Example #1
0
 private Date convertToDate(
     PrimitiveCategory inputType, Converter converter, DeferredObject argument)
     throws HiveException {
   assert (converter != null);
   assert (argument != null);
   if (argument.get() == null) {
     return null;
   }
   Date date = new Date();
   switch (inputType) {
     case STRING:
     case VARCHAR:
     case CHAR:
       String dateString = converter.convert(argument.get()).toString();
       try {
         date = formatter.parse(dateString);
       } catch (ParseException e) {
         return null;
       }
       break;
     case TIMESTAMP:
       Timestamp ts = ((TimestampWritable) converter.convert(argument.get())).getTimestamp();
       date.setTime(ts.getTime());
       break;
     case DATE:
       DateWritable dw = (DateWritable) converter.convert(argument.get());
       date = dw.get();
       break;
     default:
       throw new UDFArgumentException(
           "TO_DATE() only takes STRING/TIMESTAMP/DATEWRITABLE types, got " + inputType);
   }
   return date;
 }
 public Object set(Object o, DateWritable d) {
   if (d == null) {
     return null;
   }
   ((Date) o).setTime(d.get().getTime());
   return o;
 }
 public DateScalarSubtractDateColumn(long value, int colNum, int outputColumn) {
   this.colNum = colNum;
   this.value = new Timestamp(0);
   this.value.setTime(DateWritable.daysToMillis((int) value));
   this.outputColumn = outputColumn;
   scratchTimestamp2 = new Timestamp(0);
 }
Example #4
0
  public IntWritable evaluate(DateWritable d) {
    if (d == null) {
      return null;
    }

    calendar.setTime(d.get());
    result.set(1 + calendar.get(Calendar.MONTH));
    return result;
  }
 protected void evaluateString(ColumnVector columnVector, LongColumnVector output, int i) {
   BytesColumnVector bcv = (BytesColumnVector) columnVector;
   text.set(bcv.vector[i], bcv.start[i], bcv.length[i]);
   try {
     date.setTime(formatter.parse(text.toString()).getTime());
     output.vector[i] = baseDate - DateWritable.dateToDays(date);
   } catch (ParseException e) {
     output.vector[i] = 1;
     output.isNull[i] = true;
   }
 }
  @Test
  public void testDatePlusIntervalYearMonth() throws Exception {
    GenericUDFOPPlus udf = new GenericUDFOPPlus();

    DateWritable left = new DateWritable(Date.valueOf("2001-06-15"));
    HiveIntervalYearMonthWritable right =
        new HiveIntervalYearMonthWritable(HiveIntervalYearMonth.valueOf("2-8"));
    ObjectInspector[] inputOIs = {
      PrimitiveObjectInspectorFactory.writableDateObjectInspector,
      PrimitiveObjectInspectorFactory.writableHiveIntervalYearMonthObjectInspector
    };
    DeferredObject[] args = {
      new DeferredJavaObject(left), new DeferredJavaObject(right),
    };

    PrimitiveObjectInspector oi = (PrimitiveObjectInspector) udf.initialize(inputOIs);
    Assert.assertEquals(TypeInfoFactory.dateTypeInfo, oi.getTypeInfo());
    DateWritable res = (DateWritable) udf.evaluate(args);
    Assert.assertEquals(Date.valueOf("2004-02-15"), res.get());
  }
  @Override
  public void evaluate(VectorizedRowBatch batch) {

    if (childExpressions != null) {
      super.evaluateChildren(batch);
    }

    LongColumnVector outV = (LongColumnVector) batch.cols[outputColumn];
    ColumnVector inputCol = batch.cols[this.colNum];
    /* every line below this is identical for evaluateLong & evaluateString */
    final int n = inputCol.isRepeating ? 1 : batch.size;
    int[] sel = batch.selected;

    if (batch.size == 0) {
      /* n != batch.size when isRepeating */
      return;
    }

    /* true for all algebraic UDFs with no state */
    outV.isRepeating = inputCol.isRepeating;

    switch (inputTypes[0]) {
      case DATE:
        baseDate = (int) longValue;
        break;

      case TIMESTAMP:
        date.setTime(longValue / 1000000);
        baseDate = DateWritable.dateToDays(date);
        break;

      case STRING:
        try {
          date.setTime(formatter.parse(new String(stringValue, "UTF-8")).getTime());
          baseDate = DateWritable.dateToDays(date);
          break;
        } catch (Exception e) {
          outV.noNulls = false;
          if (batch.selectedInUse) {
            for (int j = 0; j < n; j++) {
              int i = sel[j];
              outV.isNull[i] = true;
            }
          } else {
            for (int i = 0; i < n; i++) {
              outV.isNull[i] = true;
            }
          }
          return;
        }
    }

    switch (inputTypes[1]) {
      case DATE:
        if (inputCol.noNulls) {
          outV.noNulls = true;
          if (batch.selectedInUse) {
            for (int j = 0; j < n; j++) {
              int i = sel[j];
              outV.vector[i] = evaluateDate(inputCol, i);
            }
          } else {
            for (int i = 0; i < n; i++) {
              outV.vector[i] = evaluateDate(inputCol, i);
            }
          }
        } else {
          // Handle case with nulls. Don't do function if the value is null, to save time,
          // because calling the function can be expensive.
          outV.noNulls = false;
          if (batch.selectedInUse) {
            for (int j = 0; j < n; j++) {
              int i = sel[j];
              outV.isNull[i] = inputCol.isNull[i];
              if (!inputCol.isNull[i]) {
                outV.vector[i] = evaluateDate(inputCol, i);
              }
            }
          } else {
            for (int i = 0; i < n; i++) {
              outV.isNull[i] = inputCol.isNull[i];
              if (!inputCol.isNull[i]) {
                outV.vector[i] = evaluateDate(inputCol, i);
              }
            }
          }
        }
        break;

      case TIMESTAMP:
        if (inputCol.noNulls) {
          outV.noNulls = true;
          if (batch.selectedInUse) {
            for (int j = 0; j < n; j++) {
              int i = sel[j];
              outV.vector[i] = evaluateTimestamp(inputCol, i);
            }
          } else {
            for (int i = 0; i < n; i++) {
              outV.vector[i] = evaluateTimestamp(inputCol, i);
            }
          }
        } else {
          // Handle case with nulls. Don't do function if the value is null, to save time,
          // because calling the function can be expensive.
          outV.noNulls = false;
          if (batch.selectedInUse) {
            for (int j = 0; j < n; j++) {
              int i = sel[j];
              outV.isNull[i] = inputCol.isNull[i];
              if (!inputCol.isNull[i]) {
                outV.vector[i] = evaluateTimestamp(inputCol, i);
              }
            }
          } else {
            for (int i = 0; i < n; i++) {
              outV.isNull[i] = inputCol.isNull[i];
              if (!inputCol.isNull[i]) {
                outV.vector[i] = evaluateTimestamp(inputCol, i);
              }
            }
          }
        }
        break;

      case STRING:
        if (inputCol.noNulls) {
          outV.noNulls = true;
          if (batch.selectedInUse) {
            for (int j = 0; j < n; j++) {
              int i = sel[j];
              evaluateString(inputCol, outV, i);
            }
          } else {
            for (int i = 0; i < n; i++) {
              evaluateString(inputCol, outV, i);
            }
          }
        } else {
          // Handle case with nulls. Don't do function if the value is null, to save time,
          // because calling the function can be expensive.
          outV.noNulls = false;
          if (batch.selectedInUse) {
            for (int j = 0; j < n; j++) {
              int i = sel[j];
              outV.isNull[i] = inputCol.isNull[i];
              if (!inputCol.isNull[i]) {
                evaluateString(inputCol, outV, i);
              }
            }
          } else {
            for (int i = 0; i < n; i++) {
              outV.isNull[i] = inputCol.isNull[i];
              if (!inputCol.isNull[i]) {
                evaluateString(inputCol, outV, i);
              }
            }
          }
        }
        break;
    }
  }
 protected int evaluateTimestamp(ColumnVector columnVector, int index) {
   LongColumnVector lcv = (LongColumnVector) columnVector;
   date.setTime(lcv.vector[index] / 1000000);
   return baseDate - DateWritable.dateToDays(date);
 }
 /* copied over from VectorUDFTimestampFieldLong */
 private TimestampWritable toTimestampWritable(long daysSinceEpoch) {
   Timestamp ts = new Timestamp(DateWritable.daysToMillis((int) daysSinceEpoch));
   return new TimestampWritable(ts);
 }
  /** Compare two objects with their respective ObjectInspectors. */
  public static int compare(
      Object o1,
      ObjectInspector oi1,
      Object o2,
      ObjectInspector oi2,
      MapEqualComparer mapEqualComparer) {
    if (oi1.getCategory() != oi2.getCategory()) {
      return oi1.getCategory().compareTo(oi2.getCategory());
    }

    if (o1 == null) {
      return o2 == null ? 0 : -1;
    } else if (o2 == null) {
      return 1;
    }

    switch (oi1.getCategory()) {
      case PRIMITIVE:
        {
          PrimitiveObjectInspector poi1 = ((PrimitiveObjectInspector) oi1);
          PrimitiveObjectInspector poi2 = ((PrimitiveObjectInspector) oi2);
          if (poi1.getPrimitiveCategory() != poi2.getPrimitiveCategory()) {
            return poi1.getPrimitiveCategory().compareTo(poi2.getPrimitiveCategory());
          }
          switch (poi1.getPrimitiveCategory()) {
            case VOID:
              return 0;
            case BOOLEAN:
              {
                int v1 = ((BooleanObjectInspector) poi1).get(o1) ? 1 : 0;
                int v2 = ((BooleanObjectInspector) poi2).get(o2) ? 1 : 0;
                return v1 - v2;
              }
            case BYTE:
              {
                int v1 = ((ByteObjectInspector) poi1).get(o1);
                int v2 = ((ByteObjectInspector) poi2).get(o2);
                return v1 - v2;
              }
            case SHORT:
              {
                int v1 = ((ShortObjectInspector) poi1).get(o1);
                int v2 = ((ShortObjectInspector) poi2).get(o2);
                return v1 - v2;
              }
            case INT:
              {
                int v1 = ((IntObjectInspector) poi1).get(o1);
                int v2 = ((IntObjectInspector) poi2).get(o2);
                return v1 > v2 ? 1 : (v1 < v2 ? -1 : 0);
              }
            case LONG:
              {
                long v1 = ((LongObjectInspector) poi1).get(o1);
                long v2 = ((LongObjectInspector) poi2).get(o2);
                return v1 > v2 ? 1 : (v1 < v2 ? -1 : 0);
              }
            case FLOAT:
              {
                float v1 = ((FloatObjectInspector) poi1).get(o1);
                float v2 = ((FloatObjectInspector) poi2).get(o2);
                return Float.compare(v1, v2);
              }
            case DOUBLE:
              {
                double v1 = ((DoubleObjectInspector) poi1).get(o1);
                double v2 = ((DoubleObjectInspector) poi2).get(o2);
                return Double.compare(v1, v2);
              }
            case STRING:
              {
                if (poi1.preferWritable() || poi2.preferWritable()) {
                  Text t1 = (Text) poi1.getPrimitiveWritableObject(o1);
                  Text t2 = (Text) poi2.getPrimitiveWritableObject(o2);
                  return t1 == null ? (t2 == null ? 0 : -1) : (t2 == null ? 1 : t1.compareTo(t2));
                } else {
                  String s1 = (String) poi1.getPrimitiveJavaObject(o1);
                  String s2 = (String) poi2.getPrimitiveJavaObject(o2);
                  return s1 == null ? (s2 == null ? 0 : -1) : (s2 == null ? 1 : s1.compareTo(s2));
                }
              }
            case CHAR:
              {
                HiveCharWritable t1 =
                    ((HiveCharObjectInspector) poi1).getPrimitiveWritableObject(o1);
                HiveCharWritable t2 =
                    ((HiveCharObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return t1.compareTo(t2);
              }
            case VARCHAR:
              {
                HiveVarcharWritable t1 =
                    ((HiveVarcharObjectInspector) poi1).getPrimitiveWritableObject(o1);
                HiveVarcharWritable t2 =
                    ((HiveVarcharObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return t1.compareTo(t2);
              }
            case BINARY:
              {
                BytesWritable bw1 = ((BinaryObjectInspector) poi1).getPrimitiveWritableObject(o1);
                BytesWritable bw2 = ((BinaryObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return bw1.compareTo(bw2);
              }

            case DATE:
              {
                DateWritable d1 = ((DateObjectInspector) poi1).getPrimitiveWritableObject(o1);
                DateWritable d2 = ((DateObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return d1.compareTo(d2);
              }
            case TIMESTAMP:
              {
                TimestampWritable t1 =
                    ((TimestampObjectInspector) poi1).getPrimitiveWritableObject(o1);
                TimestampWritable t2 =
                    ((TimestampObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return t1.compareTo(t2);
              }
            case INTERVAL_YEAR_MONTH:
              {
                HiveIntervalYearMonthWritable i1 =
                    ((HiveIntervalYearMonthObjectInspector) poi1).getPrimitiveWritableObject(o1);
                HiveIntervalYearMonthWritable i2 =
                    ((HiveIntervalYearMonthObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return i1.compareTo(i2);
              }
            case INTERVAL_DAY_TIME:
              {
                HiveIntervalDayTimeWritable i1 =
                    ((HiveIntervalDayTimeObjectInspector) poi1).getPrimitiveWritableObject(o1);
                HiveIntervalDayTimeWritable i2 =
                    ((HiveIntervalDayTimeObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return i1.compareTo(i2);
              }
            case DECIMAL:
              {
                HiveDecimalWritable t1 =
                    ((HiveDecimalObjectInspector) poi1).getPrimitiveWritableObject(o1);
                HiveDecimalWritable t2 =
                    ((HiveDecimalObjectInspector) poi2).getPrimitiveWritableObject(o2);
                return t1.compareTo(t2);
              }
            default:
              {
                throw new RuntimeException("Unknown type: " + poi1.getPrimitiveCategory());
              }
          }
        }
      case STRUCT:
        {
          StructObjectInspector soi1 = (StructObjectInspector) oi1;
          StructObjectInspector soi2 = (StructObjectInspector) oi2;
          List<? extends StructField> fields1 = soi1.getAllStructFieldRefs();
          List<? extends StructField> fields2 = soi2.getAllStructFieldRefs();
          int minimum = Math.min(fields1.size(), fields2.size());
          for (int i = 0; i < minimum; i++) {
            int r =
                compare(
                    soi1.getStructFieldData(o1, fields1.get(i)),
                    fields1.get(i).getFieldObjectInspector(),
                    soi2.getStructFieldData(o2, fields2.get(i)),
                    fields2.get(i).getFieldObjectInspector(),
                    mapEqualComparer);
            if (r != 0) {
              return r;
            }
          }
          return fields1.size() - fields2.size();
        }
      case LIST:
        {
          ListObjectInspector loi1 = (ListObjectInspector) oi1;
          ListObjectInspector loi2 = (ListObjectInspector) oi2;
          int minimum = Math.min(loi1.getListLength(o1), loi2.getListLength(o2));
          for (int i = 0; i < minimum; i++) {
            int r =
                compare(
                    loi1.getListElement(o1, i),
                    loi1.getListElementObjectInspector(),
                    loi2.getListElement(o2, i),
                    loi2.getListElementObjectInspector(),
                    mapEqualComparer);
            if (r != 0) {
              return r;
            }
          }
          return loi1.getListLength(o1) - loi2.getListLength(o2);
        }
      case MAP:
        {
          if (mapEqualComparer == null) {
            throw new RuntimeException("Compare on map type not supported!");
          } else {
            return mapEqualComparer.compare(
                o1, (MapObjectInspector) oi1, o2, (MapObjectInspector) oi2);
          }
        }
      case UNION:
        {
          UnionObjectInspector uoi1 = (UnionObjectInspector) oi1;
          UnionObjectInspector uoi2 = (UnionObjectInspector) oi2;
          byte tag1 = uoi1.getTag(o1);
          byte tag2 = uoi2.getTag(o2);
          if (tag1 != tag2) {
            return tag1 - tag2;
          }
          return compare(
              uoi1.getField(o1),
              uoi1.getObjectInspectors().get(tag1),
              uoi2.getField(o2),
              uoi2.getObjectInspectors().get(tag2),
              mapEqualComparer);
        }
      default:
        throw new RuntimeException("Compare on unknown type: " + oi1.getCategory());
    }
  }
  @Override
  /**
   * Method to evaluate scalar-column operation in vectorized fashion.
   *
   * @batch a package of rows with each column stored in a vector
   */
  public void evaluate(VectorizedRowBatch batch) {

    if (childExpressions != null) {
      super.evaluateChildren(batch);
    }

    // Input #2 is type date (epochDays).
    LongColumnVector inputColVector2 = (LongColumnVector) batch.cols[colNum];

    // Output is type HiveIntervalDayTime.
    IntervalDayTimeColumnVector outputColVector =
        (IntervalDayTimeColumnVector) batch.cols[outputColumn];

    int[] sel = batch.selected;
    boolean[] inputIsNull = inputColVector2.isNull;
    boolean[] outputIsNull = outputColVector.isNull;
    outputColVector.noNulls = inputColVector2.noNulls;
    outputColVector.isRepeating = inputColVector2.isRepeating;
    int n = batch.size;

    long[] vector2 = inputColVector2.vector;

    // return immediately if batch is empty
    if (n == 0) {
      return;
    }

    if (inputColVector2.isRepeating) {
      scratchTimestamp2.setTime(DateWritable.daysToMillis((int) vector2[0]));
      dtm.subtract(value, scratchTimestamp2, outputColVector.getScratchIntervalDayTime());
      outputColVector.setFromScratchIntervalDayTime(0);
      // Even if there are no nulls, we always copy over entry 0. Simplifies code.
      outputIsNull[0] = inputIsNull[0];
    } else if (inputColVector2.noNulls) {
      if (batch.selectedInUse) {
        for (int j = 0; j != n; j++) {
          int i = sel[j];
          scratchTimestamp2.setTime(DateWritable.daysToMillis((int) vector2[i]));
          dtm.subtract(value, scratchTimestamp2, outputColVector.getScratchIntervalDayTime());
          outputColVector.setFromScratchIntervalDayTime(i);
        }
      } else {
        for (int i = 0; i != n; i++) {
          scratchTimestamp2.setTime(DateWritable.daysToMillis((int) vector2[i]));
          dtm.subtract(value, scratchTimestamp2, outputColVector.getScratchIntervalDayTime());
          outputColVector.setFromScratchIntervalDayTime(i);
        }
      }
    } else {
        /* there are nulls */
      if (batch.selectedInUse) {
        for (int j = 0; j != n; j++) {
          int i = sel[j];
          scratchTimestamp2.setTime(DateWritable.daysToMillis((int) vector2[i]));
          dtm.subtract(value, scratchTimestamp2, outputColVector.getScratchIntervalDayTime());
          outputColVector.setFromScratchIntervalDayTime(i);
          outputIsNull[i] = inputIsNull[i];
        }
      } else {
        for (int i = 0; i != n; i++) {
          scratchTimestamp2.setTime(DateWritable.daysToMillis((int) vector2[i]));
          dtm.subtract(value, scratchTimestamp2, outputColVector.getScratchIntervalDayTime());
          outputColVector.setFromScratchIntervalDayTime(i);
        }
        System.arraycopy(inputIsNull, 0, outputIsNull, 0, n);
      }
    }

    NullUtil.setNullOutputEntriesColScalar(outputColVector, batch.selectedInUse, sel, n);
  }
  static Object deserialize(InputByteBuffer buffer, TypeInfo type, boolean invert, Object reuse)
      throws IOException {

    // Is this field a null?
    byte isNull = buffer.read(invert);
    if (isNull == 0) {
      return null;
    }
    assert (isNull == 1);

    switch (type.getCategory()) {
      case PRIMITIVE:
        {
          PrimitiveTypeInfo ptype = (PrimitiveTypeInfo) type;
          switch (ptype.getPrimitiveCategory()) {
            case VOID:
              {
                return null;
              }
            case BOOLEAN:
              {
                BooleanWritable r = reuse == null ? new BooleanWritable() : (BooleanWritable) reuse;
                byte b = buffer.read(invert);
                assert (b == 1 || b == 2);
                r.set(b == 2);
                return r;
              }
            case BYTE:
              {
                ByteWritable r = reuse == null ? new ByteWritable() : (ByteWritable) reuse;
                r.set((byte) (buffer.read(invert) ^ 0x80));
                return r;
              }
            case SHORT:
              {
                ShortWritable r = reuse == null ? new ShortWritable() : (ShortWritable) reuse;
                int v = buffer.read(invert) ^ 0x80;
                v = (v << 8) + (buffer.read(invert) & 0xff);
                r.set((short) v);
                return r;
              }
            case INT:
              {
                IntWritable r = reuse == null ? new IntWritable() : (IntWritable) reuse;
                int v = buffer.read(invert) ^ 0x80;
                for (int i = 0; i < 3; i++) {
                  v = (v << 8) + (buffer.read(invert) & 0xff);
                }
                r.set(v);
                return r;
              }
            case LONG:
              {
                LongWritable r = reuse == null ? new LongWritable() : (LongWritable) reuse;
                long v = buffer.read(invert) ^ 0x80;
                for (int i = 0; i < 7; i++) {
                  v = (v << 8) + (buffer.read(invert) & 0xff);
                }
                r.set(v);
                return r;
              }
            case FLOAT:
              {
                FloatWritable r = reuse == null ? new FloatWritable() : (FloatWritable) reuse;
                int v = 0;
                for (int i = 0; i < 4; i++) {
                  v = (v << 8) + (buffer.read(invert) & 0xff);
                }
                if ((v & (1 << 31)) == 0) {
                  // negative number, flip all bits
                  v = ~v;
                } else {
                  // positive number, flip the first bit
                  v = v ^ (1 << 31);
                }
                r.set(Float.intBitsToFloat(v));
                return r;
              }
            case DOUBLE:
              {
                DoubleWritable r = reuse == null ? new DoubleWritable() : (DoubleWritable) reuse;
                long v = 0;
                for (int i = 0; i < 8; i++) {
                  v = (v << 8) + (buffer.read(invert) & 0xff);
                }
                if ((v & (1L << 63)) == 0) {
                  // negative number, flip all bits
                  v = ~v;
                } else {
                  // positive number, flip the first bit
                  v = v ^ (1L << 63);
                }
                r.set(Double.longBitsToDouble(v));
                return r;
              }
            case STRING:
              {
                Text r = reuse == null ? new Text() : (Text) reuse;
                // Get the actual length first
                int start = buffer.tell();
                int length = 0;
                do {
                  byte b = buffer.read(invert);
                  if (b == 0) {
                    // end of string
                    break;
                  }
                  if (b == 1) {
                    // the last char is an escape char. read the actual char
                    buffer.read(invert);
                  }
                  length++;
                } while (true);

                if (length == buffer.tell() - start) {
                  // No escaping happened, so we are already done.
                  r.set(buffer.getData(), start, length);
                } else {
                  // Escaping happened, we need to copy byte-by-byte.
                  // 1. Set the length first.
                  r.set(buffer.getData(), start, length);
                  // 2. Reset the pointer.
                  buffer.seek(start);
                  // 3. Copy the data.
                  byte[] rdata = r.getBytes();
                  for (int i = 0; i < length; i++) {
                    byte b = buffer.read(invert);
                    if (b == 1) {
                      // The last char is an escape char, read the actual char.
                      // The serialization format escape \0 to \1, and \1 to \2,
                      // to make sure the string is null-terminated.
                      b = (byte) (buffer.read(invert) - 1);
                    }
                    rdata[i] = b;
                  }
                  // 4. Read the null terminator.
                  byte b = buffer.read(invert);
                  assert (b == 0);
                }
                return r;
              }

            case BINARY:
              {
                BytesWritable bw = new BytesWritable();
                // Get the actual length first
                int start = buffer.tell();
                int length = 0;
                do {
                  byte b = buffer.read(invert);
                  if (b == 0) {
                    // end of string
                    break;
                  }
                  if (b == 1) {
                    // the last char is an escape char. read the actual char
                    buffer.read(invert);
                  }
                  length++;
                } while (true);

                if (length == buffer.tell() - start) {
                  // No escaping happened, so we are already done.
                  bw.set(buffer.getData(), start, length);
                } else {
                  // Escaping happened, we need to copy byte-by-byte.
                  // 1. Set the length first.
                  bw.set(buffer.getData(), start, length);
                  // 2. Reset the pointer.
                  buffer.seek(start);
                  // 3. Copy the data.
                  byte[] rdata = bw.getBytes();
                  for (int i = 0; i < length; i++) {
                    byte b = buffer.read(invert);
                    if (b == 1) {
                      // The last char is an escape char, read the actual char.
                      // The serialization format escape \0 to \1, and \1 to \2,
                      // to make sure the string is null-terminated.
                      b = (byte) (buffer.read(invert) - 1);
                    }
                    rdata[i] = b;
                  }
                  // 4. Read the null terminator.
                  byte b = buffer.read(invert);
                  assert (b == 0);
                }
                return bw;
              }

            case DATE:
              {
                DateWritable d = reuse == null ? new DateWritable() : (DateWritable) reuse;
                long v = buffer.read(invert) ^ 0x80;
                for (int i = 0; i < 7; i++) {
                  v = (v << 8) + (buffer.read(invert) & 0xff);
                }
                d.set(DateWritable.timeToDate(v));
                return d;
              }

            case TIMESTAMP:
              TimestampWritable t =
                  (reuse == null ? new TimestampWritable() : (TimestampWritable) reuse);
              byte[] bytes = new byte[8];

              for (int i = 0; i < bytes.length; i++) {
                bytes[i] = buffer.read(invert);
              }
              t.setBinarySortable(bytes, 0);
              return t;

            default:
              {
                throw new RuntimeException("Unrecognized type: " + ptype.getPrimitiveCategory());
              }
          }
        }
      case LIST:
        {
          ListTypeInfo ltype = (ListTypeInfo) type;
          TypeInfo etype = ltype.getListElementTypeInfo();

          // Create the list if needed
          ArrayList<Object> r = reuse == null ? new ArrayList<Object>() : (ArrayList<Object>) reuse;

          // Read the list
          int size = 0;
          while (true) {
            int more = buffer.read(invert);
            if (more == 0) {
              // \0 to terminate
              break;
            }
            // \1 followed by each element
            assert (more == 1);
            if (size == r.size()) {
              r.add(null);
            }
            r.set(size, deserialize(buffer, etype, invert, r.get(size)));
            size++;
          }
          // Remove additional elements if the list is reused
          while (r.size() > size) {
            r.remove(r.size() - 1);
          }
          return r;
        }
      case MAP:
        {
          MapTypeInfo mtype = (MapTypeInfo) type;
          TypeInfo ktype = mtype.getMapKeyTypeInfo();
          TypeInfo vtype = mtype.getMapValueTypeInfo();

          // Create the map if needed
          Map<Object, Object> r;
          if (reuse == null) {
            r = new HashMap<Object, Object>();
          } else {
            r = (HashMap<Object, Object>) reuse;
            r.clear();
          }

          while (true) {
            int more = buffer.read(invert);
            if (more == 0) {
              // \0 to terminate
              break;
            }
            // \1 followed by each key and then each value
            assert (more == 1);
            Object k = deserialize(buffer, ktype, invert, null);
            Object v = deserialize(buffer, vtype, invert, null);
            r.put(k, v);
          }
          return r;
        }
      case STRUCT:
        {
          StructTypeInfo stype = (StructTypeInfo) type;
          List<TypeInfo> fieldTypes = stype.getAllStructFieldTypeInfos();
          int size = fieldTypes.size();
          // Create the struct if needed
          ArrayList<Object> r =
              reuse == null ? new ArrayList<Object>(size) : (ArrayList<Object>) reuse;
          assert (r.size() <= size);
          // Set the size of the struct
          while (r.size() < size) {
            r.add(null);
          }
          // Read one field by one field
          for (int eid = 0; eid < size; eid++) {
            r.set(eid, deserialize(buffer, fieldTypes.get(eid), invert, r.get(eid)));
          }
          return r;
        }
      case UNION:
        {
          UnionTypeInfo utype = (UnionTypeInfo) type;
          StandardUnion r = reuse == null ? new StandardUnion() : (StandardUnion) reuse;
          // Read the tag
          byte tag = buffer.read(invert);
          r.setTag(tag);
          r.setObject(
              deserialize(buffer, utype.getAllUnionObjectTypeInfos().get(tag), invert, null));
          return r;
        }
      default:
        {
          throw new RuntimeException("Unrecognized type: " + type.getCategory());
        }
    }
  }