[
https://issues.apache.org/jira/browse/TAJO-1355?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14368275#comment-14368275
]
Hyunsik Choi commented on TAJO-1355:
------------------------------------
Hi,
The problem is that a target list in Projectable nodes (ScanNode, JoinNode,
GroupByNode, and WinAggNode) can have some duplicated FieldEvals to actually
indicate one field.
The duplicated FieldEvals can be either fully qualified one and a simple name
one. This problem occurs occasionally in some cases. It's because of some bugs
in LogicalPlanPreProcessor, LogicalPlanner and ProjectionPushDown. We should
fix the code to make duplicated ones.
Even though this problem looks simple, its change scope is not narrow. In
addition, now, I'm not sure If we can break this problem into several subtasks
because the problems are related to one another. To know that, we need more
investigation.
The benefit of this issue is reduced row width, leading reduced memory
consumption, network transmission, and computation cost. Also, we will have a
change to cleanup and simplify Schema class, which keep two names spaces:
qualified names and simple names. In other words, after this issue, we can
change Schema to keep only one namespace.
I hope that my description is helpful to you.
> All column resolving should be as fully qualified names.
> --------------------------------------------------------
>
> Key: TAJO-1355
> URL: https://issues.apache.org/jira/browse/TAJO-1355
> Project: Tajo
> Issue Type: Improvement
> Components: planner/optimizer
> Reporter: Hyunsik Choi
>
> Currently, schemas used in a logical plan includes fully qualified names and
> simple name without qualified names. It makes many parts complicated
> unnecessarily.
> For example, Schema class can be simplified if it maintains only qualified
> names. Binding correct columns in FieldEval can be simplified if we do it.
> Also, many query optimization code would be simple.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
