Tcp
supports connection to inet
sockets and unix
sockets. These are two
different types. We use 'a where_to_connect
to specify a socket to connect to,
where the 'a
identifies the type of socket.
with_connection ~host ~port f
looks up host from a string (using DNS as needed),
connects, then calls f
, passing the connected socket and a reader and writer for it.
When the deferred returned by f
is determined, or any exception is thrown, the
socket, reader and writer are closed. The return deferred is fulfilled after f has
finished processing and the file descriptor for the socket is closed. If interrupt
is supplied then the connection attempt will be aborted if interrupt is fulfilled
before the connection has been established. Similarly, all connection attempts have a
timeout (default 30s), that can be overridden with timeout
.
It is fine for f
to ignore the supplied socket and just use the reader and writer.
The socket is there to make it convenient to call Socket
functions.
connect_sock ~host ~port
opens a TCP connection to the specified hostname
and port, returning the socket.
Any errors in the connection will be reported to the monitor that was current when connect was called.
connect ~host ~port
is a convenience wrapper around connect_sock
that returns the
socket, and a reader and writer for the socket. The reader and writer share a file
descriptor, and so closing one will affect the other by closing its underlying fd. In
particular, closing the reader before closing the writer will cause the writer to
subsequently raise an exception when it attempts to flush internally-buffered bytes to
the OS, due to a closed fd. You should close the Writer
first to avoid this
problem.
If possible, use with_connection
, which automatically handles closing.
It is fine to ignore the returned socket and just use the reader and writer. The
socket is there to make it convenient to call Socket
functions.
Where_to_listen
describes the socket that a tcp server should listen on.
Server.t
represents a TCP server listening on a socket.