Re: Unbound DATA frames in HTTP/3 proposal

2025年10月10日(金) 1:05 Martin Thomson <mt@lowentropy.net>:

> On Tue, Oct 7, 2025, at 01:00, Kazuho Oku wrote:
> > At the same time, I share the pain of Yaroslav and David. Because
> > Capsules is a TLV format nested on top of HTTP, when using Capsules, we
> > have three layers of TLV now, i.e., QUIC STREAM frames, HTTP/3 DATA
> > frames, and Capsules.
>
> There might be multiple layers of overhead, but there is no need to buffer.


Just to respond on this point, while some frames / capsules might not
require buffering, most do.

In fact, I believe that all the 4 capsule types registered [1] are
typically processed as a whole.

The DATA capsule proposed by connect-tcp and the H3 DATA frame are really
the *exceptions* that are expected to be processed before the entire
capsule / frame is received.

[1] https://www.iana.org/assignments/masque/masque.xhtml#http-capsule-types
[2] https://httpwg.org/specs/rfc9204.html#field-line-representations

  I'm not convinced of the need of this.  Not just because of the added
> complexity, but because the value is pretty marginal.  Ben's proposed
> indirection can unwind all the layers, but I challenge the relative value
> of the added complexity.  Of any of the options.
>


-- 
Kazuho Oku