Re: What happens when an ontology has data literals that are outside the range handled

On 13 Feb 2009, at 02:10, Alan Ruttenberg wrote:

> On Thu, Feb 12, 2009 at 3:05 PM, Bijan Parsia
> <bparsia@cs.manchester.ac.uk> wrote:
>>
>> On 12 Feb 2009, at 22:43, Alan Ruttenberg wrote:
>>
>>> To where can I point Jos to in our spec that defines this behavior?
>>
>> ? We don't define the behavior I described. We say that  
>> implementations must
>> support at least 16 digits. If they don't support more we are  
>> silent on what
>> they should do. I described two possibilities.
>
> This worries me. The problem is that http://www.w3.org/TR/xpath-functions/ 
>  says:
>
> For xs:decimal values the number of digits of precision returned by
> the numeric operators is ·implementation-defined·. If the number of
> digits in the result exceeds the number of digits that the
> implementation supports, the result is truncated or rounded in an
> ·implementation-defined· manner

We don't support XPath functions, so I think it's a bit of a stretch.  
I would certainly beat up on anyone who cited that in support of  
stupid behavior by their tool

> If our spec allowed that it might be the case that an ontology was
> inconsistent in some implementations and consistent in others.

I don't think our spec allows that, but it doesn't spend much time  
making it crystal clear.

> If we
> don't already disallow this then I think we should. If we do, then I'd
> appreciate if someone could point out to me what prevents this.

What allows it? We say tools must support at least 16 digits. If there  
are more than 16digits we say how to interpret that document. I don't  
see how it's *conforming* to treat the >16digit document when you only  
support 16 digits in a random way. But also, implementations are free  
to handle such things as they see fit (e.g., by rejecting, or by  
approximating). In the OWL market, I believe that there will be  
different modes available, so I think worries are moot.

This is not specific to 16 digits. What do you do if you don't support  
nominals? QCRs? I don't want us to say that e.g., reasoners MUST  
reject these (that would be weird) and if you go down the road of  
"Must raise a warning, yadda yadda", well, it's really hard to get  
that right in advance unless you do full fledged, honest to god, huge  
amount of work fine grained error handling.

Cheers,
Bijan.

Received on Friday, 13 February 2009 08:51:37 UTC