Re: NULLs: theoretical problems?

On 2007-08-12, -CELKO- <jcelko212_at_earthlink.net> wrote:
>
>
>> Proving that you did not know the difference between logical
>> and physical. Of course a NULL needs to be implemented, but
>> it is _not_ a value. Any possible implementation (and there
>> are lots) will have to be some sort of special case, but the
>> language does not need to know.
>
> I thihk it demonstrates that you have never worked on an ANSI or ISO
> committee or written a compiler :) You might want to look at the IEEE
> Floating Point Standards and their "special values" -- +inf, -inf,
> Nan, etc. Much more complex than a mere NULL! While IEEE does give
> the "bits and bytes" for their special values, we left implementation
> open but borrowed the accepted terminology from them.
>

But that was the mistake, because a NULL need not be implemented as a special value. It could be, but that is an implementation issue only. A NULL is _not_ a value, whatever you use to implement it, and you should not have called it one.

>> ["CAST (NULL AS <data type>)" to signal the SQL compiler about
>> that column's storage]
> What on earth does that mean?
>
> I thought that was pretty clear. SQL stores data; data has a srong
> data type in SQL; the compiler needs to know about it to make
> decisions and allocations.

Only if there is a situation where it can't work it out - I know, this is the next point.
>
>> [ CASE expression without explicit CAST() help] I dont't know
>> that I actually believe that.
>
> I will see if I can find one for you tomorrow -- I am babysitting my
> niece's two year old tonight and have to use her Mac.

Thankyou.

>
>> But then floating-point numbers are values that are difficult
>> to represent, whereas NULL is _not_ a value, but is easy to
>> represent. The arguments are about what it means and whether we
>> really need it.
>
> Hey, I am just providing information about SQL. I happen to think
> that NULLs can be hard to represent because they have to work with all
> kinds of data types, whereas I can burn the IEEE rules into a Math Co-
> processor at the hardware level.
>

Which sort-of proves that NULLs and FP are not really comparable.

-- 
Eric