Indeed, thanks! I could already have seen this from the SPARQL grammar:
OptionalGraphPattern is a special part within one group graph pattern (analogously to how FILTER works), whereas my braces would make it a binary operator between two group graph patterns (analogously to how UNION works). So BlazeGraph is totally correct here.

Markus

On 25/11/2018 16:16, Sylvain Boissel wrote:
Remove the { } around the 3 first lines and it should work fine.

Le dim. 25 nov. 2018 à 15:52, Markus Kroetzsch <[email protected] <mailto:[email protected]>> a écrit :

    Hi,

    I am puzzled by the behaviour of a SPARQL query. Maybe there is an
    error
    with BlazeGraph here, but hopefully I am just overlooking something.

    The query is as follows: http://tinyurl.com/y95jpmhq

    SELECT ?item ?birthdate ?spouse
    WHERE
    {
        { ?item wdt:P569 ?birthdate
          FILTER (year(?birthdate)>1900)
          ?item wdt:P26 []
        } OPTIONAL {
          ?item wdt:P26 ?spouse
          FILTER (year(?birthdate) = 1947)
        }
        # FILTER (year(?birthdate) = 1947) ## For testing: works correctly
    } LIMIT 1000

    What this should do: "Select married people born after 1900, and,
    optionally, also select their spouses, but only for people born in
    1947." What BlazeGraph does is: "Select married people born after 1900;
    never select any spouses, even if the person is born in 1947".

    The 1000 results should contain lines for 1947 births, so you can see
    they have no spouse. The commented out filter at the bottom can be used
    instead of the inner filter to verify that the condition has no typos
    and really matches some of the items.

    It seems that BlazeGraph gets the scope of ?birthdate wrong here, and
    rather processes the whole query inside out, applying the FILTER to the
    optional pattern (where ?birthdate is not bound) and then using the
    (empty) result in a binary LeftJoin operation. In reality, LeftJoin in
    the SPARQL algebra is a ternary operator that applies the FILTER to the
    Join of both sides to determine if we have an optional match or not:

    * See "Definition: LeftJoin" in Section 18.5 of the spec [1].

    Filters within optional patterns become the third parameter in the
    LeftJoin operation when translating queries as in my example:

    * See example "{ ?s :p1 ?v1 OPTIONAL {?s :p2 ?v2 FILTER(?v1<3) } }" in
    Section 18.2.3 of the spec [1].

    Is my interpretation correct or did I overlook something? Is this a
    known problem?

    Cheers,

    Markus

    [1] https://www.w3.org/TR/sparql11-query

-- Prof. Dr. Markus Kroetzsch
    Knowledge-Based Systems Group
    Center for Advancing Electronics Dresden (cfaed)
    Faculty of Computer Science
    TU Dresden
    +49 351 463 38486
    https://kbs.inf.tu-dresden.de/

    _______________________________________________
    Wikidata mailing list
    [email protected] <mailto:[email protected]>
    https://lists.wikimedia.org/mailman/listinfo/wikidata


_______________________________________________
Wikidata mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/wikidata


Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

_______________________________________________
Wikidata mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/wikidata

Reply via email to