Here are the crash reports: http://crash-stats.mozilla.com/report/list?range_unit=days&query_search=signature&query_type=contains&version=Firefox%3A3.0b3pre%2CSeaMonkey%3A2.0a1pre%2CSunbird%3A0.6a1%2CThunderbird%3A3.0a1pre&signature=nsAccessible%3A%3AInvalidateChildren%28%29&query=Access&range_value=2 It crashes when we try to use reinterpret_cast to get an nsAccEvent* from the nsIAccessibleEvent* and access the mData directly. 1) we don't need to access it directly 2) we should be using event interfaces 3) the event data is not really even currently used, although I suspect we should be firing TABLE_ROW_REMOVED events for ATK when a tree row is deleted or inserted. Because of #3, I'd like to keep the code for dealing with table events and create an interface for it for future use. We can remove now or later it if is not needed. But, we don't need property change events. I don't see the value in firing events for changes of any of those properties. If we really need them in the future we should create an nsIAccPropertyChangeEvent.

This should fix the crash because it was in the accessing of eventData that we crash.

I think the link in bug description should be http://crash-stats.mozilla.com/report/list?range_unit=weeks&query_search=signature&query_type=contains&signature=nsAccessibleWrap%3A%3AFireAccessibleEvent%28nsIAccessibleEvent%2A%29&query=FireAccessibleEvent&range_value=1 I agree to remove property change events for now, and keep table events. Although we didn't have code to use nsAccTableChangeEvent yet.