static double readDouble(DecodeBuffer* input) {
union {
int64_t f;
double t;
} transfer;
transfer.f = readI64(input);
if (transfer.f == -1) {
return -1;
}
return transfer.t;
}
/**
* read column chunk information
*/
int
readColumnChunk(
CompactProtocol *prot,
struct ColumnChunkMetadata_4C *colChunk)
{
uint32_t xfer = 0;
TType ftype;
int16_t fid;
readStructBegin(prot);
bool isset_file_offset = false;
while (true) {
xfer += readFieldBegin(prot, &ftype, &fid);
if (ftype == T_STOP) {
break;
}
switch (fid) {
case 1:
if (ftype == T_STRING) {
char *file_path;
xfer += readString(prot, &file_path);
colChunk->path = file_path;
}
break;
case 2:
if (ftype == T_I64) {
xfer += readI64(prot, &(colChunk->file_offset));
isset_file_offset = true;
}
break;
case 3:
if (ftype == T_STRUCT) {
/*read column metadata*/
xfer += readColumnMetadata(prot, colChunk);
}
break;
default:
break;
}
}
readStructEnd(prot);
if (!isset_file_offset)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk fileoffset not set")));
return xfer;
}
/**
* read column chunk metadata information
*/
int
readColumnMetadata(
CompactProtocol *prot,
struct ColumnChunkMetadata_4C *colChunk)
{
uint32_t xfer = 0;
TType ftype;
int16_t fid;
readStructBegin(prot);
bool isset_type = false;
bool isset_encodings = false;
bool isset_path_in_schema = false;
bool isset_codec = false;
bool isset_num_values = false;
bool isset_total_uncompressed_size = false;
bool isset_total_compressed_size = false;
bool isset_data_page_offset = false;
while (true) {
xfer += readFieldBegin(prot, &ftype, &fid);
if (ftype == T_STOP) {
break;
}
switch (fid) {
case 1:
if (ftype == T_I32) {
int32_t type;
xfer += readI32(prot, &type);
colChunk->type = (PrimitiveTypeName) type;
isset_type = true;
}
break;
case 2:
if (ftype == T_LIST) {
uint32_t encodingCount;
TType etype;
xfer += readListBegin(prot, &etype, &encodingCount);
colChunk->EncodingCount = encodingCount;
colChunk->pEncodings =
(enum Encoding *) palloc0(sizeof(enum Encoding) * encodingCount);
for (int i = 0; i < encodingCount; i++) {
int32_t encoding;
xfer += readI32(prot, &encoding);
colChunk->pEncodings[i] = (enum Encoding) encoding;
}
isset_encodings = true;
}
break;
case 3:
if (ftype == T_LIST) {
{
/*process path in schema, setting colchunk->depth and colchunk->pathInSchema*/
TType etype;
uint32_t lsize;
StringInfoData colNameBuf;
xfer += readListBegin(prot, &etype, &lsize);
colChunk->depth = lsize;
initStringInfo(&colNameBuf);
char *path_in_schema;
for (int i = 0; i < lsize - 1; i++) {
xfer += readString(prot, &path_in_schema);
appendStringInfo(&colNameBuf, "%s:", path_in_schema);
pfree(path_in_schema);
}
xfer += readString(prot, &path_in_schema);
appendStringInfo(&colNameBuf, "%s", path_in_schema);
colChunk->pathInSchema = colNameBuf.data;
colChunk->colName = path_in_schema;
}
isset_path_in_schema = true;
}
break;
case 4:
if (ftype == T_I32) {
int32_t compresscode;
xfer += readI32(prot, &compresscode);
colChunk->codec = (enum CompressionCodecName) compresscode;
isset_codec = true;
}
break;
case 5:
if (ftype == T_I64) {
int64_t valCnt;
xfer += readI64(prot, &valCnt);
colChunk->valueCount = valCnt;
isset_num_values = true;
}
break;
case 6:
if (ftype == T_I64) {
xfer += readI64(prot, &(colChunk->totalUncompressedSize));
isset_total_uncompressed_size = true;
}
break;
case 7:
if (ftype == T_I64) {
xfer += readI64(prot, &(colChunk->totalSize));
isset_total_compressed_size = true;
}
break;
case 8:
if (ftype == T_LIST) {
xfer += skipType(prot, ftype);
}
break;
case 9:
if (ftype == T_I64) {
xfer += readI64(prot, &(colChunk->firstDataPage));
isset_data_page_offset = true;
}
break;
case 10:
if (ftype == T_I64) {
xfer += skipType(prot, ftype);
}
break;
case 11:
if (ftype == T_I64) {
xfer += skipType(prot, ftype);
}
break;
default:
break;
}
}
readStructEnd(prot);
if (!isset_type)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk type not set")));
if (!isset_encodings)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk encoding not set")));
if (!isset_path_in_schema)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk path_in_schema not set")));
if (!isset_codec)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk compression code not set")));
if (!isset_num_values)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk value number not set")));
if (!isset_total_uncompressed_size)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk total uncompressed size not set")));
if (!isset_total_compressed_size)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk total compressed size not set")));
if (!isset_data_page_offset)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata: row group column chunk first data page not set")));
return xfer;
}
/**
* Read parquet file rowgroup information, and convert it to hawq structure
*
* @prot the reading protocol
* @rowGroupInfo the row group information
* @pfields the schema information of the file
* @pfieldCount the field count of schema
*/
int
readRowGroupInfo(
CompactProtocol *prot,
struct BlockMetadata_4C* rowGroupInfo,
struct FileField_4C* pfields,
int pfieldCount)
{
uint32_t xfer = 0;
TType ftype;
int16_t fid;
readStructBegin(prot);
bool isset_columns = false;
bool isset_total_byte_size = false;
bool isset_num_rows = false;
while (true) {
xfer += readFieldBegin(prot, &ftype, &fid);
if (ftype == T_STOP) {
break;
}
switch (fid) {
case 1:
if (ftype == T_LIST)
{
uint32_t colChunkCnt;
TType etype;
xfer += readListBegin(prot, &etype, &colChunkCnt);
rowGroupInfo->ColChunkCount = colChunkCnt;
rowGroupInfo->columns =
(struct ColumnChunkMetadata_4C*) palloc0(
colChunkCnt * sizeof(struct ColumnChunkMetadata_4C));
for (int i = 0; i < colChunkCnt; i++) {
xfer += readColumnChunk(prot, &(rowGroupInfo->columns[i]));
}
isset_columns = true;
}
break;
case 2:
if (ftype == T_I64) {
int64_t val;
xfer += readI64(prot, &val);
rowGroupInfo->totalByteSize = val;
isset_total_byte_size = true;
}
break;
case 3:
if (ftype == T_I64) {
int64_t val;
xfer += readI64(prot, &val);
rowGroupInfo->rowCount = val;
isset_num_rows = true;
}
break;
default:
break;
}
}
readStructEnd(prot);
if (!isset_columns)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata row group column chunk not set")));
if (!isset_total_byte_size)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata row group total byte size not set")));
if (!isset_num_rows)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata row group row count not set")));
/*assign r and d of fields to column chunks*/
int columnIndex = 0;
assignRDFromFieldToColumnChunk(rowGroupInfo->columns, &columnIndex, pfields,
pfieldCount);
return xfer;
}
/*
* The initialize read method, read file metadata, but just read the first 4 parts,
* including version, schema information, number of rows, and rowgroup number, but
* doesn't read each rowgroup metadata, and keyvalue part. Read metadata of next
* rowgroup before reading the actual data.
*
* @parquetMetadata parquet metadata information
* @prot footer protocol for reading
*/
int
readParquetFileMetadata(
ParquetMetadata *parquetMetadata,
CompactProtocol *prot)
{
uint32_t xfer = 0;
TType ftype;
int16_t fid;
bool isset_version = false;
bool isset_schema = false;
bool isset_num_rows = false;
bool isset_row_groups = false;
while (true) {
xfer += readFieldBegin(prot, &ftype, &fid);
if (ftype == T_STOP) {
break;
}
switch (fid) {
case 1:
/* Process version*/
if (ftype == T_I32) {
xfer += readI32(prot, &((*parquetMetadata)->version));
isset_version = true;
}
break;
case 2:
/* process schema - field information*/
if (ftype == T_LIST) {
{
uint32_t lsize;
TType ltype;
xfer += readListBegin(prot, <ype, &lsize);
readSchemaElement(prot, lsize,
&((*parquetMetadata)->pfield),
&((*parquetMetadata)->fieldCount),
&((*parquetMetadata)->colCount),
&((*parquetMetadata)->schemaTreeNodeCount));
}
isset_schema = true;
}
break;
case 3:
/* process number of rows*/
if (ftype == T_I64) {
int64_t num_rows = 0;
xfer += readI64(prot, &num_rows);
(*parquetMetadata)->num_rows = num_rows;
isset_num_rows = true;
}
break;
case 4:
/* process row group information*/
if (ftype == T_LIST) {
/* get row group count*/
uint32_t lSize;
TType etype;
xfer += readListBegin(prot, &etype, &lSize);
(*parquetMetadata)->blockCount = lSize;
isset_row_groups = true;
break;
}
break;
default:
break;
}
/*hit row groups, break out the while statement*/
if(isset_row_groups)
break;
}
if (!isset_version)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata version not set")));
if (!isset_schema)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata schema not set")));
if (!isset_num_rows)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR), errmsg("file metadata num_rows not set")));
if (!isset_row_groups)
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR),
errmsg("file metadata row group information not set")));
return xfer;
}
// Returns a new reference.
static PyObject*
decode_val(DecodeBuffer* input, TType type, PyObject* typeargs, long string_limit, long container_limit) {
switch (type) {
case T_BOOL: {
int8_t v = readByte(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
switch (v) {
case 0:
Py_RETURN_FALSE;
case 1:
Py_RETURN_TRUE;
// Don't laugh. This is a potentially serious issue.
default:
PyErr_SetString(PyExc_TypeError, "boolean out of range");
return NULL;
}
break;
}
case T_I08: {
int8_t v = readByte(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
return PyInt_FromLong(v);
}
case T_I16: {
int16_t v = readI16(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
return PyInt_FromLong(v);
}
case T_I32: {
int32_t v = readI32(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
return PyInt_FromLong(v);
}
case T_I64: {
int64_t v = readI64(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
// TODO(dreiss): Find out if we can take this fastpath always when
// sizeof(long) == sizeof(long long).
if (CHECK_RANGE(v, LONG_MIN, LONG_MAX)) {
return PyInt_FromLong((long) v);
}
return PyLong_FromLongLong(v);
}
case T_DOUBLE: {
double v = readDouble(input);
if (v == -1.0 && PyErr_Occurred()) {
return false;
}
return PyFloat_FromDouble(v);
}
case T_STRING: {
Py_ssize_t len = readI32(input);
char* buf;
if (!readBytes(input, &buf, len)) {
return NULL;
}
if (!check_length_limit(len, string_limit)) {
return NULL;
}
if (is_utf8(typeargs))
return PyUnicode_DecodeUTF8(buf, len, 0);
else
return PyString_FromStringAndSize(buf, len);
}
case T_LIST:
case T_SET: {
SetListTypeArgs parsedargs;
int32_t len;
PyObject* ret = NULL;
int i;
bool use_tuple = false;
if (!parse_set_list_args(&parsedargs, typeargs)) {
return NULL;
}
if (!checkTypeByte(input, parsedargs.element_type)) {
return NULL;
}
len = readI32(input);
if (!check_length_limit(len, container_limit)) {
return NULL;
}
use_tuple = type == T_LIST && parsedargs.immutable;
ret = use_tuple ? PyTuple_New(len) : PyList_New(len);
if (!ret) {
return NULL;
}
for (i = 0; i < len; i++) {
PyObject* item = decode_val(input, parsedargs.element_type, parsedargs.typeargs, string_limit, container_limit);
if (!item) {
Py_DECREF(ret);
return NULL;
}
if (use_tuple) {
PyTuple_SET_ITEM(ret, i, item);
} else {
PyList_SET_ITEM(ret, i, item);
}
}
// TODO(dreiss): Consider biting the bullet and making two separate cases
// for list and set, avoiding this post facto conversion.
if (type == T_SET) {
PyObject* setret;
setret = parsedargs.immutable ? PyFrozenSet_New(ret) : PySet_New(ret);
Py_DECREF(ret);
return setret;
}
return ret;
}
case T_MAP: {
int32_t len;
int i;
MapTypeArgs parsedargs;
PyObject* ret = NULL;
if (!parse_map_args(&parsedargs, typeargs)) {
return NULL;
}
if (!checkTypeByte(input, parsedargs.ktag)) {
return NULL;
}
if (!checkTypeByte(input, parsedargs.vtag)) {
return NULL;
}
len = readI32(input);
if (!check_length_limit(len, container_limit)) {
return NULL;
}
ret = PyDict_New();
if (!ret) {
goto error;
}
for (i = 0; i < len; i++) {
PyObject* k = NULL;
PyObject* v = NULL;
k = decode_val(input, parsedargs.ktag, parsedargs.ktypeargs, string_limit, container_limit);
if (k == NULL) {
goto loop_error;
}
v = decode_val(input, parsedargs.vtag, parsedargs.vtypeargs, string_limit, container_limit);
if (v == NULL) {
goto loop_error;
}
if (PyDict_SetItem(ret, k, v) == -1) {
goto loop_error;
}
Py_DECREF(k);
Py_DECREF(v);
continue;
// Yuck! Destructors, anyone?
loop_error:
Py_XDECREF(k);
Py_XDECREF(v);
goto error;
}
if (parsedargs.immutable) {
PyObject* thrift = PyImport_ImportModule("thrift.Thrift");
PyObject* cls = NULL;
PyObject* arg = NULL;
if (!thrift) {
goto error;
}
cls = PyObject_GetAttrString(thrift, "TFrozenDict");
if (!cls) {
goto error;
}
arg = PyTuple_New(1);
PyTuple_SET_ITEM(arg, 0, ret);
return PyObject_CallObject(cls, arg);
}
return ret;
error:
Py_XDECREF(ret);
return NULL;
}
case T_STRUCT: {
StructTypeArgs parsedargs;
if (!parse_struct_args(&parsedargs, typeargs)) {
return NULL;
}
return decode_struct(input, Py_None, parsedargs.klass, parsedargs.spec, string_limit, container_limit);
}
case T_STOP:
case T_VOID:
case T_UTF16:
case T_UTF8:
case T_U64:
default:
PyErr_SetString(PyExc_TypeError, "Unexpected TType");
return NULL;
}
}
// Returns a new reference.
static PyObject*
decode_val(DecodeBuffer* input, TType type, PyObject* typeargs) {
switch (type) {
case T_BOOL: {
int8_t v = readByte(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
switch (v) {
case 0: Py_RETURN_FALSE;
case 1: Py_RETURN_TRUE;
// Don't laugh. This is a potentially serious issue.
default: PyErr_SetString(PyExc_TypeError, "boolean out of range"); return NULL;
}
break;
}
case T_I08: {
int8_t v = readByte(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
return PyInt_FromLong(v);
}
case T_I16: {
int16_t v = readI16(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
return PyInt_FromLong(v);
}
case T_I32: {
int32_t v = readI32(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
return PyInt_FromLong(v);
}
case T_I64: {
int64_t v = readI64(input);
if (INT_CONV_ERROR_OCCURRED(v)) {
return NULL;
}
// TODO(dreiss): Find out if we can take this fastpath always when
// sizeof(long) == sizeof(long long).
if (CHECK_RANGE(v, LONG_MIN, LONG_MAX)) {
return PyInt_FromLong((long) v);
}
return PyLong_FromLongLong(v);
}
case T_DOUBLE: {
double v = readDouble(input);
if (v == -1.0 && PyErr_Occurred()) {
return false;
}
return PyFloat_FromDouble(v);
}
case T_STRING: {
Py_ssize_t len = readI32(input);
char* buf;
if (!readBytes(input, &buf, len)) {
return NULL;
}
return PyString_FromStringAndSize(buf, len);
}
case T_LIST:
case T_SET: {
SetListTypeArgs parsedargs;
int32_t len;
PyObject* ret = NULL;
int i;
if (!parse_set_list_args(&parsedargs, typeargs)) {
return NULL;
}
if (!checkTypeByte(input, parsedargs.element_type)) {
return NULL;
}
len = readI32(input);
if (!check_ssize_t_32(len)) {
return NULL;
}
ret = PyList_New(len);
if (!ret) {
return NULL;
}
for (i = 0; i < len; i++) {
PyObject* item = decode_val(input, parsedargs.element_type, parsedargs.typeargs);
if (!item) {
Py_DECREF(ret);
return NULL;
}
PyList_SET_ITEM(ret, i, item);
}
// TODO(dreiss): Consider biting the bullet and making two separate cases
// for list and set, avoiding this post facto conversion.
if (type == T_SET) {
PyObject* setret;
#if (PY_VERSION_HEX < 0x02050000)
// hack needed for older versions
setret = PyObject_CallFunctionObjArgs((PyObject*)&PySet_Type, ret, NULL);
#else
// official version
setret = PySet_New(ret);
#endif
Py_DECREF(ret);
return setret;
}
return ret;
}
case T_MAP: {
int32_t len;
int i;
MapTypeArgs parsedargs;
PyObject* ret = NULL;
if (!parse_map_args(&parsedargs, typeargs)) {
return NULL;
}
if (!checkTypeByte(input, parsedargs.ktag)) {
return NULL;
}
if (!checkTypeByte(input, parsedargs.vtag)) {
return NULL;
}
len = readI32(input);
if (!check_ssize_t_32(len)) {
return false;
}
ret = PyDict_New();
if (!ret) {
goto error;
}
for (i = 0; i < len; i++) {
PyObject* k = NULL;
PyObject* v = NULL;
k = decode_val(input, parsedargs.ktag, parsedargs.ktypeargs);
if (k == NULL) {
goto loop_error;
}
v = decode_val(input, parsedargs.vtag, parsedargs.vtypeargs);
if (v == NULL) {
goto loop_error;
}
if (PyDict_SetItem(ret, k, v) == -1) {
goto loop_error;
}
Py_DECREF(k);
Py_DECREF(v);
continue;
// Yuck! Destructors, anyone?
loop_error:
Py_XDECREF(k);
Py_XDECREF(v);
goto error;
}
return ret;
error:
Py_XDECREF(ret);
return NULL;
}
case T_STRUCT: {
StructTypeArgs parsedargs;
if (!parse_struct_args(&parsedargs, typeargs)) {
return NULL;
}
PyObject* ret = PyObject_CallObject(parsedargs.klass, NULL);
if (!ret) {
return NULL;
}
if (!decode_struct(input, ret, parsedargs.spec)) {
Py_DECREF(ret);
return NULL;
}
return ret;
}
case T_STOP:
case T_VOID:
case T_UTF16:
case T_UTF8:
case T_U64:
default:
PyErr_SetString(PyExc_TypeError, "Unexpected TType");
return NULL;
}
}
