diff --git a/src/Functions/FunctionsJSON.cpp b/src/Functions/FunctionsJSON.cpp index 7fec45d1f72..12b4ca76ec2 100644 --- a/src/Functions/FunctionsJSON.cpp +++ b/src/Functions/FunctionsJSON.cpp @@ -1,60 +1,1447 @@ -#include +#include + +#include + +#include + +#include +#include +#include + +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + #include +#include +#include +#include +#include +#include + +#include + + +#if !defined(ARCADIA_BUILD) +# include "config_functions.h" +#endif namespace DB { + namespace ErrorCodes { extern const int ILLEGAL_TYPE_OF_ARGUMENT; + extern const int ILLEGAL_COLUMN; + extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH; } -std::vector FunctionJSONHelpers::prepareMoves(const char * function_name, const ColumnsWithTypeAndName & columns, size_t first_index_argument, size_t num_index_arguments) -{ - std::vector moves; - moves.reserve(num_index_arguments); - for (const auto i : collections::range(first_index_argument, first_index_argument + num_index_arguments)) - { - const auto & column = columns[i]; - if (!isString(column.type) && !isInteger(column.type)) - throw Exception{"The argument " + std::to_string(i + 1) + " of function " + String(function_name) - + " should be a string specifying key or an integer specifying index, illegal type: " + column.type->getName(), - ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT}; +/// Functions to parse JSONs and extract values from it. +/// The first argument of all these functions gets a JSON, +/// after that there are any number of arguments specifying path to a desired part from the JSON's root. +/// For example, +/// select JSONExtractInt('{"a": "hello", "b": [-100, 200.0, 300]}', 'b', 1) = -100 - if (column.column && isColumnConst(*column.column)) +class FunctionJSONHelpers +{ +public: + template typename Impl, class JSONParser> + class Executor + { + public: + static ColumnPtr run(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) { - const auto & column_const = assert_cast(*column.column); - if (isString(column.type)) - moves.emplace_back(MoveType::ConstKey, column_const.getValue()); + MutableColumnPtr to{result_type->createColumn()}; + to->reserve(input_rows_count); + + if (arguments.empty()) + throw Exception{"Function " + String(Name::name) + " requires at least one argument", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; + + const auto & first_column = arguments[0]; + if (!isString(first_column.type)) + throw Exception{"The first argument of function " + String(Name::name) + " should be a string containing JSON, illegal type: " + first_column.type->getName(), + ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT}; + + const ColumnPtr & arg_json = first_column.column; + const auto * col_json_const = typeid_cast(arg_json.get()); + const auto * col_json_string + = typeid_cast(col_json_const ? col_json_const->getDataColumnPtr().get() : arg_json.get()); + + if (!col_json_string) + throw Exception{"Illegal column " + arg_json->getName(), ErrorCodes::ILLEGAL_COLUMN}; + + const ColumnString::Chars & chars = col_json_string->getChars(); + const ColumnString::Offsets & offsets = col_json_string->getOffsets(); + + size_t num_index_arguments = Impl::getNumberOfIndexArguments(arguments); + std::vector moves = prepareMoves(Name::name, arguments, 1, num_index_arguments); + + /// Preallocate memory in parser if necessary. + JSONParser parser; + if constexpr (has_member_function_reserve::value) + { + size_t max_size = calculateMaxSize(offsets); + if (max_size) + parser.reserve(max_size); + } + + Impl impl; + + /// prepare() does Impl-specific preparation before handling each row. + if constexpr (has_member_function_prepare::*)(const char *, const ColumnsWithTypeAndName &, const DataTypePtr &)>::value) + impl.prepare(Name::name, arguments, result_type); + + using Element = typename JSONParser::Element; + + Element document; + bool document_ok = false; + if (col_json_const) + { + std::string_view json{reinterpret_cast(&chars[0]), offsets[0] - 1}; + document_ok = parser.parse(json, document); + } + + for (const auto i : collections::range(0, input_rows_count)) + { + if (!col_json_const) + { + std::string_view json{reinterpret_cast(&chars[offsets[i - 1]]), offsets[i] - offsets[i - 1] - 1}; + document_ok = parser.parse(json, document); + } + + bool added_to_column = false; + if (document_ok) + { + /// Perform moves. + Element element; + std::string_view last_key; + bool moves_ok = performMoves(arguments, i, document, moves, element, last_key); + + if (moves_ok) + added_to_column = impl.insertResultToColumn(*to, element, last_key); + } + + /// We add default value (=null or zero) if something goes wrong, we don't throw exceptions in these JSON functions. + if (!added_to_column) + to->insertDefault(); + } + return to; + } + }; + +private: + BOOST_TTI_HAS_MEMBER_FUNCTION(reserve) + BOOST_TTI_HAS_MEMBER_FUNCTION(prepare) + + template + struct has_index_operator : std::false_type {}; + + template + struct has_index_operator()[0])>> : std::true_type {}; + + /// Represents a move of a JSON iterator described by a single argument passed to a JSON function. + /// For example, the call JSONExtractInt('{"a": "hello", "b": [-100, 200.0, 300]}', 'b', 1) + /// contains two moves: {MoveType::ConstKey, "b"} and {MoveType::ConstIndex, 1}. + /// Keys and indices can be nonconst, in this case they are calculated for each row. + enum class MoveType + { + Key, + Index, + ConstKey, + ConstIndex, + }; + + struct Move + { + Move(MoveType type_, size_t index_ = 0) : type(type_), index(index_) {} + Move(MoveType type_, const String & key_) : type(type_), key(key_) {} + MoveType type; + size_t index = 0; + String key; + }; + + static std::vector prepareMoves( + const char * function_name, + const ColumnsWithTypeAndName & columns, + size_t first_index_argument, + size_t num_index_arguments) + { + std::vector moves; + moves.reserve(num_index_arguments); + for (const auto i : collections::range(first_index_argument, first_index_argument + num_index_arguments)) + { + const auto & column = columns[i]; + if (!isString(column.type) && !isInteger(column.type)) + throw Exception{"The argument " + std::to_string(i + 1) + " of function " + String(function_name) + + " should be a string specifying key or an integer specifying index, illegal type: " + column.type->getName(), + ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT}; + + if (column.column && isColumnConst(*column.column)) + { + const auto & column_const = assert_cast(*column.column); + if (isString(column.type)) + moves.emplace_back(MoveType::ConstKey, column_const.getValue()); + else + moves.emplace_back(MoveType::ConstIndex, column_const.getInt(0)); + } else - moves.emplace_back(MoveType::ConstIndex, column_const.getInt(0)); + { + if (isString(column.type)) + moves.emplace_back(MoveType::Key, ""); + else + moves.emplace_back(MoveType::Index, 0); + } + } + return moves; + } + + + /// Performs moves of types MoveType::Index and MoveType::ConstIndex. + template + static bool performMoves(const ColumnsWithTypeAndName & arguments, size_t row, + const typename JSONParser::Element & document, const std::vector & moves, + typename JSONParser::Element & element, std::string_view & last_key) + { + typename JSONParser::Element res_element = document; + std::string_view key; + + for (size_t j = 0; j != moves.size(); ++j) + { + switch (moves[j].type) + { + case MoveType::ConstIndex: + { + if (!moveToElementByIndex(res_element, moves[j].index, key)) + return false; + break; + } + case MoveType::ConstKey: + { + key = moves[j].key; + if (!moveToElementByKey(res_element, key)) + return false; + break; + } + case MoveType::Index: + { + Int64 index = (*arguments[j + 1].column)[row].get(); + if (!moveToElementByIndex(res_element, index, key)) + return false; + break; + } + case MoveType::Key: + { + key = std::string_view{(*arguments[j + 1].column).getDataAt(row)}; + if (!moveToElementByKey(res_element, key)) + return false; + break; + } + } + } + + element = res_element; + last_key = key; + return true; + } + + template + static bool moveToElementByIndex(typename JSONParser::Element & element, int index, std::string_view & out_key) + { + if (element.isArray()) + { + auto array = element.getArray(); + if (index >= 0) + --index; + else + index += array.size(); + + if (static_cast(index) >= array.size()) + return false; + element = array[index]; + out_key = {}; + return true; + } + + if constexpr (has_index_operator::value) + { + if (element.isObject()) + { + auto object = element.getObject(); + if (index >= 0) + --index; + else + index += object.size(); + + if (static_cast(index) >= object.size()) + return false; + std::tie(out_key, element) = object[index]; + return true; + } + } + + return {}; + } + + /// Performs moves of types MoveType::Key and MoveType::ConstKey. + template + static bool moveToElementByKey(typename JSONParser::Element & element, const std::string_view & key) + { + if (!element.isObject()) + return false; + auto object = element.getObject(); + return object.find(key, element); + } + + static size_t calculateMaxSize(const ColumnString::Offsets & offsets) + { + size_t max_size = 0; + for (const auto i : collections::range(0, offsets.size())) + { + size_t size = offsets[i] - offsets[i - 1]; + if (max_size < size) + max_size = size; + } + if (max_size) + --max_size; + return max_size; + } + +}; + + +template typename Impl> +class ExecutableFunctionJSON : public IExecutableFunction, WithContext +{ + +public: + explicit ExecutableFunctionJSON(const NullPresence & null_presence_, bool allow_simdjson_, const DataTypePtr & json_return_type_) + : null_presence(null_presence_), allow_simdjson(allow_simdjson_), json_return_type(json_return_type_) + { + } + + String getName() const override { return Name::name; } + bool useDefaultImplementationForNulls() const override { return false; } + bool useDefaultImplementationForConstants() const override { return true; } + + ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) const override + { + if (null_presence.has_null_constant) + return result_type->createColumnConstWithDefaultValue(input_rows_count); + + ColumnsWithTypeAndName temporary_columns = null_presence.has_nullable ? createBlockWithNestedColumns(arguments) : arguments; + ColumnPtr temporary_result = chooseAndRunJSONParser(temporary_columns, json_return_type, input_rows_count); + if (null_presence.has_nullable) + return wrapInNullable(temporary_result, arguments, result_type, input_rows_count); + return temporary_result; + } + +private: + + ColumnPtr + chooseAndRunJSONParser(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) const + { +#if USE_SIMDJSON + if (allow_simdjson) + return FunctionJSONHelpers::Executor::run(arguments, result_type, input_rows_count); +#endif + +#if USE_RAPIDJSON + return FunctionJSONHelpers::Executor::run(arguments, result_type, input_rows_count); +#else + return FunctionJSONHelpers::Executor::run(arguments, result_type, input_rows_count); +#endif + } + + NullPresence null_presence; + bool allow_simdjson; + DataTypePtr json_return_type; +}; + + +template typename Impl> +class FunctionBaseFunctionJSON : public IFunctionBase +{ +public: + explicit FunctionBaseFunctionJSON( + const NullPresence & null_presence_, + bool allow_simdjson_, + DataTypes argument_types_, + DataTypePtr return_type_, + DataTypePtr json_return_type_) + : null_presence(null_presence_) + , allow_simdjson(allow_simdjson_) + , argument_types(std::move(argument_types_)) + , return_type(std::move(return_type_)) + , json_return_type(std::move(json_return_type_)) + { + } + + String getName() const override { return Name::name; } + + const DataTypes & getArgumentTypes() const override + { + return argument_types; + } + + const DataTypePtr & getResultType() const override + { + return return_type; + } + + bool isSuitableForShortCircuitArgumentsExecution(const DataTypesWithConstInfo & /*arguments*/) const override { return true; } + + ExecutableFunctionPtr prepare(const ColumnsWithTypeAndName &) const override + { + return std::make_unique>(null_presence, allow_simdjson, json_return_type); + } + +private: + NullPresence null_presence; + bool allow_simdjson; + DataTypes argument_types; + DataTypePtr return_type; + DataTypePtr json_return_type; +}; + + +/// We use IFunctionOverloadResolver instead of IFunction to handle non-default NULL processing. +/// Both NULL and JSON NULL should generate NULL value. If any argument is NULL, return NULL. +template typename Impl> +class JSONOverloadResolver : public IFunctionOverloadResolver, WithContext +{ +public: + static constexpr auto name = Name::name; + + String getName() const override { return name; } + + static FunctionOverloadResolverPtr create(ContextPtr context_) + { + return std::make_unique(context_); + } + + explicit JSONOverloadResolver(ContextPtr context_) : WithContext(context_) {} + + bool isVariadic() const override { return true; } + size_t getNumberOfArguments() const override { return 0; } + bool useDefaultImplementationForNulls() const override { return false; } + + FunctionBasePtr build(const ColumnsWithTypeAndName & arguments) const override + { + DataTypePtr json_return_type = Impl::getReturnType(Name::name, createBlockWithNestedColumns(arguments)); + NullPresence null_presence = getNullPresense(arguments); + DataTypePtr return_type; + if (null_presence.has_null_constant) + return_type = makeNullable(std::make_shared()); + else if (null_presence.has_nullable) + return_type = makeNullable(json_return_type); + else + return_type = json_return_type; + + /// Top-level LowCardinality columns are processed outside JSON parser. + json_return_type = removeLowCardinality(json_return_type); + + DataTypes argument_types; + argument_types.reserve(arguments.size()); + for (const auto & argument : arguments) + argument_types.emplace_back(argument.type); + return std::make_unique>( + null_presence, getContext()->getSettingsRef().allow_simdjson, argument_types, return_type, json_return_type); + } +}; + + +struct NameJSONHas { static constexpr auto name{"JSONHas"}; }; +struct NameIsValidJSON { static constexpr auto name{"isValidJSON"}; }; +struct NameJSONLength { static constexpr auto name{"JSONLength"}; }; +struct NameJSONKey { static constexpr auto name{"JSONKey"}; }; +struct NameJSONType { static constexpr auto name{"JSONType"}; }; +struct NameJSONExtractInt { static constexpr auto name{"JSONExtractInt"}; }; +struct NameJSONExtractUInt { static constexpr auto name{"JSONExtractUInt"}; }; +struct NameJSONExtractFloat { static constexpr auto name{"JSONExtractFloat"}; }; +struct NameJSONExtractBool { static constexpr auto name{"JSONExtractBool"}; }; +struct NameJSONExtractString { static constexpr auto name{"JSONExtractString"}; }; +struct NameJSONExtract { static constexpr auto name{"JSONExtract"}; }; +struct NameJSONExtractKeysAndValues { static constexpr auto name{"JSONExtractKeysAndValues"}; }; +struct NameJSONExtractRaw { static constexpr auto name{"JSONExtractRaw"}; }; +struct NameJSONExtractArrayRaw { static constexpr auto name{"JSONExtractArrayRaw"}; }; +struct NameJSONExtractKeysAndValuesRaw { static constexpr auto name{"JSONExtractKeysAndValuesRaw"}; }; +struct NameJSONExtractKeys { static constexpr auto name{"JSONExtractKeys"}; }; + + +template +class JSONHasImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) { return std::make_shared(); } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element &, const std::string_view &) + { + ColumnVector & col_vec = assert_cast &>(dest); + col_vec.insertValue(1); + return true; + } +}; + + +template +class IsValidJSONImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char * function_name, const ColumnsWithTypeAndName & arguments) + { + if (arguments.size() != 1) + { + /// IsValidJSON() shouldn't get parameters other than JSON. + throw Exception{"Function " + String(function_name) + " needs exactly one argument", + ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; + } + return std::make_shared(); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName &) { return 0; } + + static bool insertResultToColumn(IColumn & dest, const Element &, const std::string_view &) + { + /// This function is called only if JSON is valid. + /// If JSON isn't valid then `FunctionJSON::Executor::run()` adds default value (=zero) to `dest` without calling this function. + ColumnVector & col_vec = assert_cast &>(dest); + col_vec.insertValue(1); + return true; + } +}; + + +template +class JSONLengthImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_shared(); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + size_t size; + if (element.isArray()) + size = element.getArray().size(); + else if (element.isObject()) + size = element.getObject().size(); + else + return false; + + ColumnVector & col_vec = assert_cast &>(dest); + col_vec.insertValue(size); + return true; + } +}; + + +template +class JSONKeyImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_shared(); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element &, const std::string_view & last_key) + { + if (last_key.empty()) + return false; + ColumnString & col_str = assert_cast(dest); + col_str.insertData(last_key.data(), last_key.size()); + return true; + } +}; + + +template +class JSONTypeImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + static const std::vector> values = { + {"Array", '['}, + {"Object", '{'}, + {"String", '"'}, + {"Int64", 'i'}, + {"UInt64", 'u'}, + {"Double", 'd'}, + {"Bool", 'b'}, + {"Null", 0}, /// the default value for the column. + }; + return std::make_shared>(values); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + UInt8 type; + if (element.isInt64()) + type = 'i'; + else if (element.isUInt64()) + type = 'u'; + else if (element.isDouble()) + type = 'd'; + else if (element.isBool()) + type = 'b'; + else if (element.isString()) + type = '"'; + else if (element.isArray()) + type = '['; + else if (element.isObject()) + type = '{'; + else if (element.isNull()) + type = 0; + else + return false; + + ColumnVector & col_vec = assert_cast &>(dest); + col_vec.insertValue(type); + return true; + } +}; + + +template +class JSONExtractNumericImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_shared>(); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + NumberType value; + + if (element.isInt64()) + { + if (!accurate::convertNumeric(element.getInt64(), value)) + return false; + } + else if (element.isUInt64()) + { + if (!accurate::convertNumeric(element.getUInt64(), value)) + return false; + } + else if (element.isDouble()) + { + if constexpr (std::is_floating_point_v) + { + /// We permit inaccurate conversion of double to float. + /// Example: double 0.1 from JSON is not representable in float. + /// But it will be more convenient for user to perform conversion. + value = element.getDouble(); + } + else if (!accurate::convertNumeric(element.getDouble(), value)) + return false; + } + else if (element.isBool() && is_integer && convert_bool_to_integer) + { + value = static_cast(element.getBool()); } else + return false; + + auto & col_vec = assert_cast &>(dest); + col_vec.insertValue(value); + return true; + } +}; + + +template +using JSONExtractInt8Impl = JSONExtractNumericImpl; +template +using JSONExtractUInt8Impl = JSONExtractNumericImpl; +template +using JSONExtractInt16Impl = JSONExtractNumericImpl; +template +using JSONExtractUInt16Impl = JSONExtractNumericImpl; +template +using JSONExtractInt32Impl = JSONExtractNumericImpl; +template +using JSONExtractUInt32Impl = JSONExtractNumericImpl; +template +using JSONExtractInt64Impl = JSONExtractNumericImpl; +template +using JSONExtractUInt64Impl = JSONExtractNumericImpl; +template +using JSONExtractFloat32Impl = JSONExtractNumericImpl; +template +using JSONExtractFloat64Impl = JSONExtractNumericImpl; + + +template +class JSONExtractBoolImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_shared(); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + if (!element.isBool()) + return false; + + auto & col_vec = assert_cast &>(dest); + col_vec.insertValue(static_cast(element.getBool())); + return true; + } +}; + + +template +class JSONExtractStringImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_shared(); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + if (!element.isString()) + return false; + + auto str = element.getString(); + ColumnString & col_str = assert_cast(dest); + col_str.insertData(str.data(), str.size()); + return true; + } +}; + +template +class JSONExtractRawImpl; + +/// Nodes of the extract tree. We need the extract tree to extract from JSON complex values containing array, tuples or nullables. +template +struct JSONExtractTree +{ + using Element = typename JSONParser::Element; + + class Node + { + public: + Node() = default; + virtual ~Node() = default; + virtual bool insertResultToColumn(IColumn &, const Element &) = 0; + }; + + template + class NumericNode : public Node + { + public: + bool insertResultToColumn(IColumn & dest, const Element & element) override { - if (isString(column.type)) - moves.emplace_back(MoveType::Key, ""); + return JSONExtractNumericImpl::insertResultToColumn(dest, element, {}); + } + }; + + class LowCardinalityNode : public Node + { + public: + LowCardinalityNode(DataTypePtr dictionary_type_, std::unique_ptr impl_) + : dictionary_type(dictionary_type_), impl(std::move(impl_)) {} + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + auto from_col = dictionary_type->createColumn(); + if (impl->insertResultToColumn(*from_col, element)) + { + StringRef value = from_col->getDataAt(0); + assert_cast(dest).insertData(value.data, value.size); + return true; + } + return false; + } + private: + DataTypePtr dictionary_type; + std::unique_ptr impl; + }; + + class UUIDNode : public Node + { + public: + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + if (!element.isString()) + return false; + + auto uuid = parseFromString(element.getString()); + assert_cast(dest).insert(uuid); + return true; + } + }; + + template + class DecimalNode : public Node + { + public: + DecimalNode(DataTypePtr data_type_) : data_type(data_type_) {} + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + if (!element.isDouble()) + return false; + + const auto * type = assert_cast *>(data_type.get()); + auto result = convertToDecimal, DataTypeDecimal>(element.getDouble(), type->getScale()); + assert_cast &>(dest).insert(result); + return true; + } + private: + DataTypePtr data_type; + }; + + class StringNode : public Node + { + public: + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + if (element.isString()) + return JSONExtractStringImpl::insertResultToColumn(dest, element, {}); + else if (element.isNull()) + return false; else - moves.emplace_back(MoveType::Index, 0); + return JSONExtractRawImpl::insertResultToColumn(dest, element, {}); + } + }; + + class FixedStringNode : public Node + { + public: + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + if (!element.isString()) + return false; + auto & col_str = assert_cast(dest); + auto str = element.getString(); + if (str.size() > col_str.getN()) + return false; + col_str.insertData(str.data(), str.size()); + return true; + } + }; + + template + class EnumNode : public Node + { + public: + EnumNode(const std::vector> & name_value_pairs_) : name_value_pairs(name_value_pairs_) + { + for (const auto & name_value_pair : name_value_pairs) + { + name_to_value_map.emplace(name_value_pair.first, name_value_pair.second); + only_values.emplace(name_value_pair.second); + } + } + + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + auto & col_vec = assert_cast &>(dest); + + if (element.isInt64()) + { + Type value; + if (!accurate::convertNumeric(element.getInt64(), value) || !only_values.count(value)) + return false; + col_vec.insertValue(value); + return true; + } + + if (element.isUInt64()) + { + Type value; + if (!accurate::convertNumeric(element.getUInt64(), value) || !only_values.count(value)) + return false; + col_vec.insertValue(value); + return true; + } + + if (element.isString()) + { + auto value = name_to_value_map.find(element.getString()); + if (value == name_to_value_map.end()) + return false; + col_vec.insertValue(value->second); + return true; + } + + return false; + } + + private: + std::vector> name_value_pairs; + std::unordered_map name_to_value_map; + std::unordered_set only_values; + }; + + class NullableNode : public Node + { + public: + NullableNode(std::unique_ptr nested_) : nested(std::move(nested_)) {} + + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + ColumnNullable & col_null = assert_cast(dest); + if (!nested->insertResultToColumn(col_null.getNestedColumn(), element)) + return false; + col_null.getNullMapColumn().insertValue(0); + return true; + } + + private: + std::unique_ptr nested; + }; + + class ArrayNode : public Node + { + public: + ArrayNode(std::unique_ptr nested_) : nested(std::move(nested_)) {} + + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + if (!element.isArray()) + return false; + + auto array = element.getArray(); + + ColumnArray & col_arr = assert_cast(dest); + auto & data = col_arr.getData(); + size_t old_size = data.size(); + bool were_valid_elements = false; + + for (auto value : array) + { + if (nested->insertResultToColumn(data, value)) + were_valid_elements = true; + else + data.insertDefault(); + } + + if (!were_valid_elements) + { + data.popBack(data.size() - old_size); + return false; + } + + col_arr.getOffsets().push_back(data.size()); + return true; + } + + private: + std::unique_ptr nested; + }; + + class TupleNode : public Node + { + public: + TupleNode(std::vector> nested_, const std::vector & explicit_names_) : nested(std::move(nested_)), explicit_names(explicit_names_) + { + for (size_t i = 0; i != explicit_names.size(); ++i) + name_to_index_map.emplace(explicit_names[i], i); + } + + bool insertResultToColumn(IColumn & dest, const Element & element) override + { + ColumnTuple & tuple = assert_cast(dest); + size_t old_size = dest.size(); + bool were_valid_elements = false; + + auto set_size = [&](size_t size) + { + for (size_t i = 0; i != tuple.tupleSize(); ++i) + { + auto & col = tuple.getColumn(i); + if (col.size() != size) + { + if (col.size() > size) + col.popBack(col.size() - size); + else + while (col.size() < size) + col.insertDefault(); + } + } + }; + + if (element.isArray()) + { + auto array = element.getArray(); + auto it = array.begin(); + + for (size_t index = 0; (index != nested.size()) && (it != array.end()); ++index) + { + if (nested[index]->insertResultToColumn(tuple.getColumn(index), *it++)) + were_valid_elements = true; + else + tuple.getColumn(index).insertDefault(); + } + + set_size(old_size + static_cast(were_valid_elements)); + return were_valid_elements; + } + + if (element.isObject()) + { + auto object = element.getObject(); + if (name_to_index_map.empty()) + { + auto it = object.begin(); + for (size_t index = 0; (index != nested.size()) && (it != object.end()); ++index) + { + if (nested[index]->insertResultToColumn(tuple.getColumn(index), (*it++).second)) + were_valid_elements = true; + else + tuple.getColumn(index).insertDefault(); + } + } + else + { + for (const auto & [key, value] : object) + { + auto index = name_to_index_map.find(key); + if (index != name_to_index_map.end()) + { + if (nested[index->second]->insertResultToColumn(tuple.getColumn(index->second), value)) + were_valid_elements = true; + } + } + } + + set_size(old_size + static_cast(were_valid_elements)); + return were_valid_elements; + } + + return false; + } + + private: + std::vector> nested; + std::vector explicit_names; + std::unordered_map name_to_index_map; + }; + + static std::unique_ptr build(const char * function_name, const DataTypePtr & type) + { + switch (type->getTypeId()) + { + case TypeIndex::UInt8: return std::make_unique>(); + case TypeIndex::UInt16: return std::make_unique>(); + case TypeIndex::UInt32: return std::make_unique>(); + case TypeIndex::UInt64: return std::make_unique>(); + case TypeIndex::Int8: return std::make_unique>(); + case TypeIndex::Int16: return std::make_unique>(); + case TypeIndex::Int32: return std::make_unique>(); + case TypeIndex::Int64: return std::make_unique>(); + case TypeIndex::Float32: return std::make_unique>(); + case TypeIndex::Float64: return std::make_unique>(); + case TypeIndex::String: return std::make_unique(); + case TypeIndex::FixedString: return std::make_unique(); + case TypeIndex::UUID: return std::make_unique(); + case TypeIndex::LowCardinality: + { + auto dictionary_type = typeid_cast(type.get())->getDictionaryType(); + auto impl = build(function_name, dictionary_type); + return std::make_unique(dictionary_type, std::move(impl)); + } + case TypeIndex::Decimal256: return std::make_unique>(type); + case TypeIndex::Decimal128: return std::make_unique>(type); + case TypeIndex::Decimal64: return std::make_unique>(type); + case TypeIndex::Decimal32: return std::make_unique>(type); + case TypeIndex::Enum8: + return std::make_unique>(static_cast(*type).getValues()); + case TypeIndex::Enum16: + return std::make_unique>(static_cast(*type).getValues()); + case TypeIndex::Nullable: + { + return std::make_unique(build(function_name, static_cast(*type).getNestedType())); + } + case TypeIndex::Array: + { + return std::make_unique(build(function_name, static_cast(*type).getNestedType())); + } + case TypeIndex::Tuple: + { + const auto & tuple = static_cast(*type); + const auto & tuple_elements = tuple.getElements(); + std::vector> elements; + for (const auto & tuple_element : tuple_elements) + elements.emplace_back(build(function_name, tuple_element)); + return std::make_unique(std::move(elements), tuple.haveExplicitNames() ? tuple.getElementNames() : Strings{}); + } + default: + throw Exception{"Function " + String(function_name) + " doesn't support the return type schema: " + type->getName(), ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT}; } } - return moves; -} +}; -size_t FunctionJSONHelpers::calculateMaxSize(const ColumnString::Offsets & offsets) + +template +class JSONExtractImpl { - size_t max_size = 0; - for (const auto i : collections::range(0, offsets.size())) - { - size_t size = offsets[i] - offsets[i - 1]; - if (max_size < size) - max_size = size; - } - if (max_size) - --max_size; - return max_size; -} +public: + using Element = typename JSONParser::Element; + static DataTypePtr getReturnType(const char * function_name, const ColumnsWithTypeAndName & arguments) + { + if (arguments.size() < 2) + throw Exception{"Function " + String(function_name) + " requires at least two arguments", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; + + const auto & col = arguments.back(); + auto col_type_const = typeid_cast(col.column.get()); + if (!col_type_const || !isString(col.type)) + throw Exception{"The last argument of function " + String(function_name) + + " should be a constant string specifying the return data type, illegal value: " + col.name, + ErrorCodes::ILLEGAL_COLUMN}; + + return DataTypeFactory::instance().get(col_type_const->getValue()); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 2; } + + void prepare(const char * function_name, const ColumnsWithTypeAndName &, const DataTypePtr & result_type) + { + extract_tree = JSONExtractTree::build(function_name, result_type); + } + + bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + return extract_tree->insertResultToColumn(dest, element); + } + +protected: + std::unique_ptr::Node> extract_tree; +}; + + +template +class JSONExtractKeysAndValuesImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char * function_name, const ColumnsWithTypeAndName & arguments) + { + if (arguments.size() < 2) + throw Exception{"Function " + String(function_name) + " requires at least two arguments", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; + + const auto & col = arguments.back(); + auto col_type_const = typeid_cast(col.column.get()); + if (!col_type_const || !isString(col.type)) + throw Exception{"The last argument of function " + String(function_name) + + " should be a constant string specifying the values' data type, illegal value: " + col.name, + ErrorCodes::ILLEGAL_COLUMN}; + + DataTypePtr key_type = std::make_unique(); + DataTypePtr value_type = DataTypeFactory::instance().get(col_type_const->getValue()); + DataTypePtr tuple_type = std::make_unique(DataTypes{key_type, value_type}); + return std::make_unique(tuple_type); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 2; } + + void prepare(const char * function_name, const ColumnsWithTypeAndName &, const DataTypePtr & result_type) + { + const auto tuple_type = typeid_cast(result_type.get())->getNestedType(); + const auto value_type = typeid_cast(tuple_type.get())->getElements()[1]; + extract_tree = JSONExtractTree::build(function_name, value_type); + } + + bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + if (!element.isObject()) + return false; + + auto object = element.getObject(); + + auto & col_arr = assert_cast(dest); + auto & col_tuple = assert_cast(col_arr.getData()); + size_t old_size = col_tuple.size(); + auto & col_key = assert_cast(col_tuple.getColumn(0)); + auto & col_value = col_tuple.getColumn(1); + + for (const auto & [key, value] : object) + { + if (extract_tree->insertResultToColumn(col_value, value)) + col_key.insertData(key.data(), key.size()); + } + + if (col_tuple.size() == old_size) + return false; + + col_arr.getOffsets().push_back(col_tuple.size()); + return true; + } + +private: + std::unique_ptr::Node> extract_tree; +}; + + +template +class JSONExtractRawImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_shared(); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + ColumnString & col_str = assert_cast(dest); + auto & chars = col_str.getChars(); + WriteBufferFromVector buf(chars, WriteBufferFromVector::AppendModeTag()); + traverse(element, buf); + buf.finalize(); + chars.push_back(0); + col_str.getOffsets().push_back(chars.size()); + return true; + } + +private: + static void traverse(const Element & element, WriteBuffer & buf) + { + if (element.isInt64()) + { + writeIntText(element.getInt64(), buf); + return; + } + if (element.isUInt64()) + { + writeIntText(element.getUInt64(), buf); + return; + } + if (element.isDouble()) + { + writeFloatText(element.getDouble(), buf); + return; + } + if (element.isBool()) + { + if (element.getBool()) + writeCString("true", buf); + else + writeCString("false", buf); + return; + } + if (element.isString()) + { + writeJSONString(element.getString(), buf, format_settings()); + return; + } + if (element.isArray()) + { + writeChar('[', buf); + bool need_comma = false; + for (auto value : element.getArray()) + { + if (std::exchange(need_comma, true)) + writeChar(',', buf); + traverse(value, buf); + } + writeChar(']', buf); + return; + } + if (element.isObject()) + { + writeChar('{', buf); + bool need_comma = false; + for (auto [key, value] : element.getObject()) + { + if (std::exchange(need_comma, true)) + writeChar(',', buf); + writeJSONString(key, buf, format_settings()); + writeChar(':', buf); + traverse(value, buf); + } + writeChar('}', buf); + return; + } + if (element.isNull()) + { + writeCString("null", buf); + return; + } + } + + static const FormatSettings & format_settings() + { + static const FormatSettings the_instance = [] + { + FormatSettings settings; + settings.json.escape_forward_slashes = false; + return settings; + }(); + return the_instance; + } +}; + + +template +class JSONExtractArrayRawImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_shared(std::make_shared()); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + if (!element.isArray()) + return false; + + auto array = element.getArray(); + ColumnArray & col_res = assert_cast(dest); + + for (auto value : array) + JSONExtractRawImpl::insertResultToColumn(col_res.getData(), value, {}); + + col_res.getOffsets().push_back(col_res.getOffsets().back() + array.size()); + return true; + } +}; + + +template +class JSONExtractKeysAndValuesRawImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + DataTypePtr string_type = std::make_unique(); + DataTypePtr tuple_type = std::make_unique(DataTypes{string_type, string_type}); + return std::make_unique(tuple_type); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + if (!element.isObject()) + return false; + + auto object = element.getObject(); + + auto & col_arr = assert_cast(dest); + auto & col_tuple = assert_cast(col_arr.getData()); + auto & col_key = assert_cast(col_tuple.getColumn(0)); + auto & col_value = assert_cast(col_tuple.getColumn(1)); + + for (const auto & [key, value] : object) + { + col_key.insertData(key.data(), key.size()); + JSONExtractRawImpl::insertResultToColumn(col_value, value, {}); + } + + col_arr.getOffsets().push_back(col_arr.getOffsets().back() + object.size()); + return true; + } +}; + +template +class JSONExtractKeysImpl +{ +public: + using Element = typename JSONParser::Element; + + static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) + { + return std::make_unique(std::make_shared()); + } + + static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } + + bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) + { + if (!element.isObject()) + return false; + + auto object = element.getObject(); + + ColumnArray & col_res = assert_cast(dest); + auto & col_key = assert_cast(col_res.getData()); + + for (const auto & [key, value] : object) + { + col_key.insertData(key.data(), key.size()); + } + + col_res.getOffsets().push_back(col_res.getOffsets().back() + object.size()); + return true; + } +}; void registerFunctionsJSON(FunctionFactory & factory) { diff --git a/src/Functions/FunctionsJSON.h b/src/Functions/FunctionsJSON.h deleted file mode 100644 index 7dd724dbfc9..00000000000 --- a/src/Functions/FunctionsJSON.h +++ /dev/null @@ -1,1390 +0,0 @@ -#pragma once - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -#if !defined(ARCADIA_BUILD) -# include "config_functions.h" -#endif - - -namespace DB -{ -namespace ErrorCodes -{ - extern const int ILLEGAL_COLUMN; - extern const int ILLEGAL_TYPE_OF_ARGUMENT; - extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH; -} - - -/// Functions to parse JSONs and extract values from it. -/// The first argument of all these functions gets a JSON, -/// after that there are any number of arguments specifying path to a desired part from the JSON's root. -/// For example, -/// select JSONExtractInt('{"a": "hello", "b": [-100, 200.0, 300]}', 'b', 1) = -100 - -class FunctionJSONHelpers -{ -public: - template typename Impl, class JSONParser> - class Executor - { - public: - static ColumnPtr run(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) - { - MutableColumnPtr to{result_type->createColumn()}; - to->reserve(input_rows_count); - - if (arguments.empty()) - throw Exception{"Function " + String(Name::name) + " requires at least one argument", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; - - const auto & first_column = arguments[0]; - if (!isString(first_column.type)) - throw Exception{"The first argument of function " + String(Name::name) + " should be a string containing JSON, illegal type: " + first_column.type->getName(), - ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT}; - - const ColumnPtr & arg_json = first_column.column; - const auto * col_json_const = typeid_cast(arg_json.get()); - const auto * col_json_string - = typeid_cast(col_json_const ? col_json_const->getDataColumnPtr().get() : arg_json.get()); - - if (!col_json_string) - throw Exception{"Illegal column " + arg_json->getName(), ErrorCodes::ILLEGAL_COLUMN}; - - const ColumnString::Chars & chars = col_json_string->getChars(); - const ColumnString::Offsets & offsets = col_json_string->getOffsets(); - - size_t num_index_arguments = Impl::getNumberOfIndexArguments(arguments); - std::vector moves = prepareMoves(Name::name, arguments, 1, num_index_arguments); - - /// Preallocate memory in parser if necessary. - JSONParser parser; - if constexpr (has_member_function_reserve::value) - { - size_t max_size = calculateMaxSize(offsets); - if (max_size) - parser.reserve(max_size); - } - - Impl impl; - - /// prepare() does Impl-specific preparation before handling each row. - if constexpr (has_member_function_prepare::*)(const char *, const ColumnsWithTypeAndName &, const DataTypePtr &)>::value) - impl.prepare(Name::name, arguments, result_type); - - using Element = typename JSONParser::Element; - - Element document; - bool document_ok = false; - if (col_json_const) - { - std::string_view json{reinterpret_cast(&chars[0]), offsets[0] - 1}; - document_ok = parser.parse(json, document); - } - - for (const auto i : collections::range(0, input_rows_count)) - { - if (!col_json_const) - { - std::string_view json{reinterpret_cast(&chars[offsets[i - 1]]), offsets[i] - offsets[i - 1] - 1}; - document_ok = parser.parse(json, document); - } - - bool added_to_column = false; - if (document_ok) - { - /// Perform moves. - Element element; - std::string_view last_key; - bool moves_ok = performMoves(arguments, i, document, moves, element, last_key); - - if (moves_ok) - added_to_column = impl.insertResultToColumn(*to, element, last_key); - } - - /// We add default value (=null or zero) if something goes wrong, we don't throw exceptions in these JSON functions. - if (!added_to_column) - to->insertDefault(); - } - return to; - } - }; - -private: - BOOST_TTI_HAS_MEMBER_FUNCTION(reserve) - BOOST_TTI_HAS_MEMBER_FUNCTION(prepare) - - template - struct has_index_operator : std::false_type {}; - - template - struct has_index_operator()[0])>> : std::true_type {}; - - /// Represents a move of a JSON iterator described by a single argument passed to a JSON function. - /// For example, the call JSONExtractInt('{"a": "hello", "b": [-100, 200.0, 300]}', 'b', 1) - /// contains two moves: {MoveType::ConstKey, "b"} and {MoveType::ConstIndex, 1}. - /// Keys and indices can be nonconst, in this case they are calculated for each row. - enum class MoveType - { - Key, - Index, - ConstKey, - ConstIndex, - }; - - struct Move - { - Move(MoveType type_, size_t index_ = 0) : type(type_), index(index_) {} - Move(MoveType type_, const String & key_) : type(type_), key(key_) {} - MoveType type; - size_t index = 0; - String key; - }; - - static std::vector prepareMoves(const char * function_name, const ColumnsWithTypeAndName & columns, size_t first_index_argument, size_t num_index_arguments); - - /// Performs moves of types MoveType::Index and MoveType::ConstIndex. - template - static bool performMoves(const ColumnsWithTypeAndName & arguments, size_t row, - const typename JSONParser::Element & document, const std::vector & moves, - typename JSONParser::Element & element, std::string_view & last_key) - { - typename JSONParser::Element res_element = document; - std::string_view key; - - for (size_t j = 0; j != moves.size(); ++j) - { - switch (moves[j].type) - { - case MoveType::ConstIndex: - { - if (!moveToElementByIndex(res_element, moves[j].index, key)) - return false; - break; - } - case MoveType::ConstKey: - { - key = moves[j].key; - if (!moveToElementByKey(res_element, key)) - return false; - break; - } - case MoveType::Index: - { - Int64 index = (*arguments[j + 1].column)[row].get(); - if (!moveToElementByIndex(res_element, index, key)) - return false; - break; - } - case MoveType::Key: - { - key = std::string_view{(*arguments[j + 1].column).getDataAt(row)}; - if (!moveToElementByKey(res_element, key)) - return false; - break; - } - } - } - - element = res_element; - last_key = key; - return true; - } - - template - static bool moveToElementByIndex(typename JSONParser::Element & element, int index, std::string_view & out_key) - { - if (element.isArray()) - { - auto array = element.getArray(); - if (index >= 0) - --index; - else - index += array.size(); - - if (static_cast(index) >= array.size()) - return false; - element = array[index]; - out_key = {}; - return true; - } - - if constexpr (has_index_operator::value) - { - if (element.isObject()) - { - auto object = element.getObject(); - if (index >= 0) - --index; - else - index += object.size(); - - if (static_cast(index) >= object.size()) - return false; - std::tie(out_key, element) = object[index]; - return true; - } - } - - return {}; - } - - /// Performs moves of types MoveType::Key and MoveType::ConstKey. - template - static bool moveToElementByKey(typename JSONParser::Element & element, const std::string_view & key) - { - if (!element.isObject()) - return false; - auto object = element.getObject(); - return object.find(key, element); - } - - static size_t calculateMaxSize(const ColumnString::Offsets & offsets); -}; - - -template typename Impl> -class ExecutableFunctionJSON : public IExecutableFunction, WithContext -{ - -public: - explicit ExecutableFunctionJSON(const NullPresence & null_presence_, bool allow_simdjson_, const DataTypePtr & json_return_type_) - : null_presence(null_presence_), allow_simdjson(allow_simdjson_), json_return_type(json_return_type_) - { - } - - String getName() const override { return Name::name; } - bool useDefaultImplementationForNulls() const override { return false; } - bool useDefaultImplementationForConstants() const override { return true; } - - ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) const override - { - if (null_presence.has_null_constant) - return result_type->createColumnConstWithDefaultValue(input_rows_count); - - ColumnsWithTypeAndName temporary_columns = null_presence.has_nullable ? createBlockWithNestedColumns(arguments) : arguments; - ColumnPtr temporary_result = chooseAndRunJSONParser(temporary_columns, json_return_type, input_rows_count); - if (null_presence.has_nullable) - return wrapInNullable(temporary_result, arguments, result_type, input_rows_count); - return temporary_result; - } - -private: - - ColumnPtr - chooseAndRunJSONParser(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) const - { -#if USE_SIMDJSON - if (allow_simdjson) - return FunctionJSONHelpers::Executor::run(arguments, result_type, input_rows_count); -#endif - -#if USE_RAPIDJSON - return FunctionJSONHelpers::Executor::run(arguments, result_type, input_rows_count); -#else - return FunctionJSONHelpers::Executor::run(arguments, result_type, input_rows_count); -#endif - } - - NullPresence null_presence; - bool allow_simdjson; - DataTypePtr json_return_type; -}; - - -template typename Impl> -class FunctionBaseFunctionJSON : public IFunctionBase -{ -public: - explicit FunctionBaseFunctionJSON( - const NullPresence & null_presence_, - bool allow_simdjson_, - DataTypes argument_types_, - DataTypePtr return_type_, - DataTypePtr json_return_type_) - : null_presence(null_presence_) - , allow_simdjson(allow_simdjson_) - , argument_types(std::move(argument_types_)) - , return_type(std::move(return_type_)) - , json_return_type(std::move(json_return_type_)) - { - } - - String getName() const override { return Name::name; } - - const DataTypes & getArgumentTypes() const override - { - return argument_types; - } - - const DataTypePtr & getResultType() const override - { - return return_type; - } - - bool isSuitableForShortCircuitArgumentsExecution(const DataTypesWithConstInfo & /*arguments*/) const override { return true; } - - ExecutableFunctionPtr prepare(const ColumnsWithTypeAndName &) const override - { - return std::make_unique>(null_presence, allow_simdjson, json_return_type); - } - -private: - NullPresence null_presence; - bool allow_simdjson; - DataTypes argument_types; - DataTypePtr return_type; - DataTypePtr json_return_type; -}; - - -/// We use IFunctionOverloadResolver instead of IFunction to handle non-default NULL processing. -/// Both NULL and JSON NULL should generate NULL value. If any argument is NULL, return NULL. -template typename Impl> -class JSONOverloadResolver : public IFunctionOverloadResolver, WithContext -{ -public: - static constexpr auto name = Name::name; - - String getName() const override { return name; } - - static FunctionOverloadResolverPtr create(ContextPtr context_) - { - return std::make_unique(context_); - } - - explicit JSONOverloadResolver(ContextPtr context_) : WithContext(context_) {} - - bool isVariadic() const override { return true; } - size_t getNumberOfArguments() const override { return 0; } - bool useDefaultImplementationForNulls() const override { return false; } - - FunctionBasePtr build(const ColumnsWithTypeAndName & arguments) const override - { - DataTypePtr json_return_type = Impl::getReturnType(Name::name, createBlockWithNestedColumns(arguments)); - NullPresence null_presence = getNullPresense(arguments); - DataTypePtr return_type; - if (null_presence.has_null_constant) - return_type = makeNullable(std::make_shared()); - else if (null_presence.has_nullable) - return_type = makeNullable(json_return_type); - else - return_type = json_return_type; - - /// Top-level LowCardinality columns are processed outside JSON parser. - json_return_type = removeLowCardinality(json_return_type); - - DataTypes argument_types; - argument_types.reserve(arguments.size()); - for (const auto & argument : arguments) - argument_types.emplace_back(argument.type); - return std::make_unique>( - null_presence, getContext()->getSettingsRef().allow_simdjson, argument_types, return_type, json_return_type); - } -}; - - -struct NameJSONHas { static constexpr auto name{"JSONHas"}; }; -struct NameIsValidJSON { static constexpr auto name{"isValidJSON"}; }; -struct NameJSONLength { static constexpr auto name{"JSONLength"}; }; -struct NameJSONKey { static constexpr auto name{"JSONKey"}; }; -struct NameJSONType { static constexpr auto name{"JSONType"}; }; -struct NameJSONExtractInt { static constexpr auto name{"JSONExtractInt"}; }; -struct NameJSONExtractUInt { static constexpr auto name{"JSONExtractUInt"}; }; -struct NameJSONExtractFloat { static constexpr auto name{"JSONExtractFloat"}; }; -struct NameJSONExtractBool { static constexpr auto name{"JSONExtractBool"}; }; -struct NameJSONExtractString { static constexpr auto name{"JSONExtractString"}; }; -struct NameJSONExtract { static constexpr auto name{"JSONExtract"}; }; -struct NameJSONExtractKeysAndValues { static constexpr auto name{"JSONExtractKeysAndValues"}; }; -struct NameJSONExtractRaw { static constexpr auto name{"JSONExtractRaw"}; }; -struct NameJSONExtractArrayRaw { static constexpr auto name{"JSONExtractArrayRaw"}; }; -struct NameJSONExtractKeysAndValuesRaw { static constexpr auto name{"JSONExtractKeysAndValuesRaw"}; }; -struct NameJSONExtractKeys { static constexpr auto name{"JSONExtractKeys"}; }; - - -template -class JSONHasImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) { return std::make_shared(); } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element &, const std::string_view &) - { - ColumnVector & col_vec = assert_cast &>(dest); - col_vec.insertValue(1); - return true; - } -}; - - -template -class IsValidJSONImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char * function_name, const ColumnsWithTypeAndName & arguments) - { - if (arguments.size() != 1) - { - /// IsValidJSON() shouldn't get parameters other than JSON. - throw Exception{"Function " + String(function_name) + " needs exactly one argument", - ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; - } - return std::make_shared(); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName &) { return 0; } - - static bool insertResultToColumn(IColumn & dest, const Element &, const std::string_view &) - { - /// This function is called only if JSON is valid. - /// If JSON isn't valid then `FunctionJSON::Executor::run()` adds default value (=zero) to `dest` without calling this function. - ColumnVector & col_vec = assert_cast &>(dest); - col_vec.insertValue(1); - return true; - } -}; - - -template -class JSONLengthImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_shared(); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - size_t size; - if (element.isArray()) - size = element.getArray().size(); - else if (element.isObject()) - size = element.getObject().size(); - else - return false; - - ColumnVector & col_vec = assert_cast &>(dest); - col_vec.insertValue(size); - return true; - } -}; - - -template -class JSONKeyImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_shared(); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element &, const std::string_view & last_key) - { - if (last_key.empty()) - return false; - ColumnString & col_str = assert_cast(dest); - col_str.insertData(last_key.data(), last_key.size()); - return true; - } -}; - - -template -class JSONTypeImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - static const std::vector> values = { - {"Array", '['}, - {"Object", '{'}, - {"String", '"'}, - {"Int64", 'i'}, - {"UInt64", 'u'}, - {"Double", 'd'}, - {"Bool", 'b'}, - {"Null", 0}, /// the default value for the column. - }; - return std::make_shared>(values); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - UInt8 type; - if (element.isInt64()) - type = 'i'; - else if (element.isUInt64()) - type = 'u'; - else if (element.isDouble()) - type = 'd'; - else if (element.isBool()) - type = 'b'; - else if (element.isString()) - type = '"'; - else if (element.isArray()) - type = '['; - else if (element.isObject()) - type = '{'; - else if (element.isNull()) - type = 0; - else - return false; - - ColumnVector & col_vec = assert_cast &>(dest); - col_vec.insertValue(type); - return true; - } -}; - - -template -class JSONExtractNumericImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_shared>(); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - NumberType value; - - if (element.isInt64()) - { - if (!accurate::convertNumeric(element.getInt64(), value)) - return false; - } - else if (element.isUInt64()) - { - if (!accurate::convertNumeric(element.getUInt64(), value)) - return false; - } - else if (element.isDouble()) - { - if constexpr (std::is_floating_point_v) - { - /// We permit inaccurate conversion of double to float. - /// Example: double 0.1 from JSON is not representable in float. - /// But it will be more convenient for user to perform conversion. - value = element.getDouble(); - } - else if (!accurate::convertNumeric(element.getDouble(), value)) - return false; - } - else if (element.isBool() && is_integer && convert_bool_to_integer) - { - value = static_cast(element.getBool()); - } - else - return false; - - auto & col_vec = assert_cast &>(dest); - col_vec.insertValue(value); - return true; - } -}; - - -template -using JSONExtractInt8Impl = JSONExtractNumericImpl; -template -using JSONExtractUInt8Impl = JSONExtractNumericImpl; -template -using JSONExtractInt16Impl = JSONExtractNumericImpl; -template -using JSONExtractUInt16Impl = JSONExtractNumericImpl; -template -using JSONExtractInt32Impl = JSONExtractNumericImpl; -template -using JSONExtractUInt32Impl = JSONExtractNumericImpl; -template -using JSONExtractInt64Impl = JSONExtractNumericImpl; -template -using JSONExtractUInt64Impl = JSONExtractNumericImpl; -template -using JSONExtractFloat32Impl = JSONExtractNumericImpl; -template -using JSONExtractFloat64Impl = JSONExtractNumericImpl; - - -template -class JSONExtractBoolImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_shared(); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - if (!element.isBool()) - return false; - - auto & col_vec = assert_cast &>(dest); - col_vec.insertValue(static_cast(element.getBool())); - return true; - } -}; - - -template -class JSONExtractStringImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_shared(); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - if (!element.isString()) - return false; - - auto str = element.getString(); - ColumnString & col_str = assert_cast(dest); - col_str.insertData(str.data(), str.size()); - return true; - } -}; - -template -class JSONExtractRawImpl; - -/// Nodes of the extract tree. We need the extract tree to extract from JSON complex values containing array, tuples or nullables. -template -struct JSONExtractTree -{ - using Element = typename JSONParser::Element; - - class Node - { - public: - Node() = default; - virtual ~Node() = default; - virtual bool insertResultToColumn(IColumn &, const Element &) = 0; - }; - - template - class NumericNode : public Node - { - public: - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - return JSONExtractNumericImpl::insertResultToColumn(dest, element, {}); - } - }; - - class LowCardinalityNode : public Node - { - public: - LowCardinalityNode(DataTypePtr dictionary_type_, std::unique_ptr impl_) - : dictionary_type(dictionary_type_), impl(std::move(impl_)) {} - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - auto from_col = dictionary_type->createColumn(); - if (impl->insertResultToColumn(*from_col, element)) - { - StringRef value = from_col->getDataAt(0); - assert_cast(dest).insertData(value.data, value.size); - return true; - } - return false; - } - private: - DataTypePtr dictionary_type; - std::unique_ptr impl; - }; - - class UUIDNode : public Node - { - public: - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - if (!element.isString()) - return false; - - auto uuid = parseFromString(element.getString()); - assert_cast(dest).insert(uuid); - return true; - } - }; - - template - class DecimalNode : public Node - { - public: - DecimalNode(DataTypePtr data_type_) : data_type(data_type_) {} - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - if (!element.isDouble()) - return false; - - const auto * type = assert_cast *>(data_type.get()); - auto result = convertToDecimal, DataTypeDecimal>(element.getDouble(), type->getScale()); - assert_cast &>(dest).insert(result); - return true; - } - private: - DataTypePtr data_type; - }; - - class StringNode : public Node - { - public: - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - if (element.isString()) - return JSONExtractStringImpl::insertResultToColumn(dest, element, {}); - else if (element.isNull()) - return false; - else - return JSONExtractRawImpl::insertResultToColumn(dest, element, {}); - } - }; - - class FixedStringNode : public Node - { - public: - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - if (!element.isString()) - return false; - auto & col_str = assert_cast(dest); - auto str = element.getString(); - if (str.size() > col_str.getN()) - return false; - col_str.insertData(str.data(), str.size()); - return true; - } - }; - - template - class EnumNode : public Node - { - public: - EnumNode(const std::vector> & name_value_pairs_) : name_value_pairs(name_value_pairs_) - { - for (const auto & name_value_pair : name_value_pairs) - { - name_to_value_map.emplace(name_value_pair.first, name_value_pair.second); - only_values.emplace(name_value_pair.second); - } - } - - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - auto & col_vec = assert_cast &>(dest); - - if (element.isInt64()) - { - Type value; - if (!accurate::convertNumeric(element.getInt64(), value) || !only_values.count(value)) - return false; - col_vec.insertValue(value); - return true; - } - - if (element.isUInt64()) - { - Type value; - if (!accurate::convertNumeric(element.getUInt64(), value) || !only_values.count(value)) - return false; - col_vec.insertValue(value); - return true; - } - - if (element.isString()) - { - auto value = name_to_value_map.find(element.getString()); - if (value == name_to_value_map.end()) - return false; - col_vec.insertValue(value->second); - return true; - } - - return false; - } - - private: - std::vector> name_value_pairs; - std::unordered_map name_to_value_map; - std::unordered_set only_values; - }; - - class NullableNode : public Node - { - public: - NullableNode(std::unique_ptr nested_) : nested(std::move(nested_)) {} - - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - ColumnNullable & col_null = assert_cast(dest); - if (!nested->insertResultToColumn(col_null.getNestedColumn(), element)) - return false; - col_null.getNullMapColumn().insertValue(0); - return true; - } - - private: - std::unique_ptr nested; - }; - - class ArrayNode : public Node - { - public: - ArrayNode(std::unique_ptr nested_) : nested(std::move(nested_)) {} - - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - if (!element.isArray()) - return false; - - auto array = element.getArray(); - - ColumnArray & col_arr = assert_cast(dest); - auto & data = col_arr.getData(); - size_t old_size = data.size(); - bool were_valid_elements = false; - - for (auto value : array) - { - if (nested->insertResultToColumn(data, value)) - were_valid_elements = true; - else - data.insertDefault(); - } - - if (!were_valid_elements) - { - data.popBack(data.size() - old_size); - return false; - } - - col_arr.getOffsets().push_back(data.size()); - return true; - } - - private: - std::unique_ptr nested; - }; - - class TupleNode : public Node - { - public: - TupleNode(std::vector> nested_, const std::vector & explicit_names_) : nested(std::move(nested_)), explicit_names(explicit_names_) - { - for (size_t i = 0; i != explicit_names.size(); ++i) - name_to_index_map.emplace(explicit_names[i], i); - } - - bool insertResultToColumn(IColumn & dest, const Element & element) override - { - ColumnTuple & tuple = assert_cast(dest); - size_t old_size = dest.size(); - bool were_valid_elements = false; - - auto set_size = [&](size_t size) - { - for (size_t i = 0; i != tuple.tupleSize(); ++i) - { - auto & col = tuple.getColumn(i); - if (col.size() != size) - { - if (col.size() > size) - col.popBack(col.size() - size); - else - while (col.size() < size) - col.insertDefault(); - } - } - }; - - if (element.isArray()) - { - auto array = element.getArray(); - auto it = array.begin(); - - for (size_t index = 0; (index != nested.size()) && (it != array.end()); ++index) - { - if (nested[index]->insertResultToColumn(tuple.getColumn(index), *it++)) - were_valid_elements = true; - else - tuple.getColumn(index).insertDefault(); - } - - set_size(old_size + static_cast(were_valid_elements)); - return were_valid_elements; - } - - if (element.isObject()) - { - auto object = element.getObject(); - if (name_to_index_map.empty()) - { - auto it = object.begin(); - for (size_t index = 0; (index != nested.size()) && (it != object.end()); ++index) - { - if (nested[index]->insertResultToColumn(tuple.getColumn(index), (*it++).second)) - were_valid_elements = true; - else - tuple.getColumn(index).insertDefault(); - } - } - else - { - for (auto & [key, value] : object) - { - auto index = name_to_index_map.find(key); - if (index != name_to_index_map.end()) - { - if (nested[index->second]->insertResultToColumn(tuple.getColumn(index->second), value)) - were_valid_elements = true; - } - } - } - - set_size(old_size + static_cast(were_valid_elements)); - return were_valid_elements; - } - - return false; - } - - private: - std::vector> nested; - std::vector explicit_names; - std::unordered_map name_to_index_map; - }; - - static std::unique_ptr build(const char * function_name, const DataTypePtr & type) - { - switch (type->getTypeId()) - { - case TypeIndex::UInt8: return std::make_unique>(); - case TypeIndex::UInt16: return std::make_unique>(); - case TypeIndex::UInt32: return std::make_unique>(); - case TypeIndex::UInt64: return std::make_unique>(); - case TypeIndex::Int8: return std::make_unique>(); - case TypeIndex::Int16: return std::make_unique>(); - case TypeIndex::Int32: return std::make_unique>(); - case TypeIndex::Int64: return std::make_unique>(); - case TypeIndex::Float32: return std::make_unique>(); - case TypeIndex::Float64: return std::make_unique>(); - case TypeIndex::String: return std::make_unique(); - case TypeIndex::FixedString: return std::make_unique(); - case TypeIndex::UUID: return std::make_unique(); - case TypeIndex::LowCardinality: - { - auto dictionary_type = typeid_cast(type.get())->getDictionaryType(); - auto impl = build(function_name, dictionary_type); - return std::make_unique(dictionary_type, std::move(impl)); - } - case TypeIndex::Decimal256: return std::make_unique>(type); - case TypeIndex::Decimal128: return std::make_unique>(type); - case TypeIndex::Decimal64: return std::make_unique>(type); - case TypeIndex::Decimal32: return std::make_unique>(type); - case TypeIndex::Enum8: - return std::make_unique>(static_cast(*type).getValues()); - case TypeIndex::Enum16: - return std::make_unique>(static_cast(*type).getValues()); - case TypeIndex::Nullable: - { - return std::make_unique(build(function_name, static_cast(*type).getNestedType())); - } - case TypeIndex::Array: - { - return std::make_unique(build(function_name, static_cast(*type).getNestedType())); - } - case TypeIndex::Tuple: - { - const auto & tuple = static_cast(*type); - const auto & tuple_elements = tuple.getElements(); - std::vector> elements; - for (const auto & tuple_element : tuple_elements) - elements.emplace_back(build(function_name, tuple_element)); - return std::make_unique(std::move(elements), tuple.haveExplicitNames() ? tuple.getElementNames() : Strings{}); - } - default: - throw Exception{"Function " + String(function_name) + " doesn't support the return type schema: " + type->getName(), ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT}; - } - } -}; - - -template -class JSONExtractImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char * function_name, const ColumnsWithTypeAndName & arguments) - { - if (arguments.size() < 2) - throw Exception{"Function " + String(function_name) + " requires at least two arguments", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; - - const auto & col = arguments.back(); - auto col_type_const = typeid_cast(col.column.get()); - if (!col_type_const || !isString(col.type)) - throw Exception{"The last argument of function " + String(function_name) - + " should be a constant string specifying the return data type, illegal value: " + col.name, - ErrorCodes::ILLEGAL_COLUMN}; - - return DataTypeFactory::instance().get(col_type_const->getValue()); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 2; } - - void prepare(const char * function_name, const ColumnsWithTypeAndName &, const DataTypePtr & result_type) - { - extract_tree = JSONExtractTree::build(function_name, result_type); - } - - bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - return extract_tree->insertResultToColumn(dest, element); - } - -protected: - std::unique_ptr::Node> extract_tree; -}; - - -template -class JSONExtractKeysAndValuesImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char * function_name, const ColumnsWithTypeAndName & arguments) - { - if (arguments.size() < 2) - throw Exception{"Function " + String(function_name) + " requires at least two arguments", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH}; - - const auto & col = arguments.back(); - auto col_type_const = typeid_cast(col.column.get()); - if (!col_type_const || !isString(col.type)) - throw Exception{"The last argument of function " + String(function_name) - + " should be a constant string specifying the values' data type, illegal value: " + col.name, - ErrorCodes::ILLEGAL_COLUMN}; - - DataTypePtr key_type = std::make_unique(); - DataTypePtr value_type = DataTypeFactory::instance().get(col_type_const->getValue()); - DataTypePtr tuple_type = std::make_unique(DataTypes{key_type, value_type}); - return std::make_unique(tuple_type); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 2; } - - void prepare(const char * function_name, const ColumnsWithTypeAndName &, const DataTypePtr & result_type) - { - const auto tuple_type = typeid_cast(result_type.get())->getNestedType(); - const auto value_type = typeid_cast(tuple_type.get())->getElements()[1]; - extract_tree = JSONExtractTree::build(function_name, value_type); - } - - bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - if (!element.isObject()) - return false; - - auto object = element.getObject(); - - auto & col_arr = assert_cast(dest); - auto & col_tuple = assert_cast(col_arr.getData()); - size_t old_size = col_tuple.size(); - auto & col_key = assert_cast(col_tuple.getColumn(0)); - auto & col_value = col_tuple.getColumn(1); - - for (auto & [key, value] : object) - { - if (extract_tree->insertResultToColumn(col_value, value)) - col_key.insertData(key.data(), key.size()); - } - - if (col_tuple.size() == old_size) - return false; - - col_arr.getOffsets().push_back(col_tuple.size()); - return true; - } - -private: - std::unique_ptr::Node> extract_tree; -}; - - -template -class JSONExtractRawImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_shared(); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - ColumnString & col_str = assert_cast(dest); - auto & chars = col_str.getChars(); - WriteBufferFromVector buf(chars, WriteBufferFromVector::AppendModeTag()); - traverse(element, buf); - buf.finalize(); - chars.push_back(0); - col_str.getOffsets().push_back(chars.size()); - return true; - } - -private: - static void traverse(const Element & element, WriteBuffer & buf) - { - if (element.isInt64()) - { - writeIntText(element.getInt64(), buf); - return; - } - if (element.isUInt64()) - { - writeIntText(element.getUInt64(), buf); - return; - } - if (element.isDouble()) - { - writeFloatText(element.getDouble(), buf); - return; - } - if (element.isBool()) - { - if (element.getBool()) - writeCString("true", buf); - else - writeCString("false", buf); - return; - } - if (element.isString()) - { - writeJSONString(element.getString(), buf, format_settings()); - return; - } - if (element.isArray()) - { - writeChar('[', buf); - bool need_comma = false; - for (auto value : element.getArray()) - { - if (std::exchange(need_comma, true)) - writeChar(',', buf); - traverse(value, buf); - } - writeChar(']', buf); - return; - } - if (element.isObject()) - { - writeChar('{', buf); - bool need_comma = false; - for (auto [key, value] : element.getObject()) - { - if (std::exchange(need_comma, true)) - writeChar(',', buf); - writeJSONString(key, buf, format_settings()); - writeChar(':', buf); - traverse(value, buf); - } - writeChar('}', buf); - return; - } - if (element.isNull()) - { - writeCString("null", buf); - return; - } - } - - static const FormatSettings & format_settings() - { - static const FormatSettings the_instance = [] - { - FormatSettings settings; - settings.json.escape_forward_slashes = false; - return settings; - }(); - return the_instance; - } -}; - - -template -class JSONExtractArrayRawImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_shared(std::make_shared()); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - static bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - if (!element.isArray()) - return false; - - auto array = element.getArray(); - ColumnArray & col_res = assert_cast(dest); - - for (auto value : array) - JSONExtractRawImpl::insertResultToColumn(col_res.getData(), value, {}); - - col_res.getOffsets().push_back(col_res.getOffsets().back() + array.size()); - return true; - } -}; - - -template -class JSONExtractKeysAndValuesRawImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - DataTypePtr string_type = std::make_unique(); - DataTypePtr tuple_type = std::make_unique(DataTypes{string_type, string_type}); - return std::make_unique(tuple_type); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - if (!element.isObject()) - return false; - - auto object = element.getObject(); - - auto & col_arr = assert_cast(dest); - auto & col_tuple = assert_cast(col_arr.getData()); - auto & col_key = assert_cast(col_tuple.getColumn(0)); - auto & col_value = assert_cast(col_tuple.getColumn(1)); - - for (auto & [key, value] : object) - { - col_key.insertData(key.data(), key.size()); - JSONExtractRawImpl::insertResultToColumn(col_value, value, {}); - } - - col_arr.getOffsets().push_back(col_arr.getOffsets().back() + object.size()); - return true; - } -}; - -template -class JSONExtractKeysImpl -{ -public: - using Element = typename JSONParser::Element; - - static DataTypePtr getReturnType(const char *, const ColumnsWithTypeAndName &) - { - return std::make_unique(std::make_shared()); - } - - static size_t getNumberOfIndexArguments(const ColumnsWithTypeAndName & arguments) { return arguments.size() - 1; } - - bool insertResultToColumn(IColumn & dest, const Element & element, const std::string_view &) - { - if (!element.isObject()) - return false; - - auto object = element.getObject(); - - ColumnArray & col_res = assert_cast(dest); - auto & col_key = assert_cast(col_res.getData()); - - for (const auto & [key, value] : object) - { - col_key.insertData(key.data(), key.size()); - } - - col_res.getOffsets().push_back(col_res.getOffsets().back() + object.size()); - return true; - } -}; - -}