RFR: 8377428: VoiceOver Cursor Navigates Invisible Components [v2]

[Jeremy Wood]

On Mon, 9 Feb 2026 23:05:51 GMT, Sergey Bylokhov <serb at openjdk.org> wrote:

>> Jeremy Wood has updated the pull request incrementally with two additional commits since the last revision:
>> 
>>  - 8377428: simplify test instructions
>>    
>>    Now that there are other interesting components in the UI: we don't need to press CTRL + ALT + UP.
>>  - 8377428: test condition where axComp is null
>>    
>>    If AccessibleComponent is null, when the new CAccessibility.isShowing(..) method falls back to consulting the AccessibleStateSet.
>>    
>>    This change makes sure we follow that code path, too.
>>    
>>    When I checked CAccessibility.isShowing() in the debugger, I observed:
>>    
>>    For "row 1": we had an AccessibleComponent where isShowing() is false
>>    
>>    For "row 2": we had a no AccessibleComponent. We checked the AccessibleStateSelection and did not see SHOWING, so CAccessibility.isShowing(context) returned TRUE. This is not accurate, but it's working as designed. We had incomplete information, and we're supposed to err on the side of returning true (to minimize invasive risk). Meanwhile: the test still passes, because then _addChildren recursively inspects the JButton, and CAccessibility.isShowing(buttonContext) returns false.
>>    
>>    For "row 3": we had an AccessibleComponent where isShowing() is true.
>>    
>>    For "row 4": we had no AccessibleComponent. We checked AccessibleStateSelection and DID see SHOWING, so we returned true.
>>    
>>    This is in response to:
>>    https://github.com/openjdk/jdk/pull/29630#discussion_r2784969123
>
> src/java.desktop/macosx/classes/sun/lwawt/macosx/CAccessibility.java line 1077:
> 
>> 1075:         }
>> 1076: 
>> 1077:         AccessibleStateSet ass = context.getAccessibleStateSet();
> 
> Could you please add a test to verify that these methods from AccessibleContext are actually working? I doubt we currently have any tests that cover this code path.

I expanded the test to include 4 rows: two visible, and two with no AccessibleComponent.

I confirmed in IntelliJ's debugger that each of the 4 predicted scenarios are addressed. I detailed what I observed here: https://github.com/openjdk/jdk/pull/29630/changes/1a6b53be73e6445d79216abe94a73922b2d17a66

I'm also fine with removing these new references to AccessibleStateSet in `isShowing(..)` entirely. It's intended to be a fallback. I have seen several examples of incomplete accessibility info, but I don't think I've seen a case where the AccessibleComponent is missing. (I have seen cases where the AccessibleStateSet contains incomplete data, though. That's why I only look for the presence of the SHOWING state, but I don't interpret the absence of the SHOWING state as a clear guarantee that we're not showing...)

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/29630#discussion_r2785294585