On 17/06/2010 10:33, Buikstra, A.P.F. wrote:
Hi guys,
I’m using a construct query to extract a specific portion of RDF from
LDSR. I’m creating these queries automatically, by the following
transformation:
Select ?a
Where { ?a ?b ?c }
->
Construct {?a ?b ?c}
Where {?a ?b ?c}
There are a few trivialities, such as Filters which have to be removed,
but this turned out not to be an issue. The problem is when Union comes
into the were, the idea no longer works:
Construct { {?a ?b ?c} UNION {?a ?d ?c}}
Where { {?a ?b ?c} UNION {?a ?d ?c}}
Would give an error.
My question is, can I just do:
Construct { ?a ?b ?c .
?a ?d ?c}
Where { {?a ?b ?c} UNION {?a ?d ?c}}
It seems to work okay, but it doesn’t feel completely right. It comes
down to whether the engine would try to construct both -?a ?b ?c- and
-?a ?d ?c- given the data, or either. Is there a better way to tackle my
problem?
All help appreciated!
Arjon Buikstra
(Working with Larkc at the VU and MPIB)
The "CONSTRUCT { ?a ?b ?c . ?a ?d ?c}" works in this case because there
is a variable in one side and not the other. That means either ?a ?b ?c
or ?a ?d ?c in the CONSTRUCT has an unbound variable and so the triple
generation is for this triple (not the other one) is skipped.
This does not generalise (as you found with FILTER) but you can do some
things with multiple constructs for each branch of the union and if the
union is nested you need be able to rewrite into a disjunctive normal form.
Andy
