Chris Hegarty wrote:
:
It appears to be specific to the loopback interface. You can reproduce
the very same behavior using DatagramSocket and DatagramChannel; bind
to the loopback address and send a packet to a remote address.
I was not too happy about doing this until I found that we do the very
same thing in non raw sockets case.
The main concern is that you are hide a real "invalid argument" case. It
would be good to follow this up with the Linux folks to clarify if the
error make sense for this case.
So have you considered any other solution to this? I'm just wondering if
it would be saner to check if the network interface is the loopback
interface, and if so, just return false when the address is something
other than the loopback. As regards the current fix - it's probably okay
although it would be nicer if you didn't have to repeat the close and
return JNI_FALSE.
-Alan.
