Collation interacts badly with hyphens -- collation produces different ordering than Unicode code-point ordering -- so two indexes are created: one which can't have negative random integers, for testing collated ranges, and the other which can have negative random integers, for all other tests.
