Re: HTTP/3 Nits

[Martin Duke <[e-mail]>]

Alright, that implies an edit to 4.1.1.1.

On Fri, Oct 16, 2020, 18:44 Mark Nottingham <[e-mail]> wrote:

> HTTP/2 didn't create a registry because pseudo-headers are not an
> extensibility point on that protocol. The extensibility point that 8441
> uses is SETTINGS, which negotiates a change in the operation of the
> protocol on a connection-by-connection basis.
>
> Creating pseudo-headers as a new extensibility point is a bad idea; it
> will inevitably be misused, as the distinction between pseudo-headers and
> actual header fields is fuzzy in several dimensions. If people want to
> pursue this, I'd suggest taking it to the HTTP WG so as not to exceed the
> charter of this WG.
>
> Cheers,
>
>
> > On 17 Oct 2020, at 4:07 am, Lucas Pardue <[e-mail]>
> wrote:
> >
> > Hi Martin
> >
> > On Fri, Oct 16, 2020 at 5:33 PM Martin Duke <[e-mail]>
> wrote:
> >
> > If 4.1.1.1 is accurate, then shouldn't there be a registry for HTTP/3
> pseudoheaders? IIUC pseudoheader extensions are not possible in HTTP or
> HTTP/2, so this is an H3-specific registry.
> >
> >
> > HTTP/2 does allow pseudo-header extension. See RFC 8441 which defines
> the :protocol pseudo-header[1] to allow WebSockets over H2.
> >
> > We follow H2's example, maybe there was good reason not to have a
> registry or maybe it was an oversight?
> >
> > [1] - https://tools.ietf.org/html/rfc8441#section-5
> >
>
> --
> Mark Nottingham   https://www.mnot.net/
>
>