Add support for UDP

[lhoward]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 37.76824% with 290 lines in your changes missing coverage. Please review.

Project coverage is 89.62%. Comparing base (3de9f8a) to head (7731068).
Report is 9 commits behind head on main.

Files with missing lines	Patch %	Lines
FlyingSocks/Sources/Socket.swift	30.67%	217 Missing ⚠️
FlyingSocks/Sources/AsyncSocket.swift	27.27%	64 Missing ⚠️
FlyingSocks/Sources/Socket+Darwin.swift	50.00%	6 Missing ⚠️
FlyingSocks/Sources/SocketAddress.swift	93.75%	3 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (3de9f8a) and HEAD (7731068). Click for more details.

HEAD has 1 upload less than BASE

Flag	BASE (3de9f8a)	HEAD (7731068)
	2	1

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #129      +/-   ##
==========================================
- Coverage   96.66%   89.62%   -7.04%     
==========================================
  Files          60       60              
  Lines        3504     3932     +428     
==========================================
+ Hits         3387     3524     +137     
- Misses        117      408     +291     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[lhoward]

You mean I have to write tests? ;)

[swhitty]

This is really great thank you.

I think the changes to Socket* files are 👌🏼

I don't think we necessarily need AsyncMessageSocket as we should be able to achieve the same with adding methods to AsyncSocket. I've performed a few tests and errors should thrown (on Darwin at least) if users attempt to call the wrong methods when using SOCK_DGRAM.

I've have built a little echo server using your code with a couple of small tweaks:

From here it would be easy to add a AsyncSequence of (Socket.Address, [UInt8]).
(I used Socket.Address because its easier to print out to console but could also be sockaddr_storage)

No need to worry about tests until we settle on the right API

[lhoward]

Great, have applied your diff locally, will push an update shortly.

[lhoward]

I do need sockaddr_storage to avoid redundant string conversions in our application.

[swhitty]

this is wonderful, thank you

[swhitty]

Because Swift Testing runs tests in parallel you may need to bind to port 0 letting the OS provide a port number then retrieve it and send to it

let address: any SocketAddress = .loopback(port: 0)
let socket = try Socket(domain: Int32(type(of: address).family), type: Int32(SOCK_DGRAM))
try socket.bind(to: address)
let listenAddr = try socket.sockname() // gets port

[lhoward]

Because Swift Testing runs tests in parallel you may need to bind to port 0 letting the OS provide a port number then retrieve it and send to it

let address: any SocketAddress = .loopback(port: 0)

let socket = try Socket(domain: Int32(type(of: address).family), type: Int32(SOCK_DGRAM))

try socket.bind(to: address)

let listenAddr = try socket.sockname() // gets port

Great, I ran out of time today but I'll take a look tomorrow!

[lhoward]

BTW, UDP sockets can be connected and there can be some benefits to use this on the client side (it can save a route lookup as it can be cached in the "connection"). I'll add support for that too.

Edit: actually we can probably just use read and write, they should be identical to recv and send with no flags (shall triple check).

[lhoward]

We might want to support multihomed servers without needing to bind separate sockets to each interface.

[lhoward]

We might want to support multihomed servers without needing to bind separate sockets to each interface.

So for this I reckon maybe we want to make the message a structure rather than a tuple (as I had it originally, although I'm not wedded to this), and then add the socket address on which the packet was received and the interface index. These should be usable on both send and receive and be optional (or possibly omitted entirely from the structure if the underlying platform doesn't support).

Something like:

struct /*AsyncSocket.*/Message: Sendable {
    let data: [UInt8]
    let peerAddress: sockaddr_storage
    let localAddress: sockaddr_storage?
    let interfaceIndex: Int32?
}

Dealing with struct msghdr in Swift is a pain but I think I have some code somewhere that I can cut and paste (probably in IORingSwift).

[swhitty]

that's fine, its an additional feature so can skip a platform without breaking anything

[lhoward]

~~For Windows, AsyncSocketMessageSequence returns a (sockaddr_storage, [UInt8]). It may be more sensible to just remove this on Windows entirely so that we don't have a breaking API change when it's implemented properly.

On the other hand, use cases for a multihomed UDP server on Windows... I mean, it seems unlikely to me but, I'm sure someone can think of one.~~

Scratch that, Windows API is now consistent, it just doesn't return the local interface index and address.

public struct AsyncSocketMessageSequence: AsyncSequence, AsyncIteratorProtocol, Sendable {
    public static let DefaultMaxMessageLength: Int = 1500

    // Windows has a different recvmsg() API signature which is presently unsupported
    public typealias Element = AsyncSocket.Message

    private let socket: AsyncSocket
    private let maxMessageLength: Int

    public func makeAsyncIterator() -> AsyncSocketMessageSequence { self }

    init(socket: AsyncSocket, maxMessageLength: Int = Self.DefaultMaxMessageLength) {
        self.socket = socket
        self.maxMessageLength = maxMessageLength
    }

    public mutating func next() async throws -> Element? {
#if !canImport(WinSDK)
        try await socket.receive(atMost: maxMessageLength)
#else
        let peerAddress: sockaddr_storage
        let bytes: [UInt8]

        (peerAddress, bytes) = try await socket.receive(atMost: maxMessageLength)
        return AsyncSocket.Message(peerAddress: peerAddress, bytes: bytes)
#endif
    }
}

[lhoward]

I've made a couple of other changes which you may/may not wish to take. I was a bit annoyed by needing to use sockaddr_storage so I've had the new APIs return any SocketAddress (which is a sockaddr_storage, which now conforms to SocketAddress). I will add some more tests of this too.

And also there's an AnySocketAddress type erased wrapper which is useful if you need to pass it any to something that takes some SocketAddress.

[swhitty]

Great work

[lhoward]

Thanks Simon, really helps out my project too having this merged!

[swhitty]

I was adding some more unit tests and discovered an issue with IP6 on both Darwin and Linux:

try Socket(domain: AF_INET6, type: .datagram)

This always throws an error:
Darin: SocketError.failed(type: "SetOption", errno: 13, message: "Permission denied")
Linux: SocketError.failed(type: "SetPktInfoOption", errno: 22, message: "Invalid argument")

While I have not fixed the issue, I have a PR #131 that I would appreciate your input on, because it replaces the private func setPktInfo() API you added here allowing users to sidestep the issue.