E-mail List Archives

Re: [free-aria] Important differences between ARIA dialog, alertdialog, and role=presentation support

for

From: Bryan Garaventa
Date: May 31, 2012 12:47PM


I forgot to mention, this is in regard to the bug at
http://www.nvda-project.org/ticket/2390

----- Original Message -----
From: "Bryan Garaventa" < <EMAIL REMOVED> >
To: < <EMAIL REMOVED> >; < <EMAIL REMOVED> >
Cc: "WebAIM Discussion List" < <EMAIL REMOVED> >
Sent: Thursday, May 31, 2012 11:15 AM
Subject: Re: [WebAIM] [free-aria] Important differences between ARIA
dialog,alertdialog, and role=presentation support


> That sounds good to me as well. If we can reach a consensus about how to
> implement descriptive overrides for occasions that need it, this will help
> everyone.
>
> This is why I suggested the use of role=presentation as a way of roping
> off
> content that is unsuitable to be included in such an automated
> description.
> I know the arguments about the purpose of role=presentation, but they too
> aren't clearly defined.
>
> One problem I see with trying to automatically compensate for what
> developers fail to do properly, is that it will encourage lazy coding in
> the
> future.
>
> So if others have suggestions for a solution to this, please speak up.
>
> ----- Original Message -----
> From: "James Teh" < <EMAIL REMOVED> >
> To: < <EMAIL REMOVED> >
> Sent: Thursday, May 31, 2012 12:27 AM
> Subject: Re: [free-aria] Important differences between ARIA dialog,
> alertdialog, and role=presentation support
>
>
>> On 31/05/2012 3:21 PM, Bryan Garaventa wrote:
>>> What concerns me, is that NVDA has implemented an HTML markup
>>> requirement for ARIA support that is not documented in the spec for user
>>> agents.
>> Sure. I do recognise and understand your concern. The work around I
>> suggested is clearly very ugly. However:
>>
>>> By trying to automatically correct the mistakes of developers who should
>>> know better, we are inadvertently deviating from the spec itself, and
>>> breaking the functionality of ARIA standard compliant widgets.
>> The fundamental problem for us is that an ARIA dialog is no different to
>> any other dialog, and for all dialogs, we derive text if it is not
>> explicitly supplied. As I said, if consensus is that this behaviour is
>> incorrect, we can try to implement an exception specifically for ARIA,
>> but
>> this starts to suggest that ARIA widgets should be treated differently
>> from native widgets, which smells ugly to me. We'll have to implement
>> this
>> exception separately for each browser as well.
>>
>> All of this said, the cards are all on the table now. I am keen to hear
>> others' opinions so we can work towards an acceptable solution.
>>
>> Jamie
>>
>> --
>> James Teh
>> Director, NV Access Limited
>> Email: <EMAIL REMOVED>
>> Web site: http://www.nvaccess.org/
>> Phone: +61 7 5667 8372
>>
>>
>
> ----- Original Message -----
> From: "Bryan Garaventa"
> To: < <EMAIL REMOVED> >
> Sent: Wednesday, May 30, 2012 10:21 PM
> Subject: Re: [free-aria] Important differences between ARIA dialog,
> alertdialog, and role=presentation support
>
>
>> I'm not questioning the complexity or your workmanship. I think you guys
>> are awesome.
>>
>> What concerns me, is that NVDA has implemented an HTML markup requirement
>> for ARIA support that is not documented in the spec for user agents.
>>
>> I understand that the name (label) is announced before the description.
>> The point is, that without resorting to an HTML hack that only works in
>> NVDA, developers cannot stop NVDA from automatically announcing a
>> description.
>>
>> You are correct that many dialogs have text. However this is easily dealt
>> with using ARIA by voluntarily adding the aria-describedby attribute to
>> the dialog like so.
>>
>> <div role="dialog" aria-describedby="myDesc" >
>> <div id="myDesc">
>> Dialog description...
>> </div>
>> ... Form ...
>> </div>
>>
>> The ARIA standards exist for this reason, to advise developers how to
>> properly implement these controls. If a developer is ARIA savvy enough to
>> add role=dialog to their control, we have to assume they are smart enough
>> to know and read from the spec that supporting text should also be
>> referenced in this manner.
>>
>> By trying to automatically correct the mistakes of developers who should
>> know better, we are inadvertently deviating from the spec itself, and
>> breaking the functionality of ARIA standard compliant widgets.
>>
>> For example, both of the samples I attached to the bug are perfectly HTML
>> and ARIA standard compliant, but they don't work properly. They work fine
>> for keyboard only users, just not for screen reader users, because of
>> this
>> issue.
>>
>> I guess the best way to explain this is to recall the browser wars of
>> yore, where we had one hack for IE, one hack for Opera, one hack for
>> Safari, one hack for Firefox, and in the darkness bound them, in the land
>> of Mordor where the shadows lie...
>>
>> You see what I mean? If NVDA requires an HTML hack that is not supported
>> anywhere else because it's not documented in the user agent
>> specification,
>> we're perpetuating the same problem for screen reader support in the
>> future by introducing inconsistent accessibility.
>>
>>
>> ----- Original Message -----
>> From: "James Teh" < <EMAIL REMOVED> >
>> To: < <EMAIL REMOVED> >
>> Sent: Wednesday, May 30, 2012 4:38 PM
>> Subject: Re: [free-aria] Important differences between ARIA dialog,
>> alertdialog, and role=presentation support
>>
>>
>>> Hi.
>>>
>>> Just a few clarifications:
>>>
>>> On 31/05/2012 9:13 AM, Bryan Garaventa wrote:
>>>> If neither of the explicit label attributes are included, then the
>>>> label
>>>> is derived from the text of the dialog itself.
>>> NVDA itself only ever uses the label as provided by the browser.
>>>
>>>> From what I understand, the role of alertdialog is similar to dialog,
>>>> except that the content of an alertdialog is parsed and announced
>>>> automatically.
>>> It's announced automatically as the description, not the label. This is
>>> an important distinction. The description does not override the label.
>>> The label is announced before the role. The description is announced
>>> afterwards.
>>>
>>>> So the current stance from NVDA, is that dialog and alertdialog are
>>>> handled as the same widget type, and that all content within a widget
>>>> of
>>>> role=dialog is announced
>>> All content isn't announced. However, if aria-describedby isn't present,
>>> we do use an algorithm which attempts to determine the "text" of the
>>> dialog. This algorithm isn't perfect (and never will be), but we aim to
>>> filter out content which wouldn't normally be considered part of the
>>> dialog text/message. This is exactly what we do for native dialogs.
>>>
>>> My argument here is that everywhere else, dialogs are considered to
>>> potentially have text. The alertdialog case is obvious, but there are
>>> dialogs that ARIA wouldn't call alertdialogs that also have text. For
>>> example, consider a properties dialog which shows information that is
>>> not
>>> interactive. This information should be part of the dialog text (which
>>> means it will be announced to screen reader users automatically when the
>>> dialog opens). Thus, we treat an ARIA dialog the same as any native
>>> dialog and try to determine text if there is no explicit description. To
>>> NVDA, an ARIA dialog is no different to any other dialog (nor should it
>>> be). We can implement an ARIA specific hack to differentiate, but this
>>> seems contrary to the goal of ARIA widgets behaving the same as any
>>> native widget.
>>>
>>> Jamie
>>>
>>> --
>>> James Teh
>>> Director, NV Access Limited
>>> Email: <EMAIL REMOVED>
>>> Web site: http://www.nvaccess.org/
>>> Phone: +61 7 5667 8372
>>>
>>>
>
> > >