- From: Willy Tarreau <w@1wt.eu>
- Date: Wed, 2 Jul 2014 00:01:43 +0200
- To: "William Chan (?????????)" <willchan@chromium.org>
- Cc: Poul-Henning Kamp <phk@phk.freebsd.dk>, Jason Greene <jason.greene@redhat.com>, HTTP Working Group <ietf-http-wg@w3.org>
On Tue, Jul 01, 2014 at 02:29:25PM -0700, William Chan (?????????) wrote: > Hard fail. User visible error. End users blame the last mover, therefore we > should move first *if we believe that the functionality is important to > preserve*. In my example about extensions, assuming we think extensibility > should be preserved (I've not heard recent disagreement here, but if I'm > wrong that there's not consensus here, please raise), then we should hard > fail when the extension negotiation fails. Hence attempting to negotiate > dummy extensions and use dummy extension frames to see if they make it > through. The problem is that there's no "dummy" extension, the extension you'll try to negociate will be assigned later and you may end up accidentely negociating something that modifies the framing or anything just because at the time you do this you're not aware of this. Imagine that we implement jumbo frames this way. I'd rather not see the TCP Window Scaling mess again where many firewalls didn't look at the option but still used to analyze sequence numbers and to block traffic when they looked out of window! It's important for interoperability not to send undefined crap over a wire because what's underfined for the agent doing it may be defined for the one receiving it. Willy
Received on Tuesday, 1 July 2014 22:02:11 UTC