[
https://issues.apache.org/jira/browse/JENA-1877?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17072994#comment-17072994
]
Andy Seaborne commented on JENA-1877:
-------------------------------------
Hi - thanks for the report with details and analysis.
I think the non-optimized is right-by-the-spec. Section 18.2.4.4 "SELECT
Expressions" nests one "extend" for each expression in the SELECT clause.
SPARQL itself does not have an n-ary (extend).
The built-in no-frills execution engine agrees with the not-optimized version.
It can be accessed with command line {{--engine=ref}}; it does not scale or
have great performance.
It is possible to control the individual optimizer steps. {{--set
arq:optMergeExtends=false}} will switch off the transformation that is involved.
> Wrong results for non-optimized query
> -------------------------------------
>
> Key: JENA-1877
> URL: https://issues.apache.org/jira/browse/JENA-1877
> Project: Apache Jena
> Issue Type: Bug
> Components: ARQ
> Affects Versions: Jena 3.14.0
> Reporter: Jeremy Coulon
> Priority: Major
> Attachments: bnode01.rq, data.ttl
>
>
> I noticed that the attached query gives correct results with:
> {noformat}
> ./bin/arq --data=data.ttl --query=bnode01.rq --explain --optimize=on
> {noformat}
> while it gives wrong results with:
> {noformat}
> ./bin/arq --data=data.ttl --query=bnode01.rq --explain
> --optimize=off{noformat}
> Without optimization, ARQ algebra has 2 different 'extend' op with
> 'bnode(?s)' expression.
> With optimization, ARQ algebra merges these ops into a single op 'extend'.
>
> I tried debugging and I think that 'E_BNode.evalSpecial()' takes the same
> 'binding' object for each call in optimized mode but different 'binding'
> objects for non-optimized mode. This function relies on reference-equality.
>
> [^data.ttl]
> [^bnode01.rq]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
