Sparql "exists" function not working with "bind if"

[Kevin-Darty]

My Virtuoso is (virtuoso-t -h):

Virtuoso Open Source Edition (Column Store) (multi threaded)
Version 7.2.4.2.3217-pthreads as of Jun  3 2016
Compiled for Linux (x86_64-pc-linux-gnu)

The Sparql function exists is working fine, but not when combined with (bind if(...)).
The query:

PREFIX dbo: <http://dbpedia.org/property/>
SELECT  ?uri  ?popEstimate ?existsPopEstim ?usedExPop
WHERE {
	?uri  a <http://dbpedia.org/ontology/Place>.
	
	{?uri rdfs:label "France"@fr}
	UNION {?uri rdfs:label "Brésil"@fr}
	UNION {?uri rdfs:label "Amérique"@fr}
	
	OPTIONAL {?uri dbo:populationEstimate ?popEstimate .}	
	BIND (exists{?uri dbo:populationEstimate ?popEstimate} AS ?existsPopEstim )
	BIND (IF(?existsPopEstim , "ok", "no") AS ?usedExPop)
}
LIMIT 100

The result:

uri                                  | popEstimate | existsPopEstim | usedExPop
-------------------------------------+-------------+----------------+----------
http://dbpedia.org/resource/France   |             | 0              | ok
http://dbpedia.org/resource/Brazil   | 201032714   | 1              | ok
http://dbpedia.org/resource/Americas |             | 0              | ok 

(see https://stackoverflow.com/q/47161296/6833025)

[HughWilliams]

This issue has been reported to development to look into ...

[kidehen]

@niwizard ,

Copy of what I posted to Stackoverflow:

I notice that you are experiencing this problem via Version 7.2.4.2.3217-pthreads as of Jun 3 2016. Here's what happens with the more recent 7.2.x and 8.0 releases via live example links:

• DBpedia -- running 7.2.x (2017 release)
• URIBurner -- running 8.0 (2018 release), but doesn't have
  complete DBpedia dataset loaded.

That said, I can recreate your problem via our LOD Cloud Cache instance. Thus, I suggest:

1. Upgrade on your part to one of the 2017 releases used in my examples above
2. We revisit our Cluster Edition which is running behind the LOD Cloud instance that I used to recreate your problem.

[Kevin-Darty]

The problem is the same with a unique BIND:

BIND (IF(exists{?uri dbo:populationEstimate ?popEstimate} , "ok", "no") AS ?usedExPop)

[kidehen]

@niwizard -- share a link to a live example on DBpedia (for instance) so that I can reply with a potential workaround etc..

[Kevin-Darty]

The even simplier query:

SELECT  ?uri  ?pop ?existsPop ?usedExPop
WHERE {
	?uri  a <http://dbpedia.org/ontology/Place>.
	
	{?uri rdfs:label "France"@en}
	UNION {?uri rdfs:label "Brazil"@en}
	UNION {?uri rdfs:label "Raging River"@en}

	OPTIONAL {?uri dbo:populationTotal ?pop .}	
	BIND (exists{?uri dbo:populationTotal ?pop} AS ?existsPop)
	BIND (IF(?existsPop, "ok", "no") AS ?usedExPop)
}
LIMIT 1000

The link: https://dbpedia.org/sparql?default-graph-uri=http%3A%2F%2Fdbpedia.org&query=SELECT++%3Furi++%3Fpop+%3FexistsPop+%3FusedExPop%0D%0AWHERE+%7B%0D%0A%09%3Furi++a+%3Chttp%3A%2F%2Fdbpedia.org%2Fontology%2FPlace%3E.%0D%0A%09%0D%0A%09%7B%3Furi+rdfs%3Alabel+%22France%22%40en%7D%0D%0A%09UNION+%7B%3Furi+rdfs%3Alabel+%22Brazil%22%40en%7D%0D%0A%09UNION+%7B%3Furi+rdfs%3Alabel+%22Raging+River%22%40en%7D%0D%0A%09%0D%0A%0D%0A%09OPTIONAL+%7B%3Furi+dbo%3ApopulationTotal+%3Fpop+.%7D%09%0D%0A%09BIND+%28exists%7B%3Furi+dbo%3ApopulationTotal+%3Fpop%7D+AS+%3FexistsPop%29%0D%0A%09BIND+%28IF%28%3FexistsPop%2C+%22ok%22%2C+%22no%22%29+AS+%3FusedExPop%29%0D%0A%7D%0D%0ALIMIT+1000&format=text%2Fhtml&CXML_redir_for_subjs=121&CXML_redir_for_hrefs=&timeout=30000&debug=on&run=+Run+Query+

[kidehen]

@niwizard,

This is a datatype issue. Basically SQL Null datatype rearing its head.

1. This Query Result filters out Nulls

2. This Query Shows the Null effect

3. This Query Results Page shows the problem

That said, given --

SELECT  ?uri  ?pop ?existsPop ?usedExPop
WHERE {
	?uri  a <http://dbpedia.org/ontology/Place>.
	
	{?uri rdfs:label "France"@en}
	UNION {?uri rdfs:label "Brazil"@en}
	UNION {?uri rdfs:label "Raging River"@en}
	

	OPTIONAL {?uri dbo:populationTotal ?pop .}	
        BIND (xsd:boolean(exists{?uri dbo:populationTotal ?pop}) AS ?existsPop)
	BIND (IF(xsd:boolean(?existsPop), "ok", "no") AS ?usedExPop)
        FILTER (?pop > 0)
}
LIMIT 1000

-- we should be able to consider coercing ?existsPop to be xsd:boolean rather than requiring the explicit casting heuristics above. EXISTs should return xsd:boolean in this context.

[Kevin-Darty]

OK so you casted the exists result into xsd:boolean, thus ?existsPop is giving true or false the right way.
But ?usedExPop is still returning "ok" with:

 BIND (xsd:boolean(exists{?uri dbo:populationTotal ?pop}) AS ?existsPop)
 BIND (IF(xsd:boolean(?existsPop), "ok", "no") AS ?usedExPop)

[kidehen]

@niwizard,

Yes, as per:

SELECT  ?uri  ?pop ?existsPop ?usedExPop xsd:boolean(?usedExPop) as ?usedExPop2
WHERE {
	?uri  a `<http://dbpedia.org/ontology/Place>`.
	
	{?uri rdfs:label "France"@en}
	UNION {?uri rdfs:label "Brazil"@en}
	UNION {?uri rdfs:label "Raging River"@en}
	

	OPTIONAL {?uri dbo:populationTotal ?pop .}	
        BIND (xsd:boolean(exists{?uri dbo:populationTotal ?pop}) AS ?existsPop)
	BIND (IF(xsd:boolean(?existsPop) = "true"^^xsd:boolean, "ok", "no") AS ?usedExPop)
        # FILTER (?pop > 0)
}
LIMIT 1000

Which shows the problem re. bug to be fixed as per this query results page.

[IvanMikhailov]

I have progress with debugging, but I don't have the fix yet.