peter-toth opened a new pull request #29771:
URL: https://github.com/apache/spark/pull/29771
### What changes were proposed in this pull request?
This PR rewrites `FoldablePropagation` rule to replace foldables in a node
with foldables coming only from the node's children.
Before this PR in the case of this example:
```
val a = Seq("1").toDF("col1").withColumn("col2", lit("1"))
val b = Seq("2").toDF("col1").withColumn("col2", lit("2"))
val aub = a.union(b)
val c = aub.filter($"col1" === "2").cache()
val d = Seq("2").toDF( "col4")
val r = d.join(aub, $"col2" === $"col4").select("col4")
val l = c.select("col2")
val df = l.join(r, $"col2" === $"col4", "LeftOuter")
df.show()
```
foldable propagation happens incorrectly:
```
Join LeftOuter, (col2#6 = col4#34)
Join LeftOuter, (col2#6 = col4#34)
!:- Project [col2#6]
:- Project [1 AS col2#6]
: +- InMemoryRelation [col1#4, col2#6], StorageLevel(disk, memory,
deserialized, 1 replicas) : +- InMemoryRelation [col1#4, col2#6],
StorageLevel(disk, memory, deserialized, 1 replicas)
: +- Union
: +- Union
: :- *(1) Project [value#1 AS col1#4, 1 AS col2#6]
: :- *(1) Project [value#1 AS col1#4, 1 AS col2#6]
: : +- *(1) Filter (isnotnull(value#1) AND (value#1 = 2))
: : +- *(1) Filter (isnotnull(value#1) AND
(value#1 = 2))
: : +- *(1) LocalTableScan [value#1]
: : +- *(1) LocalTableScan [value#1]
: +- *(2) Project [value#10 AS col1#13, 2 AS col2#15]
: +- *(2) Project [value#10 AS col1#13, 2 AS
col2#15]
: +- *(2) Filter (isnotnull(value#10) AND (value#10 = 2))
: +- *(2) Filter (isnotnull(value#10) AND
(value#10 = 2))
: +- *(2) LocalTableScan [value#10]
: +- *(2) LocalTableScan [value#10]
+- Project [col4#34]
+- Project [col4#34]
+- Join Inner, (col2#6 = col4#34)
+- Join Inner, (col2#6 = col4#34)
:- Project [value#31 AS col4#34]
:- Project [value#31 AS col4#34]
: +- LocalRelation [value#31]
: +- LocalRelation [value#31]
+- Project [col2#6]
+- Project [col2#6]
+- Union false, false
+- Union false, false
:- Project [1 AS col2#6]
:- Project [1 AS col2#6]
: +- LocalRelation [value#1]
: +- LocalRelation [value#1]
+- Project [2 AS col2#15]
+- Project [2 AS col2#15]
+- LocalRelation [value#10]
+- LocalRelation [value#10]
```
and so the result is wrong:
```
+----+----+
|col2|col4|
+----+----+
| 1|null|
+----+----+
```
After this PR foldable propagation will not happen incorrectly and the
result is correct:
```
+----+----+
|col2|col4|
+----+----+
| 2| 2|
+----+----+
```
### Why are the changes needed?
To fix a correctness issue.
### Does this PR introduce _any_ user-facing change?
Yes, fixes a correctness issue.
### How was this patch tested?
Existing and new UTs.
----------------------------------------------------------------
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]
