Recently I wanted to build a generic data source layer. This data source would be able to return pretty much anything from a local cache, or if the local cache doesn't contain the requested object, it would fetch the object from a server and then cache the result locally before returning it to me. To achieve this, I figured that I would write a generic local cache, a generic remote cache and a wrapper that would combine both caches, allowing me to transparently retrieve objects without having to worry about where the object came from.
It didn't take long before I saw the first compiler warnings and remembered that generics can be extremely hard to bend to your will. Especially if your generic object uses other generic objects to transparently do generic things.
In this blog post, I will show you my approach to tackling complicated problems like this, and how I use pseudo code to design and implement a fluent API that works exactly as I wanted. In this blog post we'll go over the following topics:
- Designing an API without getting lost in the details
- Knowing how simple generics are used in Swift, and how you can use them in your code.
- Understanding you can have protocols with generic requirements, also known as associated types.
- Combining generics with protocols that have associated types
Are you ready to enter the brain-melting world of generics? Good, let's go!
Designing an API without getting lost in the details
I promised you generics. Instead, I'm going to show you how to design an API that uses generics first. This is to establish a goal, something we can work towards throughout this blogpost. Generics are complicated enough as they are and I don't want you to get confused to the point where you're not sure what we were building again.
In the introduction of this post, I mentioned that I wanted to build a generic data store that cached data locally, and would use a remote data store as a back up in case the required data didn't exist locally.
A good way to get started building something like this is to write down some pseudo-code that demonstrates how you would like to use the API or component you're building. Here's what I wrote for the caching layer:
let localDataStore = UserDataStore()
let remoteDataStore = UserApi()
let dataStore = CacheBackedDataStore(localDataStore, remoteDataStore)
dataStore.fetch(userID) { result in
// handle result
}
This is pretty straightforward, right? You can see that I want to create two stores and a wrapping store. The wrapping store is the one that's used to retrieve information and it uses a callback to inform the caller about the results. Simple and straightforward, just how I like it. Keep in mind that whatever we design has to work with more than user objects. We also want to be able to store other information in this structure, for example, documents that belong to the user.
Let's dive a bit deeper and write a pseudo-implementation for CacheBackedDataStore
:
class CacheBackedDataStore {
let localStore: LocalStore
let remoteStore: RemoteStore
func fetch(_ identifier: IdentifierType, completion: @escaping Result<T, Error>) {
localStore.fetchObject(identifier) { result in
if let result = try? result.get() {
completion(.success(result))
} else {
remoteStore.fetchObject(identifier) { result in
if let result = try? result.get() {
completion(.success(result))
} else {
// extract error and forward to the completion handler
}
}
}
}
}
}
You might notice the type T
on the result here. This type T
is where our generic adventure begins. It's the start of a rabbit hole where we're going to turn everything into objects that could be anything. At this point, we have enough "design" to get started with setting up some of our building blocks. To do this, we're going to have a look at generics in Swift.
Adding simple generics to your code
In the pseudo-code design that I showed you in the previous section, I used the type T
. Whenever we write code with generics in Swift, we typically use T
to flag a type as generic. A generic type can be pretty much anything as long as it satisfies the constraints that are specified for it. If we don't specify any constraints, T
can be anything you want. An example of a generic type that can be anything you want is Array
. Let's look at two identical ways to define an empty array in Swift:
let words1 = [String]()
let words2 = Array<String>()
Notice that the second way uses the type name Array
followed by <String>
. This informs the compiler that we're defining an array where the type of element is String
. Now let's try to imagine what the type definition for Array
might look like:
struct Array<T> {
// implementation code
}
This code declares a struct of type Array
that contains some type T
that is generic; it could be anything we want, as long as we specify it when creating an instance or when we use it as a type. In the earlier example, let words2 = Array<String>
we defined T
to be of type String
. Let's look at one more basic example before we move on:
struct SpecializedPrinter<T> {
func print(_ object: T) {
print(object)
}
}
This code declares a SpecializedPrinter
that's generic over T
and it has a function called print, that takes an object of type T
and prints it to the console. If you paste the above into a Playground, you can use this SpecializedPrinter
struct as follows:
let printer = SpecializedPrinter<String>()
printer.print("Hello!") // this is fine
printer.print(10) // this is not okay since T for this printer is String, not Int
Now that you know a bit about generics, I think we can write the first bit of code for the CacheBackedDataSource
object:
struct CacheBackedDataSource<T> {
func find(_ objectID: String, completion: @escaping (Result<T?, Error>) -> Void) {
}
}
We're not doing much here, but it's an important milestone in your journey to mastering generics in Swift. You have written a data source that claims to cache any type (T
) and will do an asynchronous lookup for an item based on a string identifier. The find(_:completion:)
function will call the completion block with a Result
object that contains an optional instance of T
, or an Error
object.
In the pseudo-code from earlier in this post, there were two properties:
let localStore: LocalStore
let remoteStore: RemoteStore
Since the caching layer should be as generic and flexible as possible, let's define LocalStore
and RemoteStore
as protocols. This will give us tons of flexibility, allowing any object to act as the local or remote store as long as they implement the appropriate functionality:
protocol LocalStore {
}
protocol RemoteStore {
}
And in these protocols, we will define methods to fetch the object we need, and in the local store, we'll define a method that persists an object.
protocol LocalStore {
func find(_ objectID: String, completion: @escaping (Result<T, Error>) -> Void)
func persist(_ object: T)
}
protocol RemoteStore {
func find(_ objectID: String, completion: @escaping (Result<T, Error>) -> Void)
}
Unfortunately, this doesn't work. Our protocols don't know what T
is since they're not generic. So how do we make these protocols generic? That's the topic of the next section.
Adding generics to protocols
While we can define a generic parameter on a struct by adding it to the type declaration between <>
, just like we did for struct CacheBackedDataSource<T>
, this is not allowed for protocols. If you want to have a protocol with a generic parameter, you need to declare the generic type as an associatedtype
on the protocol. An associatedtype
does not have to be implemented as a generic on objects that implement the protocol. I will demonstrate this shortly. For now, let's fix the local and remote store protocols so you can see associatedtype
in action:
protocol LocalStore {
associatedtype StoredObject
func find(_ objectID: String, completion: @escaping (Result<StoredObject, Error>) -> Void)
func persist(_ object: StoredObject)
}
protocol RemoteStore {
associatedtype TargetObject
func find(_ objectID: String, completion: @escaping (Result<TargetObject, Error>) -> Void)
}
Notice how we're not using a short name like T
here. This is because the associated type does not necessarily have to be generic, and we want the purpose of this type to be communicated a bit better than we typically do when you're defining a generic parameter on a struct. Let's create two structs that we conform to LocalStore
and RemoteStore
to see how associatedtype
works in the context of objects that conform to our protocols.
struct ArrayBackedUserStore: LocalStore {
func find(_ objectID: String, completion: @escaping (Result<User, Error>) -> Void) {
}
func persist(_ object: User) {
}
}
struct RemoteUserStore: RemoteStore {
func find(_ objectID: String, completion: @escaping (Result<User, Error>) -> Void) {
}
}
All that's needed to implement the protocol's associatedtype
in this example is to use the same type in all places where the protocol uses its associated type. An alternative that's a bit more verbose would be to define a typealias
inside of a conforming object and use the protocols associatedtype
where we currently use the User
object. An example of this would look like this:
struct RemoteUserStore: RemoteStore {
typealias TargetObject = User
func find(_ objectID: String, completion: @escaping (Result<TargetObject, Error>) -> Void) {
}
}
I prefer the former way where we use User
in place of TargetObject
, it's just easier to read in my opinion.
Since we're dealing with data that comes from a remote server in RemoteUserStore
, it would be quite convenient to constraint the value of TargetObject
to only allow Decodable
types to be used in place of TargetObject
. We can do this as follows:
protocol RemoteStore {
associatedtype TargetObject: Decodable
func find(_ objectID: String, completion: @escaping (Result<TargetObject, Error>) -> Void)
}
If we try to use User
in place of TargetObject
and User
isn't Decodable
, we're shown the following error by the compiler:
candidate would match and infer 'TargetObject' = 'User' if 'User' conformed to 'Decodable'
func find(_ objectID: String, completion: @escaping (Result<User, Error>) -> Void) {
We now have the following code prepared for the CacheBackedDataSource
and the local and remote store protocols:
struct CacheBackedDataSource<T> {
func find(_ objectID: String, completion: @escaping (Result<T, Error>) -> Void) {
}
}
protocol LocalStore {
associatedtype StoredObject
func find(_ objectID: String, completion: @escaping (Result<StoredObject, Error>) -> Void)
func persist(_ object: StoredObject)
}
protocol RemoteStore {
associatedtype TargetObject: Decodable
func find(_ objectID: String, completion: @escaping (Result<TargetObject, Error>) -> Void)
}
Let's add some properties for the local and remote store to the CacheBackedDataStore
:
struct CacheBackedDataSource<T> {
let localStore: LocalStore
let remoteStore: RemoteStore
func find(_ objectID: String, completion: @escaping (Result<T, Error>) -> Void) {
}
}
Unfortunately, this won't compile. The following errors are thrown by the Swift compiler:
error: protocol 'LocalStore' can only be used as a generic constraint because it has Self or associated type requirements
let localStore: LocalStore
^
error: protocol 'RemoteStore' can only be used as a generic constraint because it has Self or associated type requirements
let remoteStore: RemoteStore
^
Let's see how we can fix this error in the next session.
Using a protocol with associated type requirements as a generic constraint
Before we look deeper into the compiler errors we're currently stuck with I want to quickly recap what we've got prepared so far. Because even though the code doesn't compile, it's quite impressive already. We have a generic data source that can retrieve any object T
.
We also have a protocol for a local store that caches any type we want, the adopter of the protocol can decide what type is cached exactly. All that matters is that the object that implements the protocol has a find
method that performs a lookup based on an identifier and invokes a callback with a Result
object. It also has a persist
method that is expected to store objects that have the same type as the type object that the local store can fetch.
Lastly, we have a protocol for a remote store that fetches any kind of object, as long as it conforms to Decodable
. Similar to how local store works, the implementer of the RemoteStore
can decide what the type of the TargetObject
will be.
This is really powerful stuff and if the above is a bit confusing to you, that's okay. It's not simple or straightforward even though the code looks fairly simple. Try following along with the code we've written so far, re-read what you've learned and maybe take a short break to let it sink in. I'm sure it will eventually.
In order to use the local and remote store protocols as types on the CacheBackedStore
, we need to add generic parameters to the CacheBackedStore
, and constrain these parameters so they have to implement our protocols. Replace your current implementation of CacheBackedStore
with the following:
struct CacheBackedDataSource<Local: LocalStore, Remote: RemoteStore> {
private let localStore: Local
private let remoteStore: Remote
func find(_ objectID: String, completion: @escaping (Result<Local.StoredObject, Error>) -> Void) {
}
}
The declaration of CacheBackedStore
now has two generic parameters, Local
and Remote
. Each has to conform to its respective protocol. This means that the localStore
and remoteStore
should not be of type LocalStore
and RemoteStore
. Instead, they should be of type Local
and Remote
. Note that Result<T, Error>
has been replaced with Result<Local.StoredObject, Error>
. The find
method now uses whatever type of object the LocalStore
stores as the type for its Result
. This is really powerful because the underlying store now dictates the type of objects returned by the data source object.
There's still one problem though. Nothing prevents us from locally storing something that's completely incompatible with the remote store. Luckily we can apply constraints to the generic parameters of our struct. Update the declaration of CacheBackedDataSource
as follows:
struct CacheBackedDataSource<Local: LocalStore, Remote: RemoteStore> where Local.StoredObject == Remote.TargetObject
We can now only create CacheBackedDataSource
objects that use the same type of object for the local and remote stores. Before I show you how to create an instance of CacheBackedDataSource
, let's implement the find
method first:
func find(_ objectID: String, completion: @escaping (Result<Local.StoredObject, Error>) -> Void) {
localStore.find(objectID) { result in
do {
let object = try result.get()
completion(.success(object))
} catch {
self.remoteStore.find(objectID) { result in
do {
let object = try result.get()
self.localStore.persist(object)
completion(.success(object))
} catch {
completion(.failure(error))
}
}
}
}
}
The find
method works by calling find
on the local store. If the requested object is found, then the callback is invoked and the result is passed back to the caller. If an error occurred, for example, because the object wasn't found, the remote store is used. If the remote store finds the requested object, it's persisted in the local store and the result is passed back to the caller. If the object wasn't found or an error occurred in the remote store, we invoke the completion closure with the received error.
Note that this setup is extremely flexible. The implementation of CacheBackedDataSource
doesn't care what it's caching. It only knows how to use a local store with a fallback to a remote store. Pretty awesome, right? Let's wrap this up by creating an instance of the CacheBackedDataSource
:
let localUserStore = ArrayBackedUserStore()
let remoteUserStore = RemoteUserStore()
let cache = CacheBackedDataSource(localStore: localUserStore, remoteStore: remoteUserStore)
cache.find("someObjectId") { (result: Result<User, Error>) in
}
All you need to do is create instances of your stores, and supply them to the cache. You can then call find
on your cache and the compiler is able to understand that the result
object that's passed to the completion closure for find
is a Result<User, Error>
.
Take a look at the pseudo-code I showed you at the beginning of this post. It's very close to what we ended up implementing, and it's just as powerful as we imagined! If you've been following along, try to create some CacheBackedDataSource
objects for other types. It should be fairly straightforward.
In summary
You have learned so much in this blog post. I wouldn't be surprised if you have to read it one or two more times to make complete sense out of all these generics and type constraints. And we haven't even covered all of it! Generics are an unbelievably powerful and complex feature of the Swift language but I hope that I have been able to help you make some sense of them. Overall, you now know that adding <T>
to an object's declaration adds a generic parameter, which means that anytime you use T
inside of that object, it's whatever the user of that object decided it to be.
You also learned that you can add associatedtype
to a protocol to have it support generic types. And to top it off, you learned how you can use a protocol that has an associated type as a constraint for an object's generic parameter for maximum flexibility. If your brain hurts a bit after reading all this then again, don't worry. This stuff is hard, confusing, weird and complex. And if you have any questions, comments or need somebody to talk to because you feel lost now, don't hesitate to reach out on Twitter!
Top comments (0)