- From: Martin Duerst <duerst@w3.org>
- Date: Mon, 18 Jun 2001 10:50:54 +0900
- To: Terje Bless <link@tss.no>
- Cc: Gerald Oskoboiny <gerald@w3.org>, W3C Validator <www-validator@w3.org>
At 05:30 01/06/12 +0200, Terje Bless wrote: > >Also, we may have to do some pre-sniffing anyway in order to deal with > >UTF-16 and EBCDIC. > >I'll give you UTF-16 (kinda!), but EBCDIC is not possible to sniff for in >any meaningfull way AFAIK; for all practical purposes, it needs to be >properly labelled in the Content-Type (IOW, it's "SEP"[1]). No, not exactly. Please see http://www.w3.org/TR/REC-xml#sec-guessing-no-ext-info for how it can work for XML. I guess the same thing applies to HTML. For HTML, there are more ways to start a file, but not that many more. I know about <HTML> (in various case variants, that is) <!DOCTYPE ... Anything else (except of course for <?xml for XHTML )? >As for UTF-16, I think it's reasonable to assume that it will be properly >labelled or contain a BOM. Almost, but again see the XML rec. >Checking the first 2/3 bytes for one of the >three possible BOMs in UTF-8/UTF-16-MSB/UTF-16-LSB is a far cry from the >current mess (that alters the DOCTYPE if it sees "<FRAME"!). Yes indeed. >Is UTF-16 ASCII-compatible enough that we can assume ASCII up to the XML >Declaration ("<?xml ... ?>")? Well, yes, except that every second byte is a null byte :-). >I could live with a little content sniffing >-- to decide between HTML or XML semantics, or to determine source charset >before we convert to UTF-8 internally, etc. -- as long as it stops guessing >at doctypes based on tags present, and uses an actual SGML parser to figure >out the (provided) DOCTYPE instead of a quick+dirty regex. Once we're there >we should be able to use said SGML/XML parser to extract the necessary >charset info; using two-pass parsing if necessary. Okay. I'll work on it, as I have time. Regards, Martin.
Received on Sunday, 17 June 2001 21:51:24 UTC