▲ up

Wrapping primitives in domain objects

An interesting pattern I have seen with many developers is that they shy away from creating classes to represent domain objects, especially basic ones. This is pretty common for identifiers that are numbers, UUIDs, or just strings. Another similar pattern I see is for booleans, there is plenty of code out there that uses a boolean value to represent some binary notion that does not really map to a true or false. A tell tale is if there is a variable or function starting with isXXX. This is by no means a rule but should trigger a line of questioning whether a boolean is the correct data type. That is exactly what I wanted to explore in this post, rich data types and being able to add a whole new level of comprehension to your code and how it fits into your application architecture. This is so common it has been given a name of primitive obsession.

Wrapper classes

Using a customer ID as an example I wanted to show some ways in which we could represent this instead of its primitive type and what values we get from it. The way I manage these domain objects is what I call wrapper types or classes but there are several names I am sure, essentially I am making a class that wraps the primitive type that represents the value.

To start off this example lets assume we represent the customer ID as an integer, which is pretty common if we use the auto incrementing field of a SQL database. The use case we will use is a service that stores the customer movie recommendations and has an API to read the recommendations for a given customer ID, it may look something like this (depending on the framework you use):

@Controller
class RecommendationController(private val repo: RecommendationRepository) {
    @Get
    fun getCustomerRecommendations(@Path customerId: Int): Preferences {
        return repo.findWithId(customerId)
    }
}

The first and most common opposition I hear is what value does it add? Of course in the trivial example I have provided above it is hard to see the value, but as your application grows it becomes more valuable and it is much easier to be using it from the outset. Rather than explain all the benefits of it over the costs of an extra class I would like to show it through a set of examples similar to what I have seen/dealt with myself.

Clearer function signatures

Let us now consider a little more complicated example, because we could be giving a customer thousands of recommendations we allow the client to provide a limit to the number of recommendations we return since they likely do not require all at once. This will again be another Int given to the controller as a query parameter. Another common feature here is that when you have an account with your movie streaming company you can actually set up some profiles so lets also add that since we will likely want to be getting the recommendation for that profile only, so now our controller function looks a little more like this:

@Get
fun getCustomerRecommendations(@Path customerId: Int, @Path profileId: Int, @Query limit: Int) {
    // ....
}

Hopefully you can already see that we now have several different domain elements that are all represented as an Int. In the controller layer (the outer layer) this makes perfect sense as we want to keep these simple types to work nicely with the framework for deserialization however our internal API of the repository now will have 3 input parameters that are all of the same type. If you do not use wrapper types you have to rely on the parameter names of the API to know which goes into which place, if we were to represent these with their own classes instead the compiler will tell us when we have put the profile ID in the place of the customer ID.

How to represent the wrapper class

Now that we have our first good example why it would be worthwhile adding these classes lets see what it will look like. This will greatly vary between languages and the way I am representing it here is by no means the best or only way it will all depend on what actions you need to perform on the type. In Kotlin there are few options and I will go through them all and why I choose some over others.

Inline classes

The absolute best option in Kotlin is inline classes as this use case is exactly what the feature was built for, however as of the writing of this post it is still in experimental stages so I have not used it a lot on production code bases but definitely is fine to have a play with if you love to be on the bleeding edge.

inline class CustomerId(val value: Int)

Data classes

Typically I lean on data classes as these objects contain some piece of data as the name suggests. The benefit here is mainly that the equals function is generated for us which is something I commonly find that I need to make use of. The only element of this implementation to be cognisant of is that you will want a consistent naming convention and expect direct access to the inner field as data classes require there to be a public accessor. Personally I opt for value (and the generated Java of getValue).

data class CustomerId(val value: Int)

When the domain object is represented as a String I also adjust the toString function to just provide the value rather than the generated form which may be unexpected, such as below. This of course entirely depends on your use case and may not always add value for you.

data class CustomerName(val value: String) {
	override fun toString(): String = value
}

Normal classes

If there is not the use case for equals and you do not want to allow for direct access to the inner field I tend to just write a typical wrapper object very similar to the data class but with a slightly different API. Honestly I have not seen a lot of use for this case in a Kotlin only project but in Java and Java/Kotlin projects I have found it valuable to represent absence of a field, for example for a certain request the customer age may not be required:

class CustomerAge(private val innerValue: Int?) {
	val value: Optional<Int> = Optional.ofNullable(innerValue)
}

In an entire Kotlin project the above does not make a whole lot of sense as nullability is built into the language however I wanted to show it for completeness as you may be working in a language that does not or you may have a use case where you do not want to allow for direct access to the value for some reason.

Benefits of wrapper classes

To finish up I wanted to go over some of the benefits I have seen for wrapper classes.

Clearer intent

Wrapper classes make it crystal clear what type of data you are working with, you will never have to trace the call stack or hope that someone wrote the currently correct variable name for the arguments for the string you are working with. The class now tells you exactly what it represents and the intent on how it should be used. The most significant impact for me has been when reading older code, it makes it much easier to read function signatures and the logic as I can lean on the types to inform me whereas when things are all primitive types I rely on the developers variable naming which has many times become outdated or simply confusing.

This is most apparent when dealing with booleans. Consider the case where content that a customer can see is restricted based on their subscription plan. A pretty common thing to see is:

val isAllowed: Boolean = doesCustomerHaveSubscription(customer) && subscrptionCanAccess(customer, content)

if (isAllowed) {
	// Do something...
}

There are several other variations of this where we misrepresent a binary state as a boolean, allowed/restricted, valid/invalid are a couple. By representing this as a boolean, certainly we do get a lot of benefit because we can leverage the language construct for combining these evaluations but it is very easy to start evaluating either side of the binary choice. For example recently I have reviewed code where in one spot we are checking to see if it is allowed and at another location not far away we are checking to see if it is restricted. I can speak from experience that this is very difficult to understand and keep in a mental model.

Further, there is always some intrinsic element of the domain that dictates how these values can be combined that do not always map to the same combination as boolean logic, or you have forced it to match. For example, we typically have it that to be valid when combining evaluations, all must be valid, if one is invalid then the whole combination is invalid. Certainly this does map to having valid as true and invalid in boolean logic with the AND operator by why represent it like this when we can represent it with much clearer intent:

sealed class Evaluation {
	abstract fun combine(other: Evaluation): Evaluation
}

class Allowed: Evaluation {
	override fun combine(other: Evaluation): Evaluation {
		return when(other) {
			is Allowed -> Allowed
			is Restricted -> Restricted
		}
	}
}

class Restricted: Evaluation {
	override fun combine(other: Evalution): Evaluation {
		return Restricted
	}
}

Above I have used sealed classes which are extremely helpful for representing a known finite number of states of a value.

Then we can achieve the exact same as above but getting the intent much clearer, certainly there is more code but that is a small price to pay for ease of reading:

val contentEvaluation: Evaluation = evaluateCustomerSubscription(customer)
							.combine(evaluateSubscriptionAccess(customer, content))

if (contentEvaluation == Allowed) {
	// Do something...
}

Easier refactoring

As with a lot justifications, adding a wrapper class makes the code base easier to adapt to change, change in design or requirements. There are several examples for this but I wanted to share one that has stuck with me, let's imagine we have an application that stores a customer's address and one of the fields of the address is the zip code and since we are an American company we know that zip codes in USA are just numbers so we decide to model this as an integer. Then we may have some elements as follows around the code base:

data class Address(val streetName: String, val zipCode: Int)

// ...

fun getShippingCostForZipCode(zipCode: Int): Cost

// ...

fun generatePostalSlip(name: String, zipCode: Int): String

Now consider the case where the company wants to expand to somewhere like the UK which has letters in their zip code, suddenly we now need to represent the zip code as a String, so we need to go into all these locations and change them up. If however we had chosen to represent this with its own wrapper class then it would be pretty simple, we can just change the internal value of the class and adjust the functionality as required. I know that we have great IDE tooling that can manage this but that still does not eliminate the large diff that is typically created which will be reviewed or examined at some later point in time. Having a smaller diff makes it much easier for your colleagues to understand the actual purpose of the change. In the above example you will be conflating the adding of the UK marketplace with changing the zip code to a string.

Richer API

It is obvious that we have no good reason to add 2 to a customer ID, yet when we simply represent a customer ID as an integer this is in fact part of the API we provide. Further how many times have you seen private methods acting on a string to slice it a certain way or format it just so, and then when you have used it enough times you get the idea to add this to some distant ambiguous "util" class? This to me is a smell begging us to make some sort of class around this data to expose an API that matches the domain context and this is precisely what a wrapper class will provide for our data. One very common one is a name, we see it everywhere represented as a string and then have utility functions capitalizing the names, getting the first and second name, etc.

fun printNameForLetter(name: String) {
	println("${firstNameInitial(name)}. ${capitalizeLastName(name)}")
}

private fun firstNameInitial(name: String): String {
	// ...
}

private fun capitalizeLastName(name: String): String{
	// ...
}

Such functions are far better suited to its own class where all of this common functionality can be localized and benefited across all use cases.

class Name(value: String) {
	val fistNameInitial: String = ...
	// ...
}

The latter allows for far more reuse due to the proximity and easier to manage any changing requirements. One great feature of Kotlin it would be naive of me to skip is that it provides extension functions which bridge this gap and make for a very different design pattern. I am personally still working out the best conventions as to when add a function to the class or as an extension function. Currently my rule of thumb is if it will be used elsewhere add it to the class, simply because then when looking for the definition it is all in one place rather than across several files as extension functions.