So I am new to this concept of masque, but why do you need to have H3 involved in proxying if the objective is to proxy arbitrary applications as opposed to acting as as replacement for existing HTTP tunnels?
There is a quite a bit of complexity in H3 that is not necessary when using QUIC for general purpose proxying. As to the DATA frame, my intuition tells me that it is probably useful to retain a concept of tunneling over a single HTTP connection for semantic reasons, such as an endpoint that is written for HTTP and not for QUIC, but this need not be related to MASQUE. Kind Regards, Mikkel Fahnøe Jørgensen On 21 October 2020 at 03.32.22, Kazuho Oku ([email protected]) wrote: I agree with others that we do not need to revisit the issue we discussed in #1885. The conclusion there was that HTTP/3 will always use application-layer framing. Regarding how we might fix the issue *in the future*, my preferences goes, as stated in one of the issue comments, to adding a transport-level extension that allows "messages" (instead of bytes) to be sent over a QUIC stream, then use that to send frames in HTTP/3. Doing so reduces the wire overhead for any frame being sent, reduces the total complexity. But what we should do *now* is ship QUIC and H3, they do not have to be changed. 2020年10月21日(水) 8:07 Lucas Pardue <[email protected]>: > An extension seems entirely reasonable to me. > > > -- Kazuho Oku
