[
https://issues.apache.org/jira/browse/TAJO-939?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14064615#comment-14064615
]
ASF GitHub Bot commented on TAJO-939:
-------------------------------------
Github user hyunsik commented on a diff in the pull request:
https://github.com/apache/tajo/pull/71#discussion_r15042347
--- Diff:
tajo-core/src/main/java/org/apache/tajo/engine/planner/LogicalPlanner.java ---
@@ -383,7 +386,7 @@ private EvalExprNode
buildPlanForNoneFromStatement(PlanContext context, Stack<Ex
for (int i = 0; i < targets.length; i++) {
NamedExpr namedExpr = projection.getNamedExprs()[i];
- EvalNode evalNode = exprAnnotator.createEvalNode(plan, block,
namedExpr.getExpr());
+ EvalNode evalNode = exprAnnotator.createEvalNode(plan, block,
namedExpr.getExpr(), NameResolvingMode.RELS_ONLY);
--- End diff --
As you know, non-from statement does not need to resolve any column because
non-from statement means no table and no column reference.
The main difference among the resolving modes is the priority and scope. It
is enough to use the mode which has the correct priority and the scope. If
possible, it would be nice if we use the mode the minimum scope. RELS_ONLY is
the minimum scope one.
> Refactoring the column resolver in LogicalPlan
> ----------------------------------------------
>
> Key: TAJO-939
> URL: https://issues.apache.org/jira/browse/TAJO-939
> Project: Tajo
> Issue Type: Bug
> Components: planner/optimizer
> Reporter: Hyunsik Choi
> Assignee: Hyunsik Choi
> Fix For: 0.9.0
>
>
> The main role of the column resolver is to find the exact column in a
> relation or a temporal column to which a variable name points. We have used a
> monolithic column resolver to deal with lots of cases.
> But, resolving a name should play different roles according to at which the
> name is placed.
> For example, 1) a column name in select list always points one of fields in
> relations, 2) a column name in WHERE clause can point to one of fields in
> relations or one of aliased temporal fields in select list. If there are
> duplicated, the column name firstly chooses the field in relations. 3) a
> column name in ORDER BY clause is similar to that in WHERE clause, but it
> firstly chooses one of aliased temporal fields in select list.
> The current column resolver does not consider the above rules. As a result,
> it works incorrectly in some cases where a sql statement includes the same
> name references, actually indicating one field in relation and one aliased
> temporal field in select list. We should fix it.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
