Re: [hybi] thewebsocketprotocol #28 (new): Fragmentation

[John Tamplin <[e-mail]>]

On Mon, Nov 22, 2010 at 1:32 PM, Zhong Yu <[e-mail]> wrote:

> I think the answers to this question depend on specific programming
> models, therefore they are best addressed by applications/frameworks.
> WS protocol doesn't have to support streaming - that's the job of
> frameworks. A single layer shouldn't mix streams and messages. (So we
> have streaming framework, on top of WebSocket frames, on top of TCP
> stream, on top of IP frames, on top of electronic stream, on top of
> quantum of photons... It's turtle all the way, but there is only one
> turtle at each level)
>

I disagree completely.  WebSocket fragmentation isn't about supporting
streams but rather supporting senders that don't know the length up front
and advanced uses such as multiplexing.  The unit of communication via
WebSocket is always a message.

This question has to be answered in order to interpret a stream of frames
properly.  One of the following must be true to have a single unique
interpretation of a sequence of frames:

- control frames can't be fragmented
- control frames cannot be inserted in the middle of a fragmented message
- there must be a way to differentiate a control fragment from a data
fragment
- the receiver must keep an arbitrarily deep stack of in-progress messages

-- 
John A. Tamplin
Software Engineer (GWT), Google