dekimir commented on code in PR #16668:
URL: https://github.com/apache/iceberg/pull/16668#discussion_r3347428392
##########
core/src/test/java/org/apache/iceberg/DeleteFileIndexTestBase.java:
##########
@@ -480,6 +480,33 @@ public void
testPartitionedTableScanWithGlobalAndPartitionDeletes() {
.isEqualTo(Sets.newHashSet(unpartitionedEqDeletes.location(),
FILE_A_EQ_1.location()));
}
+ @TestTemplate
+ public void testGlobalEqDeleteWithSpecIdCollision() {
+ // PartitionSpec.unpartitioned() has specId 0, but does not reserve it so
we must handle
+ // a collision where SpecId 0 is a partitioned spec.
+
assertThat(SPEC.specId()).isEqualTo(PartitionSpec.unpartitioned().specId());
+
+ // Global equality delete written with PartitionSpec.unpartitioned()
(specId = 0)
+ DeleteFile globalEqDelete =
+ withDataSequenceNumber(4,
unpartitionedEqDeletes(PartitionSpec.unpartitioned()));
+
+ // specsById contains only the partitioned SPEC at ID 0, simulating a
table that was created
+ // with a partitioned spec as its first (and only) spec.
PartitionSpec.unpartitioned() is not
+ // in the table's spec history, but shares specId = 0 with the table's
partitioned spec.
Review Comment:
I actually think this should be (made) impossible.
`PartitionSpec.unpartitioned()` must never return a spec ID that is, in fact,
partitioned. This could be achieved either by modifying `unpartitioned()` to
ensure the returned ID is indeed unpartitioned (even if non-zero) OR by making
table creation fail when ID zero is not unpartitioned.
--
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]
