[rabbitmq-discuss] Login woes with the Ruby amqp library

[Matthias Radestock]

Tony Garnock-Jones wrote:
> On the one hand, it'd be easy enough to transmit a connection.close
> (or similar) before shutting the socket down;

I'd rather avoid that since it would be in direct violation of an
explicit (and recent) instruction to the contrary in the protocol spec,
and could cause (non-rabbit) clients to do all sorts of nasty things.

> but on the other hand, a closed connection at this point is pretty 
> unambiguously an authentication failure, IIUC.

True. Well spotted!

> I think clients -- all of them -- should probably be altered to
> report such an abrupt connection closure at that phase of connection 
> establishment as an authentication failure.

I'd suggest a slightly more circumspect wording along the lines of
"likely cause: authentication failure". After all, there are other
possible causes as well such as malformed framing, or the server or
network dying for some unrelated reason.


Matthias.