A number of accessibility tests have started crashing after r285859 added a new runtime assert to AccessibilityUIElement::create() Regressions: Unexpected crashes (12) accessibility/aria-hidden-false-works-in-subtrees.html [ Crash ] accessibility/aria-hidden-updates-alldescendants.html [ Crash ] accessibility/aria-modal-text-descendants.html [ Crash ] accessibility/auto-fill-types.html [ Crash ] accessibility/hidden-th-still-column-header.html [ Crash ] accessibility/onclick-handlers.html [ Crash ] accessibility/removed-anonymous-block-child-causes-crash.html [ Crash ] accessibility/removed-continuation-element-causes-crash.html [ Crash ] accessibility/scroll-to-make-visible-iframe-offscreen.html [ Crash ] accessibility/svg-text.html [ Crash ] accessibility/table-cell-display-block.html [ Crash ] accessibility/table-notbody.html [ Crash ]
<rdar://problem/85481838>
Created attachment 444456 [details] Patch
The current version of the patch causes a couple of failures (regressions from the previous status quo), as opposed to crashes (current status quo): Regressions: Unexpected text-only failures (2) accessibility/aria-modal-text-descendants.html [ Failure ] accessibility/table-cell-display-block.html [ Failure ]
Sorry for the breakage, Arcady. I checked GTK's usage of AccessibilityUIElement::create() and it all looked OK, but clearly I missed some. Regarding the text failures: accessibility/aria-modal-text-descendants.html [ Failure ] accessibility/table-cell-display-block.html [ Failure ] https://commits.webkit.org/r285859 prevents the creation of a UI element backed by a null pointer, which always seems like the correct thing to do. Do the text differences look easily reconcilable?
Created attachment 444583 [details] Patch
Comment on attachment 444583 [details] Patch We also need to handle nullptr in AccessibilityController::focusedElement().
Created attachment 444765 [details] Patch
Committed r286052 (244439@main): <https://commits.webkit.org/244439@main> All reviewed patches have been landed. Closing bug and clearing flags on attachment 444765 [details].