HyukjinKwon opened a new pull request #28695:
URL: https://github.com/apache/spark/pull/28695
### What changes were proposed in this pull request?
This PR proposes to check `DetectAmbiguousSelfJoin` only if there is `Join`
in the plan. Currently, the checking is too strict even to non-join queries.
For example, the codes below don't have join at all but it fails as the
ambiguous self-join:
```scala
import org.apache.spark.sql.expressions.Window
import org.apache.spark.sql.functions.sum
val df = Seq(1, 1, 2, 2).toDF("A")
val w = Window.partitionBy(df("A"))
df.select(df("A").alias("X"), sum(df("A")).over(w)).explain(true)
```
It is because `ExtractWindowExpressions` can create a `AttributeReference`
with the same metadata but a different expression ID, see:
https://github.com/apache/spark/blob/0fd98abd859049dc3b200492487041eeeaa8f737/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala#L2679
https://github.com/apache/spark/blob/71c73d58f6e88d2558ed2e696897767d93bac60f/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicLogicalOperators.scala#L63
https://github.com/apache/spark/blob/5945d46c11a86fd85f9e65f24c2e88f368eee01f/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/namedExpressions.scala#L180
Before:
```
'Project [A#19 AS X#21, sum(A#19) windowspecdefinition(A#19,
specifiedwindowframe(RowFrame, unboundedpreceding$(), unboundedfollowing$()))
AS sum(A) OVER (PARTITION BY A unspecifiedframe$())#23L]
+- Relation[A#19] parquet
```
After:
```
Project [X#21, sum(A) OVER (PARTITION BY A unspecifiedframe$())#23L]
+- Project [X#21, A#19, sum(A) OVER (PARTITION BY A
unspecifiedframe$())#23L, sum(A) OVER (PARTITION BY A unspecifiedframe$())#23L]
+- Window [sum(A#19) windowspecdefinition(A#19,
specifiedwindowframe(RowFrame, unboundedpreceding$(), unboundedfollowing$()))
AS sum(A) OVER (PARTITION BY A unspecifiedframe$())#23L], [A#19]
+- Project [A#19 AS X#21, A#19]
+- Relation[A#19] parquet
```
`X#21` holds the same metadata of DataFrame ID and column position with
`A#19` but it has a different expression ID which ends up with the checking
fails.
### Why are the changes needed?
To loose the checking and make users not surprised.
### Does this PR introduce _any_ user-facing change?
It's the changes in unreleased branches only.
### How was this patch tested?
Manually tested and unittest was added.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
