Evaluation of VALUES

[nkaralis]

Hello.

I am having trouble understanding the evaluation of the VALUES clause when encountered in this grammar rule Query ::= Prologue( SelectQuery | ConstructQuery | DescribeQuery | AskQuery ) ValuesClause.

In section 10.2.2, the specification states that, when this rule is used, the VALUES clause is to be executed over the results of the SELECT query. Provided this statement, I was expecting to be able to run queries as the one provided below.

SELECT ?s (MIN(?v) as ?min) WHERE {
    ?s <value> ?v 
}
GROUP BY ?s
VALUES ?min { 2 5 }

However, this query is rejected due to ?min being already in scope when encountered in the SELECT clause. If I understand correctly, this is due to the order in which queries are processed; paragraph 18.2.4 states that the VALUES clause is processed before the SELECT clause.

I was hoping that someone could elaborate on how the VALUES clause is evaluated in the precesence of GROUP BY. If I understand the function AggregateJoin correctly, the VALUES clause can only be joined with keys of the GROUP BY clause. The example below demonstrates this.

Is this correct?
Thank you in advance.

# data: agg-numeric.ttl https://github.com/w3c/rdf-tests/blob/main/sparql/sparql11/aggregates/agg-numeric.ttl
# Apache Jena version 5.2.0

./arq --explain --query=test_values.rq --data=agg-numeric.ttl
09:21:11 INFO  exec            :: QUERY
  PREFIX  :     <http://www.example.org/>
  
  SELECT  ?i (MIN(?v) AS ?c)
  WHERE
    { ?i  :int  ?v } # possible bindings for ?v: 1, 2, 3 ; possible bindings for ?i: :ints, :mixed
  GROUP BY ?i
  VALUES ( ?i ?v ) {
    ( :ints 5 )
    ( :ints 6 )
  }
09:21:11 INFO  exec            :: ALGEBRA
  (project (?i ?c)
    (sequence
      (extend ((?c ?.0))
        (group (?i) ((?.0 (min ?v)))
          (bgp (triple ?i <http://www.example.org/int> ?v))))
      (table (vars ?i ?v)
        (row [?i <http://www.example.org/ints>] [?v 5])
        (row [?i <http://www.example.org/ints>] [?v 6])
      )))
09:21:11 INFO  exec            :: BGP ::   ?i <http://www.example.org/int> ?v
09:21:11 INFO  exec            :: Reorder/generic ::   ?i <http://www.example.org/int> ?v
-------------
| i     | c |
=============
| :ints | 1 |
| :ints | 1 |
-------------

[afs]

Hi @nkaralis - thanks for the detailed write-up.

It looks like the scope checking is wrong in the first (?min) case.

[afs]

#3164 fixes this.

[nkaralis]

Hello.
Thank you for the update.

I am still puzzled with the evaluation of VALUES though.
In particular, from the standard, it is not entirely clear to me when its evaluation should take place.

As mentioned in my original post, paragraph 10.2.2 states that the VALUES clause is to be executed over the results of SELECT queries.
However, systems should follow paragraph 18.2.4, which states that the VALUES clause should be processed before the SELECT clause. This means that the VALUES clause is not actually evaluated over the results of the SELECT clause in the presence of aggregates and GROUP BY.
Is this correct?

[afs]

It looks like an error in the spec - I've added w3c/sparql-query#215 to make sure its not forgotten.

(VALUES is suppose to support a distributed query "bind" join but there is little real use of that as far as I'm aware)

[nkaralis]

Thank you for the clarification.