Re: use of aria-hidden to provide a text description not visible on the page.

Sorry for the lack of context. This thread was about labeling visible elements with hidden labels, for whatever reason.

For example, the Mozilla engine used to pass the labelledby relationship back to the API which would not see the label due to the fact that it was also hidden. However, in conversation, we decided that the rendering engine should pass on the computed string label (e.g. "My Label" for the dialog) to the API regardless of whether the labeling element is hidden. This is the "relatively recent" change I mentioned. It didn't specifically apply to the @aria-hidden state.

<div role="dialog" aria-labelledby="label">
	<h2 style="display:none" id="label">My Label</h2>
	… dialog contents …
</div>


On Sep 14, 2010, at 12:47 PM, David Bolter wrote:

> Sorry a bit rushed here and I don't feel like I fully understand this thread.
> 
> Aria-hidden was created to describe the fact that a node was visually hidden or not such that DOM based AT could have this information. Normally what is visually hidden or not is already very well supported by the browser accessibility API because the browser knows very intimately what is visually hidden or not.
> 
> I feel that aria-hidden might have morphed into something else here? Maybe Andi can cross check this with our implementation guide?
> 
> cheers,
> David
> 
> On 13/09/10 3:30 PM, James Craig wrote:
>> 
>> copying David Bolter and Andi Snow-Weaver.
>> 
>> On Sep 11, 2010, at 1:04 PM, Leif Halvard Silli wrote:
>> 
>>> Leif Halvard Silli, Sat, 11 Sep 2010 17:10:41 +0200:
>>> 
>>> Now I have deinstalled Jaws 10, and reinstalled Jaws 11, and the bug in 
>>> the interpretation of the test page [*] continues to be there. However, 
>>> I also found something that appears as a workaround: Aria-labelledby 
>>> can contain more than one idref. And if at least one of the idrefs 
>>> points to an element that has not been hidden, then all the elements 
>>> will be used, even if some of them are hidden. The visible element can 
>>> simply be an empty element.
>> 
>> The fact that the browser should sent a string value as the label was a relatively recent clarification in the ARIA Implementation Guide (within the last few months, I think). What Mozilla used to do was use the IDREF pointer in the accessibility API, which was problematic if that labeling element was hidden in the API. VoiceOver [WebKit actually] already calculated the string label on the user agent, and then passed the string label to the API. If I remember correctly, the PFWG decided VoiceOver [WebKit actually] was doing the right thing here, so what you're seeing is likely just a legacy implementation in Firefox. David may be able to comment on whether or not a more recent build of FF has updated that behavior.
>> 
>> James
>> 
>