ViggoC commented on code in PR #44241:
URL: https://github.com/apache/arrow/pull/44241#discussion_r1800487432
##########
java/vector/src/main/java/org/apache/arrow/vector/complex/RunEndEncodedVector.java:
##########
@@ -312,8 +302,142 @@ public TransferPair getTransferPair(Field field,
BufferAllocator allocator, Call
*/
@Override
public TransferPair makeTransferPair(ValueVector target) {
- throw new UnsupportedOperationException(
- "RunEndEncodedVector does not support makeTransferPair(ValueVector)");
+ return new TransferImpl((RunEndEncodedVector) target);
+ }
+
+ private class TransferImpl implements TransferPair {
+
+ RunEndEncodedVector to;
+ TransferPair dataTransferPair;
+ TransferPair reeTransferPair;
+
+ public TransferImpl(String name, BufferAllocator allocator, CallBack
callBack) {
+ this(new RunEndEncodedVector(name, allocator, field.getFieldType(),
callBack));
+ }
+
+ public TransferImpl(Field field, BufferAllocator allocator, CallBack
callBack) {
+ this(new RunEndEncodedVector(field, allocator, callBack));
+ }
+
+ public TransferImpl(RunEndEncodedVector to) {
+ this.to = to;
+ if (to.getRunEndsVector() instanceof ZeroVector) {
+ to.initializeChildrenFromFields(field.getChildren());
+ }
+ reeTransferPair =
getRunEndsVector().makeTransferPair(to.getRunEndsVector());
+ dataTransferPair =
getValuesVector().makeTransferPair(to.getValuesVector());
+ }
+
+ /**
+ * Transfer this vector'data to another vector. The memory associated with
this vector is
+ * transferred to the allocator of target vector for accounting and
management purposes.
+ */
+ @Override
+ public void transfer() {
+ to.clear();
+ dataTransferPair.transfer();
+ reeTransferPair.transfer();
+ if (valueCount > 0) {
+ to.setValueCount(valueCount);
+ }
+ clear();
+ }
+
+ /**
+ * Slice this vector at desired index and length and transfer the
corresponding data to the
+ * target vector.
+ *
+ * @param startIndex start position of the split in source vector.
+ * @param length length of the split.
+ */
+ @Override
+ public void splitAndTransfer(int startIndex, int length) {
+ ValueVector toDataVector = dataTransferPair.getTo();
+ ValueVector toRunEndVector = reeTransferPair.getTo();
+
+ toDataVector.clear();
+ toRunEndVector.clear();
+
+ int endIndex = startIndex + length;
+ int physicalStartIndex = getPhysicalIndex(startIndex);
+ int physicalEndIndex = getPhysicalIndex(endIndex);
+ int physicalLength = physicalEndIndex - physicalStartIndex + 1;
+ dataTransferPair.splitAndTransfer(physicalStartIndex, physicalLength);
+ if (startIndex == 0) {
+ reeTransferPair.splitAndTransfer(physicalStartIndex, physicalLength);
+ } else {
+ shiftRunEndVector(
+ toRunEndVector,
+ startIndex,
+ length,
+ physicalStartIndex,
+ physicalEndIndex,
+ physicalLength);
+ }
+ getTo().setValueCount(length);
+ }
+
+ private void shiftRunEndVector(
+ ValueVector toRunEndVector,
+ int startIndex,
+ int length,
+ int physicalStartIndex,
+ int physicalEndIndex,
+ int physicalLength) {
+ toRunEndVector.setValueCount(physicalLength);
+ toRunEndVector.getValidityBuffer().setOne(0,
toRunEndVector.getValidityBuffer().capacity());
Review Comment:
In ListVector, it slice the validityBuffer if startIndex % 8 == 0, and copy
data one by one when the first bit starts from the middle of a byte.
But for run end encoded vector, the element of run end vector can never be
null so I just set validity buffer of RunEndVector to 1.
What's your concern about this code? The memory of validity buffer should be
reused when startIndex % 8 == 0, Or we should not set the bit beyond the
physical length?
--
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]
