Re: [hybi] Review of draft-ietf-hybi-thewebsocketprotocol-13

[Willy Tarreau <[e-mail]>]

On Tue, Sep 06, 2011 at 10:51:10AM -0400, Richard L. Barnes wrote:
> No, you have to process at most three bytes, four if you include the opcode.  See sample code.
> --Richard 

I think you misunderstood me. When you receive 40kB of UTF-8 text, if
you want to stop on frame boundaries, you have to parse the data anyway.
Even if you just want to focus on the last bytes, you have to prevent
yourself from blindly forwarding the whole frame in order to check the
last bytes that you probably don't have yet, which significantly reduces
the benefits of splicing and makes the processing quite more complex.

Really, as it was stated many times, a recipient should not consider a
single frame as containing valid contents, just like data can be split
across multiple TCP packets without affecting their meaning.

I think the point you raised in fact is that having an encoding type
on a frame does not make much sense and it only makes sense on a whole
message.

Regards,
Willy