Re: Striving for Compromise (Consensus?)

On Sat, Jul 12, 2014 at 1:56 AM, Greg Wilkins <gregw@intalio.com> wrote:

>
> Roberto,
>
> So it sounds like you don't like continuations any more than the proposers
> of "Greg et al".
> So here is a path to something that will keep everybody happy,
>
> Support the proposal - at some frame size >=18 <=31 bits. Then
> continuations are gone and the two down sides for you are:
>
>    1. Can't stream headers
>    2. We advertise a max header size (it is a frame size but the same
>    thing in proposal).
>
> Now once we have that in place, we look at some new proposals to address
> those concerns.
>
> If we want to be able to stream headers, then we can put forward a
> proposal to allow header frames to be fragmented in the same way data
> frames currently are. ie we use the end-segment flag that they already
> carry to indicate end of a header block.    Then the same framing layer
> code can be used for fragmenting data and headers.      The discussion
> around that proposal would also settle if they can be interleaved or not.
> Hpack would then be updated to work with whatever results (eg by removing
> reference set or whatever else is necessary).
>

This sounds an awful lot like removing continuation & header fragmenting,
just to re-add it back in later.
It presents exactly the same debate we're having now: Do we preserve the
ability to fragment headers? Do we allow the interleaving of those
fragments?