Re: END_SEGMENT and headers

Are you mixing up the HTTP on HTTP2 usecase with the framing part of HTTP2?
-=R


On Wed, Apr 16, 2014 at 4:30 PM, David Krauss <potswa@gmail.com> wrote:

> You mean all on the same stream? No, that is not allowed, because the
> headers (and CONTINUATIONs) must be an uninterrupted sequence of frames.
> There must be exactly one HEADERS frame per stream because it changes the
> stream’s state.
>
> Before the headers are complete, the server cannot dispatch the stream to
> an application, so there is no application there to receive the metadata…
>
>
> On 2014–04–17, at 7:21 AM, Roberto Peon <grmocg@gmail.com> wrote:
>
> A message might be comprised of HEADERS   DATA(200bytes)
> HEADERS+CONTINUATION   DATA(20bytes)   HEADERS.
>
> Your suggestion of making END_HEADERS mean END_SEGMENT would disallow
> expression of such a message.
> -=R
>
>
> On Wed, Apr 16, 2014 at 4:17 PM, David Krauss <potswa@gmail.com> wrote:
>
>> What? This proposal does not disturb segmentation. Metadata may appear in
>> the middle of a message, but not in the middle of the headers!
>>
>> On 2014–04–17, at 5:45 AM, Roberto Peon <grmocg@gmail.com> wrote:
>>
>> > I misread that!
>> >
>> > I withdraw that this is a good improvement, since there are protocols
>> that may want to have metadata in the middle of a message.
>> >
>> > -=R
>>
>>
>
>