W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > May to August 1996

Re: proposed HTTP changes for charset

From: Drazen Kacar <dave@fly.cc.fer.hr>
Date: Sun, 7 Jul 1996 01:16:39 +0200 (MET DST)
Message-Id: <199607062316.BAA22810@fly.cc.fer.hr>
To: "Roy T. Fielding" <fielding@liege.ICS.UCI.EDU>
Cc: yergeau@alis.ca, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
Roy T. Fielding wrote:

>   3) None of the issues you have raised involve a technical problem
>      with the HTTP/1.1 protocol -- they are POLITICAL problems that
>      are an artifact of historical reality, a reality which the IETF
>      is not capable of changing.
>   4) Labeling the charset with its real value if it is different than
>      iso-8859-1 *always* works, both in old an new practice, because
>      any user agent incapable of handling a charset value is also
>      incapable of handling a charset other than iso-8859-1.  The only
>      time problems occur is when iso-8859-1 data is labeled as such
>      and then delivered to an older client.
> I see no point in continuing this discussion unless you can demonstrate
> a real problem that needs to be solved and can be solved within the
> constraints of HTTP/1.1.

Demonstration of a real problem following...

Suppose we have a server that delivers a page with

Content-type: text/html; charset=iso-8859-2

On the other side of a connection we'll probably (50-60% in my logs) have
Netscape 2.0 on Windows CEE. CEE is Central & Eastern Europe version,
Latin 2 fonts come with OS. Netscape 2.0 can switch code page when it
receives charset parameter (so I've been told). Everything should work,
but it doesn't. Why?

Because Latin 2 does *not* mean the same for ISO and Microsoft. Microsoft
delivers their systems with something they sometimes call CP1250 and
sometimes Latin 2. That code page has all of the ISO 8859-2 characters, but
some of them are at different positions. Positions from 128 to 159 are
filled with something, but that's not the problem. The problem is that
they swapped two 32-character blocks. They wanted to have copyright (or
trademark, I don't recall any more) sign at the same position as in
Latin 1.

I couldn't find any charset with 1250 in its name in IANA registry, but
there is iso-8859-2-windows-latin-2, and I suppose that's the name of
the code page, since nothing else fits. I don't use PCs (except as text
terminals for Unix) and I'm not 100% sure, but I think that Netscape
can't recognize that in charset parameter and it would show the page
with default charset, which is ISO 8859-1. Wrong, again.

<note>Netscape 3.0 beta has a workaround for this, with lots of bugs
at this stage. Bug reports filled and delivered. But this is just one

The typical server here will send a page with CP1250 (without charset),
the page would inform the user that he should manually switch to Latin 2
encoding, and offer 2 or 3 links for other encodings (those pages would
again be sent without charset parameter).

I hacked my server a bit, wrote several CGI programs and it's a little
smarter than others. It can convert HTML pages to 5 different code pages or
3 different ASCII approximations on the fly. I'll probably add some more
output representations. I think Macs use the 6th code page for Latin 2
and two more approximations would be handy.
The conversion is automatic if browser sends Accept-charset header.
Lynx 2.5 is the only one at the moment. Other browsers will receive
some kind of menu.

Too many code pages are in use (ISO 646 has a fair amount of users) and
browsers are currently incapable to deal with them. Servers (or proxies)
could. Not with labeling Content-type, because it would only pass the
potato to the browser. Servers could convert, but they MUST know which
code page user on the other side has installed. HTTP 1.1 spec says that
absence of Accept-charset means that any charset is acceptable and almost
all browsers don't bother to send it. I'd like to change that to
something like this:

No Accept-charset       --   HTTP 1.1 agent is capable of representing
                             ISO 8859-1 only.
Accept-charset: *       --   Any charset is acceptable. I doubt that this
			     will be true for browsers, but it would be
			     useful for robots.
If the agent can use charsets other than ISO 8859-1, then it MUST, MUST
and MUST send Accept-charset header with those charsets listed.

Life is a sexually transmitted disease.

Received on Saturday, 6 July 1996 16:24:54 EDT

This archive was generated by hypermail pre-2.1.9 : Wednesday, 24 September 2003 06:32:04 EDT