Copilot commented on code in PR #12503:
URL: https://github.com/apache/gluten/pull/12503#discussion_r3573008078
##########
cpp/velox/tests/VeloxGpuShuffleWriterTest.cc:
##########
@@ -603,6 +603,8 @@ TEST_P(GpuRoundRobinPartitioningShuffleWriterTest,
preAllocForceRealloc) {
makeNullableFlatVector<int64_t>({0, 1}),
makeNullableFlatVector<float>({0, 0.142857}),
makeNullableFlatVector<bool>({false, true}),
+ makeNullableFlatVector<facebook::velox::Timestamp>(
+ {facebook::velox::Timestamp(5, 0), facebook::velox::Timestamp(4,
0)}),
makeNullableFlatVector<StringView>({"", "alice"}),
Review Comment:
In GPU tests, `inputVectorNoNull_` still contains an `UNKNOWN` column (only
`children1_`/`children2_` are `pop_back()`’d in `SetUp()`), but `intHasNull`
here does not. This makes the schemas inconsistent across splits and also
reintroduces an `UNKNOWN` column into GPU shuffle inputs (cudf doesn’t support
it). Build `inputNoNull` from `childrenNoNull_` with the trailing `UNKNOWN`
removed so both splits have the same, supported schema.
##########
cpp/velox/tests/VeloxGpuShuffleWriterTest.cc:
##########
@@ -663,6 +665,8 @@ TEST_P(GpuRoundRobinPartitioningShuffleWriterTest,
preAllocForceReuse) {
makeNullableFlatVector<int64_t>({0, 1}),
makeNullableFlatVector<float>({0, 0.142857}),
makeNullableFlatVector<bool>({false, true}),
+ makeNullableFlatVector<facebook::velox::Timestamp>(
+ {facebook::velox::Timestamp(5, 0), facebook::velox::Timestamp(4,
0)}),
Review Comment:
Same issue as above: `inputVectorNoNull_` still includes the trailing
`UNKNOWN` column in GPU tests, but `stringHasNull` here doesn’t. This can break
split/write due to schema mismatch and keeps an unsupported `UNKNOWN` type in
the first batch. Construct `inputNoNull` from `childrenNoNull_` with the
trailing `UNKNOWN` removed.
--
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]
