| 1 | #include <type_traits> |
| 2 | #include <DataTypes/DataTypeNumberBase.h> |
| 3 | #include <Columns/ColumnVector.h> |
| 4 | #include <Columns/ColumnConst.h> |
| 5 | #include <IO/ReadHelpers.h> |
| 6 | #include <IO/WriteHelpers.h> |
| 7 | #include <Common/NaNUtils.h> |
| 8 | #include <Common/typeid_cast.h> |
| 9 | #include <Common/assert_cast.h> |
| 10 | #include <Formats/FormatSettings.h> |
| 11 | #include <Formats/ProtobufReader.h> |
| 12 | #include <Formats/ProtobufWriter.h> |
| 13 | |
| 14 | |
| 15 | namespace DB |
| 16 | { |
| 17 | |
| 18 | template <typename T> |
| 19 | void DataTypeNumberBase<T>::serializeText(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings &) const |
| 20 | { |
| 21 | writeText(assert_cast<const ColumnVector<T> &>(column).getData()[row_num], ostr); |
| 22 | } |
| 23 | |
| 24 | template <typename T> |
| 25 | void DataTypeNumberBase<T>::deserializeText(IColumn & column, ReadBuffer & istr, const FormatSettings &) const |
| 26 | { |
| 27 | T x; |
| 28 | |
| 29 | if constexpr (is_integral_v<T> && is_arithmetic_v<T>) |
| 30 | readIntTextUnsafe(x, istr); |
| 31 | else |
| 32 | readText(x, istr); |
| 33 | |
| 34 | assert_cast<ColumnVector<T> &>(column).getData().push_back(x); |
| 35 | } |
| 36 | |
| 37 | template <typename T> |
| 38 | static inline void writeDenormalNumber(T x, WriteBuffer & ostr) |
| 39 | { |
| 40 | if constexpr (std::is_floating_point_v<T>) |
| 41 | { |
| 42 | if (std::signbit(x)) |
| 43 | { |
| 44 | if (isNaN(x)) |
| 45 | writeCString("-nan" , ostr); |
| 46 | else |
| 47 | writeCString("-inf" , ostr); |
| 48 | } |
| 49 | else |
| 50 | { |
| 51 | if (isNaN(x)) |
| 52 | writeCString("nan" , ostr); |
| 53 | else |
| 54 | writeCString("inf" , ostr); |
| 55 | } |
| 56 | } |
| 57 | else |
| 58 | { |
| 59 | /// This function is not called for non floating point numbers. |
| 60 | (void)x; |
| 61 | } |
| 62 | } |
| 63 | |
| 64 | |
| 65 | template <typename T> |
| 66 | void DataTypeNumberBase<T>::serializeTextJSON(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings & settings) const |
| 67 | { |
| 68 | auto x = assert_cast<const ColumnVector<T> &>(column).getData()[row_num]; |
| 69 | bool is_finite = isFinite(x); |
| 70 | |
| 71 | const bool need_quote = (is_integral_v<T> && (sizeof(T) == 8) && settings.json.quote_64bit_integers) |
| 72 | || (settings.json.quote_denormals && !is_finite); |
| 73 | |
| 74 | if (need_quote) |
| 75 | writeChar('"', ostr); |
| 76 | |
| 77 | if (is_finite) |
| 78 | writeText(x, ostr); |
| 79 | else if (!settings.json.quote_denormals) |
| 80 | writeCString("null" , ostr); |
| 81 | else |
| 82 | writeDenormalNumber(x, ostr); |
| 83 | |
| 84 | if (need_quote) |
| 85 | writeChar('"', ostr); |
| 86 | } |
| 87 | |
| 88 | template <typename T> |
| 89 | void DataTypeNumberBase<T>::deserializeTextJSON(IColumn & column, ReadBuffer & istr, const FormatSettings &) const |
| 90 | { |
| 91 | bool has_quote = false; |
| 92 | if (!istr.eof() && *istr.position() == '"') /// We understand the number both in quotes and without. |
| 93 | { |
| 94 | has_quote = true; |
| 95 | ++istr.position(); |
| 96 | } |
| 97 | |
| 98 | FieldType x; |
| 99 | |
| 100 | /// null |
| 101 | if (!has_quote && !istr.eof() && *istr.position() == 'n') |
| 102 | { |
| 103 | ++istr.position(); |
| 104 | assertString("ull" , istr); |
| 105 | |
| 106 | x = NaNOrZero<T>(); |
| 107 | } |
| 108 | else |
| 109 | { |
| 110 | static constexpr bool is_uint8 = std::is_same_v<T, UInt8>; |
| 111 | static constexpr bool is_int8 = std::is_same_v<T, Int8>; |
| 112 | |
| 113 | if (is_uint8 || is_int8) |
| 114 | { |
| 115 | // extra conditions to parse true/false strings into 1/0 |
| 116 | if (istr.eof()) |
| 117 | throwReadAfterEOF(); |
| 118 | if (*istr.position() == 't' || *istr.position() == 'f') |
| 119 | { |
| 120 | bool tmp = false; |
| 121 | readBoolTextWord(tmp, istr); |
| 122 | x = tmp; |
| 123 | } |
| 124 | else |
| 125 | readText(x, istr); |
| 126 | } |
| 127 | else |
| 128 | { |
| 129 | readText(x, istr); |
| 130 | } |
| 131 | |
| 132 | if (has_quote) |
| 133 | assertChar('"', istr); |
| 134 | } |
| 135 | |
| 136 | assert_cast<ColumnVector<T> &>(column).getData().push_back(x); |
| 137 | } |
| 138 | |
| 139 | template <typename T> |
| 140 | void DataTypeNumberBase<T>::deserializeTextCSV(IColumn & column, ReadBuffer & istr, const FormatSettings &) const |
| 141 | { |
| 142 | FieldType x; |
| 143 | readCSV(x, istr); |
| 144 | assert_cast<ColumnVector<T> &>(column).getData().push_back(x); |
| 145 | } |
| 146 | |
| 147 | template <typename T> |
| 148 | Field DataTypeNumberBase<T>::getDefault() const |
| 149 | { |
| 150 | return NearestFieldType<FieldType>(); |
| 151 | } |
| 152 | |
| 153 | template <typename T> |
| 154 | void DataTypeNumberBase<T>::serializeBinary(const Field & field, WriteBuffer & ostr) const |
| 155 | { |
| 156 | /// ColumnVector<T>::ValueType is a narrower type. For example, UInt8, when the Field type is UInt64 |
| 157 | typename ColumnVector<T>::ValueType x = get<NearestFieldType<FieldType>>(field); |
| 158 | writeBinary(x, ostr); |
| 159 | } |
| 160 | |
| 161 | template <typename T> |
| 162 | void DataTypeNumberBase<T>::deserializeBinary(Field & field, ReadBuffer & istr) const |
| 163 | { |
| 164 | typename ColumnVector<T>::ValueType x; |
| 165 | readBinary(x, istr); |
| 166 | field = NearestFieldType<FieldType>(x); |
| 167 | } |
| 168 | |
| 169 | template <typename T> |
| 170 | void DataTypeNumberBase<T>::serializeBinary(const IColumn & column, size_t row_num, WriteBuffer & ostr) const |
| 171 | { |
| 172 | writeBinary(assert_cast<const ColumnVector<T> &>(column).getData()[row_num], ostr); |
| 173 | } |
| 174 | |
| 175 | template <typename T> |
| 176 | void DataTypeNumberBase<T>::deserializeBinary(IColumn & column, ReadBuffer & istr) const |
| 177 | { |
| 178 | typename ColumnVector<T>::ValueType x; |
| 179 | readBinary(x, istr); |
| 180 | assert_cast<ColumnVector<T> &>(column).getData().push_back(x); |
| 181 | } |
| 182 | |
| 183 | template <typename T> |
| 184 | void DataTypeNumberBase<T>::serializeBinaryBulk(const IColumn & column, WriteBuffer & ostr, size_t offset, size_t limit) const |
| 185 | { |
| 186 | const typename ColumnVector<T>::Container & x = typeid_cast<const ColumnVector<T> &>(column).getData(); |
| 187 | |
| 188 | size_t size = x.size(); |
| 189 | |
| 190 | if (limit == 0 || offset + limit > size) |
| 191 | limit = size - offset; |
| 192 | |
| 193 | if (limit) |
| 194 | ostr.write(reinterpret_cast<const char *>(&x[offset]), sizeof(typename ColumnVector<T>::ValueType) * limit); |
| 195 | } |
| 196 | |
| 197 | template <typename T> |
| 198 | void DataTypeNumberBase<T>::deserializeBinaryBulk(IColumn & column, ReadBuffer & istr, size_t limit, double /*avg_value_size_hint*/) const |
| 199 | { |
| 200 | typename ColumnVector<T>::Container & x = typeid_cast<ColumnVector<T> &>(column).getData(); |
| 201 | size_t initial_size = x.size(); |
| 202 | x.resize(initial_size + limit); |
| 203 | size_t size = istr.readBig(reinterpret_cast<char*>(&x[initial_size]), sizeof(typename ColumnVector<T>::ValueType) * limit); |
| 204 | x.resize(initial_size + size / sizeof(typename ColumnVector<T>::ValueType)); |
| 205 | } |
| 206 | |
| 207 | |
| 208 | template <typename T> |
| 209 | void DataTypeNumberBase<T>::serializeProtobuf(const IColumn & column, size_t row_num, ProtobufWriter & protobuf, size_t & value_index) const |
| 210 | { |
| 211 | if (value_index) |
| 212 | return; |
| 213 | value_index = static_cast<bool>(protobuf.writeNumber(assert_cast<const ColumnVector<T> &>(column).getData()[row_num])); |
| 214 | } |
| 215 | |
| 216 | |
| 217 | template <typename T> |
| 218 | void DataTypeNumberBase<T>::deserializeProtobuf(IColumn & column, ProtobufReader & protobuf, bool allow_add_row, bool & row_added) const |
| 219 | { |
| 220 | row_added = false; |
| 221 | T value; |
| 222 | if (!protobuf.readNumber(value)) |
| 223 | return; |
| 224 | |
| 225 | auto & container = typeid_cast<ColumnVector<T> &>(column).getData(); |
| 226 | if (allow_add_row) |
| 227 | { |
| 228 | container.emplace_back(value); |
| 229 | row_added = true; |
| 230 | } |
| 231 | else |
| 232 | container.back() = value; |
| 233 | } |
| 234 | |
| 235 | |
| 236 | template <typename T> |
| 237 | MutableColumnPtr DataTypeNumberBase<T>::createColumn() const |
| 238 | { |
| 239 | return ColumnVector<T>::create(); |
| 240 | } |
| 241 | |
| 242 | template <typename T> |
| 243 | bool DataTypeNumberBase<T>::isValueRepresentedByInteger() const |
| 244 | { |
| 245 | return is_integral_v<T>; |
| 246 | } |
| 247 | |
| 248 | template <typename T> |
| 249 | bool DataTypeNumberBase<T>::isValueRepresentedByUnsignedInteger() const |
| 250 | { |
| 251 | return is_integral_v<T> && is_unsigned_v<T>; |
| 252 | } |
| 253 | |
| 254 | |
| 255 | /// Explicit template instantiations - to avoid code bloat in headers. |
| 256 | template class DataTypeNumberBase<UInt8>; |
| 257 | template class DataTypeNumberBase<UInt16>; |
| 258 | template class DataTypeNumberBase<UInt32>; |
| 259 | template class DataTypeNumberBase<UInt64>; |
| 260 | template class DataTypeNumberBase<UInt128>; |
| 261 | template class DataTypeNumberBase<Int8>; |
| 262 | template class DataTypeNumberBase<Int16>; |
| 263 | template class DataTypeNumberBase<Int32>; |
| 264 | template class DataTypeNumberBase<Int64>; |
| 265 | template class DataTypeNumberBase<Float32>; |
| 266 | template class DataTypeNumberBase<Float64>; |
| 267 | |
| 268 | } |
| 269 | |