Re: n-triples for datatype values [was: Agenda for RDFCore WG Telecon 2002-10-18]

On Sunday, November 3, 2002, at 05:28 AM, Graham Klyne wrote:

> My problem is that rational cannot be handled as an optimization of 
> any of these others:  there are some values that rational can 
> represent that cannot be represented exactly by the others (e.g. 1/3). 
>  (This was an objection I had to XML schema datatypes, to which I 
> never really received a satisfactory response, but that's another 
> story.)
>
> This idea of only allowing "extended by standards body work" seems to 
> me at odds with the W3C's approach to extensibility in other areas.  
> In the case of RDF, we're allowing that certain entailments may not be 
> available if the datatype mapping (lexical->value) is not understood 
> by the software, but the framework is still usable for applications 
> that don't need to understand those datatypes.  That seems like a good 
> approach to me.
>
> So, if in your software you see "foo"^^datatypeURI with some datatype 
> URI you don't understand, you just keep the whole thing together and 
> treat it as an opaque blob, equal to somne other instance of 
> "foo"^^datatypeURI but different from "bar"^^datatypeURI or 
> "foo"^^someOtherURI.
>

I agree with everything so long as we are careful about the very last.  
  You cannot know whether "foo"^^datatypeURI is mathematically different 
from "bar"^^datatypeURI or from "foo"^^someOtherURI if different from 
means daml:equivalentTo, etc. Anything which deals with equality (like 
cardinality, uniqueness) has to have a concept of equality such that 
you don't know whether things of unknown datatype  are unequal.  So 
while you can say that in the RDF triples sense, 3/1 the rational 
number and 3 the integer are different nodes, it isn't very useful for 
inference.

Tim


> #g
> --
>
> At 05:31 PM 11/1/02 -0500, Tim Berners-Lee wrote:
>> I think fixed in this case means "a small number extended by 
>> standards body
>> work",
>> rather than 'anyone can make a new one".  Interoperability at the 
>> atomic
>> datatype
>> level is important.  But maybe I am being near-sighted.  In languages 
>> like
>> python
>> it is important to have a very well-defined common set of atomic 
>> datatypes,
>> but the again the ability to make new ones is rather neat.
>>
>> I guess I could imagine the implementations of code for integer, real,
>> floating point
>> and rational arithmetic being handled as optimizations.
>>
>> Tim
>>
>> ----- Original Message -----
>> From: "Graham Klyne" <Graham.Klyne@MIMEsweeper.com>
>> To: "Jos De_Roo" <jos.deroo.jd@belgium.agfa.com>
>> Cc: <w3c-rdfcore-wg@w3.org>; "Tim Berners-Lee" <timbl@w3.org>
>> Sent: Friday, November 01, 2002 4:58 AM
>> Subject: Re: n-triples for datatype values [was: Agenda for RDFCore WG
>> Telecon 2002-10-18]
>>
>>
>> > At 01:31 AM 11/1/02 +0100, Jos De_Roo wrote:
>> > > > I feel that "^^", being syntactic, should only be usable with a
>> > > > fixed set of type URIs.
>> > >
>> > >that's indeed better
>> >
>> > I have a concern with that.  For example rational values as 
>> described in
>> > CC/PP.  I'm rather concerned that the type system would be closed.
>> >
>> > [later]
>> >
>> > Or does "fixed" in this context mean non-variable?  I have no 
>> problem with
>> > that.
>> >
>> > #g
>> >
>> >
>> > -------------------
>> > Graham Klyne
>> > <GK@NineByNine.org>
>> >
>
> -------------------
> Graham Klyne
> <GK@NineByNine.org>

Received on Tuesday, 5 November 2002 11:53:03 UTC