Thanks to Rich for analysing the bug introduced by the previous commit. The
attached patch should fix it.
This commit made for the bug 635987 introduced a bug to replication.
Author: Noriko Hosoi<nhosoi(a)redhat.com>
Date: Tue Sep 21 15:12:07 2010 -0700
If you do have a tombstone filter, descendants will be NULL,
and idl_intersection of candidates and descendents will wipe
out all of the candidates, leaving just the one entry, e->ep_id.
Changed to call idl_intersection only when the filter is not
for tombstone or entryrdn_get_noancestorid (false, by default).