We use github's CI instead of travis now.

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

TestNextEventAfterFailedSubscribe tests whether a specific error is sent
on event error channel. This is not a good test in the sense that it is
for internal behavior, not the API.

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

The daemon key may not always be 'charon' (e.g. it is charon-systemd on
my system), so instead just make sure there is a non-empty value set.

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

Currently, recv() does not check the number of bytes read into the
buffer during its Read() calls. This means that in some cases a partial
read may occur, which will cause the next recv() call to read the packet
length incorrectly. In particular, this may be a huge number and cause a
huge buffer allocation. To fix this, simply use io.ReadFull to make sure
we do not get partial reads.

Related: #34

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

In 6dc3797, a read deadline was added
so that listen() would not block forever on recv() calls when the
listener was closed. This was necessary because the event listener was
Close()'d by cancelling a context. If recv() was blocking, the loop
would not ever notice the closed context.

In the current logic, Close() will call el.conn.Close() which is all we
really need to stop the event loop. To do this, however, just return on
any error from recv(). With this, strip out the perr logic as there is
no longer a need for it. And, strip out the ctx and cancel logic because
as noted, closing the connection and returning on error from recv() is
sufficient.

Overall, the reduction of complexity should make this code easier to
maintain going forward.

Related: #34

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

In general, the const err approach is not great. In the case of
transport, the errTransport "wrapping" sometimes masks more useful io
errors. Remove the errTransport, and do not "wrap" transport errors.

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

Signed-off-by: Nick Rosbrook <nr@enr0n.net>

Signed-off-by: Nick Rosbrook <nr@enr0n.net>