AX: AOM: Implement relation type properties

• Component: Accessibility
• Status: RESOLVED FIXED

[Nan Wang]

AOM Phase 1:
Implement relation type properties (aria-activedescendant, etc.).

Spec: https://wicg.github.io/aom/spec/

<rdar://problem/35077972>

[Nan Wang]

Created attachment 328063 [details]
Patch

[Nan Wang]

Created attachment 328064 [details]
patch

[EWS Watchlist]

Attachment 328064 [details] did not pass style-queue:


ERROR: Tools/WebKitTestRunner/InjectedBundle/AccessibilityUIElement.h:199:  'ariaDetailsElementAtIndex' is incorrectly named. It should be named 'protector' or 'protectedUnsigned'.  [readability/naming/protected] [4]
ERROR: Tools/WebKitTestRunner/InjectedBundle/AccessibilityUIElement.h:200:  'ariaErrorMessageElementAtIndex' is incorrectly named. It should be named 'protector' or 'protectedUnsigned'.  [readability/naming/protected] [4]
Total errors found: 2 in 15 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Ryosuke Niwa]

Comment on attachment 328064 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328064&action=review

> Source/WebCore/accessibility/AccessibilityObject.cpp:2194
> +        auto elements = AccessibleNode::effectiveElementsValueForElement(*element, propertyKey);
> +        if (elements.size() == 1)
> +            return elements.first();

What happens when elements contain more than one elements?
Should that ever happen? If so, is it really correct to return nullptr in that case?
Where is this behavior spec'ed?

> Source/WebCore/accessibility/AccessibilityObject.cpp:3469
> +    bool idEmpty = id.isEmpty();

Nit: should be idIsEmpty.

> Source/WebCore/accessibility/AccessibleNode.cpp:337
> +AccessibleNode* AccessibleNode::singleRelationValueForProperty(Element& element, AXPropertyName propertyName)

Please return RefPtr<AccessibleNode> for pointer safety.

> Source/WebCore/accessibility/AccessibleNode.cpp:340
> +    if (accessibleNodes.size() == 1)

Is it possible for accessibleNodes to have more than one items? If so, is it really correct to return nullptr.
If that's not possible, we should assert that accessibleNodes.size() is either 0 or 1,
and this check should instead be replaced by if (accessibleNodes.size()).

> Source/WebCore/accessibility/AccessibleNode.cpp:345
> +Vector<AccessibleNode*> AccessibleNode::relationsValueForProperty(Element& element, AXPropertyName propertyName)

Please return Vector<Ref<AccessibleNode>> instead.

> Source/WebCore/accessibility/AccessibleNode.cpp:353
> +Vector<Element*> AccessibleNode::effectiveElementsValueForElement(Element& element, AXPropertyName propertyName)

Please return Vector<Ref<Element>> instead.

> Source/WebCore/accessibility/AccessibleNode.cpp:383
> +AccessibleNode* AccessibleNode::activeDescendant() const

Please return RefPtr<AccessibleNode>

> Source/WebCore/accessibility/AccessibleNode.cpp:482
> +AccessibleNode* AccessibleNode::details() const

Please return RefPtr<AccessibleNode>

> Source/WebCore/accessibility/AccessibleNode.cpp:498
> +AccessibleNode* AccessibleNode::errorMessage() const

Please return RefPtr<AccessibleNode>

[Nan Wang]

Comment on attachment 328064 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328064&action=review

>> Source/WebCore/accessibility/AccessibilityObject.cpp:2194
>> +            return elements.first();
> 
> What happens when elements contain more than one elements?
> Should that ever happen? If so, is it really correct to return nullptr in that case?
> Where is this behavior spec'ed?

I implemented this as a convenient function for relation types that only contain one AccessibleNode. Maybe we should add an assert here to make sure it only contains one element.
The more than one elements case should be addressed in another bug. In that case we could have elementsValueForProperty(property)
https://wicg.github.io/aom/spec/#accessiblenodelist-interface

>> Source/WebCore/accessibility/AccessibleNode.cpp:340
>> +    if (accessibleNodes.size() == 1)
> 
> Is it possible for accessibleNodes to have more than one items? If so, is it really correct to return nullptr.
> If that's not possible, we should assert that accessibleNodes.size() is either 0 or 1,
> and this check should instead be replaced by if (accessibleNodes.size()).

Yes I think an assert here is more appropriate. This is to cover the properties that only should contain one AccessibleNode.

[Ryosuke Niwa]

Comment on attachment 328064 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328064&action=review

r=me with the assertion change.

>>> Source/WebCore/accessibility/AccessibilityObject.cpp:2194
>>> +            return elements.first();
>> 
>> What happens when elements contain more than one elements?
>> Should that ever happen? If so, is it really correct to return nullptr in that case?
>> Where is this behavior spec'ed?
> 
> I implemented this as a convenient function for relation types that only contain one AccessibleNode. Maybe we should add an assert here to make sure it only contains one element.
> The more than one elements case should be addressed in another bug. In that case we could have elementsValueForProperty(property)
> https://wicg.github.io/aom/spec/#accessiblenodelist-interface

Okay. Please add an assert to check that there is either 0 or 1 element.

[Nan Wang]

Committed r225511: <https://trac.webkit.org/changeset/225511>