- From: Rick Jelliffe <ricko@allette.com.au>
- Date: Sun, 8 Jun 1997 04:29:42 +1000
- To: <w3c-sgml-wg@w3.org>
> From: Gavin Nicol <gtn@eps.inso.com> > I would say to apply the heuristics *if* no outside encoding > specification is applied, but in cases where it is available (and > these will become more common), that should override everything. Yes, Gavin's examples are very compelling, especially his point about documents composed of external entities in different charsets (CASE 1): a client should prefer external encoding specifications to inherited encoding specifications. I don't think my proposal & the Japanese one handles that case, which is an important one. I have previously made the comment that XML generators should be quick to stick in BOM and encoding PIs, but that transmitters and clients should err on the side of caution and remove them if there is any doubt (or duplication). The best place for this to occur may well be at the server, to use and strip the charset pseudo-attribute in the encoding PIs into the charset parameter of the MIME header (and then reintroduced if the file is saved to disk at the client end). PI -> MIME -> PI The reason for the encoding PI was that XML documents will have a life before and after transmission: not all of them will be generated on the fly. I think Gavin has provided a good argument why MIME charset may be preferable to encoding PI's charset for transmission. The chain of detection should be for servers and when the MIME header is incomplete. But lets not complicate things. Lets just make the 1.0 draft just have the chain of detection for a local file, and leave it up to implementers of servers and clients to figure out a policy for making use of this information w.r.t. http. XML is a markup language, not a high-level networking protocol. Rick Jelliffe
Received on Saturday, 7 June 1997 14:29:33 UTC