The old method of redefining destroy meant that clearing the HABTM join table would happen as long as the call to destroy succeeded.  Which meant if there was a before_destroy that stopped the instance being destroyed using normal means (returning false, raising ActiveRecord::Rollback) rather than exceptional means the join table would be cleared even though the instance wasn't destroyed.  Doing it in an after_destroy hook avoids this and has the advantage of happening inside the DB transaction too.