Re: [Masque] [Webtransport] Layered or integrated protocols

[Lucas Pardue <[e-mail]>]

Good points on this thread. Too many to respond to directly.

WebTransport and MASQUE remind me of the UK comedy duo known as the Chuckle
Brothers [1]. They were related, looked vaguely similar, and had a
catchphrase "to me, to you" that was used often when they were trying to
deliver hard to manage items and shuffling back and forth.

There's a lot of similarity between the WebTransport and MASQUE. I think
we've been hitting problems that might have occurred in either as we
progress at different rates. And the solutions designed in one WG or spec
aren't as synchronised as we might like.

A Capsule semantic, that can be mapped to HTTP versions, can go a long way
to aligning WebTransport and MASQUE. It's currently defined in the HTTP/3
DGRAM spec out of necessity of invention, but it could be extracted it
people really wanted. When reading draft-thomson-webtrans-hwtq-00 my early
thoughts were that you could define capsule types to model the proposed
QUIC subset - this is basically what the datagram capsules are doing for
QUIC DATAGRAM frames.

There's benefits to in-stack and on-stack. I think that on-stack, in
simplistic terms, either requires treating everything as content (e.g. DATA
frames) and running a sub-protocol on top, or it requires  H2/H3 stacks to
treat new frames as opaque blobs that they throw to an upper layer to deal
with. The problem with the latter is where those frames have some
interaction or dependency that is not normally exposed to upper layers. I'd
hope our designs can allow implementations of in-stack and on-stack.

Cheers
Lucas

[1] - https://en.m.wikipedia.org/wiki/Chuckle_Brothers