Re: ISSUE-22: Proposal based on sh:hasShape

On 6/12/2015 13:37, Peter F. Patel-Schneider wrote:
> Changing the order of triples in a shapes or data graph document, or even
> just running the system again, might end up producing different answers if
> care is not taken in fleshing out this approach, even in cases where there
> is no negation or even disjunction.

Does anyone have a test case that demonstrates such a non-deterministic 
scenario? AND, OR and XOR all use rdf:Lists and are therefore ordered. 
sh:valueShape needs to walk through all values of the property, so order 
shouldn't matter there either. (Apologies if the answer to these 
questions is already somewhere on the mailing list - I found this topic 
hard to follow)

>    Picking a bad propagation rule for a
> "Duh" is going to produce something that I think will be unusable---I don't
> even know if there is any good set of propagation rules.

Maybe it would help if we made sh:hasShape return three possible values: 
true, false and unknown. Then the writers of custom SPARQL queries can 
better specify how they are going to treat the undefined cases. I 
believe we need the "unknown" value anyway, e.g. when a JavaScript 
implementation encounters a SPARQL query that it cannot handle.

Thanks,
Holger

Received on Sunday, 14 June 2015 04:42:55 UTC