Re: Seeking clarification for "undefined" as a literal value

Rich, I'm currently looking at this to make sure Chrome does the right
thing.

What do you think about the confusion Joanie pointed out regarding
aria-current?

Aaron

On Fri, Jun 30, 2017 at 2:06 PM Aaron Leventhal <aleventhal@google.com>
wrote:

>
>
> In CORE-AAM, the "undefined" literal string is only mentioned in this
> section, which only discusses unknown or disallowed values, whereas
> "undefined" can be an allowed value. Further, "undefined" as a literal
> string is only addressed in a sub section of that about boolean states, yet
> "undefined" can be used on non-boolean properties.
>
> If a WAI-ARIA property contains an unknown or disallowed value, the user
> agent should expose to platform accessibility APIs
> <http://rawgit.com/w3c/aria/master/core-aam/core-aam.html#dfn-accessibility-api>
>  as follows:
>
>    - When exposing as an object attribute, expose the unknown value — do
>    not vet it against possible values.
>    - When exposing as a platform API Boolean state
>    <http://rawgit.com/w3c/aria/master/core-aam/core-aam.html#dfn-state>:
>       - For values of "" (empty string), "undefined" or no attribute
>       <http://rawgit.com/w3c/aria/master/core-aam/core-aam.html#dfn-attribute>
>        present:
>          - Follow the guidance in the Default values for missing required
>          attributes
>          <http://rawgit.com/w3c/aria/master/core-aam/core-aam.html#authorErrorDefaultValuesTable>
>           table below, if applicable.
>          - Otherwise, treat as false.
>       - Treat any other value as true.
>    - Otherwise, ignore the value and treat the property as not present.
>
>
> On Fri, Jun 30, 2017 at 12:04 PM Rich Schwerdtfeger <richschwer@gmail.com>
> wrote:
>
>> Yes. The reason for undefined was that we could not assume that something
>> was false by default - especially when in the case of a checkbox it could
>> be “mixed”.
>>
>> If a property is required for a role it could be undefined. … You might
>> be waiting for a server to tell you what the value is such as for
>> preferences for a given user.
>>
>> I am a little confused about where you think the core-aam is unclear.
>>
>> Here is the current core-aam. The spec. clear states what the default
>> values must be for states and properties and we have mappings associated
>> with each value (which indicates the default):
>>
>> http://rawgit.com/w3c/aria/master/core-aam/core-aam.html
>>
>> see aria-disabled. It indicates the default. The core-aam reflects the
>> spec.
>>
>>
>>
>> Rich
>>
>>
>> Rich Schwerdtfeger
>>
>>
>>
>>
>> On Jun 30, 2017, at 10:13 AM, Aaron Leventhal <aleventhal@google.com>
>> wrote:
>>
>> To be clear, I've never seen the "undefined" literal used in the real
>> world, but it's good to get this right anyway.
>>
>> Are these correct:
>>
>> - The author can explicitly set the "undefined" literal unless the
>> property is required for the role.
>>
>> - Setting "undefined" is equivalent to using the default value
>>
>> - The default value itself may be "undefined" (like in aria-checked) or
>> not (e.g. "false" for aria-disabled). Therefore, using
>> "undefined"/undefined may result in undefined or a specific value depending
>> on the property.
>>
>> IMO the CORE-AAM could use more clarification. I don't find it covers
>> these cases -- at least it's not clear to me.
>>
>> Aaron
>>
>>
>>
>> On Fri, Jun 30, 2017 at 9:09 AM Joanmarie Diggs <jdiggs@igalia.com>
>> wrote:
>>
>>> This does have interesting implications for (at least) aria-current.
>>> According to the ARIA spec:
>>>
>>>     The aria-current attribute is an enumerated type. Any value not
>>>     included in the list of allowed values should be treated by
>>>     assistive technologies as if the value true had been provided. If
>>>     the attribute is not present or its value is an empty string, the
>>>     default value of false applies and the aria-current state must not
>>>     be exposed by user agents or assistive technologies.
>>>
>>> Thus the language is consistent with what Rich said, which means: If the
>>> value of aria-current is undefined (in the sense of a value having not
>>> been provided), the default of false applies, the element is not
>>> current, and the aria-current state must not be exposed. BUT, if the
>>> value of aria-current is "undefined" (a string literal), then we have a
>>> value not included in the list of allowed values, which should be
>>> treated as if aria-current were set to true (which means the
>>> aria-current state must be exposed by user agents).
>>>
>>> The fact that the results of undefined and "undefined" are expected to
>>> be the complete opposite gives me a headache.
>>>
>>> Also, looking at the Core AAM, aria-current is undefined is "not
>>> mapped":
>>>
>>> https://rawgit.com/w3c/aria/master/core-aam/core-aam.html#ariaCurrentUndefined
>>> .
>>> I guess Joseph thinks undefined means not defined? (To be honest, that's
>>> how my brain works too.) And I guess we need to add both flavors of
>>> undefined to Core AAM?
>>>
>>> --joanie
>>>
>>> On 06/29/2017 01:47 PM, Rich Schwerdtfeger wrote:
>>> > Aaron,
>>> >
>>> > The spec. clearly states that (see aria-checked) that “undefined” is
>>> the
>>> > (default). If nothing is specified that is what is assumed the value
>>> is.
>>> > If the author has not set aria-checked on the role that supports the
>>> > aria-checked states then the default is undefined. When it says default
>>> > it is very clear. Had default not been indicated then I agree there
>>> > would be confusion.
>>> >
>>> > So, “undefined” is a valid value. … so is leaving the attribute off
>>> > altogether.
>>> >
>>> > In aria 1.0 here is a test example for aria-grabbed being set to
>>> > “undefined” from the Candidate Recommendation test harness:
>>> >
>>> >
>>> https://dvcs.w3.org/hg/pfwg/raw-file/default/ARIA/1.0/tests/test-files/roles-properties-global/roles-properties-global-main-aria-grabbed-undefined.html
>>> >
>>> > This passed candidate recommendation and therefor we had working
>>> > implementations.
>>> >
>>> > The important thing to remember is the default value. So, programmers
>>> > can leave it off altogether.
>>> >
>>> > Rich
>>> >
>>> >
>>> > Rich Schwerdtfeger
>>> >
>>> >
>>> >
>>> >
>>> >> On Jun 29, 2017, at 1:32 PM, Aaron Leventhal <aleventhal@google.com
>>> >> <mailto:aleventhal@google.com>> wrote:
>>> >>
>>> >> As I understand ARIA, a possible value of "undefined" means the
>>> >> attribute is not present. If undefined is also the default, then a
>>> >> value of "" is equivalent. However, I would not expect a user agent to
>>> >> process the literal string "undefined" as undefined.
>>> >>
>>> >> Was there an expectation somewhere that the literal string "undefined"
>>> >> should be treated as attribute not present?
>>> >>
>>> >> I feel that the ARIA 1.1 spec could be more clear here:
>>> >>
>>> >> 'The "undefined" value, when allowed on a state or property, is an
>>> >> explicit indication that the state or property is not set. The value
>>> >> is used on states and properties that support tokens, and the
>>> >> "undefined" value is a string that is one of the allowed tokens. It is
>>> >> also used on some states and properties that accept true/false values,
>>> >> when "undefined" has a different meaning than "false".'
>>> >>
>>> >> Perhaps when undefined is discussed it should not be put in quotes --
>>> >> to programmers this means literal string. Mostly, CORE-AAM does this,
>>> >> but it does have one place under disallowed values that discusses
>>> >> "undefined" as a literal string. It does not, however, discuss the
>>> >> "undefined" literal as an allowed value.
>>> >>
>>> >> Can someone provide more clarity for our implementation? I'd like to
>>> >> see more clarity in both specs.
>>> >>
>>> >> Thank you,
>>> >>
>>> >> - Aaron
>>> >>
>>> >>
>>> >>
>>> >
>>>
>>>
>>