|
|
Posted by Ed Murphy on 04/06/07 02:55
Hugo Kornelis wrote:
> On Thu, 29 Mar 2007 18:08:00 -0700, Ed Murphy wrote:
>
>> --CELKO-- wrote:
>>
>>> All data types have to be NULL-able in SQL. Having a BOOLEAN type
>>> would lead to 4 valued logic with inconsistent rules about how NULLs
>>> propagate. And the various vendor extension do not work or port
>>> either.
>> It seems like dropping UNKNOWN would leave a sensible set of rules:
>>
>> and | T N F or | T N F not |
>> ----+------ ---+------ ----+--
>> T | T N F T | T T T T | F
>> N | N N F N | T N N N | N
>> F | F F F F | T N F F | T
>>
>> Am I overlooking anything?
>
> Hi Ed,
>
> You've overlooked the basic rule of NULL propagation: any expression
> involving NULL results in NULL. In the tables above, there are
> exceptions to this rule, such as NULL AND FALSE resulting in FALSE, and
> TRUE OR NULL resulting in TRUE.
That rule is oversimplified. Really, it should be "any expression
whose value _depends_ on a NULL input results in NULL", i.e. could
replacing the NULL with different non-NULL values lead to different
values of the expression? NULL + 2 qualifies; NULL = 'ABC' qualifies;
but NULL AND FALSE does not, and neither does TRUE OR NULL.
(As usual, IS NULL and IS NOT NULL remain special cases.)
Navigation:
[Reply to this message]
|