bindings of variables are not carried into FILTERs in (NOT) EXISTS clauses

[pfps]

My understanding is that variable bindings carry into FILTER constructs. However, this sometime does not work in rdflib. I have a short example that shows this, with a FILTER inside a FILTER EXISTS. My belief is that the binding of ?this from the outermost portion should be visible in the innermost portion, meaning that the ?this = ?valu will only filter out solutions where the two variables have different bindings. However, all solutions appear to be filtered out, which would happen if the ?this binding(s) from the outermost portion are not making their way into the innermost portion.

This is very similar to the example in SPARQL 1.1 Query Language section 8.3.3.

bug2-rdflib.txt

[joernhees]

so you're complaining that the following query doesn't return any results:

 SELECT ?this WHERE { 
  VALUES ?this { <y> <n> } 
  FILTER (EXISTS {
    VALUES ?valu { <y> <o> } 
    FILTER ( ?this = ?valu )
  })
 }

This is actually by design in SPARQL: The inner-most query is evaluated first.

The consequence is that we have the following execution order:

  EXISTS {
    VALUES ?valu { <y> <o> } 
    FILTER ( ?this = ?valu )  # ?this is an unknown variable here
  }

As the ?this in the above is still unbound the equality always fails, so the result of the EXISTS is false. Then the following is executed:

 SELECT ?this WHERE { 
  VALUES ?this { <y> <n> } 
  FILTER (false)
 }

Which correctly doesn't result in a binding for ?this.

[pfps]

I don't think that this reasoning is correct.

Look at Section 8.3.3 of SPARQL 1.1 Query Language, at
https://www.w3.org/TR/sparql11-query/#idp899488.

The example there is

PREFIX : http://example.com/
SELECT \* WHERE {
        ?x :p ?n
        FILTER NOT EXISTS {
                ?x :q ?m .
                FILTER(?n = ?m)
        }
}

which is analogous to my example and the wording says that ?n is bound in the
inner filter.

peter

On 03/28/2016 03:39 AM, Jörn Hees wrote:

so you're complaining that the following query doesn't return any results:

|SELECT ?this WHERE { VALUES ?this { } FILTER (EXISTS { VALUES ?valu {
} FILTER ( ?this = ?valu ) }) } |

This is actually by design in SPARQL: The inner-most query is evaluated first.

The consequence is that we have the following execution order:

|EXISTS { VALUES ?valu { } FILTER ( ?this = ?valu ) # ?this is an
unknown variable here } |

As the |?this| in the above is still unbound the equality always fails, so the
result of the |EXISTS| is false. Then the following is executed:

|SELECT ?this WHERE { VALUES ?this { } FILTER (false) } |

Which correctly doesn't result in a binding for |?this|.

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#615 (comment)

[pfps]

Is this a bug or is my understanding of SPARQL incorrect?

[joernhees]

To be honest, i'm not sure, which is why re-opened this...

The link you point to seems to support your view-point, but jena seems to support mine:

$ sparql -v --data=empty.n3 --query=<(echo '
>  SELECT ?this WHERE {
>   VALUES ?this { <y> <n> }
>   FILTER (EXISTS {
>     VALUES ?valu { <y> <o> }
>     FILTER ( ?this = ?valu )
>   })
>  }
> ')

  1 SELECT  ?this
  2 WHERE
  3   { VALUES ?this { <y> <n> }
  4     FILTER EXISTS { VALUES ?valu { <y> <o> }
  5                     FILTER ( ?this = ?valu )
  6                   }
  7   }

13:39:05 INFO  exec                 :: QUERY
  SELECT  ?this
  WHERE
    { VALUES ?this { <y> <n> }
      FILTER EXISTS { VALUES ?valu { <y> <o> }
                      FILTER ( ?this = ?valu )
                    }
    }
13:39:05 INFO  exec                 :: ALGEBRA
  (project (?this)
    (filter (exists
               (table empty))
      (table (vars ?this)
        (row [?this <file:///dev/fd/y>])
        (row [?this <file:///dev/fd/n>])
      )))
--------
| this |
========
--------

That said, the SPARQL parsing and its conversion to an algebra isn't the easiest code and i just didn't find the time to look into this properly.

[pfps]

I put together the example from the SPARQL 1.1 QL spec and it has the incorrect behaviour as well.

[Again problems with uploading - I'll put it into a separate comment.]
Uploading q.txt…

[pfps]

import rdflib
from rdflib import Namespace
from rdflib.term import Literal
from rdflib.namespace import XSD

EX = Namespace("http://example.com/")

g = rdflib.Graph()
g.add ( ( EX.a, EX.p, Literal(1,datatype=XSD.integer) ) ) 
g.add ( ( EX.a, EX.q, Literal(1,datatype=XSD.integer) ) ) 
g.add ( ( EX.a, EX.q, Literal(2,datatype=XSD.integer) ) ) 

g.add ( ( EX.b, EX.p, Literal(3.0,datatype=XSD.decimal) ) ) 
g.add ( ( EX.b, EX.q, Literal(4.0,datatype=XSD.decimal) ) ) 
g.add ( ( EX.b, EX.q, Literal(5.0,datatype=XSD.decimal) ) ) 


print "MINUS"
query = """PREFIX : <http://example.com/>
SELECT * WHERE {
    ?x :p ?n
    MINUS {
        ?x :q ?m .
        FILTER(?n = ?m)
    }
}"""
result = g.query(query)
for row in result : print row
assert len(result) == 2

print "FILTER NOT EXISTS"
query = """PREFIX : <http://example.com/>
SELECT * WHERE {
    ?x :p ?n
    FILTER NOT EXISTS {
        ?x :q ?m .
        FILTER(?n = ?m)
    }
}"""
result = g.query(query)
for row in result : print row
assert len(result) == 1

[gromgull]

I believe this was broken by: 6200934