> As it's a hash, of course, it is not guaranteed to generate unique
> values for unique rows. In practice I'd be very(!) surprised if you
> got a clash, but it it was crucially important that keys were
> one-to-one with records you might want to find an alternative.

Yeah, I did a little reading between there and here -- it does seem quite
unlikely, but possible. Is there an alternative that's in the same ballpark,
but not what angus suggested? I'm guessing not -- unless there was something
inherent in my dataset that I could take advantage of.

I'll probably go with MD5 -- for my needs I'd be able to detect a duplicate
hash should it ever occur and deal with it then.

Chris