Re: Permissible states for extension frames #591

[Amos Jeffries <[e-mail]>]

On 12/08/2014 4:51 p.m., Greg Wilkins wrote:
> On 12 August 2014 13:00, Martin Thomson wrote:
> 
>> Greg, this issue has nothing whatsoever to do with CONTINUATIONS, other
>> than to note that extensions would not be permitted within a sequence of
>> continuations.
> 
> 
> 
> Martin,
> 
> I don't know how it is not related to continuations?    The question was
> are we OK with allowing extension frames to appear anywhere for any stream?
> 
> I'm not OK with that because of continuation frames, which I believe have
> to be sent in a contiguous block.   We have to specify that extensions
> frames are OK anywhere except between HEADERS and CONTINUATION and between
> CONTINUATION frames.
> 
> I would also be OK with extensions frames anywhere, but only if
> continuations allowed interleaving.... the consensus was not to do that, so
> we can't allows extension frames anywhere and have to have a more complex
> specification for them.
> 
> I'm sorry if it looks like I'm trying to fight the already lost
> continuation battle.   I'm not, I'm just making sure that we keep the
> specification consistent with the complexities of the continuation
> mechanism.

I dont see why we need to do that at all.

The specs for extensions is already that they cannot redefine existing
frame semantics or layout without an explicit ACK from the other end.
Such as the non-interleaving of HEADERS+CONTINUATION.

To me that means it is perfectly legal to define an extension for making
CONTINUATION interleave, BUT must not be done without explicit
negotiation of that extension.

Amos