github-actions[bot] commented on code in PR #66007:
URL: https://github.com/apache/doris/pull/66007#discussion_r3679293696
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/trees/plans/logical/LogicalSetOperation.java:
##########
@@ -311,9 +311,13 @@ private static DataType
getAssignmentCompatibleTypeLegacy(DataType left, DataTyp
}
ImmutableList.Builder<StructField> commonFields =
ImmutableList.builder();
for (int i = 0; i < leftFields.size(); i++) {
- boolean nullable = leftFields.get(i).isNullable() ||
rightFields.get(i).isNullable();
DataType commonType = getAssignmentCompatibleType(
leftFields.get(i).getDataType(),
rightFields.get(i).getDataType());
+ // Legacy set operations need a nullable common child whenever
coercing either
+ // input can itself produce NULL, even if both source fields
are declared required.
+ boolean nullable = leftFields.get(i).isNullable() ||
rightFields.get(i).isNullable()
+ || Cast.castNullable(false,
leftFields.get(i).getDataType(), commonType)
Review Comment:
[P2] Keep an unchanged datetime sibling required
When the structs differ in another field, this loop also evaluates unchanged
children. For
example, combining
`STRUCT<d:DATETIMEV2(3) NOT NULL,n:INT NOT NULL>` with
`STRUCT<d:DATETIMEV2(3) NOT NULL,n:BIGINT NOT NULL>` widens only `n`, but
`Cast.castNullable(false, d, d)` still returns true, so `d` becomes
nullable. The BE uses an
identity wrapper for that identical child, so it cannot introduce NULL;
nevertheless a later
INSERT/cast into the original required nested layout is rejected by
`CheckCast`, and its same
unconditional check also rejects a direct required-to-required widening of
these structs.
This is the inverse of the earlier thread's genuinely converted child.
Please make exact-type
casts preserve nullability at the shared `castNullable` boundary (or guard
every changed
caller, including `CheckCast` and the parallel reducers) and add two-field
set/direct-cast
cases.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
