You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@pinot.apache.org by "KKcorps (via GitHub)" <gi...@apache.org> on 2023/04/13 20:06:27 UTC

[GitHub] [pinot] KKcorps commented on a diff in pull request #10594: [feature] [null support # 8] Support null in all null related transform functions

KKcorps commented on code in PR #10594:
URL: https://github.com/apache/pinot/pull/10594#discussion_r1165976967


##########
pinot-core/src/main/java/org/apache/pinot/core/operator/transform/function/CaseTransformFunction.java:
##########
@@ -241,226 +255,768 @@ public TransformResultMetadata getResultMetadata() {
   }
 
   /**
-   * Evaluate the ValueBlock for the WHEN statements, returns an array with the
-   * index(1 to N) of matched WHEN clause ordered by match priority, 0 means nothing
-   * matched, so go to ELSE.
+   * Evaluate the ValueBlock for the WHEN statements, returns an array with the index(1 to N) of matched WHEN clause
+   * -1 means there is no match.
    */
-  private int[] getSelectedArray(ValueBlock valueBlock) {
+  private int[] getSelectedArray(ValueBlock valueBlock, boolean nullHandlingEnabled) {
     int numDocs = valueBlock.getNumDocs();
     if (_selectedResults == null || _selectedResults.length < numDocs) {
       _selectedResults = new int[numDocs];
-    } else {
-      Arrays.fill(_selectedResults, 0, numDocs, 0);
-      Arrays.fill(_selections, false);
     }
+    Arrays.fill(_selectedResults, 0, numDocs, -1);
+    Arrays.fill(_selections, false);
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
     int numWhenStatements = _whenStatements.size();
-    for (int i = numWhenStatements - 1; i >= 0; i--) {
+    for (int i = 0; i < numWhenStatements; i++) {
       TransformFunction whenStatement = _whenStatements.get(i);
-      int[] conditions = whenStatement.transformToIntValuesSV(valueBlock);
-      for (int j = 0; j < numDocs & j < conditions.length; j++) {
-        _selectedResults[j] = Math.max(conditions[j] * (i + 1), _selectedResults[j]);
+      int[] conditions = getWhenConditions(whenStatement, valueBlock, nullHandlingEnabled);
+      for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+        if (conditions[docId] == 1) {
+          unselectedDocs.clear(docId);
+          _selectedResults[docId] = i;
+        }
+      }
+      if (unselectedDocs.isEmpty()) {
+        break;
       }
     }
     // try to prune clauses now
     for (int i = 0; i < numDocs; i++) {
-      _selections[_selectedResults[i]] = true;
-    }
-    int numSelections = 0;
-    for (boolean selection : _selections) {
-      if (selection) {
-        numSelections++;
+      if (_selectedResults[i] != -1) {
+        _selections[_selectedResults[i]] = true;
       }
     }
-    _numSelections = numSelections;
     return _selectedResults;
   }
 
+  // Returns an array of valueBlock length to indicate whether a row is selected or not.
+  // When nullHandlingEnabled is set to true, we also check whether the row is null and set to false if null.
+  private static int[] getWhenConditions(TransformFunction whenStatement, ValueBlock valueBlock,
+      boolean nullHandlingEnabled) {
+    if (!nullHandlingEnabled) {
+      return whenStatement.transformToIntValuesSV(valueBlock);
+    }
+    Pair<int[], RoaringBitmap> result = whenStatement.transformToIntValuesSVWithNull(valueBlock);
+    RoaringBitmap bitmap = result.getRight();
+    int[] intResult = result.getLeft();
+    if (bitmap != null) {
+      bitmap.forEach((IntConsumer) (i) -> {
+        if (bitmap.contains(i)) {
+          intResult[i] = 0;
+        }
+      });
+    }
+    return intResult;
+  }
+
   @Override
   public int[] transformToIntValuesSV(ValueBlock valueBlock) {
     if (_resultMetadata.getDataType().getStoredType() != DataType.INT) {
       return super.transformToIntValuesSV(valueBlock);
     }
-    int[] selected = getSelectedArray(valueBlock);
+    int[] selected = getSelectedArray(valueBlock, false);
     int numDocs = valueBlock.getNumDocs();
     initIntValuesSV(numDocs);
-    int numElseThenStatements = _elseThenStatements.size();
-    for (int i = 0; i < numElseThenStatements; i++) {
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    for (int i = 0; i < numThenStatements; i++) {
       if (_selections[i]) {
-        TransformFunction transformFunction = _elseThenStatements.get(i);
+        TransformFunction transformFunction = _thenStatements.get(i);
         int[] intValues = transformFunction.transformToIntValuesSV(valueBlock);
-        if (_numSelections == 1) {
-          System.arraycopy(intValues, 0, _intValuesSV, 0, numDocs);
-        } else {
-          for (int j = 0; j < numDocs; j++) {
-            if (selected[j] == i) {
-              _intValuesSV[j] = intValues[j];
-            }
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            unselectedDocs.clear(docId);
+            _intValuesSV[docId] = intValues[docId];
           }
         }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _intValuesSV[docId] = (int) DataSchema.ColumnDataType.INT.getNullPlaceholder();
+        }
+      } else {
+        int[] intValues = _elseStatement.transformToIntValuesSV(valueBlock);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _intValuesSV[docId] = intValues[docId];
+        }
       }
     }
     return _intValuesSV;
   }
 
+  @Override
+  public Pair<int[], RoaringBitmap> transformToIntValuesSVWithNull(ValueBlock valueBlock) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.INT) {
+      return super.transformToIntValuesSVWithNull(valueBlock);
+    }
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    initIntValuesSV(numDocs);
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        Pair<int[], RoaringBitmap> intValuesNullPair = transformFunction.transformToIntValuesSVWithNull(valueBlock);
+        int[] intValues = intValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = intValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _intValuesSV[docId] = intValues[docId];
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _intValuesSV[docId] = (int) DataSchema.ColumnDataType.INT.getNullPlaceholder();
+          bitmap.add(docId);
+        }
+      } else {
+        Pair<int[], RoaringBitmap> intValuesNullPair = _elseStatement.transformToIntValuesSVWithNull(valueBlock);
+        int[] intValues = intValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = intValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _intValuesSV[docId] = intValues[docId];
+          if (nullBitmap != null && nullBitmap.contains(docId)) {
+            bitmap.add(docId);
+          }
+        }
+      }
+    }
+    return ImmutablePair.of(_intValuesSV, bitmap);
+  }
+
   @Override
   public long[] transformToLongValuesSV(ValueBlock valueBlock) {
     if (_resultMetadata.getDataType().getStoredType() != DataType.LONG) {
       return super.transformToLongValuesSV(valueBlock);
     }
-    int[] selected = getSelectedArray(valueBlock);
+    int[] selected = getSelectedArray(valueBlock, false);
     int numDocs = valueBlock.getNumDocs();
     initLongValuesSV(numDocs);
-    int numElseThenStatements = _elseThenStatements.size();
-    for (int i = 0; i < numElseThenStatements; i++) {
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    for (int i = 0; i < numThenStatements; i++) {
       if (_selections[i]) {
-        TransformFunction transformFunction = _elseThenStatements.get(i);
+        TransformFunction transformFunction = _thenStatements.get(i);
         long[] longValues = transformFunction.transformToLongValuesSV(valueBlock);
-        if (_numSelections == 1) {
-          System.arraycopy(longValues, 0, _longValuesSV, 0, numDocs);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _longValuesSV[docId] = longValues[docId];
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+      if (!unselectedDocs.isEmpty()) {
+        if (_elseStatement == null) {
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _longValuesSV[docId] = (long) DataSchema.ColumnDataType.LONG.getNullPlaceholder();
+          }
         } else {
-          for (int j = 0; j < numDocs; j++) {
-            if (selected[j] == i) {
-              _longValuesSV[j] = longValues[j];
-            }
+          long[] longValues = _elseStatement.transformToLongValuesSV(valueBlock);
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _longValuesSV[docId] = longValues[docId];
           }
         }
       }
     }
     return _longValuesSV;
   }
 
+  @Override
+  public Pair<long[], RoaringBitmap> transformToLongValuesSVWithNull(ValueBlock valueBlock) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.LONG) {
+      return super.transformToLongValuesSVWithNull(valueBlock);
+    }
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    initLongValuesSV(numDocs);
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        Pair<long[], RoaringBitmap> longValuesNullPair = transformFunction.transformToLongValuesSVWithNull(valueBlock);
+        long[] longValues = longValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = longValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _longValuesSV[docId] = longValues[docId];
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _longValuesSV[docId] = (long) DataSchema.ColumnDataType.LONG.getNullPlaceholder();
+          bitmap.add(docId);
+        }
+      } else {
+        Pair<long[], RoaringBitmap> longValuesNullPair = _elseStatement.transformToLongValuesSVWithNull(valueBlock);
+        long[] longValues = longValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = longValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _longValuesSV[docId] = longValues[docId];
+          if (nullBitmap != null && nullBitmap.contains(docId)) {
+            bitmap.add(docId);
+          }
+        }
+      }
+    }
+    return ImmutablePair.of(_longValuesSV, bitmap);
+  }
+
   @Override
   public float[] transformToFloatValuesSV(ValueBlock valueBlock) {
     if (_resultMetadata.getDataType().getStoredType() != DataType.FLOAT) {
       return super.transformToFloatValuesSV(valueBlock);
     }
-    int[] selected = getSelectedArray(valueBlock);
+    int[] selected = getSelectedArray(valueBlock, false);
     int numDocs = valueBlock.getNumDocs();
     initFloatValuesSV(numDocs);
-    int numElseThenStatements = _elseThenStatements.size();
-    for (int i = 0; i < numElseThenStatements; i++) {
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    for (int i = 0; i < numThenStatements; i++) {
       if (_selections[i]) {
-        TransformFunction transformFunction = _elseThenStatements.get(i);
+        TransformFunction transformFunction = _thenStatements.get(i);
         float[] floatValues = transformFunction.transformToFloatValuesSV(valueBlock);
-        if (_numSelections == 1) {
-          System.arraycopy(floatValues, 0, _floatValuesSV, 0, numDocs);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _floatValuesSV[docId] = floatValues[docId];
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+      if (!unselectedDocs.isEmpty()) {
+        if (_elseStatement == null) {
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _floatValuesSV[docId] = (float) DataSchema.ColumnDataType.FLOAT.getNullPlaceholder();
+          }
         } else {
-          for (int j = 0; j < numDocs; j++) {
-            if (selected[j] == i) {
-              _floatValuesSV[j] = floatValues[j];
-            }
+          float[] floatValues = _elseStatement.transformToFloatValuesSV(valueBlock);
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _floatValuesSV[docId] = floatValues[docId];
           }
         }
       }
     }
     return _floatValuesSV;
   }
 
+  @Override
+  public Pair<float[], RoaringBitmap> transformToFloatValuesSVWithNull(ValueBlock valueBlock) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.FLOAT) {
+      return super.transformToFloatValuesSVWithNull(valueBlock);
+    }
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    initFloatValuesSV(numDocs);
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        Pair<float[], RoaringBitmap> floatValuesNullPair =
+            transformFunction.transformToFloatValuesSVWithNull(valueBlock);
+        float[] floatValues = floatValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = floatValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _floatValuesSV[docId] = floatValues[docId];
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _floatValuesSV[docId] = (float) DataSchema.ColumnDataType.FLOAT.getNullPlaceholder();
+          bitmap.add(docId);
+        }
+      } else {
+        Pair<float[], RoaringBitmap> floatValuesNullPair = _elseStatement.transformToFloatValuesSVWithNull(valueBlock);
+        float[] floatValues = floatValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = floatValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _floatValuesSV[docId] = floatValues[docId];
+          if (nullBitmap != null && nullBitmap.contains(docId)) {
+            bitmap.add(docId);
+          }
+        }
+      }
+    }
+    return ImmutablePair.of(_floatValuesSV, bitmap);
+  }
+
   @Override
   public double[] transformToDoubleValuesSV(ValueBlock valueBlock) {
     if (_resultMetadata.getDataType().getStoredType() != DataType.DOUBLE) {
       return super.transformToDoubleValuesSV(valueBlock);
     }
-    int[] selected = getSelectedArray(valueBlock);
+    int[] selected = getSelectedArray(valueBlock, false);
     int numDocs = valueBlock.getNumDocs();
     initDoubleValuesSV(numDocs);
-    int numElseThenStatements = _elseThenStatements.size();
-    for (int i = 0; i < numElseThenStatements; i++) {
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    for (int i = 0; i < numThenStatements; i++) {
       if (_selections[i]) {
-        TransformFunction transformFunction = _elseThenStatements.get(i);
+        TransformFunction transformFunction = _thenStatements.get(i);
         double[] doubleValues = transformFunction.transformToDoubleValuesSV(valueBlock);
-        if (_numSelections == 1) {
-          System.arraycopy(doubleValues, 0, _doubleValuesSV, 0, numDocs);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _doubleValuesSV[docId] = doubleValues[docId];
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+      if (!unselectedDocs.isEmpty()) {
+        if (_elseStatement == null) {
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _doubleValuesSV[docId] = (double) DataSchema.ColumnDataType.DOUBLE.getNullPlaceholder();
+          }
         } else {
-          for (int j = 0; j < numDocs; j++) {
-            if (selected[j] == i) {
-              _doubleValuesSV[j] = doubleValues[j];
-            }
+          double[] doubleValues = _elseStatement.transformToDoubleValuesSV(valueBlock);
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _doubleValuesSV[docId] = doubleValues[docId];
           }
         }
       }
     }
     return _doubleValuesSV;
   }
 
+  @Override
+  public Pair<double[], RoaringBitmap> transformToDoubleValuesSVWithNull(ValueBlock valueBlock) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.DOUBLE) {
+      return super.transformToDoubleValuesSVWithNull(valueBlock);
+    }
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    initDoubleValuesSV(numDocs);
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        Pair<double[], RoaringBitmap> doubleValuesNullPair =
+            transformFunction.transformToDoubleValuesSVWithNull(valueBlock);
+        double[] doubleValues = doubleValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = doubleValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _doubleValuesSV[docId] = doubleValues[docId];
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _doubleValuesSV[docId] = (double) DataSchema.ColumnDataType.DOUBLE.getNullPlaceholder();
+          bitmap.add(docId);
+        }
+      } else {
+        Pair<double[], RoaringBitmap> doubleValuesNullPair =
+            _elseStatement.transformToDoubleValuesSVWithNull(valueBlock);
+        double[] doubleValues = doubleValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = doubleValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _doubleValuesSV[docId] = doubleValues[docId];
+          if (nullBitmap != null && nullBitmap.contains(docId)) {
+            bitmap.add(docId);
+          }
+        }
+      }
+    }
+    return ImmutablePair.of(_doubleValuesSV, bitmap);
+  }
+
   @Override
   public BigDecimal[] transformToBigDecimalValuesSV(ValueBlock valueBlock) {
-    if (_resultMetadata.getDataType() != DataType.BIG_DECIMAL) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.BIG_DECIMAL) {
       return super.transformToBigDecimalValuesSV(valueBlock);
     }
-    int[] selected = getSelectedArray(valueBlock);
+    int[] selected = getSelectedArray(valueBlock, false);
     int numDocs = valueBlock.getNumDocs();
     initBigDecimalValuesSV(numDocs);
-    int numElseThenStatements = _elseThenStatements.size();
-    for (int i = 0; i < numElseThenStatements; i++) {
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    for (int i = 0; i < numThenStatements; i++) {
       if (_selections[i]) {
-        TransformFunction transformFunction = _elseThenStatements.get(i);
+        TransformFunction transformFunction = _thenStatements.get(i);
         BigDecimal[] bigDecimalValues = transformFunction.transformToBigDecimalValuesSV(valueBlock);
-        if (_numSelections == 1) {
-          System.arraycopy(bigDecimalValues, 0, _bigDecimalValuesSV, 0, numDocs);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _bigDecimalValuesSV[docId] = bigDecimalValues[docId];
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+      if (!unselectedDocs.isEmpty()) {
+        if (_elseStatement == null) {
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _bigDecimalValuesSV[docId] = (BigDecimal) DataSchema.ColumnDataType.BIG_DECIMAL.getNullPlaceholder();
+          }
         } else {
-          for (int j = 0; j < numDocs; j++) {
-            if (selected[j] == i) {
-              _bigDecimalValuesSV[j] = bigDecimalValues[j];
-            }
+          BigDecimal[] bigDecimalValues = _elseStatement.transformToBigDecimalValuesSV(valueBlock);
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _bigDecimalValuesSV[docId] = bigDecimalValues[docId];
           }
         }
       }
     }
     return _bigDecimalValuesSV;
   }
 
+  @Override
+  public Pair<BigDecimal[], RoaringBitmap> transformToBigDecimalValuesSVWithNull(ValueBlock valueBlock) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.BIG_DECIMAL) {
+      return super.transformToBigDecimalValuesSVWithNull(valueBlock);
+    }
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    initBigDecimalValuesSV(numDocs);
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        Pair<BigDecimal[], RoaringBitmap> bigDecimalValuesNullPair =
+            transformFunction.transformToBigDecimalValuesSVWithNull(valueBlock);
+        BigDecimal[] bigDecimals = bigDecimalValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = bigDecimalValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _bigDecimalValuesSV[docId] = bigDecimals[docId];
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _bigDecimalValuesSV[docId] = (BigDecimal) DataSchema.ColumnDataType.BIG_DECIMAL.getNullPlaceholder();
+          bitmap.add(docId);
+        }
+      } else {
+        Pair<BigDecimal[], RoaringBitmap> bigDecimalNullPair =
+            _elseStatement.transformToBigDecimalValuesSVWithNull(valueBlock);
+        BigDecimal[] bigDecimals = bigDecimalNullPair.getLeft();
+        RoaringBitmap nullBitmap = bigDecimalNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _bigDecimalValuesSV[docId] = bigDecimals[docId];
+          if (nullBitmap != null && nullBitmap.contains(docId)) {
+            bitmap.add(docId);
+          }
+        }
+      }
+    }
+    return ImmutablePair.of(_bigDecimalValuesSV, bitmap);
+  }
+
   @Override
   public String[] transformToStringValuesSV(ValueBlock valueBlock) {
     if (_resultMetadata.getDataType().getStoredType() != DataType.STRING) {
       return super.transformToStringValuesSV(valueBlock);
     }
-    int[] selected = getSelectedArray(valueBlock);
+    int[] selected = getSelectedArray(valueBlock, false);
     int numDocs = valueBlock.getNumDocs();
     initStringValuesSV(numDocs);
-    int numElseThenStatements = _elseThenStatements.size();
-    for (int i = 0; i < numElseThenStatements; i++) {
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    for (int i = 0; i < numThenStatements; i++) {
       if (_selections[i]) {
-        TransformFunction transformFunction = _elseThenStatements.get(i);
+        TransformFunction transformFunction = _thenStatements.get(i);
         String[] stringValues = transformFunction.transformToStringValuesSV(valueBlock);
-        if (_numSelections == 1) {
-          System.arraycopy(stringValues, 0, _stringValuesSV, 0, numDocs);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _stringValuesSV[docId] = stringValues[docId];
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+      if (!unselectedDocs.isEmpty()) {
+        if (_elseStatement == null) {
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _stringValuesSV[docId] = (String) DataSchema.ColumnDataType.STRING.getNullPlaceholder();
+          }
         } else {
-          for (int j = 0; j < numDocs; j++) {
-            if (selected[j] == i) {
-              _stringValuesSV[j] = stringValues[j];
-            }
+          String[] stringValues = _elseStatement.transformToStringValuesSV(valueBlock);
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _stringValuesSV[docId] = stringValues[docId];
           }
         }
       }
     }
     return _stringValuesSV;
   }
 
+  @Override
+  public Pair<String[], RoaringBitmap> transformToStringValuesSVWithNull(ValueBlock valueBlock) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.STRING) {
+      return super.transformToStringValuesSVWithNull(valueBlock);
+    }
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    initStringValuesSV(numDocs);
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        Pair<String[], RoaringBitmap> stringValuesNullPair =
+            transformFunction.transformToStringValuesSVWithNull(valueBlock);
+        String[] stringValues = stringValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = stringValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _stringValuesSV[docId] = stringValues[docId];
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _stringValuesSV[docId] = (String) DataSchema.ColumnDataType.STRING.getNullPlaceholder();
+          bitmap.add(docId);
+        }
+      } else {
+        Pair<String[], RoaringBitmap> stringNullPair = _elseStatement.transformToStringValuesSVWithNull(valueBlock);
+        String[] stringValues = stringNullPair.getLeft();
+        RoaringBitmap nullBitmap = stringNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _stringValuesSV[docId] = stringValues[docId];
+          if (nullBitmap != null && nullBitmap.contains(docId)) {
+            bitmap.add(docId);
+          }
+        }
+      }
+    }
+    return ImmutablePair.of(_stringValuesSV, bitmap);
+  }
+
   @Override
   public byte[][] transformToBytesValuesSV(ValueBlock valueBlock) {
     if (_resultMetadata.getDataType().getStoredType() != DataType.BYTES) {
       return super.transformToBytesValuesSV(valueBlock);
     }
-    int[] selected = getSelectedArray(valueBlock);
+    int[] selected = getSelectedArray(valueBlock, false);
     int numDocs = valueBlock.getNumDocs();
     initBytesValuesSV(numDocs);
-    int numElseThenStatements = _elseThenStatements.size();
-    for (int i = 0; i < numElseThenStatements; i++) {
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    for (int i = 0; i < numThenStatements; i++) {
       if (_selections[i]) {
-        TransformFunction transformFunction = _elseThenStatements.get(i);
-        byte[][] bytesValues = transformFunction.transformToBytesValuesSV(valueBlock);
-        if (_numSelections == 1) {
-          System.arraycopy(bytesValues, 0, _bytesValuesSV, 0, numDocs);
+        TransformFunction transformFunction = _thenStatements.get(i);
+        byte[][] byteValues = transformFunction.transformToBytesValuesSV(valueBlock);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _bytesValuesSV[docId] = byteValues[docId];
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+      if (!unselectedDocs.isEmpty()) {
+        if (_elseStatement == null) {
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _bytesValuesSV[docId] = (byte[]) DataSchema.ColumnDataType.BYTES.getNullPlaceholder();
+          }
         } else {
-          for (int j = 0; j < numDocs; j++) {
-            if (selected[j] == i) {
-              _bytesValuesSV[j] = bytesValues[j];
-            }
+          byte[][] bytesValues = _elseStatement.transformToBytesValuesSV(valueBlock);
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            _bytesValuesSV[docId] = bytesValues[docId];
           }
         }
       }
     }
     return _bytesValuesSV;
   }
+
+  @Override
+  public Pair<byte[][], RoaringBitmap> transformToBytesValuesSVWithNull(ValueBlock valueBlock) {
+    if (_resultMetadata.getDataType().getStoredType() != DataType.BYTES) {
+      return super.transformToBytesValuesSVWithNull(valueBlock);
+    }
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    initBytesValuesSV(numDocs);
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        Pair<byte[][], RoaringBitmap> byteValuesNullPair =
+            transformFunction.transformToBytesValuesSVWithNull(valueBlock);
+        byte[][] byteValues = byteValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = byteValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            _bytesValuesSV[docId] = byteValues[docId];
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+    }
+    if (!unselectedDocs.isEmpty()) {
+      if (_elseStatement == null) {
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _bytesValuesSV[docId] = (byte[]) DataSchema.ColumnDataType.BYTES.getNullPlaceholder();
+          bitmap.add(docId);
+        }
+      } else {
+        Pair<byte[][], RoaringBitmap> byteValuesNullPair = _elseStatement.transformToBytesValuesSVWithNull(valueBlock);
+        byte[][] byteValues = byteValuesNullPair.getLeft();
+        RoaringBitmap nullBitmap = byteValuesNullPair.getRight();
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          _bytesValuesSV[docId] = byteValues[docId];
+          if (nullBitmap != null && nullBitmap.contains(docId)) {
+            bitmap.add(docId);
+          }
+        }
+      }
+    }
+    return ImmutablePair.of(_bytesValuesSV, bitmap);
+  }
+
+  @Override
+  public RoaringBitmap getNullBitmap(ValueBlock valueBlock) {
+    int[] selected = getSelectedArray(valueBlock, true);
+    int numDocs = valueBlock.getNumDocs();
+    int numThenStatements = _thenStatements.size();
+    BitSet unselectedDocs = new BitSet();
+    unselectedDocs.set(0, numDocs);
+    final RoaringBitmap bitmap = new RoaringBitmap();
+    for (int i = 0; i < numThenStatements; i++) {
+      if (_selections[i]) {
+        TransformFunction transformFunction = _thenStatements.get(i);
+        RoaringBitmap nullBitmap = transformFunction.getNullBitmap(valueBlock);
+        for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+          if (selected[docId] == i) {
+            if (nullBitmap != null && nullBitmap.contains(docId)) {
+              bitmap.add(docId);
+            }
+            unselectedDocs.clear(docId);
+          }
+        }
+        if (unselectedDocs.isEmpty()) {
+          break;
+        }
+      }
+      if (!unselectedDocs.isEmpty()) {
+        if (_elseStatement == null) {
+          for (int docId = unselectedDocs.nextSetBit(0); docId >= 0; docId = unselectedDocs.nextSetBit(docId + 1)) {
+            bitmap.add(docId);

Review Comment:
   We can replace this with bitmap.or(unselectedDocs)?



-- 
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: commits-unsubscribe@pinot.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@pinot.apache.org
For additional commands, e-mail: commits-help@pinot.apache.org