Using Linux's `epoll(7)` interface, a `EPOLLHUP` condition is
signalled for the reading end of a pipe or socket when the other end
is closed for writing. This may happen in combination with `EPOLLIN`
being signalled (if further data is available for reading), or with
`EPOLLIN`.
If `EPOLLHUP` is signalled without `EPOLLIN` it indicates an immediate
EOF condition, which will result in the next `read()` suceeding with 0
bytes read. It is thus not required to handle `EPOLLHUP` specially in
the reader, but we need to indicate readiness upon encountering it.
Looking at the `kqueue` and `windows` mio backends, they seem to be
turn a detected EOF and connection reset into `mio::Ready::hup()`
events, so it may be reasonable to speculate that this change is an
improvement for these platforms as well.
Fixes#117.
Note that this brings in a lot of machinery just for the sake of
wrapping file descriptors as PollEvented. This should maybe be moved
into the public API of some crate.
* Move to `std::net` as it's all purely UDP related
* Rename to `UdpCodec` and `UdpFramed` to give a consistent `Udp` prefix
* Add `RecvDgram`, rename `SendDGram` to `SendDgram`
* Touch up some style here and there
This commit makes a few changes to the Sink implementation on Framed:
* Backpressure is implemented for `start_send`. If the write buffer is
over 8KiB and can't be flushed, no new items are accepted.
* 0 length writes to the upstream transport are translated into a
`WriteZero` error, as with `io::Write::write_all`. `write_all` checks
for and ignores `Interrupted` errors, but I do not think this is
necessary for non-blocking writes.
* In `poll_complete`, the upstream transport is not flushed until
*after* writing the entire write buffer.
Alex Crichton