Re: Proposal for three tier RDB2RDF standard - and relationship with SQL core

On 7/21/10, Daniel Miranker <miranker@cs.utexas.edu> wrote:
> 		
> We are already down a path for a multipart standard
>
> - with and with out schema
> - SQL or RIF etc.
>
>
> Please consider the following framework for the RDB2RDF standard.
> (where framework means a starting point for discussion.)
>
>
> I suggest three elements for an implementation of the RDB2RDF
> standard wrt an implementation claiming to be standard compliant;
> kind of like levels in SQL and many other systems.
>
>
> 1. Required
> -- All implementations must provide this part in a consistent way
>
> 2. Optional
> -- If an implementation includes these parts, the details are specified
>
> 3. Suggested
> -- For implementations of these parts, the standard has some
> suggestions, but these are not binding on an implementation
>
>
> // I see I just crossed emails with Richards email "the minimum we
> need"....
>
>
> In this structure I see a basis for putting some of our issues to
> rest, and benefit from the concomitant concrete for forward progress.
>
>
> A) Required
>
> I suggest we can probably agree that there should be
> - a standard mapping of SQL data types to RDF,
> and
> - a standard way of forming URIs
>
> and, that this is required of all implementations.
>
> B) Optional
>
> An implementation may omit schema mapping and thus still be compliant.
>
> Every system that does implement schema mapping will do so in a
> standard way.
>
>
> C) Suggested
>
> Example - how to deal with aggregation and/or other things that might
> be introduced by looking
> at advanced features of SQL, RIF, SPARQL 2 vs. SPARQL
>
>
> I introduce this third part as a kind of engineering solution for
> those parts of the system that
> it is sort-of-obvious wrt what this means for engineers, but the
> formalization is either tough
> or premature.
>
>
> To elaborate using aggregation as an example:  The SQL side of the
> group has a clear idea, syntax and semantics,
> of what it will mean to include aggregation in mapping rules
> expressed in a SQL syntax.
>
> On the Semantic Web language side of the group (actually beyond),
> aggregation is something that is in flux.
>
> So, "suggested", is a bucket where we can go on record for a direction
> for functional parts of the standard where any of the following is true:
>
> - Either due to limits on our own time, or lack of maturity on things
> out of our control, we can only sketch
> an implementation example.
>
> - We're all in agreement on something, but the formal semantics is
> tricky.  Where tricky means, either, it is beyond convenient
> systems for defining semantics (e.g. beyond the scope of pure
> datalog.)  and/or until we implement some of these we really
> don't know.  It would never the less be useful to get everyone on
> roughly the same page.

I like very much this idea of having different levels. In fact, it
would be nice to have in one of the levels a mapping language with
simple syntax and semantics, which is extended in other levels with
extra functionalities.

Please take a look at the new version of:

http://www.w3.org/2001/sw/rdb2rdf/wiki/Database-Instance-Only_and_Database-Instances-and-Schema_Mapping

It now includes formal definitions of the syntax and semantics of the
Datalog fragment that we have been mentioning. I think this language
would be a good alternative to provide a formal definition of the
mapping language in one the levels, as its syntax and semantics can be
defined in a few pages, and it has the same expressive power as
first-order logic (and relational algebra).

All the best,

Marcelo

Received on Thursday, 22 July 2010 02:08:50 UTC