Martynas,
Two things:
1:: IIRC it is the queries that back various features that need to be
more stable. Depending on the internal details of ARQ isn't a good place
to be. Note that SpinDRF has it's own function and property function
evaluation subsystem with slightly different semantics.
This is the problem you are encountering.
2:: The effect is initial binding isn't well defined.
Syntax rewriting is going to be more stable (and works remotely which
initial bindings can't).
Andy
On 22/05/2020 19:55, Martynas Jusevičius wrote:
Andy,
which part needs to be replaced with syntax rewriting -- SPIN
functions or the initial bindings?
Do you think it's possible to rewrite SPINRDF in a more SPARQL
compliant and less ARQ-dependent way?
On Thu, May 21, 2020 at 10:54 AM Andy Seaborne <a...@apache.org> wrote:
On 20/05/2020 22:11, Martynas Jusevičius wrote:
https://github.com/spinrdf/spinrdf/issues/22
On Wed, May 20, 2020 at 10:57 PM Martynas Jusevičius
<marty...@atomgraph.com> wrote:
Andy,
I was able to isolate a standalone example:
https://github.com/namedgraph/spinrdf-test
With 3.0.1 it succeeds. Change Jena version to 3.16.0-SNAPSHOT by
uncommenting the other <version> in pom.xml and it will fail:
https://github.com/namedgraph/spinrdf-test/blob/master/pom.xml#L28
I don't understand how or where spl:objectCount() and spl:instanceOf()
are executed, but i think it demonstrates clearly that some behaviour
has changed between those versions, even though SpinRDF code hasn't.
Many SpinRDF functions are written with implicit assumptions about how
ARQ - not SPARQL - executes. They don't work when the optimizer is off,
reconfigured, or if the reference query engine is used.
In addition, the external injection of initial bindings
(Long term - that should be replaced with syntax rewriting of the query
- see QueryTransformOps)
The result is that each version of SpinRDF needs to be tuned to the
version of Jena it uses.
Andy
Will post on https://github.com/spinrdf/spinrdf/issues as well.
Martynas
On Wed, May 20, 2020 at 12:37 PM Martynas Jusevičius
<marty...@atomgraph.com> wrote:
Hi,
I've narrowed down the differences to the counts produced by
spl:objectCount() SPIN function.
SELECT *
WHERE
{ ?this a ?TYPE_CLASS
{ BIND(<http://spinrdf.org/spl#objectCount>(?this, ?predicate)
AS ?objCount)
FILTER ( bound(?minCount) && ( ?objCount < ?minCount ) )
}
UNION
{ BIND(<http://spinrdf.org/spl#objectCount>(?this, ?predicate)
AS ?objCount)
FILTER ( bound(?maxCount) && ( ?objCount > ?maxCount ) )
}
UNION
{ FILTER bound(?valueType)
?this ?predicate ?value
FILTER ( ! <http://spinrdf.org/spl#instanceOf>(?value, ?valueType) )
}
}
Bindings: ( ?predicate, spin:body )( ?comment, "the body of the
Template" )( ?minCount, 0 )( ?TYPE_CLASS, spin:Template )( ?maxCount,
1 )
3.15.0-SNAPSHOT
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
| this
| TYPE_CLASS |
objCount | predicate | value |
=================================================================================================================================================================================================
| <http://spinrdf.org/spin#Templates>
| <http://spinrdf.org/spin#Template> |
2 | <http://spinrdf.org/spin#body> | |
| <http://spinrdf.org/spin#ConstructTemplates>
| <http://spinrdf.org/spin#Template> |
2 | <http://spinrdf.org/spin#body> | |
|
<https://github.com/AtomGraph/Processor/blob/develop/http-tests/custom#DefaultSubjectUpdateTemplate>
| <http://spinrdf.org/spin#Template> | 2 |
<http://spinrdf.org/spin#body> | |
| <http://spinrdf.org/spin#AskTemplates>
| <http://spinrdf.org/spin#Template> |
2 | <http://spinrdf.org/spin#body> | |
|
<https://github.com/AtomGraph/Processor/blob/develop/http-tests/custom#DefaultSubjectQueryTemplate>
| <http://spinrdf.org/spin#Template> | 2 |
<http://spinrdf.org/spin#body> | |
| <http://spinrdf.org/spin#UpdateTemplates>
| <http://spinrdf.org/spin#Template> |
2 | <http://spinrdf.org/spin#body> | |
| <http://spinrdf.org/spl#Argument>
| <http://spinrdf.org/spin#Template> |
2 | <http://spinrdf.org/spin#body> | |
| <http://spinrdf.org/spin#SelectTemplates>
| <http://spinrdf.org/spin#Template> |
2 | <http://spinrdf.org/spin#body> | |
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
3.0.1
----------------------------------------------------
| this | TYPE_CLASS | objCount | predicate | value |
====================================================
----------------------------------------------------
I'll try to create an example now.
On Tue, May 19, 2020 at 6:14 PM Andy Seaborne <a...@apache.org> wrote:
That's a long jump.
(and it's a SpinRDF question)
>> Are there some obvious suspects here or do I need to create a
>> reproducible example?
reproducible example (noting that the query has custom functions which
have names that suggest they are not proper "functions" (i.e. their
return is not a function of their arguments alone and they access the
data graph .. which might be a query).
I can't see why it might now have answers when it didn't before.
Andy
On 19/05/2020 13:04, Martynas Jusevičius wrote:
With 3.0.1, no such internal *violations* are produced.
On Tue, May 19, 2020 at 2:03 PM Martynas Jusevičius
<marty...@atomgraph.com> wrote:
Hi,
After upgrading our code as well as SPINRDF from 3.0.1 to
3.16.0-SNAPSHOT, some constraint violation-related tests are failing.
With 3.16.0-SNAPSHOT, it seems that a bunch of constraint violations
are coming from the SPIN vocabulary iself, e.g. "Attribute spin:body :
[0,1]" on spin:ConstructTemplates, spin:SelectTemplates etc.
With 3.0.1, no such internal validations are produced.
The models being validated use a basic RDFS inference (subclassing etc.)
I have tracked the source of the difference to query execution in
SPINConstraints.runQueryOnClass():
https://github.com/spinrdf/spinrdf/blob/master/src/main/java/org/spinrdf/constraints/SPINConstraints.java#L614
Since SPINRDF code hasn't changed, only the Jena versions, could there
be differences in query execution? The specific query that produces
different results is this:
CONSTRUCT
{
_:c0 <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
<http://spinrdf.org/spin#ConstraintViolation> .
_:c0 <http://spinrdf.org/spin#violationRoot> ?this .
_:c0 <http://spinrdf.org/spin#violationPath> ?predicate .
}
WHERE
{ ?this a ?TYPE_CLASS
{ FILTER ( bound(?minCount) && (
<http://spinrdf.org/spl#objectCount>(?this, ?predicate) < ?minCount )
) }
UNION
{ FILTER ( bound(?maxCount) && (
<http://spinrdf.org/spl#objectCount>(?this, ?predicate) > ?maxCount )
) }
UNION
{ FILTER bound(?valueType)
?this ?predicate ?value
FILTER ( ! <http://spinrdf.org/spl#instanceOf>(?value, ?valueType) )
}
}
The query string and initial bindings are the same with both versions,
for example:
( ?predicate, spin:body )( ?comment, "the body of the Template" )(
?minCount, 0 )( ?TYPE_CLASS, spin:Template )( ?maxCount, 1 )
Are there some obvious suspects here or do I need to create a
reproducible example?
Thanks,
Martynas
