Re: Implementation of boolean types.
Date: 16 Jul 2005 10:27:01 -0700
Message-ID: <1121534821.846347.260300_at_o13g2000cwo.googlegroups.com>
Mikito Harakiri wrote:
> Tony Andrews wrote:
> > That would be a dumb treatment, really. NULLs propagate in
> > computations for numbers, because it makes sense to do so: we really
> > don't know what 1+NULL is equal to.
>
> Not quite: 1+NULL = NULL while sum(1,1,NULL) = 2
> No elementary consistency.
yeah, that's really weird. the 1+null case shows null having 'unknown' semantics. the sum(1,1,null) case shows null having 'missing value' semantics. So which is it? Clearly, there is no particular consistency.
I would prefer a treatment with better consistency. In particular, I see value in using 'missing value' semantics, because how they should behave is quite clear, whereas unknown introduces 3vl which I find unattractive, from the standpoint that it introduces a lot of complexity for little payback. I also think that unknown semantics are better handled by user-defined special values.
Marshall Received on Sat Jul 16 2005 - 19:27:01 CEST