At 03:16 AM 26/10/2007, you wrote:

>On 22 Oct 2007, at 15:16, Helen Borrie wrote:
>
>> Even though Richard thinks he has resolved the performance problem
>> by indexing the zipcode field(s), it doesn't address a worse one:
>> the use of floating-point values as join criteria or, indeed, any
>> equivalence search such as the pivot in what he refers to as
>> "dimensional data". Such keys need to be stored as fixed types to
>> avoid a (quite unknown) number of false negatives.
>
>Is this caused by floating point representation issues?

Yes.

>And would this affect integer representations?

No.

./hb