Re: Information Preserving and ISSUE-42

On 18 May 2011, at 15:33, Juan Sequeda wrote:

> On Wed, May 18, 2011 at 2:30 AM, Enrico Franconi <franconi@inf.unibz.it> wrote:
> On 18 May 2011, at 06:59, Juan Sequeda wrote:
> 
> > So does that mean that you would translate all NULL values to a triple with rdb2rdf:NULL? That makes sense and would make the current direct mapping information preserving.
> 
> Definitely not.
> Take the case of a RDB with a Person table, having ID, name, and age.
> If you query (in SQL with a simple conjunctive query) all the people with the same age of a specific person in the table having a NULL value as age, you do *not* get the people whose age is a NULL value. However, with your naive translation in RDF of this database, you would get the wrong answer with the same query translated in SPARQL.
> 
> Yes, if you have a query like the one you are stating. But for the direct mapping, the input is the complete table. So it shouldn't be a problem... right?

I don't get what you are saying, sorry :-(
A simple BGP is enough to cover this example. 
In other words, If I do translate the NULL value as a constant, and I try to use a BGP to write a query giving me the people with the same age as a person with a NULL value as an age, I will fail miserably if I don't explicitly mention in the query the special case of the NULL value (which, by the way, goes beyond BGPs).
cheers
--e.