Some tests currently present in LayoutTests/accessibility (e.g aria-help, aria-describedby-on-input) need helpText() to be implemented in DRT and WKTR, which in the mac platform relies in the AXHelp accessibility attribute, normally exposing the value that AccessibilityObject::helpText() provides, which might be associated to the aria-help attribute. However, it turns out (now I finally see it) that both the helpText concept and the aria-help attribute are Mac specific things that do not have a clear match in the ATK world, so it's pointless both to expose those 'aria-help' attributes and implement the corresponding helpText() function in DRT and WKTR. Furthermore, the way that such an 'aria-help' attribute is now being exposed in the ATK wrapper (through an object attribute) is wrong too because of two reasons: 1. There's not such a mapping to be expected for ATs as Orca, nor documented in [1], so it's useless other than for passing the tests. But those tests relying on helpText only make sense in the Mac, so it's pointless. 2. We should not call functions that might modify the accessibility tree in webkitAccessibleGetAttributes(), such as helpText() -see bug 121558 for more info-, but check and expose information that we already have at hand, such as the value of certain attributes, the toolkit name or the ARIA state of the object. Calling helpText() here will cause to call textUnderElement(), making it slower than what it should and potentially dangerous, since textUnderElement() might modify the accessibility tree. So, my proposal now is: 1. Get rid of the exposure of "aria-help" as an object attribute in WebKitAccesibilityWrapperAtk 2. Remove the whole support for helpText() from DRT (based in DumpRenderTreeSupportGTK) and from WKTR (based on the "aria-help" object attribute) 3. Move currently passing tests such as aria-help or aria-described-on-input to platform/mac/accessibility. I hope to provide a patch soon. [1] http://www.w3.org/TR/wai-aria-implementation/#mapping_state-property