Re: [hybi] Testsuite update case 7.1.5

["Arman Djusupov" <[e-mail]>]

Hello Tobias

> From: Tobias Oberstein [mailto:[e-mail]]
> > connection was closed in the middle of fragmented message.
> 
> If there is a chance during AUTH48 and support by the WG, then maybe add
> 
> close code 1015 = "closed in middle of message"

I would rather give this error code a more generic description such as "Aborted/Interrupted", because an endpoint may need to urgently close the connection while receiving a long inbound message. This way the endpoint can inform the remote side that, for whatever reason, it has to terminate all inbound and outbound transfers immediately.

But, in such cases, a much simpler option is to expect the sender to drop the connection without a close handshake. This is actually what the sender would also want, i.e. to stop sending and release all allocated resources immediately. The network error would make it clear to the remote side that the connection was aborted. This is what most implementations will have to do when requested to close in the middle of a transfer.

For example, when the application code requires the connection to be closed while WebSocket is in the process of sending a large non-fragmented message, there is no other option but to drop the connection (unless the sender is prepared to wait until the current message is sent to the remote side). Since the situation that we try to handle can also happen in the middle of a large fragment or in the middle of a single-fragment message, why should a fragmented message be treated any differently that a non-fragmented one? Fragmentation is just the way that permits us to send a message when its final length is unknown.

With best regards,
Arman