Re: [hybi] HELLO frames (was Re: It's time to ship)

[Willy Tarreau <[e-mail]>]

On Mon, Jan 10, 2011 at 12:50:12AM +0100, Julian Reschke wrote:
> On 10.01.2011 00:46, Willy Tarreau wrote:
> >On Mon, Jan 10, 2011 at 12:42:33AM +0100, Julian Reschke wrote:
> >>On 10.01.2011 00:24, Willy Tarreau wrote:
> >>>...
> >>>In fact I'd be more worried about load balancers and content switches
> >>>in general. They have to support 100 because of the very common
> >>>"Expect: 100-Continue" that is present in many POST requests, and it's
> >>>highly likely that they'll process 101 as a 100 just as haproxy did,
> >>>because 101 was part of RFC2817 and not 2616.
> >>>...
> >>
> >>This is incorrect, see
> >><http://greenbytes.de/tech/webdav/rfc2616.html#rfc.section.10.1.2>.
> >
> >OK, but that's not what I'd call a definition. It's just as little verbose
> >as CONNECT, it lets one imagine it's just to use a new HTTP version. Both
> >were given an example for the first time in 2817 in my opinion.
> 
> Well, there's also the text in 
> <http://greenbytes.de/tech/webdav/rfc2616.html#rfc.section.14.42>.

Yes I know, but the connection between 101 and Upgrade was not easy to
establish. I discovered that particularity here with websocket for the
first time after 10 years of everyday use of HTTP.

> So... we recently adopted language from 2817 into HTTPbis; could you 
> please check that, and provide feedback to the WG if you feel we need to 
> improve what we have?

I remember having checked that part in http-bis, as I was one of those
asking for the merging of CONNECT and 101 into http-bis, and found it to
be OK. I could recheck though. But it's always harder to spot things that
are difficult to understand once you know them. I'd better ask someone
else to indicate me if he understands how to use it.

Cheers,
Willy