Github user sounakr commented on a diff in the pull request:
https://github.com/apache/carbondata/pull/2417#discussion_r201243892
--- Diff:
core/src/main/java/org/apache/carbondata/core/datastore/chunk/store/ColumnPageWrapper.java
---
@@ -53,11 +60,75 @@ public int fillVector(int[] filteredRowId,
ColumnVectorInfo[] vectorInfo, int ch
throw new UnsupportedOperationException("internal error");
}
- @Override
- public byte[] getChunkData(int rowId) {
- return columnPage.getBytes(rowId);
+ @Override public byte[] getChunkData(int rowId) {
+ ColumnType columnType = columnPage.getColumnSpec().getColumnType();
+ DataType srcDataType = columnPage.getColumnSpec().getSchemaDataType();
+ DataType targetDataType = columnPage.getDataType();
+ if (columnPage.getNullBits().get(rowId)) {
+ // if this row is null, return default null represent in byte array
+ return CarbonCommonConstants.MEMBER_DEFAULT_VAL_ARRAY;
+ }
+ if ((columnType == ColumnType.COMPLEX_PRIMITIVE) &&
this.isAdaptiveComplexPrimitive()) {
+ if (srcDataType == DataTypes.DOUBLE || srcDataType ==
DataTypes.FLOAT) {
+ double doubleData = columnPage.getDouble(rowId);
+ if (srcDataType == DataTypes.FLOAT) {
+ float out = (float) doubleData;
--- End diff --
in continuation of the above comment, for floating codec lazyColumnPage has
only double based comparison with statistics min and max value. So always we
will get lazy page returned value as double for floating codec. So the
conversion cannot be avoided.
Also currently float is not implemented in carbon core. Always float values
will be converted to double and saved in disk.
---
