http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54296
--- Comment #2 from François Dumont <fdumont at gcc dot gnu.org> 2012-08-26 10:55:43 UTC --- I will have a closer look but what I can say for the moment is that the tested source code doesn't seem to match the latest trunk state, the line numbers don't match. Can you have a check with latest version ? For info, there is a check on the address of the key to avoid just what you describe that is to say deallocate a node and keep on using the key part of it.