github-actions[bot] commented on code in PR #63192:
URL: https://github.com/apache/doris/pull/63192#discussion_r3252833924
##########
be/src/format/parquet/vparquet_column_reader.cpp:
##########
@@ -103,6 +123,1069 @@ static void fill_array_offset(FieldSchema* field,
ColumnArray::Offsets64& offset
}
}
+static constexpr int64_t UNIX_EPOCH_DAYNR = 719528;
+static constexpr int64_t MICROS_PER_SECOND = 1000000;
+
+static int64_t variant_date_value(const VecDateTimeValue& value) {
+ return value.daynr() - UNIX_EPOCH_DAYNR;
+}
+
+static int64_t variant_date_value(const DateV2Value<DateV2ValueType>& value) {
+ return value.daynr() - UNIX_EPOCH_DAYNR;
+}
+
+static int64_t variant_datetime_value(const VecDateTimeValue& value) {
+ int64_t timestamp = 0;
+ value.unix_timestamp(×tamp, cctz::utc_time_zone());
+ return timestamp * MICROS_PER_SECOND;
+}
+
+static int64_t variant_datetime_value(const DateV2Value<DateTimeV2ValueType>&
value) {
+ int64_t timestamp = 0;
+ value.unix_timestamp(×tamp, cctz::utc_time_zone());
+ return timestamp * MICROS_PER_SECOND + value.microsecond();
+}
+
+static int64_t variant_datetime_value(const TimestampTzValue& value) {
+ int64_t timestamp = 0;
+ value.unix_timestamp(×tamp, cctz::utc_time_zone());
+ return timestamp * MICROS_PER_SECOND + value.microsecond();
+}
+
+static int find_child_idx(const FieldSchema& field, std::string_view name) {
+ for (int i = 0; i < field.children.size(); ++i) {
+ if (field.children[i].lower_case_name == name) {
+ return i;
+ }
+ }
+ return -1;
+}
+
+static bool is_variant_wrapper_typed_value_child(const FieldSchema& field) {
+ auto type = remove_nullable(field.data_type);
+ return type->get_primitive_type() == TYPE_STRUCT ||
type->get_primitive_type() == TYPE_ARRAY;
+}
+
+static bool is_variant_wrapper_field(const FieldSchema& field,
+ bool
allow_scalar_typed_value_only_wrapper,
+ bool allow_value_only_wrapper = false) {
+ auto type = remove_nullable(field.data_type);
+ if (type->get_primitive_type() != TYPE_STRUCT &&
type->get_primitive_type() != TYPE_VARIANT) {
+ return false;
+ }
+
+ bool has_metadata = false;
+ bool has_value = false;
+ const FieldSchema* typed_value = nullptr;
+ for (const auto& child : field.children) {
+ if (child.lower_case_name == "metadata") {
+ if (child.physical_type != tparquet::Type::BYTE_ARRAY) {
+ return false;
+ }
+ has_metadata = true;
+ continue;
+ }
+ if (child.lower_case_name == "value") {
+ if (child.physical_type != tparquet::Type::BYTE_ARRAY) {
+ return false;
+ }
+ has_value = true;
+ continue;
+ }
+ if (child.lower_case_name == "typed_value") {
+ typed_value = &child;
+ continue;
+ }
+ return false;
+ }
+ if (has_metadata && has_value) {
+ return type->get_primitive_type() == TYPE_VARIANT || typed_value !=
nullptr;
+ }
+ if (has_value) {
+ return typed_value != nullptr || allow_value_only_wrapper;
+ }
+ return typed_value != nullptr && (allow_scalar_typed_value_only_wrapper ||
+
is_variant_wrapper_typed_value_child(*typed_value));
+}
+
+static Status get_binary_field(const Field& field, std::string* value, bool*
present) {
+ if (field.is_null()) {
+ *present = false;
+ return Status::OK();
+ }
+ *present = true;
+ switch (field.get_type()) {
+ case TYPE_STRING:
+ *value = field.get<TYPE_STRING>();
+ return Status::OK();
+ case TYPE_CHAR:
+ *value = field.get<TYPE_CHAR>();
+ return Status::OK();
+ case TYPE_VARCHAR:
+ *value = field.get<TYPE_VARCHAR>();
+ return Status::OK();
+ case TYPE_VARBINARY: {
+ auto ref = field.get<TYPE_VARBINARY>().to_string_ref();
+ value->assign(ref.data, ref.size);
+ return Status::OK();
+ }
+ default:
+ return Status::Corruption("Parquet VARIANT binary field has unexpected
Doris type {}",
+ field.get_type_name());
+ }
+}
+
+static PathInData append_path(const PathInData& prefix, const PathInData&
suffix) {
+ if (prefix.empty()) {
+ return suffix;
+ }
+ if (suffix.empty()) {
+ return prefix;
+ }
+ PathInDataBuilder builder;
+ builder.append(prefix.get_parts(), false);
+ builder.append(suffix.get_parts(), false);
+ return builder.build();
+}
+
+static Status insert_empty_object_marker(const PathInData& path, VariantMap*
values) {
+ JsonBinaryValue empty_object;
+ RETURN_IF_ERROR(empty_object.from_json_string("{}"));
+ (*values)[path] = FieldWithDataType {.field =
Field::create_field<TYPE_JSONB>(JsonbField(
+ empty_object.value(),
empty_object.size())),
+ .base_scalar_type_id = TYPE_JSONB};
+ return Status::OK();
+}
+
+static bool is_empty_object_marker(const FieldWithDataType& value) {
+ if (value.field.get_type() != TYPE_JSONB) {
+ return false;
+ }
+ const auto& jsonb = value.field.get<TYPE_JSONB>();
+ const JsonbDocument* document = nullptr;
+ Status st =
+ JsonbDocument::checkAndCreateDocument(jsonb.get_value(),
jsonb.get_size(), &document);
+ if (!st.ok() || document == nullptr || document->getValue() == nullptr ||
+ !document->getValue()->isObject()) {
+ return false;
+ }
+ return document->getValue()->unpack<ObjectVal>()->numElem() == 0;
+}
+
+static Status collect_empty_object_markers(const rapidjson::Value& value,
PathInDataBuilder* path,
+ VariantMap* values) {
+ if (!value.IsObject()) {
+ return Status::OK();
+ }
+ if (value.MemberCount() == 0) {
+ return insert_empty_object_marker(path->build(), values);
+ }
+ for (auto it = value.MemberBegin(); it != value.MemberEnd(); ++it) {
+ if (it->value.IsObject()) {
+ path->append(std::string_view(it->name.GetString(),
it->name.GetStringLength()), false);
+ RETURN_IF_ERROR(collect_empty_object_markers(it->value, path,
values));
+ path->pop_back();
+ }
+ }
+ return Status::OK();
+}
+
+static Status add_empty_object_markers_from_json(const std::string& json,
const PathInData& prefix,
+ VariantMap* values) {
+ if (json.find("{}") == std::string::npos) {
+ return Status::OK();
+ }
+ rapidjson::Document document;
+ document.Parse(json.data(), json.size());
+ if (document.HasParseError()) {
+ return Status::Corruption("Invalid Parquet VARIANT decoded JSON");
+ }
+ PathInDataBuilder path;
+ path.append(prefix.get_parts(), false);
+ return collect_empty_object_markers(document, &path, values);
+}
+
+static Status parse_json_to_variant_map(const std::string& json, const
PathInData& prefix,
+ VariantMap* values) {
+ auto parsed_column = ColumnVariant::create(0, false);
+ ParseConfig parse_config;
+ StringRef json_ref(json.data(), json.size());
+ RETURN_IF_CATCH_EXCEPTION(
+ variant_util::parse_json_to_variant(*parsed_column, json_ref,
nullptr, parse_config));
+ Field parsed = (*parsed_column)[0];
+ if (!parsed.is_null()) {
+ auto& parsed_values = parsed.get<TYPE_VARIANT>();
+ for (auto& [path, value] : parsed_values) {
+ (*values)[append_path(prefix, path)] = std::move(value);
+ }
+ }
+ RETURN_IF_ERROR(add_empty_object_markers_from_json(json, prefix, values));
+ return Status::OK();
+}
+
+static Status variant_map_to_json(VariantMap values, std::string* json) {
+ auto variant_column = ColumnVariant::create(0, false);
+ RETURN_IF_CATCH_EXCEPTION(
+
variant_column->insert(Field::create_field<TYPE_VARIANT>(std::move(values))));
+ DataTypeSerDe::FormatOptions options;
+ variant_column->serialize_one_row_to_string(0, json, options);
+ return Status::OK();
+}
+
+static bool path_has_prefix(const PathInData& path, const PathInData& prefix) {
+ const auto& parts = path.get_parts();
+ const auto& prefix_parts = prefix.get_parts();
+ if (parts.size() < prefix_parts.size()) {
+ return false;
+ }
+ for (size_t i = 0; i < prefix_parts.size(); ++i) {
+ if (parts[i] != prefix_parts[i]) {
+ return false;
+ }
+ }
+ return true;
+}
+
+static bool has_descendant_path(const VariantMap& values, const PathInData&
prefix) {
+ const size_t prefix_size = prefix.get_parts().size();
+ return std::ranges::any_of(values, [&](const auto& entry) {
+ const auto& path = entry.first;
+ return path.get_parts().size() > prefix_size && path_has_prefix(path,
prefix);
+ });
+}
+
+static void erase_shadowed_empty_object_markers(VariantMap* values,
+ const VariantMap&
shadowing_values) {
+ for (auto it = values->begin(); it != values->end();) {
+ if (is_empty_object_marker(it->second) &&
+ (has_descendant_path(*values, it->first) ||
+ has_descendant_path(shadowing_values, it->first))) {
+ it = values->erase(it);
+ continue;
+ }
+ ++it;
+ }
+}
+
+static void erase_shadowed_empty_object_markers(VariantMap* value_values,
+ VariantMap* typed_values) {
+ erase_shadowed_empty_object_markers(value_values, *typed_values);
+ erase_shadowed_empty_object_markers(typed_values, *value_values);
+}
+
+static Status check_no_shredded_value_typed_duplicates(const VariantMap&
value_values,
+ const VariantMap&
typed_values,
+ const PathInData&
prefix) {
+ const size_t prefix_size = prefix.get_parts().size();
+ for (const auto& value_entry : value_values) {
+ const auto& value_path = value_entry.first;
+ if (!path_has_prefix(value_path, prefix)) {
+ continue;
+ }
+ if (value_path.get_parts().size() == prefix_size) {
+ if (is_empty_object_marker(value_entry.second) &&
+ !has_descendant_path(typed_values, value_path)) {
+ continue;
+ }
+ if (!typed_values.empty()) {
+ return Status::Corruption(
+ "Parquet VARIANT residual value conflicts with
typed_value at path {}",
+ value_path.get_path());
+ }
+ continue;
+ }
+ for (const auto& typed_entry : typed_values) {
+ const auto& typed_path = typed_entry.first;
+ if (!path_has_prefix(typed_path, prefix)) {
+ continue;
+ }
+ if (typed_path.get_parts().size() == prefix_size) {
+ if (is_empty_object_marker(typed_entry.second) &&
+ !has_descendant_path(value_values, typed_path)) {
+ continue;
+ }
+ return Status::Corruption(
+ "Parquet VARIANT residual value and typed_value
contain duplicate field {}",
+ value_path.get_parts()[prefix_size].key);
+ }
+ if (value_path.get_parts()[prefix_size] ==
typed_path.get_parts()[prefix_size]) {
+ if (value_path == typed_path &&
is_empty_object_marker(value_entry.second) &&
+ is_empty_object_marker(typed_entry.second)) {
+ continue;
+ }
+ return Status::Corruption(
+ "Parquet VARIANT residual value and typed_value
contain duplicate field {}",
+ value_path.get_parts()[prefix_size].key);
+ }
+ }
+ }
+ return Status::OK();
+}
+
+static bool has_direct_typed_parent_null(const std::vector<const NullMap*>&
null_maps, size_t row) {
+ return std::ranges::any_of(null_maps, [&](const NullMap* null_map) {
+ DCHECK_LT(row, null_map->size());
+ return (*null_map)[row];
+ });
+}
+
+static void insert_direct_typed_leaf_range(const IColumn& column, size_t
start, size_t rows,
+ const std::vector<const NullMap*>&
parent_null_maps,
+ IColumn* variant_leaf) {
+ auto& nullable_leaf = assert_cast<ColumnNullable&>(*variant_leaf);
+ const IColumn* value_column = &column;
+ const NullMap* leaf_null_map = nullptr;
+ if (const auto* nullable_column =
check_and_get_column<ColumnNullable>(&column)) {
+ value_column = &nullable_column->get_nested_column();
+ leaf_null_map = &nullable_column->get_null_map_data();
+ }
+
+ nullable_leaf.get_nested_column().insert_range_from(*value_column, start,
rows);
+ auto& null_map = nullable_leaf.get_null_map_data();
+ null_map.reserve(null_map.size() + rows);
+ for (size_t i = 0; i < rows; ++i) {
+ const size_t row = start + i;
+ const bool leaf_is_null = leaf_null_map != nullptr &&
(*leaf_null_map)[row];
+ null_map.push_back(leaf_is_null ||
has_direct_typed_parent_null(parent_null_maps, row));
+ }
+}
+
+static bool is_temporal_variant_leaf_type(PrimitiveType type) {
+ switch (type) {
+ case TYPE_TIMEV2:
+ case TYPE_DATE:
+ case TYPE_DATETIME:
+ case TYPE_DATEV2:
+ case TYPE_DATETIMEV2:
+ case TYPE_TIMESTAMPTZ:
+ return true;
+ default:
+ return false;
+ }
+}
+
+static DataTypePtr direct_variant_leaf_type(const DataTypePtr& data_type) {
+ const auto& type = remove_nullable(data_type);
+ if (is_temporal_variant_leaf_type(type->get_primitive_type())) {
+ return std::make_shared<DataTypeInt64>();
+ }
+ return type;
+}
+
+static bool contains_temporal_variant_leaf_type(const DataTypePtr& data_type) {
+ const auto& type = remove_nullable(data_type);
+ if (is_temporal_variant_leaf_type(type->get_primitive_type())) {
+ return true;
+ }
+ if (type->get_primitive_type() == TYPE_ARRAY) {
+ return contains_temporal_variant_leaf_type(
+ assert_cast<const
DataTypeArray*>(type.get())->get_nested_type());
+ }
+ return false;
+}
+
+static int64_t direct_temporal_variant_value(PrimitiveType type, const
IColumn& column,
+ size_t row) {
+ switch (type) {
+ case TYPE_TIMEV2:
+ return static_cast<int64_t>(
+ std::llround(assert_cast<const
ColumnTimeV2&>(column).get_data()[row]));
+ case TYPE_DATE:
+ return variant_date_value(assert_cast<const
ColumnDate&>(column).get_data()[row]);
+ case TYPE_DATETIME:
+ return variant_datetime_value(assert_cast<const
ColumnDateTime&>(column).get_data()[row]);
+ case TYPE_DATEV2:
+ return variant_date_value(assert_cast<const
ColumnDateV2&>(column).get_data()[row]);
+ case TYPE_DATETIMEV2:
+ return variant_datetime_value(assert_cast<const
ColumnDateTimeV2&>(column).get_data()[row]);
+ case TYPE_TIMESTAMPTZ:
+ return variant_datetime_value(
+ assert_cast<const ColumnTimeStampTz&>(column).get_data()[row]);
+ default:
+ DORIS_CHECK(false);
+ return 0;
+ }
+}
+
+static void insert_direct_typed_temporal_leaf_range(
+ PrimitiveType type, const IColumn& column, size_t start, size_t rows,
+ const std::vector<const NullMap*>& parent_null_maps, IColumn*
variant_leaf) {
+ auto& nullable_leaf = assert_cast<ColumnNullable&>(*variant_leaf);
+ const IColumn* value_column = &column;
+ const NullMap* leaf_null_map = nullptr;
+ if (const auto* nullable_column =
check_and_get_column<ColumnNullable>(&column)) {
+ value_column = &nullable_column->get_nested_column();
+ leaf_null_map = &nullable_column->get_null_map_data();
+ }
+
+ auto& data =
assert_cast<ColumnInt64&>(nullable_leaf.get_nested_column()).get_data();
+ data.reserve(data.size() + rows);
+ auto& null_map = nullable_leaf.get_null_map_data();
+ null_map.reserve(null_map.size() + rows);
+ for (size_t i = 0; i < rows; ++i) {
+ const size_t row = start + i;
+ data.push_back(direct_temporal_variant_value(type, *value_column,
row));
+ const bool leaf_is_null = leaf_null_map != nullptr &&
(*leaf_null_map)[row];
+ null_map.push_back(leaf_is_null ||
has_direct_typed_parent_null(parent_null_maps, row));
+ }
+}
+
+static void append_json_string(std::string_view value, std::string* json) {
+ auto column = ColumnString::create();
+ VectorBufferWriter writer(*column);
+ writer.write_json_string(value);
+ writer.commit();
+ json->append(column->get_data_at(0).data, column->get_data_at(0).size);
+}
+
+static bool is_column_selected(const FieldSchema& field_schema,
+ const std::set<uint64_t>& column_ids) {
+ return column_ids.empty() || column_ids.find(field_schema.get_column_id())
!= column_ids.end();
+}
+
+static bool has_selected_column(const FieldSchema& field_schema,
+ const std::set<uint64_t>& column_ids) {
+ if (is_column_selected(field_schema, column_ids)) {
+ return true;
+ }
+ return std::any_of(field_schema.children.begin(),
field_schema.children.end(),
+ [&column_ids](const FieldSchema& child) {
+ return has_selected_column(child, column_ids);
+ });
+}
+
+static bool is_direct_variant_leaf_type(const DataTypePtr& data_type) {
+ const auto& type = remove_nullable(data_type);
+ switch (type->get_primitive_type()) {
+ case TYPE_BOOLEAN:
+ case TYPE_TINYINT:
+ case TYPE_SMALLINT:
+ case TYPE_INT:
+ case TYPE_BIGINT:
+ case TYPE_LARGEINT:
+ case TYPE_DECIMALV2:
+ case TYPE_DECIMAL32:
+ case TYPE_DECIMAL64:
+ case TYPE_DECIMAL128I:
+ case TYPE_DECIMAL256:
+ case TYPE_STRING:
+ case TYPE_CHAR:
+ case TYPE_VARCHAR:
+ return true;
+ case TYPE_TIMEV2:
+ case TYPE_DATE:
+ case TYPE_DATETIME:
+ case TYPE_DATEV2:
+ case TYPE_DATETIMEV2:
+ case TYPE_TIMESTAMPTZ:
+ return true;
+ case TYPE_ARRAY: {
+ const auto* array_type = assert_cast<const DataTypeArray*>(type.get());
+ return
!contains_temporal_variant_leaf_type(array_type->get_nested_type()) &&
+ is_direct_variant_leaf_type(array_type->get_nested_type());
+ }
+ default:
+ return false;
+ }
+}
+
+static bool can_direct_read_typed_value(const FieldSchema& field_schema, bool
allow_variant_wrapper,
+ const std::set<uint64_t>& column_ids) {
+ if (!has_selected_column(field_schema, column_ids)) {
+ return true;
+ }
+ if (allow_variant_wrapper && is_variant_wrapper_field(field_schema,
false)) {
+ const int value_idx = find_child_idx(field_schema, "value");
+ const int typed_value_idx = find_child_idx(field_schema,
"typed_value");
+ return (value_idx < 0 ||
+ !has_selected_column(field_schema.children[value_idx],
column_ids)) &&
+ typed_value_idx >= 0 &&
+
can_direct_read_typed_value(field_schema.children[typed_value_idx], false,
+ column_ids);
+ }
+
+ const auto& type = remove_nullable(field_schema.data_type);
+ if (type->get_primitive_type() == TYPE_STRUCT) {
+ return std::all_of(field_schema.children.begin(),
field_schema.children.end(),
+ [&column_ids](const FieldSchema& child) {
+ return can_direct_read_typed_value(child, true,
column_ids);
+ });
+ }
+ return is_direct_variant_leaf_type(field_schema.data_type);
+}
+
+static bool has_selected_direct_typed_leaf(const FieldSchema& field_schema,
+ bool allow_variant_wrapper,
+ const std::set<uint64_t>&
column_ids) {
+ if (!has_selected_column(field_schema, column_ids)) {
+ return false;
+ }
+ if (allow_variant_wrapper && is_variant_wrapper_field(field_schema,
false)) {
+ const int typed_value_idx = find_child_idx(field_schema,
"typed_value");
+ DCHECK_GE(typed_value_idx, 0);
+ return
has_selected_direct_typed_leaf(field_schema.children[typed_value_idx], false,
+ column_ids);
+ }
+
+ const auto& type = remove_nullable(field_schema.data_type);
+ if (type->get_primitive_type() == TYPE_STRUCT) {
+ return std::any_of(field_schema.children.begin(),
field_schema.children.end(),
+ [&column_ids](const FieldSchema& child) {
+ return has_selected_direct_typed_leaf(child,
true, column_ids);
+ });
+ }
+ return is_direct_variant_leaf_type(field_schema.data_type);
+}
+
+static bool can_use_direct_typed_only_value(const FieldSchema& variant_field,
+ const std::set<uint64_t>&
column_ids) {
+ const int value_idx = find_child_idx(variant_field, "value");
+ const int typed_value_idx = find_child_idx(variant_field, "typed_value");
+ return (value_idx < 0 ||
!has_selected_column(variant_field.children[value_idx], column_ids)) &&
+ typed_value_idx >= 0 &&
+
has_selected_direct_typed_leaf(variant_field.children[typed_value_idx], false,
+ column_ids) &&
+
can_direct_read_typed_value(variant_field.children[typed_value_idx], false,
column_ids);
+}
+
+static void fill_variant_field_info(FieldWithDataType* value) {
+ FieldInfo info;
+ variant_util::get_field_info(value->field, &info);
+ DCHECK_LE(info.num_dimensions, std::numeric_limits<uint8_t>::max());
+ value->base_scalar_type_id = info.scalar_type_id;
+ value->num_dimensions = static_cast<uint8_t>(info.num_dimensions);
+}
+
+static Status field_to_variant_field(const FieldSchema& field_schema, const
Field& field,
+ FieldWithDataType* value, bool* present) {
+ if (field.is_null()) {
+ *present = false;
+ return Status::OK();
+ }
+ *present = true;
+ const DataTypePtr& type = remove_nullable(field_schema.data_type);
+ switch (type->get_primitive_type()) {
+ case TYPE_BOOLEAN:
+ case TYPE_TINYINT:
+ case TYPE_SMALLINT:
+ case TYPE_INT:
+ case TYPE_BIGINT:
+ case TYPE_LARGEINT:
+ case TYPE_DECIMALV2:
+ case TYPE_DECIMAL32:
+ case TYPE_DECIMAL64:
+ case TYPE_DECIMAL128I:
+ case TYPE_DECIMAL256:
+ case TYPE_STRING:
+ case TYPE_CHAR:
+ case TYPE_VARCHAR:
+ case TYPE_ARRAY:
+ value->field = field;
+ fill_variant_field_info(value);
+ value->precision = type->get_precision();
+ value->scale = type->get_scale();
+ return Status::OK();
+ case TYPE_FLOAT: {
+ const auto float_value = field.get<TYPE_FLOAT>();
+ if (!std::isfinite(float_value)) {
+ return Status::NotSupported(
+ "Parquet VARIANT non-finite floating point typed_value is
not supported");
+ }
+ value->field = field;
+ fill_variant_field_info(value);
+ return Status::OK();
+ }
+ case TYPE_DOUBLE: {
+ const auto double_value = field.get<TYPE_DOUBLE>();
+ if (!std::isfinite(double_value)) {
+ return Status::NotSupported(
+ "Parquet VARIANT non-finite floating point typed_value is
not supported");
+ }
+ value->field = field;
+ fill_variant_field_info(value);
+ return Status::OK();
+ }
+ case TYPE_TIMEV2:
+ value->field = Field::create_field<TYPE_BIGINT>(
+ static_cast<int64_t>(std::llround(field.get<TYPE_TIMEV2>())));
+ value->base_scalar_type_id = TYPE_BIGINT;
+ return Status::OK();
+ case TYPE_DATE:
+ value->field =
Field::create_field<TYPE_BIGINT>(variant_date_value(field.get<TYPE_DATE>()));
+ value->base_scalar_type_id = TYPE_BIGINT;
+ return Status::OK();
+ case TYPE_DATETIME:
+ value->field = Field::create_field<TYPE_BIGINT>(
+ variant_datetime_value(field.get<TYPE_DATETIME>()));
+ value->base_scalar_type_id = TYPE_BIGINT;
+ return Status::OK();
+ case TYPE_DATEV2:
+ value->field =
+
Field::create_field<TYPE_BIGINT>(variant_date_value(field.get<TYPE_DATEV2>()));
+ value->base_scalar_type_id = TYPE_BIGINT;
+ return Status::OK();
+ case TYPE_DATETIMEV2:
+ value->field = Field::create_field<TYPE_BIGINT>(
+ variant_datetime_value(field.get<TYPE_DATETIMEV2>()));
+ value->base_scalar_type_id = TYPE_BIGINT;
+ return Status::OK();
+ case TYPE_TIMESTAMPTZ:
+ value->field = Field::create_field<TYPE_BIGINT>(
+ variant_datetime_value(field.get<TYPE_TIMESTAMPTZ>()));
+ value->base_scalar_type_id = TYPE_BIGINT;
+ return Status::OK();
+ case TYPE_VARBINARY:
+ return Status::NotSupported("Parquet VARIANT binary typed_value is not
supported");
+ default:
+ return Status::Corruption("Unsupported Parquet VARIANT typed_value
Doris type {}",
+ type->get_name());
+ }
+}
+
+static Status typed_value_to_json(const FieldSchema& typed_value_field, const
Field& field,
+ const std::string& metadata, std::string*
json, bool* present);
+
+static Status serialize_field_to_json(const DataTypePtr& data_type, const
Field& field,
+ std::string* json) {
+ MutableColumnPtr column = data_type->create_column();
+ column->insert(field);
+
+ auto json_column = ColumnString::create();
+ VectorBufferWriter writer(*json_column);
+ auto serde = data_type->get_serde();
+ DataTypeSerDe::FormatOptions options;
+ RETURN_IF_ERROR(serde->serialize_one_cell_to_json(*column, 0, writer,
options));
+ writer.commit();
+ *json = json_column->get_data_at(0).to_string();
+ return Status::OK();
+}
+
+static Status scalar_typed_value_to_json(const FieldSchema& field_schema,
const Field& field,
+ std::string* json, bool* present) {
+ FieldWithDataType value;
+ RETURN_IF_ERROR(field_to_variant_field(field_schema, field, &value,
present));
+ if (!*present) {
+ return Status::OK();
+ }
+ if (value.field.is_null()) {
+ *json = "null";
+ return Status::OK();
+ }
+
+ DataTypePtr json_type;
+ if (value.base_scalar_type_id != PrimitiveType::INVALID_TYPE) {
+ json_type =
DataTypeFactory::instance().create_data_type(value.base_scalar_type_id, false,
+
value.precision, value.scale);
+ } else {
+ json_type = remove_nullable(field_schema.data_type);
+ }
+ return serialize_field_to_json(json_type, value.field, json);
+}
+
+static Status variant_to_json(const FieldSchema& variant_field, const Field&
field,
+ const std::string* inherited_metadata,
std::string* json,
+ bool* present) {
+ if (field.is_null()) {
+ *present = false;
+ return Status::OK();
+ }
+
+ const auto& fields = field.get<TYPE_STRUCT>();
+ const int metadata_idx = find_child_idx(variant_field, "metadata");
+ const int value_idx = find_child_idx(variant_field, "value");
+ const int typed_value_idx = find_child_idx(variant_field, "typed_value");
+
+ std::string metadata;
+ bool has_metadata = false;
+ if (inherited_metadata != nullptr) {
+ metadata = *inherited_metadata;
+ has_metadata = true;
+ }
+ if (metadata_idx >= 0) {
+ bool metadata_present = false;
+ RETURN_IF_ERROR(get_binary_field(fields[metadata_idx], &metadata,
&metadata_present));
+ has_metadata = metadata_present;
+ }
+
+ std::string typed_json;
+ bool typed_present = false;
+ if (typed_value_idx >= 0) {
+
RETURN_IF_ERROR(typed_value_to_json(variant_field.children[typed_value_idx],
+ fields[typed_value_idx], metadata,
&typed_json,
+ &typed_present));
+ }
+
+ std::string value_json;
+ bool value_present = false;
+ if (value_idx >= 0) {
+ std::string value;
+ RETURN_IF_ERROR(get_binary_field(fields[value_idx], &value,
&value_present));
+ if (value_present) {
+ if (!has_metadata) {
+ return Status::Corruption("Parquet VARIANT value is present
without metadata");
+ }
+ RETURN_IF_ERROR(parquet::decode_variant_to_json(
+ StringRef(metadata.data(), metadata.size()),
+ StringRef(value.data(), value.size()), &value_json));
+ }
+ }
+
+ if (value_present && typed_present) {
+ VariantMap value_values;
+ RETURN_IF_ERROR(parse_json_to_variant_map(value_json, PathInData(),
&value_values));
+ VariantMap typed_values;
+ RETURN_IF_ERROR(parse_json_to_variant_map(typed_json, PathInData(),
&typed_values));
+ erase_shadowed_empty_object_markers(&value_values, &typed_values);
+ auto root_value = value_values.find(PathInData());
+ if (root_value != value_values.end() &&
!is_empty_object_marker(root_value->second)) {
+ return Status::Corruption(
+ "Parquet VARIANT has conflicting non-object value and
typed_value");
+ }
+ RETURN_IF_ERROR(
+ check_no_shredded_value_typed_duplicates(value_values,
typed_values, PathInData()));
+ value_values.merge(std::move(typed_values));
+ RETURN_IF_ERROR(variant_map_to_json(std::move(value_values), json));
+ *present = true;
+ return Status::OK();
+ }
+
+ if (typed_present) {
+ *json = std::move(typed_json);
+ *present = true;
+ return Status::OK();
+ }
+ if (value_present) {
+ *json = std::move(value_json);
+ *present = true;
+ return Status::OK();
+ }
+
+ *present = false;
+ return Status::OK();
+}
+
+static Status shredded_field_to_json(const FieldSchema& field_schema, const
Field& field,
+ const std::string& metadata, std::string*
json, bool* present,
+ bool
allow_scalar_typed_value_only_wrapper) {
+ if (is_variant_wrapper_field(field_schema,
allow_scalar_typed_value_only_wrapper, true)) {
+ return variant_to_json(field_schema, field, &metadata, json, present);
+ }
+ return typed_value_to_json(field_schema, field, metadata, json, present);
+}
+
+static Status typed_array_to_json(const FieldSchema& typed_value_field, const
Field& field,
+ const std::string& metadata, std::string*
json, bool* present) {
+ if (field.is_null()) {
+ *present = false;
+ return Status::OK();
+ }
+ if (typed_value_field.children.empty()) {
+ return Status::Corruption("Parquet VARIANT array typed_value has no
element schema");
+ }
+
+ const auto& elements = field.get<TYPE_ARRAY>();
+ const auto& element_schema = typed_value_field.children[0];
+ json->clear();
+ json->push_back('[');
+ for (size_t i = 0; i < elements.size(); ++i) {
+ if (i != 0) {
+ json->push_back(',');
+ }
+ std::string element_json;
+ bool element_present = false;
+ RETURN_IF_ERROR(shredded_field_to_json(element_schema, elements[i],
metadata, &element_json,
+ &element_present, true));
+ if (element_present) {
+ json->append(element_json);
+ } else {
+ return Status::Corruption("Parquet VARIANT array typed_value
element is missing");
+ }
+ }
+ json->push_back(']');
+ *present = true;
+ return Status::OK();
+}
+
+static Status typed_struct_to_json(const FieldSchema& typed_value_field, const
Field& field,
+ const std::string& metadata, std::string*
json, bool* present) {
+ if (field.is_null()) {
+ *present = false;
+ return Status::OK();
+ }
+
+ const auto& fields = field.get<TYPE_STRUCT>();
+ json->clear();
+ json->push_back('{');
+ bool first = true;
+ for (int i = 0; i < typed_value_field.children.size(); ++i) {
+ std::string child_json;
+ bool child_present = false;
+ RETURN_IF_ERROR(shredded_field_to_json(typed_value_field.children[i],
fields[i], metadata,
+ &child_json, &child_present,
false));
+ if (!child_present) {
+ continue;
+ }
+ if (!first) {
+ json->push_back(',');
+ }
+ append_json_string(typed_value_field.children[i].name, json);
+ json->push_back(':');
+ json->append(child_json);
+ first = false;
+ }
+ json->push_back('}');
+ *present = true;
+ return Status::OK();
+}
+
+static Status typed_value_to_json(const FieldSchema& typed_value_field, const
Field& field,
+ const std::string& metadata, std::string*
json, bool* present) {
+ const DataTypePtr& typed_type =
remove_nullable(typed_value_field.data_type);
+ switch (typed_type->get_primitive_type()) {
+ case TYPE_STRUCT:
+ return typed_struct_to_json(typed_value_field, field, metadata, json,
present);
+ case TYPE_ARRAY:
+ return typed_array_to_json(typed_value_field, field, metadata, json,
present);
+ default:
+ return scalar_typed_value_to_json(typed_value_field, field, json,
present);
+ }
+}
+
+static Status typed_value_to_variant_map(const FieldSchema& typed_value_field,
const Field& field,
+ const std::string& metadata,
PathInDataBuilder* path,
+ VariantMap* values, bool* present);
+
+static Status variant_to_variant_map(const FieldSchema& variant_field, const
Field& field,
+ const std::string* inherited_metadata,
PathInDataBuilder* path,
+ VariantMap* values, bool* present) {
+ if (field.is_null()) {
+ *present = false;
+ return Status::OK();
+ }
+ const auto& fields = field.get<TYPE_STRUCT>();
+ const int metadata_idx = find_child_idx(variant_field, "metadata");
+ const int value_idx = find_child_idx(variant_field, "value");
+ const int typed_value_idx = find_child_idx(variant_field, "typed_value");
+
+ std::string metadata;
+ bool has_metadata = false;
+ if (inherited_metadata != nullptr) {
+ metadata = *inherited_metadata;
+ has_metadata = true;
+ }
+ if (metadata_idx >= 0) {
+ bool metadata_present = false;
+ RETURN_IF_ERROR(get_binary_field(fields[metadata_idx], &metadata,
&metadata_present));
+ has_metadata = metadata_present;
+ }
+
+ VariantMap value_values;
+ bool value_present = false;
+ const PathInData current_path = path->build();
+ if (value_idx >= 0) {
+ std::string value;
+ RETURN_IF_ERROR(get_binary_field(fields[value_idx], &value,
&value_present));
+ if (value_present) {
+ if (!has_metadata) {
+ return Status::Corruption("Parquet VARIANT value is present
without metadata");
+ }
+ std::string value_json;
+ RETURN_IF_ERROR(parquet::decode_variant_to_json(
+ StringRef(metadata.data(), metadata.size()),
+ StringRef(value.data(), value.size()), &value_json));
+ RETURN_IF_ERROR(parse_json_to_variant_map(value_json,
current_path, &value_values));
+ }
+ }
+
+ VariantMap typed_values;
+ bool typed_present = false;
+ if (typed_value_idx >= 0) {
+
RETURN_IF_ERROR(typed_value_to_variant_map(variant_field.children[typed_value_idx],
+ fields[typed_value_idx],
metadata, path,
+ &typed_values,
&typed_present));
+ }
+
+ erase_shadowed_empty_object_markers(&value_values, &typed_values);
+ auto current_value = value_values.find(current_path);
+ if (value_present && typed_present && current_value != value_values.end()
&&
+ !is_empty_object_marker(current_value->second)) {
+ return Status::Corruption(
+ "Parquet VARIANT has conflicting non-object value and
typed_value");
+ }
+ RETURN_IF_ERROR(
+ check_no_shredded_value_typed_duplicates(value_values,
typed_values, current_path));
+ values->merge(std::move(value_values));
+ values->merge(std::move(typed_values));
+ *present = value_present || typed_present;
+ return Status::OK();
+}
+
+static Status shredded_field_to_variant_map(const FieldSchema& field_schema,
const Field& field,
+ const std::string& metadata,
PathInDataBuilder* path,
+ VariantMap* values, bool* present)
{
+ if (is_variant_wrapper_field(field_schema, false, true)) {
+ return variant_to_variant_map(field_schema, field, &metadata, path,
values, present);
+ }
+ return typed_value_to_variant_map(field_schema, field, metadata, path,
values, present);
+}
+
+static Status typed_value_to_variant_map(const FieldSchema& typed_value_field,
const Field& field,
+ const std::string& metadata,
PathInDataBuilder* path,
+ VariantMap* values, bool* present) {
+ if (field.is_null()) {
+ *present = false;
+ return Status::OK();
+ }
+ const DataTypePtr& typed_type =
remove_nullable(typed_value_field.data_type);
+ if (typed_type->get_primitive_type() == TYPE_STRUCT) {
+ const auto& fields = field.get<TYPE_STRUCT>();
+ *present = true;
+ bool has_present_child = false;
+ for (int i = 0; i < typed_value_field.children.size(); ++i) {
+ path->append(typed_value_field.children[i].name, false);
+ bool child_present = false;
+
RETURN_IF_ERROR(shredded_field_to_variant_map(typed_value_field.children[i],
fields[i],
+ metadata, path,
values, &child_present));
+ has_present_child |= child_present;
+ path->pop_back();
+ }
+ if (!has_present_child) {
+ RETURN_IF_ERROR(insert_empty_object_marker(path->build(), values));
+ }
+ return Status::OK();
+ }
+ if (typed_type->get_primitive_type() == TYPE_ARRAY) {
+ std::string value_json;
+ RETURN_IF_ERROR(
+ typed_value_to_json(typed_value_field, field, metadata,
&value_json, present));
+ if (*present) {
+ RETURN_IF_ERROR(parse_json_to_variant_map(value_json,
path->build(), values));
+ }
+ return Status::OK();
+ }
+
+ FieldWithDataType value;
+ RETURN_IF_ERROR(field_to_variant_field(typed_value_field, field, &value,
present));
+ if (*present) {
+ (*values)[path->build()] = std::move(value);
+ }
+ return Status::OK();
+}
+
+static Status append_direct_typed_column_to_batch(const FieldSchema&
field_schema,
+ const IColumn& column,
size_t start, size_t rows,
+ PathInDataBuilder* path,
ColumnVariant* batch,
+ bool allow_variant_wrapper,
+ const std::set<uint64_t>&
column_ids,
+ std::vector<const NullMap*>
parent_null_maps) {
+ if (!has_selected_column(field_schema, column_ids)) {
+ return Status::OK();
+ }
+
+ const IColumn* value_column = &column;
+ if (const auto* nullable_column =
check_and_get_column<ColumnNullable>(&column)) {
+ parent_null_maps.push_back(&nullable_column->get_null_map_data());
+ value_column = &nullable_column->get_nested_column();
+ }
+
+ if (allow_variant_wrapper && is_variant_wrapper_field(field_schema,
false)) {
+ const int typed_value_idx = find_child_idx(field_schema,
"typed_value");
+ DCHECK_GE(typed_value_idx, 0);
+ const auto& typed_struct = assert_cast<const
ColumnStruct&>(*value_column);
+ return append_direct_typed_column_to_batch(
+ field_schema.children[typed_value_idx],
typed_struct.get_column(typed_value_idx),
+ start, rows, path, batch, false, column_ids, parent_null_maps);
+ }
+
+ const auto& type = remove_nullable(field_schema.data_type);
+ if (type->get_primitive_type() == TYPE_STRUCT) {
+ const auto& struct_column = assert_cast<const
ColumnStruct&>(*value_column);
+ for (int i = 0; i < field_schema.children.size(); ++i) {
+ if (!has_selected_column(field_schema.children[i], column_ids)) {
+ continue;
+ }
+ path->append(field_schema.children[i].name, false);
+ RETURN_IF_ERROR(append_direct_typed_column_to_batch(
+ field_schema.children[i], struct_column.get_column(i),
start, rows, path, batch,
+ true, column_ids, parent_null_maps));
+ path->pop_back();
+ }
+ return Status::OK();
+ }
+
+ DataTypePtr variant_leaf_type =
make_nullable(direct_variant_leaf_type(field_schema.data_type));
+ MutableColumnPtr variant_leaf = variant_leaf_type->create_column();
+ variant_leaf->insert_default();
+ if (is_temporal_variant_leaf_type(type->get_primitive_type())) {
+ insert_direct_typed_temporal_leaf_range(type->get_primitive_type(),
*value_column, start,
+ rows, parent_null_maps,
variant_leaf.get());
+ } else {
+ insert_direct_typed_leaf_range(*value_column, start, rows,
parent_null_maps,
+ variant_leaf.get());
+ }
+ if (!batch->add_sub_column(path->build(), std::move(variant_leaf),
variant_leaf_type)) {
+ return Status::Corruption("Failed to add Parquet VARIANT typed
subcolumn {}",
+ path->build().get_path());
+ }
+ return Status::OK();
+}
+
+#ifdef BE_TEST
+namespace parquet_variant_reader_test {
+bool can_direct_read_typed_value_for_test(const FieldSchema&
typed_value_field) {
Review Comment:
This array branch reconstructs typed arrays by serializing the typed column
to JSON and reparsing it into a `VariantMap`, which drops the scalar type
metadata that the scalar typed path preserves in `FieldWithDataType`. A normal
full projection of a shredded VARIANT with schema `metadata, value,
typed_value: array<decimal(...)>` takes the row-wise path because `column_ids`
is empty and the optional `value` leaf is selected, then decimal array elements
are parsed back from JSON numbers instead of inserted with the original decimal
precision/scale. That makes `SELECT v` produce a different VARIANT shape/type
fidelity than the direct typed path for the same typed data. Please build the
`VariantMap` for typed arrays without a JSON round trip (or otherwise preserve
element type metadata) and add coverage for a typed array of DECIMAL or another
non-JSON-native scalar under full VARIANT projection.
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
