Intention Revealing Interfaces [w/ Examples] - Domain-Driven Design w/ TypeScript
We cover this topic in The Software Essentialist online course. Check it out if you liked this post.
Have you ever heard someone say that your code should be so well-understood that it should be read like a book?
Those are the same developers who are usually against writing comments. Their argument is that comments are non-ideal because as soon as code gets changed, comments go out of date really quickly.
I tend to agree with that.
I think there's a more important reason why your code should be well-understood.
When new developers join your company and start to peer through your codebase, new developers might struggle with how to use your code.
More critically, even if it's undertstood how to use something, new developers might struggle with when it's the right choice to use that particular class or method.
In previous articles, we talked about the fact that understanding the domain is key to implementing the single responsibility principle properly and writing decoupled code.
In this article, we're going to explore what it means to write Intention Revealing Interfaces and how it enables you to write code that improves comprehension as to how and when a particular class or method should be used.
In Eric Evans' seminal book about Domain-Driven Design, he describes the need for something called an Intention Revealing Interface with respect to choosing the names for our classes, methods, etc.
Very simply put,
Name classes and operations to describe their effect and purpose, without reference to the means by which they do what they promise. - Los Techies 1
In my own words, an Intention-Revealing interface is:
Writing code... that reveals... the intention... to others!
And doing that without regard for expressing how it does it (ie: encapsulation).
Also, when we say interfaces here, we’re not necessarily referring to the interface
that you and I best know: the OOP construct. In this case, we’re just referring to an interface as whatever the client (be that a consumer of an SDK, and API, or a junior developer joining our company) using our code will see. We're referring to it more as "how the client uses our code".
The client doesn't need to know
I was having a discussion with a colleague the other day about if it was OK to run additional logic in a getter
without letting the client explictly know if we're doing that.
Of course it is.
Let's say we wanted to return the most popular tracks
that were topping the charts.
class Chart {
get popularTracks (): Track [] {
return this.props.tracks
// Filter out tracks with no listens
.filter((track) => track.listens.length > 0)
// Sort them descending
.sort((a, b) => b.listens.length - b.listens.length)
}
}
This is basic OOP encapsulation. Consider the additional mental overhead if everytime we wanted to use a method or access a property, we had to look up the documentation for it to see if it was really the right one for us to use.
Creating Domain Events
In the previous article about where Domain Events get created, we demonstrated that sometimes we want to dispatch Domain Events within method calls or intercepted getters
and setters
.
Consider if we wanted to do some change tracking on our Artist
domain entity from White Label.
How would we go about creating a ArtistEdited
domain event?
class Artist extends AggregateRoot {
private dirty: boolean = false;
...
set name (name: string): void {
this.props.name = name;
// intercept our setter
this.markDirty();
}
private markDirty (): void {
if (!this.dirty) {
this.dirty = true;
this.addDomainEvent(new ArtistEditedEvent(this));
}
}
}
That ability to intercept changes to our domain model with getters and setters and create Domain Events that signify relevant occurrences in our domain, while saying no more than necessary to the client of the API (through encapsulation) moves us closer to a rich declarative domain model.
How to know when our ability to express intention is lost
Here’s what Evans says on how to know when we've run amuck.
If a developer must consider the implementation of a component in order to use it, the value of encapsulation is lost. - Evans (DDD)
If developers need to know how something's works before they use it, then we’ve lost encapsulation.
Alright, so how do we write code that tells other developers what is done without having to show them exactly how it’s done?
Two ways.
- Encapsulate things that aren’t necessary for other developers to know about, and
- We name things really well.
We’ve talked about all the benefits of encapsulation in object-oriented programming, but in Domain-Driven Design, there’s an even more essential reason why it’s important.
One of the primary goals of Domain-Driven Design is to "embed the domain terminology in the code" 2.
Ideally, we would like to end up with an entire system that’s more declarative than imperative, can be understood by non-coders and domain experts, and educates newly onboarded developers about how the business actually works because the software is a in-code representation and solution of the actual problem domain.
That means we should check our variable names, methods, classes and Domain Events to ensure that we’ve effectively "capture[d] the domain model, in domain terms, through interactions with domain experts” 2.
Let’s make it perfectly understood exactly what this class is for
If someone other than the original developer must infer the purpose of an object or operation based on its implementation, that new developer may infer a purpose that the operation or class fulfills only by chance.
We should aim to be writing code that is irrefutable about what it is and what it does.
In fact, this is the primary theme of my short ebook “Name, Construct & Structure” and my approach to organizing features with use cases.
For example, a good practice for naming classes is to ensure that the actual name dictates what type of construct it is:
Therefore: UserController
, UserRepository
, UserMapper
, User
.
Check out the free book if you're interested in learning more about how to name things.
Let’s make it perfectly understood what this method needs
Have you ever seen code like this? Here’s an example I got from Los Techies.
class MakeOffer extends UseCase {
execute (items) {
const vinyl = items[0];
const trader1 = items[1];
const trader2 = items[2];
}
}
Apparently this type of thing is pretty common in the Java and C# world as blob parameters.
In this MakeOffer
class, it’s probably well understood that the single responsibility is to make an offer, but the parameter items
in the execute
method doesn’t reveal anything about what we were intended to pass in.
Let’s make it perfectly understood what this class is not for
If [code used in one way] was not the intent, the code may work for the moment, but the conceptual basis of the design will have been corrupted, and the two developers will be working at cross-purposes. - Evans (DDD)
This is exactly what we strive so very hard to avoid.
Consider we were building the Billing
subdomain in our Vinyl Trading App and we had a Use Case called BillCustomer
.
class BillCustomerUseCase implements UseCase<any, any> {
execute (customerItem): void {
// algorithm
}
}
At this point, you might ask me something like:
“uhh.. Khalil, why isn’t there something in there to dictate the amount of money you’re going to bill the customer? Don’t you need to pass that in as well?”
Sure. That’s one way you could look at it.
You could look at it as if the BillCustomerUseCase
is a general use case that takes in an amount and a customer
in order to bill them.
But yes, correct. That’s not very explicit from the design.
Perhaps, you might see it a different way.
customerItem
. does that mean I pass in a customer? Or iscustomerItem
the item that we’re billing the customer with? Wait. If I pass in an item, likevinyl
, how is it going to know whichcustomer
to bill? Is there an object for this? Is there something that contains both theitemId
and thecustomerId
? It would have been really helpful if I knew what the type needed here was...
It's much harder to reason about code without types, and that's why typescript is such a great choice for domain modeling, but there's another problem with this code.
Let’s say that someone got this working to do transactional charging; meaning you get charged for every transaction of a single item.
Nice.
Then, another developer comes along, takes one look at the code, and says:
This is definitely for billing subscriptions.
Maybe that other developer considers the absence of an amount
parameter to signify that this use case meant for charging customers on recurring charges.
At that point, both developers are using this function for two separate reasons.
This is what Evans was warning us about when he said that the “conceptual basis of the design will have been corrupted” if two developers are working “at cross-purpose[s]”.
We’ve already seen what happens when developers add code to something that changes for two very different reasons. That coupling becomes a big problem when those two different features evolve in different ways. An adjustment to one feature has the repeated risk to break the other.
This violation of the single responsibility principle could have been avoided if we had used intention-revealing interfaces (ie: better names) and types for the parameters.
interface Product {
amount: number;
}
// Vinyl can be purchased
class PurchasedVinyl implements Product {
public amount: number;
constructor (amount: number) {
this.amount = amount;
}
}
class AnnualSubscription extends Subscription implements Product {
public amount: number;
constructor (amount: number) {
super();
this.amount = amount;
}
}
class BillCustomerUseCase implements UseCase<any, any> {
// Explicit parameters
execute (customer: Customer, product: Product): void {
// Execute algorithm
}
}
Notice the improvement here? We're explicitly specifying what's required in the parameters here: a customer
and a product
. And it just so happens that with this design, we do charge transactionally, and a Subscription
is a type of product
that we could charge for.
Constantly ask yourself if the terminology sounds too "techy"
Try really hard not to use words that sound "techy" like Processor
, Creator
or and Manager
for the names of your domain layer constructs.
If you notice that seems to be the case, there's a chance that you might have run amuck.
The language we use in the code should be similar to the language that the domain-experts use in the real world.
1 Another great read on intention revealing interfaces. https://lostechies.com/jimmybogard/2007/11/29/intention-concealing-interfaces-blob-parameters/
2 Here's a great article by Steven A. Lowe on the 3 basic principles of Domain-Driven Design. He's currently working on his "Head First Domain-Driven Design" book. https://techbeacon.com/app-dev-testing/get-your-feet-wet-domain-driven-design-3-guiding-principles
Discussion
Liked this? Sing it loud and proud 👨🎤.
Stay in touch!
Join 15000+ value-creating Software Essentialists getting actionable advice on how to master what matters each week. 🖖
View more in Domain-Driven Design
You may also enjoy...
A few more related articles