making the test suite results meaningful

So all the malformed literal tests cannot be trusted.


Each test or set of tests should have one or more related tests with the
opposite result to demonstrate that test results are not the result of some
larger failure of a system.


For example, although this is a case where it is a bit hard to imagine systems
failing in this way, there should be tests showing that embedded triples with
different subjects, predicates, or objects are not collapsed into the same
entity, such as


prefix : <http://example.com/ns#>

<< :a :b _:c >> :p1 :o1.
<< :a :b _:d >> :p2 :o2.

|does not entail|

prefix : <http://example.com/ns#>

<< :a :b _:e >> :p1 :o1; :p2 :o2.


and
 

prefix : <http://example.com/ns#>

<< :a :b :c >> :p1 :o1.
<< :a :b :d >> :p2 :o2.

|does not entail|

prefix : <http://example.com/ns#>

<< :a :b :c >> :p1 :o1; :p2 :o2.




Ideally this test should be repeated for subjects, predicates, and objects and
for blank nodes, IRIs, and literals.


As far as testing for identity of embedded triples, this should be expanded to
tests involving blank nodes and literals.

peter



On 11/27/20 8:26 AM, Pierre-Antoine Champin wrote:
> On 26/11/2020 23:21, Peter F. Patel-Schneider wrote:
>
>> Probably the most important of my proposed tests is whether malformed literals
>> are correctly handled outside of embedded triples, but this test does not
>> appear to have been incorporated into the test suite.
> oops, I missed that one. That's fixed.
>> This test could be transformed into an entailment test (instead of a
>> consistency test) as follows:
>>
>>
>>
>> Are malformed literals handled correctly outside of embedded triples?
>>
>> prefix : <http://example.com/ns#>
>> prefix xsd: <http://www.w3.org/2001/XMLSchema#>
>> :a :b "c"^^xsd:integer .
>>
>> entails
>>
>> prefix : <http://example.com/ns#>
>>
>> :a :b :c .
>
> You are right, but that would not solve my problem when it comes to checking
> entailment through SPARQL.
>
> Looking briefly at https://www.w3.org/TR/sparql11-entailment/, it seems to
> me that inconsistent graphs are not expected to yield arbitrary triples...
> Anyway, the four systems that I have tested don't do that.
>
>>
>>
>> peter
>>

Received on Friday, 27 November 2020 14:30:52 UTC