- From: Nick Kew <nick@webthing.com>
- Date: Mon, 14 Feb 2005 14:31:33 +0000
- To: QA Dev <public-qa-dev@w3.org>
Terje Bless wrote: > Going through some old w-v mail I ran across another instance of custom DTDs > getting treated poorly by v.w.o due to exceeding the QUANTITY values in the > SGML Declaration we use for all non-XML documents. I tried to raise this on #validator, but Terje asked me to post instead. I guess this puts it on-record. I see two reasons to change here. One is that we're validating documents, not DTDs. Error messages referring to a DTD are downright confusing to users "Hey, line 123 is nothing like that: this service is broken". That wouldn't be so bad if it was only custom DTDs - and therefore advanced users - that were affected, but in fact the HTML 4.0 DTD generates IIRC four errors, dragging newbies straight into the confusion zone. The only sensible solution IMO is to restrict validation to the HTML and suppress messages from the DTD. And even Jukka hasn't proclaimed that wrong (yet). Now the Quantity errors come from custom DTDs. Liam was first to deal with this, and his solution makes sense to me. Basically, if we take the W3C sgmldecl as authoritative (which I don't really suppose anyone intended), then it is authoritative for the W3C DTDs. Not custom DTDs. So imposing limits on custom DTDs is unhelpful and unnecessary. If we do impose the W3C arbitrary limits, then the correct error message to associate with them is not "invalid" but "unsupported by this service". > Should we try to dig up the original discussions of the HTML WG on this? > > Should we use the WDG limits, or should we just set QUANTITY NONE? I think probably the latter makes most sense. > We could in theory keep the current SGMLDECL for W3C-specified HTML and use a > modified one for custom DTDs (as the WDG does, but is a poor fit with our > current implementation). Is this an option worth exploring? Is there an HTML document written to a W3C DTD for which this would affect validation results? I don't see how to construct it. And if not, the answer seems obvious. Or am I missing something? -- Nick Kew
Received on Monday, 14 February 2005 14:31:58 UTC