[
https://issues.apache.org/jira/browse/TRAFODION-2130?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15399410#comment-15399410
]
ASF GitHub Bot commented on TRAFODION-2130:
-------------------------------------------
Github user selvaganesang commented on a diff in the pull request:
https://github.com/apache/incubator-trafodion/pull/624#discussion_r72800580
--- Diff: core/sql/optimizer/NormRelExpr.cpp ---
@@ -3044,14 +3044,23 @@ Join::pullUpGroupByTransformation()
// a copy of join as the child. The original tree has not
changed.
// The predicates in the new groupBy and the new Join will have
// changed according to the comments above.
+//
------------------------------------------------------------------------------*/
GroupByAgg* Join::pullUpGroupByTransformation(NormWA& normWARef)
{
CollHeap *stmtHeap = CmpCommon::statementHeap() ;
- // Determine a set of unique columns for the left sub-tree
+ // Determine a set of unique columns for the left sub-tree.
+
+ // Note: Scans and joins synthesize uniqueness constraints even for
+ // columns that are not in the characteristic outputs. Other
+ // operators such as groupby or union don't. We make use of these
+ // extra uniqeness constraints here. Any needed columns not yet
+ // added to the characteristic outputs will be added later, in
+ // method getMoreOutputsIfPossible().
+
ValueIdSet leftUniqueCols ;
- if (NOT (child(0)-getGroupAttr()->findUniqueCols(leftUniqueCols)))
+ if (NOT (child(0)->getGroupAttr()->findUniqueCols(leftUniqueCols)))
--- End diff --
Interesting. Just curious what did the C++ compiler do earlier?
> Incorrect subquery transformation for tables w/o key
> ----------------------------------------------------
>
> Key: TRAFODION-2130
> URL: https://issues.apache.org/jira/browse/TRAFODION-2130
> Project: Apache Trafodion
> Issue Type: Bug
> Components: sql-cmp
> Affects Versions: 1.2-incubating
> Reporter: Hans Zeller
> Assignee: Hans Zeller
>
> Some of the subquery unnesting transformations rely on a unique key on the
> result of the main query. It turns out, however, that the transformation
> happens even if there is no unique key. Example:
> select *
> from (values (1,1), (1,1), (2,2)) T(a,b)
> where t.a in (select max(a)
> from (values (1,1), (1,1), (3,3)) S(a,b)
> where S.b = T.b);
> This should return two rows, but it actually just returns one, because it
> does not include a unique key for the main query, T.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
