Swift Combine chaining .mapError()
I agree with iamtimmo that you don't need .subscribe(on:)
. I also think this method is the wrong place for .receive(on:)
, because nothing in the method requires the main thread. If you have code elsewhere that subscribes to this publisher and wants results on the main thread, then that is where you should use the receive(on:)
operator. I'm going to omit both .subscribe(on:)
and .receive(on:)
in this answer.
Anyway, let's address your questions.
- Is there better way to map errors (and replace line above) using chaining in Combine?
“Better” is subjective. The problem you're trying to solve here is that you only want to apply that mapError
to an error produced by the decode(type:decoder:)
operator. You can do that using the flatMap
operator to create a mini-pipeline inside the full pipeline:
return session.dataTaskPublisher(for: url)
.mapError { NetworkError.request(underlyingError: $0) }
.map { $0.data }
.flatMap {
Just($0)
.decode(type: ReposList.self, decoder: JSONDecoder())
.mapError { .unableToDecode(underlyingError: $0) } }
.eraseToAnyPublisher()
Is this “better”? Meh.
You could extract the mini-pipeline into a new version of decode
:
extension Publisher {
func decode<Item, Coder>(type: Item.Type, decoder: Coder, errorTransform: @escaping (Error) -> Failure) -> Publishers.FlatMap<Publishers.MapError<Publishers.Decode<Just<Self.Output>, Item, Coder>, Self.Failure>, Self> where Item : Decodable, Coder : TopLevelDecoder, Self.Output == Coder.Input {
return flatMap {
Just($0)
.decode(type: type, decoder: decoder)
.mapError { errorTransform($0) }
}
}
}
And then use it like this:
return session.dataTaskPublisher(for: url)
.mapError { NetworkError.request(underlyingError: $0) }
.map { $0.data }
.decode(
type: ReposList.self,
decoder: JSONDecoder(),
errorTransform: { .unableToDecode(underlyingError: $0) })
.eraseToAnyPublisher()
- Is there any way to include first
guard let
withFail(error:)
in chain?
Yes, but again it's not clear that doing so is better. In this case, the transformation of query
into a URL
is not asynchronous, so there's little reason to use Combine. But if you really want to do it, here's a way:
return Just(query)
.setFailureType(to: NetworkError.self)
.map { URL(string: searchUrl + $0).map { Result.success($0) } ?? Result.failure(.badUrl) }
.flatMap { $0.publisher }
.flatMap {
session.dataTaskPublisher(for: $0)
.mapError { .request(underlyingError: $0) } }
.map { $0.data }
.decode(
type: ReposList.self,
decoder: JSONDecoder(),
errorTransform: { .unableToDecode(underlyingError: $0) })
.eraseToAnyPublisher()
This is convoluted because Combine doesn't have any operators that can turn a normal output or completion into a typed failure. It has tryMap
and similar, but those all produce a Failure
type of Error
instead of anything more specific.
We can write an operator that turns an empty stream into a specific error:
extension Publisher where Failure == Never {
func replaceEmpty<NewFailure: Error>(withFailure failure: NewFailure) -> Publishers.FlatMap<Result<Self.Output, NewFailure>.Publisher, Publishers.ReplaceEmpty<Publishers.Map<Publishers.SetFailureType<Self, NewFailure>, Result<Self.Output, NewFailure>>>> {
return self
.setFailureType(to: NewFailure.self)
.map { Result<Output, NewFailure>.success($0) }
.replaceEmpty(with: Result<Output, NewFailure>.failure(failure))
.flatMap { $0.publisher }
}
}
Now we can use compactMap
instead of map
to turn query
into a URL
, producing an empty stream if we can't create a URL
, and use our new operator to replace the empty stream with the .badUrl
error:
return Just(query)
.compactMap { URL(string: searchUrl + $0) }
.replaceEmpty(withFailure: .badUrl)
.flatMap {
session.dataTaskPublisher(for: $0)
.mapError { .request(underlyingError: $0) } }
.map { $0.data }
.decode(
type: ReposList.self,
decoder: JSONDecoder(),
errorTransform: { .unableToDecode(underlyingError: $0) })
.eraseToAnyPublisher()
I don't think your approach is unreasonable. A benefit of the first mapError()
(at // 1
) is that you don't need to know much about the possible errors from the request.
return session.dataTaskPublisher(for: url)
.mapError { NetworkError.request(underlyingError: $0) } // 1
.map { $0.data }
.decode(type: ReposList.self, decoder: JSONDecoder())
.mapError { $0 as? NetworkError ?? .unableToDecode(underlyingError: $0) }
.subscribe(on: DispatchQueue.global()) // 2 - not needed
.receive(on: DispatchQueue.main)
.eraseToAnyPublisher()
}
I don't think you need the subscribe(on:)
at // 2
, since URLSession.DataTaskPublisher starts on a background thread already. The subsequent receive(on:)
is required.
An alternative approach would be to run through the "happy path" first and map all of the errors later, as in the following. You'll need to understand which errors come from which publishers/operators to correctly map to your NetworkError enum.
return session.dataTaskPublisher(for: url)
.map { $0.data }
.decode(type: ReposList.self, decoder: JSONDecoder())
.mapError({ error -> NetworkError in
// map all the errors here
})
.receive(on: DispatchQueue.main)
.eraseToAnyPublisher()
To handle your second question, you can use the tryMap()
and flatMap()
to map your query
into a URL and then into a URLSession.DataTaskPublisher
instance. I haven't tested this particular code, but a solution would be along these lines.
Just(query)
.tryMap({ query in
guard let url = URL(string: searchUrl + query) else { throw NetworkError.badUrl }
return url
})
.flatMap({ url in
URLSession.shared.dataTaskPublisher(for: url)
.mapError { $0 as Error }
})
.map { $0.data }
//
// ... operators from the previous examples
//
.eraseToPublisher()