Re: Systematic approach to layout table heuristics

Thanks, looks really promising!

Leif Halvard Silli

James Craig, Fri, 21 Feb 2014 10:29:36 -0800:
> Just copy what the open source browsers are doing already. For example:
> AccessibilityTable::isDataTable()

> On Feb 21, 2014, at 2:19 AM, Leif Halvard Silli  wrote:
>> After filing bug 24679 [1], which suggests to add more features to the 
>> list of ”possible indicators”[2] for layout tables vs non-layout 
>> tables, I was asked to provide data for the proposals. But how should 
>> one go about providing data?
>> Web data is of what ultimately matters. But when identifying possible 
>> indicators, it seems that knowing what features AT actually use for 
>> discerning between layout and non-layout tables, would give us a good 
>> shortlist of candidate indicators. Secondly, if we can spot some trends 
>> in the (hopefully correct) data we already have, then that - as well - 
>> could help us identify candidates. 
>> Today, its appears possible to glean following trends in the spec:
>> 1) Conformance, completeness amd semantics indicate non-layout
>>   usage. The spec lists borders via CSS or @border=1, <th>,  
>>   <thead>, <caption>, @headers, @scope.
>> 2) Non-conforming ways to disable borders (border=0/cellspacing=0
>>   cellpadding=0) plus role=presentation indicate layout usage.
>> 3) @summary is ”not a good indicator” (this is probably based
>>   *both* on Web data *and* AT behavior analysis
>>   Trends & AT data applied to: table@border
>> To me, HTML5’s table over “possible indicators” of (non-)layout usage, 
>> seems correct, but not complete. However, in bug 24678,[3] Steve asks 
>> for data about *one* non-layout indicator, namely border=1. But which 
>> data? Web data? Assistive technology data? The fact that VoiceOver + 
>> Safari matches spec w.r.t. borders as non-layout indication, ought to 
>> put it on the shortlist of non-layout features. In another, somewhat 
>> related, bug, Steve indicated that web data did not support that 
>> table@border=1 indicates data tables.[4] My own toe dipping into same 
>> pool told me the opposite.
>>   Trends & AT data applied to: <colgroup> with <col/> children.
>> It seems in line with the conformance & completeness trend that 
>> VoiceOver+Safari treats <colgroup> with <col/> children[*] as indicator 
>> of non-layout usage. What do other ATs do? What does Web data say? 
>> Should the spec change accordingly? (PS: I know one HTML generator 
>> which, for layout tables in XHTML1/HTML4, uses cellpadding=0 
>> cellspacing=0, but which for HTML5 removes cellpadding/cellspacing and 
>> *adds* <colgroup> with children, with bad results in VoiceOver as 
>> result.)
>>   Trends & AT data applied to: lack of cellpadding=0 cellspacing=0[*]
>> Is it risky to delete cellpadding=0 cellspacing=0 from a table? 
>> Could it cause an AT to treat a layout table, as a non-layout table? 
>> This question seems relevant since authors are probably simply deleting 
>> these attributes without adding @role=presentation in their place.
>>   Trends & AT data applied to: presence of sortable attribute
>> This is good candidate based on its strong link to data tables - it is 
>> a semantic feature. But is is also possible to take the view that we 
>> need implementation and usage before putting into the spec.
>> Finally: How important are heuristics for identifying data tables? Does 
>> the situation remind about how UAs detect quirks vs no-quirks? (Except 
>> in XHTML and @srcdocs (other exceptions?), no-quirks is triggered when 
>> UAs detect a DOCTYPE that meets certain conformance and completeness 
>> criteria.)
>> [1]

>> [2] 

>> [3]

>> [4]

>> [*] I did not check omitting <col/> child or omitting parent <colgroup>
>> -- 
>> leif halvard silli

Received on Friday, 21 February 2014 21:20:01 UTC