JingsongLi commented on code in PR #10208:
URL: https://github.com/apache/paimon/pull/10208#discussion_r4178165866
##########
paimon-core/src/main/java/org/apache/paimon/table/sink/TableWriteImpl.java:
##########
@@ -264,6 +268,10 @@ private void checkNullability(InternalRow row, RowKind
rowKind) {
}
private InternalRow wrapDefaultValue(InternalRow row) {
+ if (defaultValueRowOutdated) {
+ defaultValueRow = DefaultValueRow.create(writeType);
Review Comment:
[P2] Preserve ROW defaults when rebuilding for a nested partial write
The Java `BatchTableWrite.withWriteType(...).write(...)` path also wraps
rows (as exercised by `NestedSubfieldDataEvolutionTableTest`). For a table with
one column `nest ROW<a INT,b STRING> DEFAULT {42,z}`, enabling row
tracking/data evolution/nested fields and writing with
`table.rowType().projectByPaths(["nest.a"])` retains the full `{42,z}` default
in the projected field metadata. This new `create(writeType)` consequently
throws `Row field count mismatch. Expected: 1, Actual: 2`, even for an
explicitly nonnull `GenericRow.of(GenericRow.of(100))`. I reproduced this with
a real FileSystemCatalog: commit `(10,x)`, then update only `nest.a`; this head
fails before writing, while disabling only `defaultValueRowOutdated` restores
the previous behavior and commits/reads `(100,x)`. Please project/translate the
cached defaults consistently with the nested write type (or otherwise retain
support for this wrapping path), and add a Java partial-write regression test;
bypassing the wrapper
in Spark/Flink partial writers does not cover this API.
--
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]
