This looks pretty good! Could you redo the spacing on the C/C++ changes using tabs instead of spaces, in order to match the code already there?

I might suggest that the REUSEADDR should go directly into the existing open_datagram_socket function, since the corresponding TCP function already sets SO_REUSEADDR.

I didn't include the RESUSEADDR into the existing open_datagram_socket function because it alters the function parameters and causes one of the tests to fail because it relies on the last args parameter. I'm not sure if changing this will break any other projects or not.

I will address the other comments this week.

I wouldn't even make it configurable, this should be the behavior in order to remain consistent with the TCP version of the function. I would merge this for the 1.2.0 release so that it won't be a surprise 1.0.x change.

Any thing else I need to fix?

Thanks for the ping! I hadn't checked back to see the updated commits. Could you squash the updates into the original commit?

Ok. Done! Thanks for replying. Let me know if there's anything else.

Yup. You're right. Should be fixed now.

This looks really good. I definitely want to merge the SO_REUSEADDR code.

However, it turns out you may not need any special code to support SO_BINDTODEVICE after all... sorry it took me so many iterations before reading up on alternative approaches. The Ruby documentation for setsockopt shows how to generate a packed struct here:
http://ruby-doc.org/stdlib-1.8.7/libdoc/socket/rdoc/BasicSocket.html#method-i-setsockopt-label-Parameters
and looking at the struct ifreq which has the interface char array right at the beginning:
https://bytes.com/topic/python/answers/30563-passing-argument-setsockopt-socket-sol_socket-so_bindtodevice-intr
and
http://stackoverflow.com/questions/9127311/how-can-you-choose-the-network-interface-via-which-udp-broadcast-packet-is-sent

you should be able to do the device binding like this (note the explicit nul byte in the Ruby string):

EM.run do
  EM.connect "example.com", 80, Module.new {
    define_method :post_init do
      set_sock_opt Socket::SOL_SOCKET, Socket::SO_BINDTODEVICE, "eth0\0"
      EM.stop
    end
  }
end

I tried this with "eth0\0" and got a successful response from set_sock_opt, and with "eth4\0" on a system that had no eth4, and got an appropriate error: lib/em/connection.rb:280:inset_sock_opt: No such device - setsockopt (Errno::ENODEV)

Granted that it is slightly gross to fake the struct, it is consistent with Ruby's own setsockopt. Actually I think I could replace the EM implementation of set_sock_opt with a wrapper around Ruby's own setsockopt on the Ruby side and remove this C function entirely!

Ok I verified that it works. New change is ready. I think I was using 1.0.0 which I believe was not working for this correctly.

Btw, I would agree with your wrapper idea. That way the code is not duplicated.

Fixed :) Thanks!