[SPARK-16594] Physical Plan Differences when Table Scan Having Duplicate Columns

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Incomplete
• Affects Version/s: 2.0.0
• Component/s: SQL
• Labels: bulk-closed
• Assignee: Unassigned
• Reporter: Xiao Li
• Created: 17/Jul/16 16:54
• Updated: 21/May/19 04:36
• Resolved: 21/May/19 04:36

Description

Currently, we keep two implementations for planning scans over data sources. There is one difference between two implementation when deciding whether a `Project` is needed or not.

• Data Source Table Scan: https://github.com/apache/spark/blob/b1e5281c5cb429e338c3719c13c0b93078d7312a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala#L322-L395

• `SELECT b, b FROM oneToTenPruned`: *Add `ProjectExec`.*

• Hive Table Scan and In-memory Table Scan:
  https://github.com/apache/spark/blob/865ec32dd997e63aea01a871d1c7b4947f43c111/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlanner.scala#L71-L99

• `SELECT b, b FROM oneToTenPruned`: *No `ProjectExec` is added.*

*Note.* When alias is being used, we will always add `ProjectExec` in all the scan types.

• `SELECT b as alias_b, b FROM oneToTenPruned`: *Add `ProjectExec`.*

Because selecting the same column twice without adding `alias` is very weird, no clue why the code needs to behave differently here.

Issue Links

• links to: [Github] Pull Request #14240 (gatorsmile)