Re: [hybi] Comments about draft-13

[Iñaki Baz Castillo <[e-mail]>]

2011/9/8 Alexey Melnikov <[e-mail]>:
>> 2011/9/2 Iñaki Baz Castillo <[e-mail]>:
>>
>>>
>>> 3) Section 4.2.2 does include nothing about a 4XX HTTP response code
>>> for the case in which the WS clients offers N WS subprotocols and the
>>> server supports no one of them. Current text states that the server
>>> MUST reply 101 without Sec-WebSocket-Protocol, so the client would
>>> then MUST Fail_The_Connection.
>>>
>>> There has been a long thread recently suggesting that in that case
>>> (and just in *that* case) the server whould *reject* the GET request
>>> with a *specific* HTTP status code which would mean "I don't support
>>> any of your suggested WS protocols, so I cannot accept the WS session
>>> because I have no idea what we are supposed to speak".
>>>
>>> This is a new protocol, so feel *free* to create a new "HTTP" response
>>> code for this case, maybe:
>>>
>>> 488 WebSocket SubProtocol Not Supported
>>>
>>> As I explained in other thread, this status code could help the human
>>> user and the client application developer to realize that his software
>>> is not up-to-date with the service requeriments ("your programm is too
>>> old, upgrade please").
>>>
>>
>>
>> Hi, any new consideration about this?
>> Thanks a lot.
>>
> After thinking more about this: I would prefer to leave the current
> behaviour as is:
>
> C: I would like to use either subprotocol "foo" or "bar"
> S: 101 (WebSocket handshake), but speak neither
>
> at this point the client can decide to close the connection, or decide that
> it can cope with lack of the subprotocols.
>
> You propose:
> C: I would like to use either subprotocol "foo" or "bar"
> S: 488 speak neither
>
> At this point if the client can cope with lack of support for both "foo" and
> "bar", it has to redo the handshake without mentioning them.
> Otherwise it can close the connection.

No, the client cannot close the connection because the server replied
488 so there is no established WS connection.

>
> I don't think there is a significant difference between 2 cases: the number
> of round trips is the same and the same behaviour can be achieved easily in
> either case.

This is a negotiation. If Client wants to speak "foo" or "bar" but the
server does not support any of them, the server should reply 488 (such
response could/MUST include a Sec-WebSocket-Protocol with the
subprotocols the server supports).

I see no reason for the server to reply 101 in the case it does not
support any subprotocol offered by the client.



-- 
Iñaki Baz Castillo
<[e-mail]>