Re: A numerical methods viewpoint on OO/FP/Relational

In article <_Xx17.26702$C81.2082162_at_bgtnsc04-news.ops.worldnet.att.net>, Richard MacDonald says...
>
>"Mikito Harakiri" <nospam_at_newsranger.com> wrote in message
>news:SD717.8804$Kf3.99960_at_www.newsranger.com...
>> In article <3b44a4a2$1_at_tobjects.newsource.com>, peter_douglass says...
>> >
>> >> Relational model:
 

>> >> The relational model is declarative. It defines the variables
>> >> and defines the constraints that the variables must satisfy.
>> >> It doesn't say how these constraints must be satisfied.
>> >> It simply accepts or rejects values for the variables depending
>> >> on whether these values satisfy the constraints or not.
 

>> >> Note, however, that the relational model also has a
>> >> provision for explicitly defining dependent variables as
>> >> functions of other variables.
>>
>> This definition is too broad for Relational, you essentially describe
 Constraint
>> Databases. Relational narrows it's scope to very simplistic (and highly
>> successful, as well:-) kinds of predicates.
>
>Can you explain this? What is the "break-point" between relational
>databases and constraint databases? I always thought constraint
>databases would deal with declarative or cyclical constraints, while
>relational would deal with dependents as a function of independents,
>possibly nested.
>
In relational only simplistic predicates like

Employee=Smith & SSN=111223333

are allowed. In constraint databases we can store facts like

X*X + Y*Y < 4

Even though, relational can also pretend to store facts about geometric figures like

Radius=2 & CenterX=0 & CenterY=0

it can't answer queries about spatial relationship. (I'm talking spatial as the most impressive application of constraint databases). Natural question is "If constrait DBs are so great why don't we see them around already?". Simple: because constraint databases are so general, they easily hit NP-complete problematics. So we are talking about something even worse than full table scan inefficiency in relational. For "reasonable" problems, however, innovative indexing techniques must be developed... Received on Sat Jul 07 2001 - 19:13:16 CEST