How can I remove distinct from subquery?

Question

Hello, I'll expose my problem, I don't know if may be is a bug.
Suppose We have
Table A:
field A1 primary key
field A2
field A3
Table B
field B1 primary key
field B2 foreig key to A(A1)
field B3
I use reverse tool to build objects from DB schema.
Everything runs ok but when I want to recover A objects
with sum(B3) from B,it doesn't runs fine.
To achieve this I've made a class C with fields of class A and a double
for sum. The code I've use is:
KodoQuery query = (KodoQuery)getPersistenceManager()
.newQuery(A.class, "this == b.b2");
query.declareVariables("B b");
query.declareImports("x.x.B");
query.setResultClass(C.class);
query.setResult("this as a, sum(b.b3) as s");
query.setGrouping("a");
Collection c = query.execute();
I use PostGreSQL 7.4, and the sql that Kodo generates is like this:
SELECT s.A1, SUM(s.C) AS c
FROM ( SELECT DISTINCT a.A1 AS A1, s.B3 AS C
FROM PUBLIC.A a INNER JOIN PUBLIC.B b ON (a.A1 = b.B2) ) s
GROUP BY s.A1
But the problem is 'DISTINCT' keyword. If I have records on B
with same B3 value then:
B1=1,B2=1,B3=25
B1=2,B2=1,B3=25
The query returns SUM(B3)=25 which is not valid!!!, it must be 50.
It fails with repeated B3 values because distinct keyword.
I bypass this adding a third field to class C that I will not use, and
introduce it into query:
KodoQuery query = (KodoQuery)getPersistenceManager()
.newQuery(A.class, "this == b.b2");
query.declareVariables("B b");
query.declareImports("x.x.B");
query.setResultClass(C.class);
query.setResult("this as a, sum(b.b3) as s, sum(b.b1) as t");
query.setGrouping("a");
Collection c = query.execute();
With this, I force to put primary key into subquery, and disables
DISTINCT, but I never will use this field :(. SQL generated by KODO:
SELECT s.A1, SUM(s.C) AS c, SUM(s.T) as t
FROM ( SELECT DISTINCT a.A1 AS A1, s.B3 AS C, s.B1 as T
FROM PUBLIC.A a INNER JOIN PUBLIC.B b ON (a.A1 = b.B2) ) s
GROUP BY s.A1
Questions:
1.- What may I doing wrong?
2.- How can I disable Distinct from subquerys?
And, no, I don't want to select by class B and then search A objects.
I want a C object with A object plus SUM of values from B.
I am achieving this with this patch.
I know that when there aren't B objects I don't receive results.
Is there any way to build an OUTER JOIN with JDOQL ?
In order to bypass this, I execute the query another time but reseting to
0, but I will prefer to be able to build an OUTER JOIN ;)
PD: Little class descriptions:
class A{ int a1, int a2, int a3} (autogenerated by kodo)
class B{ int b1, int b2, double b3} (autogenerated by kodo)
class C{ A a, double c, int t} (generated by me)
Sorry for my english :(
thanks.

Answer 1

I'll address your outer join question first, since it's the easiest. Kodo will
create outer joins automatically as needed to satisfy your JDOQL. In this case,
your JDOQL does not warrant an outer join. Your filter criteria on the query is
"this == b.b2". That means the query cannot and should not return any values
when there are no A's that match that join. Why not try this equivalent but
simpler query?
Query q = pm.newQuery (B.class);
q.setResult ("b2 as a, sum(b3) as s");
q.setGrouping ("b2");
q.setResultClass (C.class);
That will return results grouped by the A objects from the b2 field, and the sum
of the b3 field for all B's in each group. It should use an outer join.
About DISTINCT:
The pattern of a distinct subselect is caused by two conditions:
1. Your result clause is non-distinct. That is, you did:
q.setResult("this as a, sum(b3) as s")
rather than:
q.setResult("distinct this as a, sum(b3) as s")
This means that you want to allow duplicates in the projected values you get back.
2. You use an unbound variable ("b") in a way that could lead to duplicates
caused by relational joins. JDO always eliminates duplicates caused by database
joins.
So in order to eliminate the possible duplicates caused by #2, Kodo issues the
query as a DISTINCT subselect. Then in order to allow duplicates of the
projected results (#1), Kodo uses a non-distinct outer select.
The simplest way to eliminate the DISTINCT subselect is to use the alternate
query on B.class I mentioned above. That will result in no DISTINCTs at all.
Another way is to use the "distinct" keyword in your setResult() call, as I
showed in #1. That will result in a DISTINCT select without any subselect.
Without reverting to a SQL query (which you can still execute through JDO APIs,
as described in the documentation), I believe those are the options available.