Snocket Interface

Named pipes and Berkeley sockets have different API when accepting a connection. For named pipes the file descriptor created by createNamedPipe is supposed to be used for the first connected client. Named pipe accept loop looks this way:

acceptLoop k = do
  h <- createNamedPipe name
  connectNamedPipe h
  -- h is now in connected state
  forkIO (k h)
  acceptLoop k

For Berkeley sockets equivalent loop starts by creating a socket which accepts connections and accept returns a new socket in connected state

acceptLoop k = do
    s <- socket ...
    bind s address
    listen s
    loop s
  where
    loop s = do
      (s' , _addr') <- accept s
      -- s' is in connected state
      forkIO (k s')
      loop s

To make common API for both we use a recursive type Accept, see berkeleyAccept below. Creation of a socket / named pipe is part of Snocket, but this means we need to have different recursion step for named pipe & sockets. For sockets its recursion step will always return accept syscall; for named pipes the first callback will reuse the file descriptor created by open and only subsequent calls will create a new file descriptor by createNamedPipe, see namedPipeSnocket.

We support either sockets or named pipes.

Abstract communication interface that can be used by more than Socket. Snockets are polymorphic over monad which is used, this feature is useful for testing and/or simulations.

Create a Snocket for the given Family. In the bind method set ReuseAddr and ReusePort.

System dependent LocalSnocket

System dependent LocalSnocket type

Local address, on Unix is associated with AF_UNIX family, on

Windows with `named-pipes`.

Socket file descriptor.

We use unsafeFdSocket but FileDescriptor constructor is not exposed. This forbids any usage of FileDescriptor (at least in a straightforward way) using any low level functions which operate on file descriptors.