Re: WGLC for Datagram Extension

[Christian Huitema <[e-mail]>]

The way I understand it, we want to say that QUIC makes no effort to 
partition the space of datagrams into subcategories. Applications send 
and receive datagrams, and the conceptual API is, "here is a datagram 
that should be sent or just arrived on this QUIC connection". I 
application want to treat some datagrams as duck calls and some other 
datagrams as music packets, they can do that, but they have to do it 
themselves by decorating the datagrams appropriately. And whether they 
do that using stream IDs or packet codes or sequence numbers is not 
QUIC's business.

On 9/16/2021 5:05 PM, Tommy Pauly wrote:
> Agreed that the word “strongly” can simply be removed.
>
> Applications using QUIC can choose to associate particular datagrams with data sent on a stream—like HTTP/3 choosing to add a value calculated based on stream IDs into the payload of the DATAGRAM frame—but such associations do not belong to the transport protocol.
>
> https://github.com/quicwg/datagram/issues/52 <https://github.com/quicwg/datagram/issues/52>
>
> Thanks,
> Tommy
>
>> On Sep 16, 2021, at 4:48 PM, Martin Thomson <[e-mail]> wrote:
>>
>> On Fri, Sep 17, 2021, at 07:00, Eliot Lear wrote:
>>>> DATAGRAM frames belong to a QUIC connection as a whole, and are not
>>>> strongly associated with any stream ID at the QUIC layer
>>> What does "strongly associated" mean in this context?  Apologies if this
>>> is well trodden ground.
>> This is unfortunately so well-trodden that this text was added without consideration for people who weren't involved in the trampling process.
>>
>> I think that "strongly" can be struck here, it's working too hard.  And smart people will latch onto it.
>>
>> Context:
>>
>> When we use DATAGRAMs in HTTP (and likely in other contexts) there will be a need to bind each DATAGRAM to a (request) stream.  That's necessary to ensure that flows of DATAGRAMs can be routed by gateways and the like along with the stream.  There were lots of debates about how to manage that binding and the layer at which it would be documented.  This text is likely intended to record the conclusion that this document definitely isn't where that sort of binding occurs, but for someone without that history.  It doesn't really achieve that though and because it doesn't need to (why would you think that any association exists?), it ends up being distracting.
>>
>