- From: Leif Halvard Silli <lhs@malform.no>
- Date: Tue, 23 Sep 2008 11:31:37 +0200
- To: James Craig <jcraig@apple.com>
- CC: Al Gilman <alfred.s.gilman@ieee.org>, Chris Wilson <Chris.Wilson@microsoft.com>, W3C WAI-XTECH <wai-xtech@w3.org>, public-html@w3.org, Gez Lemon <gez.lemon@gmail.com>
James Craig 2008-09-23 04.32: > Al Gilman wrote: >> + note: I think we should be able to come up with a definition >> of header hierarchy where there is consensus that this is present >> in the table genre that we have to cover. It's widespread. > [snip] In over 99% of the cases, a clearly-defined algorithm for > calculating implicit scope is going to replace the need for explicit > author-defined associations. What 99%? Those when authors have not used any TH cells at all? HTML 4 already has a such clearly-defined algorithm for finding heading information, when TH cells *are* present, see section 11.4.3. (And even better: that algorithm incoroprates @headers.) Just as a @scope-like algorithm can be generalised to work also when there is no @scope and not even any TH cells, the general @headers-incoprating algorithm of HTML, can also be extended to work under the same circumstances. > In most of the other edge cases @scope will > suffice, and in very rare cases, a more explicit association could be > achieved via @headers or @aria-labelledby. > > It is my understanding that @headers, while valuable, is almost never > used. Like the long-term longdesc testing, has there been any attempt to > determine where and how often @headers is used? Have you are anyone calculated how far we would come if we dropped @scope and instead went 100% for @headers? Note: the @headers-incorporating algorithm of HTML 4, is part of the @headers feature there. You must take that into account when counting how often you see the @headers attribute. > In most cases where @headers is necessary, the author would do better to > change the information architecture of the table into a more > understandable form, instead of "accessifying" an already overly complex > table grid. [snip] What you say here would still apply even if we go for @headers instead of @scope: It is more vital - covers more broadly - if one first uses TH cells properly before considering @headers. The HTML 5 draft's point of view, is that @scope is natural, and @headers is extra. That view is false. We have 3 choices: @scope, @headers or both. Currently the draft proposes to use both, allthough @headers is reduced to the very stupid use of it that you argues against. (Just as it is stupid to use @scope if the algorithm makes it unneccessary, it is also stupid to use @headers if the @headers-like algorithm of HTMl 4 makes it unneccessary.) [That said, when using @headers, we are at least doing something that AT can take advantage of.] If we really want to simplify, we could just as well drop @scope entirely, and only go for @headers. That would allow us to have one attribute less. Al said: >> + note: @scope-like markup requires more burdensome processing >> on the client side to get the associations to the user than >> does @headers-like markup. I would assume that Al does not consider the general algorithm of HTML 4 as "@scope-like". @scope and "@scope-like" solutions start with the header cells and "propagate" the headerness to the other cells. The @headers-incorporating algorithm of HTML 4, as well as the very @headers attribute, starts with the cell and looks for the header cells. So it should be simpler. >> ++ browser provides via DOM a method to learn the "immediate critical >> context" (in bottom-up @headers-like direction) for cells that combines >> the results of @scope-implications analysis with @headers data. These >> are cumulative; @headers does not cancel @scope. > > Assuming @headers sticks around and isn't replaced by something more > general like @aria-labelledby, I don't agree with this scenario. Headers > associated through scope (implicit or explicit) should be cumulative, > but if an author has explicitly defined @headers, that specific > association should cancel @scope. Here I agree with you. This is also the way HTML 4 sees it. -- leif halvard silli
Received on Tuesday, 23 September 2008 09:32:27 UTC