Re: proposed set of SHACL tests

On 15/01/2016 3:08 PM, Holger Knublauch wrote:
> Peter,
> I did run a copy of your tests and noticed that my engine currently 
> has trouble when a sh:Shape is a blank node (I will investigate later).

This is now fixed and shapes that are bnodes should work.


> If you turn the shape in simpleShape.ttl into a URI node for now, you 
> will only have 2 remaining test failures. Those are due to the 
> (already) discussed issue that the validation is performed over the 
> shape definitions, and the SHACL graph currently requires sh:class to 
> point at classes.
> Holger
> PS: If you are still having github access issues, maybe talk to Eric?
> On 12/01/2016 3:16 AM, Peter F. Patel-Schneider wrote:
>> On 01/10/2016 07:14 PM, Holger Knublauch wrote:
>>> Hi Peter,
>>> thanks for playing with the test framework. I am still catching up 
>>> on a number
>>> of fronts and may not have time to look into the details this week. 
>>> A quick
>>> scan through your results indicates
>>> - My current test framework also validates the shape definitions. A 
>>> symptom of
>>> this is that, for example, values of sh:class are expected to be 
>>> instances of
>>> rdfs:Class. The idea in my API is that this schema-level testing can be
>>> switched off using the "filtered" argument of
>>> ModelConstraintValidator.validateModel. ValueTestClass calls this 
>>> with false,
>>> i.e. everything will be validated including shapes.
>> The problem is that these results look just like data validation 
>> results,
>> which is problematic in my view.
>>> - Inferencing is switched off by default, as defined by the spec. So 
>>> you
>>> should not expect rdfs:range and rdfs:domain to impact validation.
>> Well that was my expectation.  However it does not appear to be the 
>> case.  In
>> particular, the use of a node as the target of an rdf:type link 
>> appears to
>> make the node be an instance of rdfs:Class.
>> peter

Received on Monday, 18 January 2016 04:49:07 UTC