W3C home > Mailing lists > Public > ietf-http-wg@w3.org > April to June 2011

Re: #282: Recommend minimum sizes for protocol elements

From: Willy Tarreau <w@1wt.eu>
Date: Mon, 2 May 2011 10:34:02 +0200
To: Mark Nottingham <Mnot@mnot.net>
Cc: Poul-Henning Kamp <phk@phk.freebsd.dk>, HTTP Working Group <ietf-http-wg@w3.org>
Message-ID: <20110502083402.GN10529@1wt.eu>
On Mon, May 02, 2011 at 05:14:59PM +1000, Mark Nottingham wrote:
> > One of the more amusing ways to bust caches is to include "Vary:
> > User-agent" in a response.

Amusing indeed.

> > One would think that there would be two identical browsers on the
> > planet but I guess we're all individualists in that respect.

I've designed a very simple UA logging algorithm for a customer. The
principle is simply to log an ID when the UA has already been seen,
and otherwise to log both the UA and its new assigned ID. And doing
so revealed that we got something like 6000 IDs in a week among about
10M different visitors. For sure servers can't rely on the full UA.
 
> > I think that we should recommend that clients not send
> > a U-A header at all, to enourage servers to DTRT with respect
> > to content portablity.

I know some customers who make the UA mandatory to serve requests, not for
stats, but because they know that (right now) 100% of requests without a
valid UA come from stupid bots.

> Agreed, but from discussing this (at length) with browser vendors, they're
> unwilling to change UA and other headers much, because of potential interop
> problems with sites that depend upon them.

Indeed, some applications need to identify small sets of known devices,
especially the ones with limited capabilities (IE6 announcing support for
compression which is not completely true, or other devices with small
screens). But I think the main reason is that browser vendors like to
provide a way to see their browsers reported in stats. BTW, one of the
reasons that Opera is not much represented is that it lets you pretend
to be another browser, and I've seen many people set it to MSIE :-)

> There has been some headway in recent browsers; IE9 and FF4 have somewhat shorter UA headers than before.

Indeed, just the browser name, its version and optionally the device type
should be plenty, as right now developers don't mention other criteria
when targetting a specific combination.

Regards,
Willy
Received on Monday, 2 May 2011 08:34:31 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Friday, 27 April 2012 06:51:40 GMT