package async_extra

The default buffer capacity for UDP-oriented buffers is 1472, determined as the typical Ethernet MTU (1500 octets) less the typical UDP header length (28). Using buffers of this size, one avoids accidentally creating messages that will be dropped on send because they exceed the MTU, and can receive the largest corresponding UDP message.

While this number is merely typical and not guaranteed to work in all cases, defining it in one place makes it easy to share and change. For example, another MTU in common use is 9000 for Jumbo frames, so the value of default_capacity might change to 8972 in the future.

A typical receive loop implicitly calls Iobuf.flip_lo before calling its callback to prepare a packet buffer for reading by the callback and Iobuf.reset afterward to prepare for the next iteration.

sendto_sync sock buf addr does not try again if sock is not ready to write. Instead, it returns EWOULDBLOCK immediately.

Short writes are distinguished by buf not being empty afterward.

See also Iobuf.sendto_nonblocking_no_sigpipe and Bigstring.sendto_nonblocking_no_sigpipe.

• raises Failure

  on internal errors but return Unix.error via Unix.Syscall_result.Unit.t rather than raising Unix_error.

send_sync sock buf has identical semantics to sendto_sync, but is intended for connected UDP sockets (and therefore does not require a "to" address).

See also Iobuf.send_nonblocking_no_sigpipe and Bigstring.send_nonblocking_no_sigpipe.

• raises Failure

  on internal errors but return Unix.error via Unix.Syscall_result.Unit.t rather than raising Unix_error.

sendto sock buf addr retries if sock is not ready to write.

• raises Unix_error

  in the case of Unix output errors and Failure on internal errors.

send sock buf retries if sock is not ready to write.

• raises Unix_error

  in the case of Unix output errors and Failure on internal errors.

bind address creates a socket bound to address, and, if address is a multicast address, joins the multicast group.

Loops, including recvfrom_loop, terminate normally when the socket is closed.

recvfrom_loop_with_buffer_replacement callback calls callback synchronously on each message received. callback returns the packet buffer for subsequent iterations, so it can replace the initial packet buffer when necessary. This enables immediate buffer reuse in the common case and fallback to allocation if we want to save the packet buffer for asynchronous processing.

recvmmsg_loop ~socket callback iteratively receives up to max_count packets at a time on socket and passes them to callback. Each packet is up to Iobuf.capacity bytes.

callback bufs ~count processes count packets synchronously.

Config.init config is used as a prototype for bufs and as one of the elements.