🎉 End-of-year Sale! Save 25% when you subscribe today.

The Combine Framework and Effects: Part 2

Episode #81 • Nov 18, 2019 • Free Episode

Now that we’ve explored the Combine framework and identified its correspondence with the Effect type, let’s refactor our architecture to take full advantage of it.

Previous episode
The Combine Framework and Effects: Part 2
Next episode
FreeThis episode is free for everyone.

Subscribe to Point-Free

Access all past and future episodes when you become a subscriber.

See plans and pricing

Already a subscriber? Log in

Introduction

So that’s the basics of the Combine framework. There is a ton more to say, but we’ve learned just enough to be dangerous. And we’ve learned the correspondence between Combine and the Effect type.

To recap: in the Combine world we have publishers and in the Effect world we have Effect, and in the Combine world we have subscribers, and in the Effect world we have run. Luckily, Combine comes with a bunch of bells and whistles, though, like sink, which works just like run on Effect. And further, Combine comes with Future, which are created a lot like Effects, but with the caveats that they are eager and need to be wrapped in a Deferred publisher, and that they can only receive a single value, which means we must use another Combine concept, subjects, to simply set up more long-living event streams.

That’s the basic correspondence, so the question is can we refactor the Composable Architecture that we have been building to leverage Combine’s functionality rather than building it ourselves from scratch?

Let’s refactor away the Effect type so that we can leverage Combine and avoid reinventing the wheel.

Effect as a Combine publisher

The refactor is pretty straightforward, with only a few twists and turns along the way. Perhaps the simplest way to start is to simply comment out our Effect type and replace it with a type alias pointing to a publisher. However, since Publisher is a protocol we can’t use it directly, we have to actually pick a concrete publisher conformance. We could use the AnyPublisher concrete conformance that Apple gives us, but it will be handy to have our own named type so that we can add effect-specific helpers and extensions to it without polluting AnyPublisher.

So, let’s try to create a publisher conformance from scratch:

public struct Effect: Publisher {
}

Type ‘Effect’ does not conform to protocol ‘Publisher’

We need to implement some conformances, so let’s see what’s required:

public struct Effect: Publisher {
  public typealias Output = <#type#>
  public typealias Failure = <#type#>

}

We need an Output and Failure type. Remember that an effect’s only purpose is to ultimately produce an action that is fed back into the store. Even if the effect errors in some way, like a network request when the device is offline, it still needs to produce an action. So the effect could put a Result value inside an action to denote the failure, but the effect publisher itself cannot fail. This means we should use Never for the publisher’s Failure type:

public struct Effect: Publisher {
  public typealias Output = <#type#>
  public typealias Failure = Never

}

The output on the other hand is something that users of this type need to determine, so it should be a generic:

public struct Effect<Output>: Publisher {
  public typealias Failure = Never

}

With that done there’s one more requirement:

public func receive<S: Subscriber>(
  subscriber: S
) where Failure == S.Failure, Output == S.Input {
  <#code#>
}

This is called when a subscriber attaches to this publisher, and this is where we need to do our work to send this subscriber values. However, we don’t actually want to do any custom work inside here. We just want to serve as a wrapper around publishers, much like AnyPublisher is.

So let’s just hold onto an AnyPublisher under the hood and delegate to it:

let publisher: AnyPublisher<Output, Failure>

public func receive<S: Subscriber>(subscriber: S)
where Failure == S.Failure, Output == S.Input {
  self.publisher.receive(subscriber: subscriber)
}

Right now the Effect publisher is really no different from the AnyPublisher, other than its failure has been specialized to Never. However, by having our own type we are going to have more control over how it’s transformed, which is something we will see very soon.

This has created a number of compiler errors, the first of which is:

effect.run(self.send)

Value of type ‘Effect>Action>’ has no member ‘run’

Instead of running our effect we need to call sink on it, and we can still pass the send method to the receive block:

effect.sink(receiveValue: self.send)

Now we have a warning:

Result of call to ‘sink(receiveValue:)’ is unused

We technically already have an instance variable for a cancellable in our store:

private var cancellable: Cancellable?

We added this back in our episodes on modularity when we discussed the concept of viewing a store. We showed that it’s possible to transform a store that operates on global state and actions into a store that operates on local state and actions. However, in order for changes to the global store to be reflected in the local store we had to subscribe to the global store and replay those changes in the local store.

So we don’t want to repurpose this cancellable value, it has a very important responsibility right now. In fact, maybe we should go ahead and rename it to better reflect its purpose:

private var viewCancellable: Cancellable?

We need to keep track of these effect cancellables separately from this view cancellable. We could introduce a new instance variable:

private var effectCancellable: Cancellable?

But really we need a bunch of cancellables. Each reducer can return an array of effects, and so we could potentially be dealing with many effects each time we send an action. So, let’s upgrade to an array:

private var effectCancellables: [Cancellable] = []

And now we just need to retain the cancellable from calling sink so that it lives long for our effect to do its job:

effectCancellables.append(
  effect.sink(receiveValue: self.send)
)

However, this isn’t quite right. This effectCancellables array is going to continue to grow as the application progresses. We are never removing cancellables from this array. So, we need some way to know when an effect finishes and then we should remove its cancellable from the array.

This is possible using a different overload of sink that allows us to tap into the completion event of the publisher:

effectCancellables.append(
  effect.sink(
    receiveCompletion: { _ in
      <#code#>
    },
    receiveValue: self.send
  )
)

Inside this receiveCompletion we could try to remove the cancellable from the array, but we don’t actually have access to it here. We have a bit of a chicken-and-egg problem, where the cancellable is created by calling sink but we need access to the cancellable from inside one of the closures that defines the sink.

To work around this we need to extract out the cancellable into an implicitly unwrapped optional, which allows us to get a variable for a type before it holds a value, and then later we get to assign the variable.

It looks like this:

var effectCancellable: Cancellable!
effectCancellable = effect.sink(
  receiveCompletion: { _ in

  },
  receiveValue: self.send
)
self.effectCancellables.append(
)

Now we have access to the array of cancellables inside receiveCompletion, and so we can try to remove the cancellable when the publisher finishes. There are many ways to remove a value from an array, such as removing the first or last or at a particular index, but to remove a particular value from an array we have to search the entire array and find the ones we want to remove:

self.effectCancellables.removeAll(where: { $0 == effectCancellable })

But this won’t work because the Cancellable protocol does not inherit from the Equatable protocol, and so we cannot do this equality check. However, the AnyCancellable wrapper is a class, and so is equatable thanks to object identity.

So let’s upgrade everything to AnyCancellable. The property:

private var effectCancellables: [AnyCancellable] = []
…
func send(_ action: Action) {
  …
  var effectCancellable: AnyCancellable!
  effectCancellable = effect.sink(
    receiveCompletion: { _ in
      self.effectCancellables.removeAll(
        where: { $0 == effectCancellable }
      )
    },
    receiveValue: self.send
  )

And finally append that cancellable to our array.

self.effectCancellables.append(effectCancellable)

One small gotcha is that we have retained self in the cancellable’s completion handler, and the cancellable is being retained on self:

effectCancellable = effect.sink(
  receiveCompletion: { _ in
    self.effectCancellables.removeAll(
      where: { $0 == effectCancellable }
    )
  },

This means we have a retain cycle, which we can break by using weak self instead:

receiveCompletion: { [weak self] _ in
  self?.effectCancellables.removeAll(where: {
    $0 == effectCancellable
  })
}

And that technically fixes this portion of our architecture. However, there’s a quick improvement we can make to this. The AnyCancellable class conforms to the Hashable protocol, which means we can use them in a set, which will give us a very easy way to remove them:

private var effectCancellables: Set<AnyCancellable> = []
…
func send(_ action: Action) {
  …
  var effectCancellable: AnyCancellable!
  effectCancellable = effect.sink(
    receiveCompletion: { [weak self] _ in
      self?.effectCancellables.remove(effectCancellable)
    },
    receiveValue: self.send
  )
  effectCancellables.insert(effectCancellable)

Pulling back reducers with publishers

We have now introduced our own publisher for our architecture, called Effect. And we have upgraded the Store to handle these effects in its send method, including all of the complexities around handling their associated cancellables and completion.

The next error is in the pullback method, so let’s remind ourselves what this does. It takes reducers that work on very local state and actions and can pull them back to work on more global state and actions. Where effects fit into this is when a local reducer produces a local effect, we need to transform that into a more global effect. A local effect can return a local action back into the store, so we need to wrap that local action in a more global one.

So let’s see what that work looks like using our new publisher.

The pullback function currently has a compile error.

return localEffects.map { localEffect in
  Effect { callback in
    localEffect.run { localAction in
      var globalAction = globalAction
      globalAction[keyPath: action] = localAction
      callback(globalAction)
    }
  }
}

Value of type ‘Effect<LocalAction>’ has no member ‘run’

We are trying to call the run method, which existed on our old effect type, but no longer exists on our publisher. Technically, we could call sink here instead:

return localEffects.map { localEffect in
  Effect { callback in
    localEffect.sink { localAction in
      var globalAction = globalAction
      globalAction[keyPath: action] = localAction
      callback(globalAction)
    }
  }
}

But sink returns an AnyCancellable, which is something we’d need to track, and it’s not even clear how we would do so because we are in a pullback, which is in pure function reducer world with no Store in sight to manage these details.

We’re also trying to create an Effect with a callback closure, and we no longer have that interface at our disposal.

If we think about what this code is really doing, it simply trying to transform an effect that can produce local actions into an effect that can produce global actions. This is precisely what the map operation allows us to do on generic types, and lucky for us the Publisher type supports a map operation!

We can take advantage of the fact that publishers have map by replacing all of this manual effect conversion code with a simple map:

localEffect.map { localAction in
  var globalAction = globalAction
  globalAction[keyPath: action] = localAction
  callback(globalAction)
}

However, instead of feeding the globalAction to the callback we can now simply return that value to the map:

localEffect.map { localAction in
  var globalAction = globalAction
  globalAction[keyPath: action] = localAction
  return globalAction
}

Now this is basically right, but it’s not compiling. The error message isn’t great, but if we add a return type to our closure it will get a little better:

localEffect.map { localAction -> GlobalAction in
  var globalAction = globalAction
  globalAction[keyPath: action] = localAction
  return globalAction
}

Cannot convert value of type ‘Publishers.Map<Effect<LocalAction>, GlobalAction>’ to closure result type ‘Effect<GlobalAction>’

This is a long error message, but it is correctly describing what is wrong. We expect that the type returned from the map is an Effect<GlobalAction> since we need to return an array of those things from this function. But it didn’t find that type, instead it found this strange Publishers.Map<Effect<LocalAction>, GlobalAction> type, which is a mouthful to say.

This brings us to yet another important lesson when dealing with the Combine framework. Publishers come with many operations, things like map, zip, flatMap, filter and more. But they don’t return the exact same type of publisher they acted upon, they only return something conforming to the Publisher protocol. This is due to a limitation of Swift’s type system, which although powerful it cannot express the idea of the map operation on a publisher returning a publisher of the same type, such as map on Effect returning another Effect.

The type returned by the map operation is something known as a Map publisher.

Publishers.Map<Effect<LocalAction>, GlobalAction>

The first generic refers to the publisher that we are mapping on and the second generic refers to the new value that the mapped publisher can emit. So this says that we have mapped on an Effect to transform its values of LocalActions into GlobalActions.

We somehow need to convert this publisher into a plain old Effect.

This is the exact same problem that Combine’s AnyPublisher has, and the way they deal with it is that they allow any publisher to transform itself into an AnyPublisher:

.eraseToAnyPublisher()

We basically want this functionality, but for our Effect type instead. It is possible to make an eraseToEffect method. It is expressed as an extension on the Publisher protocol so that any publisher can be erased, but it should only work on publishers that can’t fail:

extension Publisher where Failure == Never {
  public func eraseToEffect() -> Effect<Output> {

  }
}

To implement this method all we have to do is erase ourselves to the AnyPublisher and then wrap that it in the effect type:

extension Publisher where Failure == Never {
  public func eraseToEffect() -> Effect<Output> {
    Effect(publisher: self.eraseToAnyPublisher())
  }
}

That’s all there is to it, but also it seems strange to do so much wrapping in eraser types. We aren’t doing this for no reason, it will be handy soon, but for now just take our word for it.

Going back to our pullback operation, we can now erase the Map publisher to be the Effect publisher:

localEffect.map { localAction -> GlobalAction in
  var globalAction = globalAction
  globalAction[keyPath: action] = localAction
  return globalAction
}
.eraseToEffect()

And this finally makes the compiler happy.

This eraseToEffect dance is unfortunately going to be common practice when it comes to dealing with Combine. Since our reducers are defined in terms of Effect, and any time you transform it you get something that is not an Effect, we are many times going to have to invoke eraseToEffect on our effects before returning them. It’s annoying, but necessary.

It’s pretty cool, though, that publishers have a map operation. We first talked about map on just the 13th episode of Point-Free…seems so long ago! But in that episode we showed that the map operation is a very universal thing, and there isn’t a lot of choice when it comes to defining it. Back them we remarked how the Swift standard library comes with two maps: one defined on arrays and one on optionals. Since then the Result type was introduced and it also has a map operation. And now we even have a map operation in the Combine framework. This means that in the Apple ecosystem alone there are 4 distinct map operations!

Finishing the architecture refactor

The next error we have in this file is in our logging higher-order reducer. Here we are trying to return an effect that wraps a few print statements.

return [
  Effect { _ in
    print("Action: \(action)")
    print("Value:")
    dump(newValue)
    print("---")
  }
] + effects

Cannot convert value of type ‘() -> ()’ to expected argument type ’AnyPublisher<, Never>’

We no longer have the callback closure-based initializer on Effect, but we have previously encountered a similar solution in the world of Combine.

We want to return a publisher that encapsulates the execution of the work passed in, but we want to make sure that we don’t run that work until the publisher is subscribed to. As we saw before, a simple way to do this is to wrap it in a Deferred publisher:

return [
  Deferred { _ in
    print("Action: \(action)")
    print("Value:")
    dump(newValue)
    print("---")
  }
] + effects

Contextual closure type ‘() -> _’ expects 0 arguments, but 1 was used in closure body

Now the Deferred requires us to return a publisher, but we don’t want to actually do anything since this is a fire-and-forget effect. Luckily there’s a special publisher called Empty that simply never emits any values, and we can even make it complete immediately:

return [
  Deferred { _ in
    print("Action: \(action)")
    print("Value:")
    dump(newValue)
    print("---")
    return Empty(completeImmediately: true)
  }
] + effects

This isn’t compiling because Swift needs a little help with type inference:

return [
  Deferred { () -> Empty<Action, Never> in
    print("Action: \(action)")
    print("Value:")
    dump(newValue)
    print("---")
    return Empty(completeImmediately: true)
  }
] + effects

And now we’re getting a much better error message.

Cannot convert value of type ‘Deferred<Empty<Action, Never>>’ to expected element type ‘Effect<Action>’

So we need to erase this to the Effect type:

return [
  Deferred { () -> Empty<Action, Never> in
    print("Action: \(action)")
    print("Value:")
    dump(newValue)
    print("---")
    return Empty(completeImmediately: true)
  }
  .eraseToEffect()
] + effects

The compiler is happy, but the work we needed to do was a little gross. Let’s introduce a helper that does this same work but in a nicer way. We will create it as a static function on effect that takes a void-to-void closure and returns an effect:

extension Effect {
  public static func fireAndForget(
    work: @escaping () -> Void
  ) -> Effect {
  }
}

And we can capture that work we did in our helper.

extension Effect {
  public static func fireAndForget(
    work: @escaping () -> Void
  ) -> Effect {
    return Deferred { () -> Empty<Output, Never> in
      work()
      return Empty(completeImmediately: true)
    }
    .eraseToEffect()
  }
}

And now this is perfect to use for our printing effect:

return [
  .fireAndForget {
    print("Action: \(action)")
    print("Value:")
    dump(newValue)
    print("---")
  }
] + effects

This is the main reason we introduced the custom Effect publisher conformance instead of just relying on AnyPublisher. It will give us opportunities to make transforming publishers a bit nicer at the call site.

The ComposableArchitecture module still isn’t compiling yet because we have this file of effects that we extracted out last time. It contains some handy base effects for network requests, JSON decoding, and forcing values to be delivered on a particular queue.

All of these effects are great, but the Combine framework actually provides API’s that accomplish all of these tasks. We even named our methods after the Combine framework’s names.

So we don’t actually need to fix these compile errors, we are simply going to comment everything out because we can just lean on Combine for these effects instead of recreating them ourselves.

And now the ComposableArchitecture module is compiling.

Refactoring synchronous effects

Now that the ComposableArchitecture module is finally building, we are set up to start refactoring our application to use this new style of side effects. We have a few modules to get building, so let’s see how that goes.

Let’s attack them one at a time.

We can start with the simplest one, PrimeModal. If we switch to that target and build we see that magically everything still builds. This is because the primeModalReducer doesn’t do any side effects, and so it doesn’t care if we change the definition of the effect type.

The next simplest module is the FavoritePrimes module, and it fails to compile since we construct some effects for saving and loading favorite primes.

The saveEffect is a fire-and-forget effect that is currently using our old callback-style effect. We can simply swap this out for our new fireAndForget helper to get this part compiling:

private func saveEffect(
  favoritePrimes: [Int]
) -> Effect<FavoritePrimesAction> {
  return .fireAndForget {
    …
  }
}

The loadEffect is a synchronous effect that needs to feed a result back into the system, and it is also currently using the old callback-style effect. Similar to how we created the fireAndForget helper on Effect we can create a synchronous effect helper.

It’s just a function that takes some work that produces an action, and we want to return an effect that wraps that work.

extension Effect {
  public static func sync(work: @escaping () -> Output) -> Effect {
  }
}

Again we want to wrap the work in a Deferred so that we don’t execute the work until a subscription is made:

extension Effect {
  public static func sync(work: @escaping () -> Output) -> Effect {
    return Deferred {
    }
  }
}

And then in here we want to return a publisher that holds the result of the work. Luckily there’s yet another concrete publisher that Combine gives us that represents an emission of a single value, and it’s called Just:

extension Effect {
  public static func sync(work: @escaping () -> Output) -> Effect {
    return Deferred {
      Just(work())
    }
  }
}

And of course we always need to erase it to the Effect type:

extension Effect {
  public static func sync(work: @escaping () -> Output?) -> Effect {
    return Deferred {
      Just(work())
    }
    .eraseToEffect()
  }
}

And this finally gets us into compiling order. And we can now use it to define our loadEffect:

private let loadEffect = Effect<FavoritePrimesAction>.sync {
  guard …
  else { return }
  return .loadedFavoritePrimes(favoritePrimes)
}

This isn’t quite right because the Effect<FavoritePrimesAction>.sync closure must return an actual FavoritePrimesAction, but here we are returning an optional. We are returning nil in the guard because something failed, and we just don’t want the effect to emit in that situation. One thing we could do is change this to an Effect<FavoritePrimesAction?>.

private let loadEffect = Effect<FavoritePrimesAction?>.sync {
  guard …
  else { return nil }
  return .loadedFavoritePrimes(favoritePrimes)
}

But our reducer can’t yet use loadEffect directly because it needs to work with effects that return non-optional actions. In order to get rid of the nil values we can use compactMap:

loadEffect
  .compactMap { $0 }

And finally, we need to re-erase to an effect.

.compactMap { $0 }
.eraseToEffect()

This compactMap operators works on publishers just like it does on arrays, it simply filters out the nil values leaving behind only the honest ones.

And just like that the FavoritePrimes module is building. Now unfortunately our entire application still isn’t building, but fortunately due to our efforts in modularizing this app we have a playground that allows us to run this screen in full isolation before we even get everything else building. Let’s hop over and give it a spin.

We can delete a favorite prime and save the change…

Fatal error: Unexpectedly found nil while implicitly unwrapping an Optional value: file PrimeTime/ComposableArchitecture/ComposableArchitecture.swift, line 80

Ouch! Looks we have a crash in our ComposableArchitecture module. We must have overlooked something. If we go to line 53 we will see that we are crashing on this line:

self?.effectCancellables.remove(effectCancellable)

This can only crash if effectCancellable is nil because we are using an implicitly unwrapped optional, which always means we have the possibility of crashing no matter how careful we are.

The problem here is that we were assuming that sink will return before any of these closures were invoked so that we could get a hold of the cancellable. It turns out that for publishers that complete immediately the closures are invoked right away, before sink even returns, and that is why we are crashing. One thing we could do is make effectCancellable an optional so that we could safely handle it:

var effectCancellable: AnyCancellable?
effectCancellable = effect.sink(
  receiveCompletion: { [weak self] _ in
    guard let effectCancellable = effectCancellable else { return }
    self?.effectCancellables.remove(effectCancellable)
  },
  receiveValue: self.send
)
if let effectCancellable = effectCancellable {
  effectCancellables.insert(effectCancellable)
}

Now we just need to check if effectCancellable is present before removing it from the set and before inserting it into the set. However, this isn’t going to be correct. If a publisher completes immediately it will be inserted into the set, but we’ll never get the chance to remove it because the receiveCompletion was already executed, before we even inserted.

So, we need to be able to understand if the receiveCompletion closure was already called by the time we get to the insert line so that we can just skip it. To do that we yet another mutable variable to track that state:

var effectCancellable: AnyCancellable?
var didComplete = false
effectCancellable = effect.sink(
  receiveCompletion: { [weak self] _ in
    didComplete = true
    guard let effectCancellable = effectCancellable else { return }
    self?.effectCancellables.remove(effectCancellable)
  },
  receiveValue: self.send
)
if !didComplete, let effectCancellable = effectCancellable {
  effectCancellables.insert(effectCancellable)
}

And now we will only insert the cancellable into the set if the publisher does not complete immediately. If we hop back over to our playground we will see that now everything works again.

Refactoring asynchronous effects

Alright, both FavoritePrimes module effects are now fully powered by combine.

We even got a couple of helpers out of it, including fireAndForget and sync.

It’s also cool to see our modularization efforts pay off: we didn’t have to get the entire application building in order to run this screen in isolation. This is the power of modularization! We are free to do broad refactors incrementally with feedback along the way for each screen.

We have one more screen, though: the Counter module, which has the complicated async effect which makes an API request. Let’s see what all needs to change to upgrade this to Combine.

The first error we will look at is in the wolframAlpha function, which currently is using some of the effect helpers we built last time, like this dataTask and decode helper:

return dataTask(with: components.url(relativeTo: nil)!)
  .decode(as: WolframAlphaResult.self)

Fortunately Combine provides all of these helpers for us, but unfortunately it also takes a bit more work to make use of them. We can start by using Foundation’s new dataTaskPublisher method on URLSession to get a publisher that represents a network request:

URLSession.shared
  .dataTaskPublisher(for: components.url(relativeTo: nil)!)

Let’s assign this to a temporary variable so that we can understand what its type looks like:

let tmp = URLSession.shared
  .dataTaskPublisher(for: components.url(relativeTo: nil)!)

Its type is URLSession.DataTaskPublisher, which is some concrete publisher defined in Foundation. If we check its Output and Failure, we’ll find the following type aliases.

typealias Output = (data: Data, response: URLResponse)
…
typealias Failure = URLError

Remember that ultimately we need to return an Effect from this method, and so if we hit this with the eraseToAnyPublisher method we will get a better view into its actual type:

let tmp = URLSession.shared
  .dataTaskPublisher(for: components.url(relativeTo: nil)!)
  .eraseToAnyPublisher()
// AnyPublisher<
//   URLSession.DataTaskPublisher.Output,
//   URLSession.DataTaskPublisher.Failure
// >

We want to decode this data into our WolframAlphaResult model, and luckily Combine comes with a nice decoding helper:

let tmp = URLSession.shared
  .dataTaskPublisher(for: components.url(relativeTo: nil)!)
  .decode(type: WolframAlphaResult.self, decoder: JSONDecoder())
  .eraseToAnyPublisher()

However, this won’t compile because the .decode method is only defined on publishers whose output matches the thing we are trying to decode, which in this case is Data. So, before we can invoke decode we must map on the publisher to pluck out just the data from the tuple:

let tmp = URLSession.shared
  .dataTaskPublisher(for: components.url(relativeTo: nil)!)
  .map { data, _ in data }
  .decode(type: WolframAlphaResult.self, decoder: JSONDecoder())
  .eraseToAnyPublisher()

Now the type of our publisher is:

// AnyPublisher<
//   WolframAlphaResult,
//   Publishers.Decode<Upstream, Output, Coder>.Failure
// >

This is getting close. The main problem here is that our publisher can error, but our effect must return a publisher whose error type is Never, which means it can never error. The easiest way to transform a publisher that can error into one that can never error is using the replaceError method, which allows you to simply replace any error that occurs with an output value:

.replaceError(with: <#WolframAlphaResult#>)

The problem here is that we need to replace the error with an honest WolframAlphaResult, but we don’t have such a value. We would rather just use nil to represent that we couldn’t construct a WolframAlphaResult, but in order for that to our our publisher needs to have an output of an optional WolframAlphaResult.

The easiest way to do that is to instead decode an optional result and replace the error with nil:

.decode(type: WolframAlphaResult?.self, decoder: JSONDecoder())
.replaceError(with: nil)

And finally our chain of transformations has given us a publisher of type:

AnyPublisher<
  WolframAlphaResult?, Publishers.ReplaceError<Upstream>.Failure
>

This failure is difficult to read, but if we follow it along to its definition, it is Never, which is exactly what we need it to be to erase to an Effect and return it.

return URLSession.shared
  .dataTaskPublisher(for: components.url(relativeTo: nil)!)
  .map { data, _ in data }
  .decode(type: WolframAlphaResult.self, decoder: JSONDecoder())
  .map(Optional.some)
  .replaceError(with: nil)
  .eraseToEffect()

And now this effect is building. It’s intense, but at least the Combine framework gave us all the tools to get the job done.

The next error is in nthPrime, which is calling the wolframAlpha effect, but then mapping on it. As we’ve seen before, mapping on a publisher changes its type, and so we have to erase that change:

.eraseToEffect()

And now this function is happy. The only compiler error left is in our counter reducer where we use this effect. Since we are mapping on this publisher we yet again have to hit it with eraseToEffect:

nthPrime(state.count)
  .map(CounterAction.nthPrimeResponse)
  .receive(on: .main)
  .eraseToEffect()

However, it looks like things still aren’t building. The error message isn’t great, but the problem is here:

.receive(on: .main)

The receive(on:) method takes a Scheduler:

.receive(on: <#Scheduler#>)

Which is a protocol, not a concrete type. Both DispatchQueues and RunLoops conform to this protocol, so we just have to be more explicit with the type to use it:

.receive(on: DispatchQueue.main)

And now the module is finally building. Even better, the entire app target is building, which means we can run our app yet again. If we give it a spin we will see that everything still works the same, but now it’s running off of Combine instead of our own custom effect type. I don’t know about you, but the app just feels nicer now that I know Combine is powering it under the hood 😄.

What’s the point?

So that completes our Combine framework refactor of the toy application we have been building for the past many weeks. It was pretty straightforward once we had some knowledge of how Combine works.

We like to end every episode on Point-Free by asking “what’s the point?!”, which is our time to bring things down the earth and discuss why these ideas are important. However, this episode was pretty practical from the outset. We noticed that the Effect type looked eerily similar to what a lot of reactive programming libraries give us, such as ReactiveSwift, RxSwift and most recently Combine from Apple. So, we wondered if maybe we could lean on one of the frameworks instead of rolling our own type, and indeed we could. The most important qualities of our Effect type was that it could represent asynchronous work and that it was transformable. Well, all of these reactive libraries provide exactly this too, so might as well take advantage.

But perhaps the “meta-point” of this episode is that by having a very simple, focused type for accomplishing one thing, and doing so in a transformable way, you can later be open to lots of nice refactors in the future. The fact that we were able to completely swap our our effects type for another type, and everything continued to just work is pretty amazing. So amazing in fact that you might ask why can’t we abstract over the shape of effects so that a user of this library could bring their own effect type? Maybe they want to use ReactiveSwift or RxSwift instead of Combine, and maybe they want to just roll their own simple type.

Unfortunately Swift’s type system is not powerful enough to express this idea, but at least we can see the shape of such an abstraction and try to use it as inspiration for how we build our library.

So, that’s it for this episode. We took a little detour from our episodes building out the Composable Architecture so that we could address the strangeness of the Effect type from last time. But next week we pick up where we left off last time: testing! We want to show how testable this architecture is, even when effects are involved.

Until next time!


References

Downloads

Get started with our free plan

Our free plan includes 1 subscriber-only episode of your choice, access to 64 free episodes with transcripts and code samples, and weekly updates from our newsletter.

View plans and pricing