Dangling SwAccessibleContext -> SwAccessibleMap

[Michael Stahl]

On 17.02.2016 17:40, Stephan Bergmann wrote:
> <http://ci.libreoffice.org/job/lo_tb_master_linux_dbg/5478/> crashed in 
> JunitTest_sw_unoapi_1 with a SEGV as below.  Smells like the 
> SwAccessibleContext (base class of SwAccessibleParagraph)'s m_pMap 
> (pointing at an SwAccessibleMap) is dangling. The lifecycle of any 
> SwAccessibleMap instances is controlled by SwViewShellImp (member 
> m_pAccessibleMap, deleted during dtor).
> 
> It wouldn't even easily work to notify all SwAccessibleContext instances 
> registered at SwViewShellImp::m_pAccessibleMap when the latter is 
> destroyed (so that the former could clear their m_pMap members), as what 
> gets registered in the map are not the SwAccessibleContext instances 
> themselves, but their GetFrame() objects (cf. 
> SwAccessibleContext::RemoveFrameFromAccessibleMap).
> 
> Any friend of the Writer and/or accessibility code got any idea?

sounds familiar - you probably have this message in the log too?
"Frame map should be empty after disposing the root frame"

for some reason a recursive dispose of the root frame's accessibility
object left some SwAccessible corresponding to a paragraph still alive.

this has been happening for years but unfortunately it is rather hard to
reproduce and would need some extra logging to track down...