private int getLastIdx() {
Set<Integer> indices = Sets.newHashSet();
for (Type type : footer.getTypesList()) {
indices.addAll(type.getSubtypesList());
}
return Collections.max(indices);
}
private long getRawDataSizeOfColumn(int colIdx) {
OrcProto.ColumnStatistics colStat = footer.getStatistics(colIdx);
long numVals = colStat.getNumberOfValues();
Type type = footer.getTypes(colIdx);
switch (type.getKind()) {
case BINARY:
// old orc format doesn't support binary statistics. checking for binary
// statistics is not required as protocol buffers takes care of it.
return colStat.getBinaryStatistics().getSum();
case STRING:
case CHAR:
case VARCHAR:
// old orc format doesn't support sum for string statistics. checking for
// existence is not required as protocol buffers takes care of it.
// ORC strings are deserialized to java strings. so use java data model's
// string size
numVals = numVals == 0 ? 1 : numVals;
int avgStrLen = (int) (colStat.getStringStatistics().getSum() / numVals);
return numVals * JavaDataModel.get().lengthForStringOfLength(avgStrLen);
case TIMESTAMP:
return numVals * JavaDataModel.get().lengthOfTimestamp();
case DATE:
return numVals * JavaDataModel.get().lengthOfDate();
case DECIMAL:
return numVals * JavaDataModel.get().lengthOfDecimal();
case DOUBLE:
case LONG:
return numVals * JavaDataModel.get().primitive2();
case FLOAT:
case INT:
case SHORT:
case BOOLEAN:
case BYTE:
return numVals * JavaDataModel.get().primitive1();
default:
LOG.debug("Unknown primitive category.");
break;
}
return 0;
}
private List<Integer> getColumnIndicesFromNames(List<String> colNames) {
// top level struct
Type type = footer.getTypesList().get(0);
List<Integer> colIndices = Lists.newArrayList();
List<String> fieldNames = type.getFieldNamesList();
int fieldIdx = 0;
for (String colName : colNames) {
if (fieldNames.contains(colName)) {
fieldIdx = fieldNames.indexOf(colName);
}
// a single field may span multiple columns. find start and end column
// index for the requested field
int idxStart = type.getSubtypes(fieldIdx);
int idxEnd;
// if the specified is the last field and then end index will be last
// column index
if (fieldIdx + 1 > fieldNames.size() - 1) {
idxEnd = getLastIdx() + 1;
} else {
idxEnd = type.getSubtypes(fieldIdx + 1);
}
// if start index and end index are same then the field is a primitive
// field else complex field (like map, list, struct, union)
if (idxStart == idxEnd) {
// simple field
colIndices.add(idxStart);
} else {
// complex fields spans multiple columns
for (int i = idxStart; i < idxEnd; i++) {
colIndices.add(i);
}
}
}
return colIndices;
}