pgkos.bugzilla basically did, what I proposed at comment #95 half a year
ago: He identified the critical sections and implemented synchronization
to prevent the race condition. And that's exactly the right way to fix
this bug! If pgkos.bugzilla determined the CS right, the rest should be
a child's
slumbergod is right: The bug has not been tracked down, yet. By now it
seems pretty clear that the bug is caused by different threads that
operate on the same data(structures). Unfortunately bugs, which are
related to multithreading are hard to track down, since they are non-
deterministic and
