Re: [hybi] Framing take IV

[Ian Hickson <[e-mail]>]

On Wed, 4 Aug 2010, Greg Wilkins wrote:
> On 4 August 2010 10:53, Ian Hickson <[e-mail]> wrote:
> 
> at the f2f there was a very loud hum for a single framing mechanism. 
> Following that, there were loud hums for fragmentation and fixed length 
> frames, even without consideration of multiplexing.
> 
> I believe the feeling was that fragmentation makes sending and receiving 
> big messages simpler.  It also prepares the ground for multiplexing 
> extensions. There was very little support for variable length lengths, 
> as that was just seen as more complex than fragmentation.
> 
> But if you have a counter proposal that meets the single framing hum, 
> then please post it.

With all due respect, I'd much rather base work on this protocol on actual 
technical arguments, and not on a self-selected group of people humming, 
as much fun as that may be.

Here's a proposal for length-based frames:

 - one byte to indicate the frame type (used by the WebSocket API to 
   decide what kind of event to fire)
 - a BER-encoded packed integer, most-significant-digits first, high bit 
   used to indicate a continuation byte, giving the payload length.
 - payload

I'm still not convinced that length-based frames are safe enough for the 
Web, but I have been convinced by the various discussions of security bugs 
related to framing that sentinel-based frames are likely just as insecure. 
I really think we should find a solution that is more secure than either, 
though; I don't think that we should go with length-based frames on the 
basis that the alternatives are equally likely to lead to security bugs.

In the absence of a better solution, though, the above (basically the same 
as what the protocol has now, except with sentinel-based frame type 
removed) seems to fit the bill fine.

-- 
Ian Hickson               U+1047E                )\._.,--....,'``.    fL
http://ln.hixie.ch/       U+263A                /,   _.. \   _\  ;`._ ,.
Things that are impossible just take longer.   `._.-(,_..'--(,_..'`-.;.'