}
// Use it like this:
const myEmptyArray: string[] = EmptyArray.create<string>();
What About Objects?
Imagine that you are working on a video game. In it, some levels might have a boss.
When checking if the current level has a boss, you might see something like this:
if(currentLevel.boss != null) {
currentLevel.boss.fight(player);
}
We might find other places that do this null check:
if(currentLevel.boss != null) {
currentLevel.completed = currentLevel.boss.isDead();
}
If we introduce a null object, then we can remove all these null checks.
First, we need an interface to represent our Boss
:
interface IBoss {
fight(player: Player);
isDead();
}
Then, we can create our concrete boss class:
class Boss implements IBoss {
fight(player: Player) {
// Do some logic and return a bool.
}
isDead() {
// Return whether boss is dead depending on how the fight went.
}
}
Next, we'll create an implementation of the IBoss
interface that represents a "null" Boss
:
class NullBoss implements IBoss {
fight(player: Player) {
// Player always wins.
}
isDead() {
return true;
}
}
The NullBoss
will automatically allow the player to "win", and we can remove all our null checks!
In the following code example, if the boss is an instance of NullBoss
or Boss
there are no extra checks to be made.
currentLevel.boss.fight(player);
currentLevel.completed = currentLevel.boss.isDead();
Note: This section in the book contains more techniques to attack this code smell!
How To Keep Your Code Healthy
This post was an excerpt from Refactoring TypeScript which is designed as an approachable and practical tool to help developers get better at building quality software.
Keep In Touch
Don't forget to connect with me on:
Navigating Your Software Development Career Newsletter
An e-mail newsletter that will help you level-up in your career as a software developer! Ever wonder:
✔ What are the general stages of a software developer?
✔ How do I know which stage I'm at? How do I get to the next stage?
✔ What is a tech leader and how do I become one?
✔ Is there someone willing to walk with me and answer my questions?
Sound interesting? Join the community!
Top comments (48)
I don't particularly like the NullBoss pattern.
It means you have to introduce fictional objects to work around a problem, which moves the problem elsewhere. Now you can't make a
currentLevel.hasBoss
function without making checks for your special object, or adding a property to it calledisActuallyARealBoss
.In that sort of instance, I'd use an array, since there could be multiple bosses. As for being able to assign an empty array, I'd make sure I was using a language that allowed me to do that, because I value my mental health!
I agree that functions shouldn't generally return "thing or null".
One of the benefits of this pattern is that you typically would avoid the need for a
hasBoss
method (unless there's an exceptional case where you would need that check). Otherwise, using null objects should remove the need to do that.I would agree that an array of bosses would work well, if the scenario warrants it.
Thanks! Great feedback.
What would the hasBoss function be used for? My understanding is the null object pattern would handle it for you. If you want current level to do something, then just do it and let the Boss or NullBoss handle what to do in each situation
This way of handling nullable objects reminds me a special type in Functional programming called Option (or Maybe). It keeps two values: None and Some(val). While Some always keeps a real not null value and you can map it, None has nothing but it is not null.
I would also apply this pattern instead of creating null versions of objects because it is more generic and covers all nullable cases. For example,
fp-ts
provides such utilities asfromNullable
.Agree. This was a solution I debated whether to include in the book or not. Due to not wanting to take forever with the book, I decided to exclude it.
But I agree, there are many situations where using the option pattern works very well 👍
I don't understand why do we need a whole class to create an array
const myEmptyArray: string[] = EmptyArray.create<string>();
in C# you can juststring[] myEmptyArray = new string[] {};
orList<string> myEmptyList = new List<string>();
Sure. There's a semantic preference and also, in some cases, a performance boost too.
In C#, for example, using
Enumerable.Empty<T>()
will use a singleton underneath the covers so you aren't allocating an empty array on the heap.In very specific situations that could help. Otherwise, it's mostly a semantic preference.
That enumerator was not good example. You can't compare mutable with immutable - in this case it is not about semantic.
"What's wrong with that? Well, it allows your code (in this case, the Product class) to get into an inconsistent state."
So your real problem is that you are not following the basic rules of OOP. Your Product class should be responsible for not letting you create one with null ID in the first place.
And the saddest part is that neither of your examples above highlight the root cause / solution for the issue. I appreciate the gesture but unless this is not addressed, you are just treating a sympthom and being part of a much larger problem.
(The solution is correct, but it is applied at the wrong place)
What if a property or a property of a property is still optional? E.g.,
user.address.zipCode
. There is a proposal to avoid intermediate checks.user.address?.zipCode
. However,zipCode
is still nullable. Having null/undefined in any place is code smell and potential hole in the application. That's why it is always better to have a not null result. The problem is persistent not only in OOP but in FP as well, however, FP has a good way to handle it. Instead of working with null it provides you None.I would disagree. Following the "basic rules of OOP" is not in-and-of-itself an end. It is a means to an end, just like any programming paradigm. Who cares if you follow OOP?
What's the end? To have software that is easy to understand and therefore can be maintained easily, deliver business value, etc.
Also, there's nothing in OOP that says a class' property cannot be
null
.The article was about treating issues around nulls and the fact that TypeScript allows for null variables. That applies whether you are doing OOP or not, using classes or not, etc.
Thanks for the comments! 😊
Thank you for proving my point. Much appreciated
I agree with Hegi. Just add a null check once in the class constructor and you won’t need to do a check in all parts of the code that work with an instance of this class.
I disagree with the notion that "just adding a null check in the constructor" is always the best solution.
The fact is that any solution you choose is going to have trade-offs.
I've written about using constructor validation here and I agree that there are some contexts when using this technique is the best choice.
That being said, there are trade-offs to using constructor validation:
Mainly: You have to throw an exception to notify the caller that they did something wrong.
The caller (and developer using your class) has to have implicit knowledge up-front that your class will throw when used incorrectly. That can lead to developers potentially using your class wrong.
This also depends on whether your code is a publically accessible library or custom/internal business logic for a company.
With a public library, you aren't guaranteed to be able to educate users of the fact that your class will throw (other than in documentation, them finding out "the hard way", etc.)
What if that person wants to use your library in a performance-sensitive context? Throwing exceptions has a huge performance impact... so they may not opt for your library now.
The same applies to custom business code: does it need to be used in a performance-sensitive context? If so, throwing exceptions is the first offender to that most devs will look at removing.
Now, looking at the technique in this article, it allows the caller to not have to know about nulls and having to check them. Sure, it's a different technique, but now you have hidden/encapsulated the need to check for nulls, it's more performant since you don't need to catch exceptions, etc.
Is it a bit more complicated (internally)? Sure. But that's the price you pay in order to make code that's easier to use or has less of an entry barrier.
SOOO... which way is best?
Well, it depends! And that comes back to my comments with @hegi - just saying that one technique works best in all contexts across the board leads to issues when perhaps the context in question is something like a performance-sensitive, etc.
This is the same thinking people have when they say that "microservices is the best architecture and should always use it." That attitude and approach can (and has) destroy companies when used in the wrong context.
Thanks for your thoughts @alexander !
So instead of making guards using
null
, you propose to create useless empty array and objects, right? Just to spam memory with a empty values of nothing?As I am seeing from the comments you propose to use
NullObject
pattern, like everywhere, to fill any of gaps, which makes the code leaking to any executing instead of breaking the flow when it needs.What?
No. Instead of making 10 null checks, you do 1.This pattern never makes sense if you just want to avoid a single null check. But when you do the same checking at multiple levels, this vastly simplifies logic at times.
Review your comment in 10 years.
Immutable code where you can drop all null checking is fantastic to work with. And I have been using the Null object (/DefaultObject pattern) for a decade in various settings. Still does wonders for avoiding nasty NPEs in Java and works quite well in JS as well, though I am pragmatic: I do not create a full 1:1 mapping for every class. Just the ones where it makes sense.
Hi, the talk in my comments is not about mutability , but about clean and not overused code with default values where execution must be terminated, but not being executed with empty values.
Hi James, how do you instantiate the NullBoss class? I reckon it still needs to do the null check in the first place, doesn't it?
The
NullBoss
doesn't do any null checking because it replaces the concept of null altogether.Usually with kind of pattern, you would have a factory of some kind create your objects for you.
Does that help?
Could you provide the practical example with
NullBoss
?Cause the current example is not like dummy date, that can be used as mock. But the example which is far from reality, correct me if I wrong.
I think the whole idea of using NullObject is taken from Option type, which holds None or Some. When they is no
null
you don't think it exists. E.g. while Some holds a real not null value, which can be mapped, None always returns None. This approach is called "Railway programming".null
is avoided and the program doesn't crash. Just different way of thinking when there is no null.As I know, the pattern was a long time before Optionals (some, none).
And what author proposes – it’s a big difference from it.
Using optionals you stop the execution flow, while the topic about redundant objects creation.
2 examples I've seen in Minecraft:
Air blocks - they are blocks like any other, rather than nulls to represent a block not being present in a location
Also added in the last few versions (replacing null) - "Empty" item stacks - there is a singleton "ItemStack.EMPTY" and a method "ItemStack#isEmpty()"
Like in all OOP, you still need to make the decision on which class to create somewhere. What you gain by taking the decision at a higher-up place is the ability to avoid this decision in all other places.
This came as divine intervention. I'm currently struggling with a null check hell case because of some inconsistently structured api data, and I was thinking about what would be the clean way to deal with it. Then this article came up. Thanks! :)
If you're using JS, take a look at lodash.com/docs/#get
Sweet! The book covers the special case pattern also.
Sometimes null is useful but generally, it plagues the code. Null being part of the type system like typescript allows is really neat (props to Kotlin, Rust and C++ too). Having the compiler yell at you for checking for null when something can never be null makes code a lot more readable without leaving the writer wondering whether this or that will actually be null or not.
Nice article.
This part though, I don't understand:
I'm not sure what this is supposed to mean. You can make empty arrays in C# and Java just fine (though you generally use an empty collection instead of raw arrays).
The code you provided just does nothing, I'm not sure what problem you tried to solve by writing it.
Great post!
I've just been reading about "The Billion Dollar mistake" from a SQL point of view, interesting to hear another perspective on the same.
Oh ya... NULL in SQL can be even trickier!
nice post! but one typo in examples
should be
Good catch! Thanks
Optional types are where it's at, and there's some prebuilt ones that are more or less free to use in terms of complexity.
Rearchitecting your entire application to deny the fact that nulls exist as a valid value strikes me as basically the nuclear option when compared to dealing with them head on.
It is not necessarily about
null
s, but what to do in the absence of a value. You can often find multiple types of these null classes to cover different cases. And it does mean you have to rearchitect anything. For instance, when designing a Card component in a React application, that was dealt aUser
object, we assumed it would always exist. Suddenly, it was potentially optional, but that would imply changinguser: User
touser: User | undefined
through multiple levels. Simply creating aNullUser
that was returned at the top level selector function allowed us to make 0 (zero) changes further down.