Swift Networking, or swift-networking
, is a library for building a flexible network stack inside a Swift application. It can be used on any Apple platform. You can use it to provide rich network features such as authenticating, de-duping, throttling and more.
Browse the documentation for main, 0.4.0.
Swift Networking is a Swift Package, which provides some core tools used to make HTTP network requests.
Its philosophy is centered around the idea that at a high level, clients send requests and await a response for each request.
This, like almost everything in programming, is a transformation of data types, and so lends itself to a functional programming style. With this in-mind, the library provides components which can be composed together to perform this transformation. All of the built-in components are well tested, with test helpers to make it easy to test your own custom components.
This library makes use of URLSession, as it provides the terminal component which is ultimately responsible for sending the request. Swift Networking abstracts this detail away, while also providing a lot more convenience than URLSession. Furthermore, this library provides many useful building blocks which are not provided by URLSession alone.
If we consider that when a client makes a network request, it is essentially a function: (Request) async throws -> Response
, which can be represented through a protocol called NetworkingComponent
. We can provide a conformance to this protocol on URLSession
, and use it make network requests. However, before the request is given to URLSession
there is opportunity to transform it further, perhaps it needs to be modified, or we wish to collect metrics, or maybe even return the Response from another system.
Taking this concept a bit further, we can consider a chain of components,
┌────────────────────────────────────────────┐
│ Network Stack │ ┌────────┐
┌─────────┐ │ ┌─────────┐ ┌─────────┐ ┌──────────┐ │ │ │
│ Request │─ ─ ─▶│ │ A │───▶│ B │───▶│ Terminal │ │─ ─ ─▶│ Server │
└─────────┘ │ └─────────┘ └─────────┘ └──────────┘ │ │ │
└────────────────────────────────────────────┘ └────────┘
in this diagram, the application (or client) has 3 components: A
, B
and Terminal
. When a request is provided to the overall stack, it starts in component A, which eventually gives it to component B, which eventually gives it to the terminal component, which is the last component and responsible for getting the response. The response from the server travels back through the stack.
To make this composition easy, the library provides a NetworkingModifier
protocol, which works exactly like ViewModifier
from SwiftUI. All networking components except the terminal one are actually modifiers, so that in addition to the Request
value, they also receive the "upstream" networking component. This is how it is possible for component A to pass the request onto component B etc.
Each of the built-in components provide public extensions on NetworkingComponent
, just as ViewModifiers in SwiftUI typically provide an API through extensions to the View
protocol. This results in a declarative network stack, something like this:
let network = URLSession.shared
.removeDuplicates()
.logged()
Updating our diagram from above, we can see that the network stack enables us to connect components together, feed in requests, and get responses out.
┌─────────┐
│ Request │─ ─ ─▶┌────────────────────────────────────────────┐
└─────────┘ │ Network Stack │ ┌────────┐
│ ┌─────────┐───▶┌─────────┐───▶┌──────────┐ │ │ │
│ │ Logged │ │ De-dupe │ │URLSession│ │◀─ ─ ─▶│ Server │
┌──────────┐ │ └─────────┘◀───└─────────┘◀───└──────────┘ │ │ │
│ Response │◀ ─ ─ └────────────────────────────────────────────┘ └────────┘
└──────────┘
I highly recommend that you make use of pointfreeco/swift-dependencies for dependency injection, and client management. This means that you should create a "Network Client" like this:
import Dependencies
import Foundation
import Networking
struct NetworkClient {
var network: () -> any NetworkingComponent
init(network: @escaping () -> any NetworkingComponent) {
self.network = network
}
}
extension NetworkClient: DependencyKey {
static let liveValue: Self = {
let network = URLSession.shared
.duplicatesRemoved()
.automaticRetry()
return .init(network: {
network
})
}()
static let testValue: Self = .init(
network: unimplemented("\(Self.self).network")
)
}
extension DependencyValues {
var networkClient: NetworkClient {
get { self[NetworkClient.self] }
set { self[NetworkClient.self] = newValue }
}
}
This is a somewhat barebones network stack, which can be used by accessing @Dependency(\.networkClient) var networkClient
.
The library ships with the following built-in components.
-
Authentication
Can we used to handle network authentication. This is probably the most complex component, and it's usage in an application requires a delegate conformance. Currently supported are Basic and Bearer authentication methods. Future enhancement will be to support OAuth etc. -
Cached
Can be used to cache network responses in memory. A future enhancement would be to support different cache backend systems. -
CheckedStatusCode
This is a simple component to sanitise error handling to pick out some basic cases. Currently it is open for customisation, so it more useful as an internal component, but a future enhancement could allow it to be used for custom error handling. -
Delayed
Delay requests by a fixedDuration
. This uses the Swift continuous clock, and it very testable. -
DuplicatesRemoved
The network stack allows concurrent network requests, meaning that it is possible for multiple requests to be active at the same time. This component will prevent any duplicate requests firing, and share the response of the only request executed with all duplicate requests. Be careful with this, it might mask underlying application errors. -
Logged
Although fully customisable, this component has sensible defaults to log info about requests as they start and finish using aLogger
. Additionally, underlying types have properties to enable pretty printing. -
Instrument
This can be included in your network stack to instrument its performance. It can report the overall elapsed time of each request, including a breakdown for each component (which support instrumentation). Currently this just logs metrics to the console. A future improvement would allow a richer reporting mechanism, including session statistics. -
Numbered
Adds a monotonically increasing number of every request sent in the current session, i.e. from when the stack is initialised. This is quite handy for logging and debugging. It's also worth noting, that the basic HTTP request type,HTTPRequestData
uniquely identifies each request too. -
Retry
Automatically retry failed requests. By default each request is retried up to 3 times each after a constant delay of 3 seconds. This can be configured for each request however, with constant, immediate or exponential strategies available. Or create your own by conforming toRetryingStrategy
. -
Server
A building block component to configure all requests which are sent via the stack. For example, set default request headers, base URL, scheme etc. Typically this allows your application to create just the specific aspects of each request, such as query parameters or body values. Yet all requests will get the default request parameters as configured by the stack. This component can also be chained together, so typically it is used many times which makes each line/invocation a readable and maintainable point of your configuration. Generally speaking, it is best to add the server components after logging, so that they are included in the logged info. -
Throttled
Can be used to limit the number of concurrent requests. This is very helpful to protect your backend from situations where user behaviour could flood the servers. Additionally requests are added to an internal queue. -
URLSession
Currently the only terminal component is for URLSession. Future transports which would fit the request/response could be supported in the future.
The library provides structs called HTTPRequestData
and HTTPResponseData
. Internally these make use of Apple's HTTPRequest
and HTTPResponse
value types.
These are the building blocks of the library, and are used to make requests with the stack, like this:
// Access the network client
@Dependency(\.networkClient) var networkClient
// Create a http request data value, more on this later...
let request = HTTPRequestData(path: "hello")
do {
// Await the data response
let response = try await networkClient.data(request)
// Access basic properties
let originalRequest = response.request
let payloadData: Data = response.data // might be empty.
} catch as NetworkingError {
// Access basic properties
let originalRequest = error.request
if let response = error.response {
// in some cases we might have an `HTTPResponseData` value,
// which allows access to underlying response info
}
}
While this is fine, it's very low level and not recommended for most use-cases. Instead applications typically which to decode payload Data
value into a specific Coadable
type. This is where the Request
type can be used.
Request
is a generic value which composes the HTTPRequestData
value, along with the ability to decode Data
into some Body
type. If the desired Body
type conforms to Decodable
this is automatic, but full customization is supported. It's even possible to decode the data to a decodable intermediate "data-transport-object" before converting that the desired Body
for use as an application domain type.
// Access the network client
@Dependency(\.networkClient) var networkClient
// Create a http request data value, more on this later...
let http = HTTPRequestData(path: "hello")
// Create a request, assuming that MyExpectedBody is Decodable
let request = Request<MyExpectedBody>(http: http) // This convenience uses default JSON decoder
// Await the value response
let (value, response) = try await networkClient.value(request)
While this works okay, it's a bit fiddly having to create seemingly two request values. Instead, it is recommended to use a constrained extension on the Request
type.
extension Request where Body == MyExpectedBody {
static func myBodyValue() -> Self {
Request(http: .init(
path: "hello"
))
}
}
With this in place, our code becomes,
// Access the network client
@Dependency(\.networkClient) var networkClient
// Await the value response
let (value, response) = try await networkClient.value(.myBodyValue())
The above example shows usage of a NetworkClient. None of the examples specify anything beyond a path of "hello", which is missing some key info, such as the authority. It defaults to "GET" HTTP method, and "https" scheme. All of these properties can be get/set on the request,
var request = HTTPRequestData(method: .post, authority: "my-server.com")
request.headerFields[.accepts] = "application/json"
request.path = "message"
request.greeting = "Hello World"
print(request.debugDescription) // POST https://my-server.com/message?greeting=Hello%20World
But of course, doing this for every request is not desirable, properties of the server should only be configured once, and this is why it is recommended to create an API Client instead of, or in addition to, a Network Client.
Lets assume that we want to figure out the client's geographic location by using http://ipinfo.io/. This is a service which performs geographic information for an IP address. If your application needs to connect to multiple servers, such as 3rd party servers in addition to your own 1st party server, it is a good reason to have a single Network Client with multiple API Clients. In this example, we can create an IpInfo client,
import Dependencies
import Foundation
struct IpInfoClient {
typealias FetchIpInfoData = @Sendable (String?) async throws -> IpInfoData
var fetch: FetchIpInfoData
init(fetch: @escaping FetchIpInfoData) {
self.fetch = fetch
}
}
extension IpInfoClient: TestDependencyKey {
static var testValue: Self = .init(
fetch: unimplemented("\(Self.self).fetch")
)
}
extension DependencyValues {
var ipInfoClient: IpInfoClient {
get { self[IpInfoClient.self] }
set { self[IpInfoClient.self] = newValue }
}
}
// And the Live Network Client (could be a separate module)
import LiveNetwork // Lets assume we have a "live" network client as described above in this module
import Networking // Import this library, swift-networking
extension IpInfoClient: DependencyKey {
static let liveValue: Self = {
// Get the network client - to access the standard network stack
@Dependency(\.networkClient) var client
let network = client
.network()
.logged(using: .app(category: "Network"))
.server(headerField: .authorization, "Bearer \(<Secret API Key>)") // Don't store secrets in SCM though!
.server(headerField: .accept, "application/json")
.server(authority: "ipinfo.io")
return IpInfoClient(
fetch: { address in
try await network.value(.ipInfo(of: address)).body
}
)
}()
}
extension Request where Body == IpInfoData {
static func ipInfo(of address: String?) -> Self {
Request(
http: HTTPRequestData(path: address ?? "/")
)
}
}
To use this in an application, we just need to import some modules, and access the api client.
import IpInfoClient
import Dependencies
@Dependency(\.ipInfoClient) var ipInfoClient
// get the geographic info for the client's IP address
let ipInfo = try await ipInfoClient.fetch(nil)