NH> It seems to me that the ideal would be some sort of a hash
NH> involving a second array, as long in bits as there are integer
NH> numbers.  Each integer would set a bit.  If that bit had been
NH> previously set, bingo, a duplicate.

Something else just dawned on me.  After all the values (a few thousand 
at most) are inserted into the 4+ gigabyte array (with each value being 
represented by just 1 bit in the array) each bit in the array must be 
examined.  That makes any previously supposed advantage evaporate (and 
turn into a big-time deficit).

JB>if you sort the array and then parse it for dupes
JB>you can do that in O(n)...

I've never heard of a sorting algorithm working in O(n) time.

JB) ... time using a radix sort.
JB>and using  hopefully mucn less memory (need space for 4 times the
JB>size of the array)

JB>qsort O(n.log(n)) could be faster than radix sort for small arrays.

Define small?

What is advantageous about radix sort for large arrays?  Does large 
equal everything that's not small?

þ CMPQwk 1.42 999

--- Maximus/2 3.01