On 03/14/2011 09:25 PM, Merlin Moncure wrote:
>> Unless the point is to guarantee uniqueness of the "long-long value"s.
>
> md5 will do that too: the main thing you lose going to hash indexing
> is ordering.
MD5 will *probably* guarantee the uniqueness of the values. Personally
I'm not a big fan of betting on that, and tend to like to test for
equality against the hash first and then, if the hashes are equal,
against the values.
I have several md5 collisions in one of my tables, so it's far from
impossible. In fact, it's much more likely than you'd expect thanks to
the birthday paradox.
--
Craig Ringer