- From: Julian Reschke <julian.reschke@gmx.de>
- Date: Mon, 04 Aug 2008 21:21:52 +0200
- To: Frank Ellermann <hmdmhdfmhdjmzdtjmzdtzktdkztdjz@gmail.com>
- CC: ietf-http-wg@w3.org
Frank Ellermann wrote: > ... > So far it is simple, and no reason to talk about it in the > spec. Now back to our Accept-Charset cases cases, assume > that Koi8-R, BOCU-1, and UTF-8 are available: > > All match *;q=0.7. Toss a coin ? Take always the last, > i.e. for utf-8;q=0.7,*;q=0.7 take the * and toss a coin ? > > This is _relatively_ harmless (unless you get BOCU-1 and > your UA can't handle it, obviously). > > But it is strange when the <qvalue> of * isn't one of the > smallest non-zero <qvalue>s, and I'm surprised that folks > here refuse to see it. It is a problem if it occurs in practice. Does it? > But why ? If everybody does something else the wildcard > can be simply deprecated. Or if desired fixed in some > way, e.g., state that a wildcard <qvalue> greater or > equal than the smallest non-zero <qvalue> elsewhere MUST > be interpreted as *;q=0.001. Or SHOULD. Or something > better than "dunno, who cares, toss a coin, get BOCU-1". Of course we can spend time on edge cases like these. But if we do, I'd prefer to talk about repeating Content-Length or Content-Type headers first, because those may have security implications. As a matter of fact, we *did* discuss those a few months ago, and decided not to include error handling instructions into the spec. So why start here, without any evidence that we're working on a problem that indeed affects people? BR, Julian
Received on Monday, 4 August 2008 19:22:35 UTC