Any update on this? Will it make it's way into 9k? I know it's nothing MAJOR but seems nice to fix :)

I'm getting ENOTCONN on every CRuby version I tried (2.0.0, 2.1.10, 2.2.5, 2.3.3).
@PragTob could you please check this again?

Still happening for me:

tobi@comfy ~ $ ruby -v
ruby 2.3.1p112 (2016-04-26 revision 54768) [x86_64-linux]
tobi@comfy ~ $ cat server.rb 
require 'socket'

server = TCPServer.new 9090
server.shutdown
tobi@comfy ~ $ ruby server.rb 

Maybe cause I'm on Linux?

tobi@comfy ~ $ uname -a
Linux comfy 4.4.0-47-generic #68-Ubuntu SMP Wed Oct 26 19:39:52 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

also still happening on JRuby (sorry had to fixup my JRuby install after some asdf problems):

tobi@comfy ~ $ ruby -v
jruby 9.1.6.0 (2.3.1) 2016-11-09 0150a76 OpenJDK 64-Bit Server VM 25.111-b14 on 1.8.0_111-8u111-b14-2ubuntu0.16.04.2-b14 +jit [linux-x86_64]
tobi@comfy ~ $ ruby server.rb 
Errno::ENOTCONN: Transport endpoint is not connected - Socket is not connected
  <main> at server.rb:4
tobi@comfy ~ $ 

@PragTob I checked on Linux and yes, CRuby behaves differently on Linux and macOS. On macOS behavior is the same as on JRuby. So I feel confused about what behaviour is expected.

I guess this would be a bug for CRuby itself then, in which they can figure out what the correct behavior is. I tend to think no error is correct but who am I to tell? :)

@PragTob I investigated this a little bit more and wanna say that this is completely legal to return ENOTCONN. On BSD/macOS shutdown call on a socket in LISTENING state raises error and doesn't change socket state, but on Linux it handled in a special way (as far as I understood it actually acts as close). Why don't you just use close over shutdown?

cc: @enebo @headius @kares

@etehtsea see the initial issue report:

That's a pity right now, as that's why my test suite fails for JRuby. Funnily changing shutdown to close (in my test suite, not this example script) works for JRuby but then CRuby fails... :-/

Sadly I don't have access to the application anymore, as it was a code challenge in a private github repo and I don't pay for github :D

That said the goal of JRuby is to be CRuby compatible so I think it should do the same. With CRuby being inconsistent it might be worthwhile to ask them though

@PragTob uhmm. This is strange that you had issues with close, but anyway, according to documentation:

shutdown([how]) => 0
Calls shutdown(2) system call.

shutdown method is no more than wrapper around shutdown(2) system call and my comment above is about its behavior. So this inconsistency is not CRuby fault. JRuby mostly uses java APIs and I think there is no sense to emulate system calls behavior across different OSes.

shutdown(Input|Output) methods methods that used in JRuby's shutdown implementation don't close sockets in LISTENING state. This is why TCPServer#shutdown is considered unsupported under JRuby.

@headius am I right?

@etehtsea Yes, I would say you are right.

We do stretch to try to emulate CRuby on a "bug-for-bug basis" but asking to shut down a socket that's not connected seems like a bridge too far.

This is also very old now, so I'm going to call it stale and mark it Won't Fix.