Re: NULLs: theoretical problems?

From: Keith H Duggar <>
Date: Sat, 01 Sep 2007 14:15:22 -0700
Message-ID: <>

Bob Badour wrote:
> IOW, you are suggesting a consistent SUM--unlike SQL--that correctly
> sums to UNKNOWN when any term is missing. Yes, in that case the key
> difference is static versus dynamic checking. Compared to SQL, both of
> those options are automatic versus manual/no checking.
> One could also argue that Jan's proposal limits the locus of effect. A
> null could propagate throughout a whole system before anyone notices it,
> in which case it could prove time-consuming to locate the origin.

Indeed. A colleague once commented to me that "NaN is the original computer virus."

> I assume a proof system would also automatically detect or require
> constraints like "x != 1" in:
> lim [(x^2 - 1) / (x - 1)], x != 1
> x->1
> or more simply
> f(x) = [(x^2 - 1) / (x - 1)], x != 1

In the case of the IMPS I mentioned before, all functions that are not propositions ie all functions whose range is other than {TRUE, FALSE} are partial functions. Thus, for the case above f(x) would simply be a partial function on domain Sort(x) that was undefined for x == 1. Either this can be stated explicity (as you did above) or you can let the proof system automatically deduce the definedness. So exactly as you said, it can automatically detect or allow one to specify such constraints.

KHD Received on Sat Sep 01 2007 - 23:15:22 CEST

Original text of this message