- From: Chris Lilley <chris@w3.org>
- Date: Mon, 09 Aug 1999 04:07:13 +0200
- To: Jonathan Rosenne <rosenne@qsm.co.il>
- CC: "Martin J. Duerst" <duerst@w3.org>, Nir Dagan <nir@nirdagan.com>, www-html@w3.org, www-international@w3.org, Eli Marmor <marmor@Elmar.co.il>
Jonathan Rosenne wrote: > > Eli Marmor's notes are following. > > Jony > > Martin J. Duerst wrote: > > >At 05:20 99/08/05 -0400, Nir Dagan wrote: > >> > >> The HTML4.0 spec. [1] suggests that the dir attribute in LINK > >> refers to the directionality of the linked resource. > >> This is very different from the lang attribute that refers to the > >> language of the title attribute (and hreflang refers to that > >> of the linked resource) The basic problem arrises because XML (and SGML) has a tree structured system of elements, which have attributes; attributes themselves cannot have attributes and cannot contain markup. This is a statement of the obvious, practicallya FAQ, but even so the current problem appears to stemfrom departing from the XML/SGML model. This is a legacy issue back from when HTML was in its infancy. The problems are: 1) The element which denotes the link should contain only information regarding the link. Attributes on the link element refer to the linked-to thing. 2) Content is being hidden in attributes. In this case, the title information. If more information (of a structured nature) is required then XML/SGML structure (elements, and attributes) should be used to encode it in preference to convention which simulate attributes-on-attributes. If title was made a child element of link, then the following desirable results would happen: a) the title could contain markup, such as emphasis, language changes for subsections of the title, and so on. b) attributes on the title element would naturally refer to the title content. Thus, xml:lang on the title element would mean the language of the title. c) accessibility would be improved, since content was no longer being hidden away in an attribute. d) there would no longer be an ambiguity about whether language (xml:lang) was referring to the link or to the title; no need to have a keyed system (hreflang and dir refer to the language and directionality of the link, but lang and (not specified) refer to the language and directionality of the title. Otherwise, to take this to its logical conclusion one would need to add titledir, longdescdir, and longdesclang - all on the same element - thus building a simulation of a tree structure by associating attributes with one another - simulating attributes-of-attributes. Compare these two models 1) <link lang="en" dir="rtl" hreflang="he" titledir="ltr" longdescdir="ltr" longdesclang="fr" href="foo" title="bar" longdesc="baz"/> The parser is supposed to figure out that foo and baz are urls but bar is content, and the application needs to know that lang and titledir are a pair, and hreflang and dir are another pair, and so forth. 2) <link xlink:href="foo" xml:lang="he" i18n:dir="rtl"> <title xml:lang="en" i18n:dir="ltr">bar</title> <longdesc xlink:href="baz" xml:lang="fr" i18n:dir="ltr"/> <desc xml:lang="fr" i18n:dir="ltr">baz <x>with markup</x> even</desc> </link> The parser figures out where the urls are by reference to the xlink spec, and figures out what the language and directionality of each piece of text is by looking at the element that it is on. > >> 1. It would make more sense that dir would refer to the title > >> attribute's directionality. > > > >Yes indeed, like DIR everywhere else. Rather than arguing the bes way of pairing up attributes, it is better to avoid the necessity altogether. > It is FORBIDDEN to put instructions about how to display a page in > another page that the linked page doesn't have a link (like in CSS) > to. I agree with that also; though it can be useful to know some things ahead of time, the linked to resource should also be completely labelled itself. > >> 2. In order to refer to directionality of linked resources > >> we need a new attribute, e.g., hrefdir > > > >i don't think this is needed. 'hreflang' and 'charset' are > >very helpful because they help selecting the right link and > >can provide information that may not be in the target document. > >But my guess is that 'dir' is always used in the target document > >(otherwise, it won't be readable), and therefore, it's not needed > >on link. > > Here I must disagree with Nir (and agree with Martin), because of > the same reason I wrote above. Again, regardless of whether the information (which should certainly be in the linked to document) should be duplicated somewhere in the link, it is better to use xml structure (and to use xml:link for its intended purpose, indicating the language of the element it is on) rather than invent new naming schemes for grouping clusters of duplicated attributes. -- Chris
Received on Monday, 9 August 1999 00:45:14 UTC