| 1 | #pragma once |
| 2 | |
| 3 | #include <cassert> |
| 4 | #include <vector> |
| 5 | #include <algorithm> |
| 6 | #include <type_traits> |
| 7 | #include <functional> |
| 8 | |
| 9 | #include <Common/Exception.h> |
| 10 | #include <Common/UInt128.h> |
| 11 | #include <Core/Types.h> |
| 12 | #include <Core/Defines.h> |
| 13 | #include <Core/UUID.h> |
| 14 | #include <common/DayNum.h> |
| 15 | #include <common/strong_typedef.h> |
| 16 | |
| 17 | |
| 18 | namespace DB |
| 19 | { |
| 20 | |
| 21 | namespace ErrorCodes |
| 22 | { |
| 23 | extern const int BAD_TYPE_OF_FIELD; |
| 24 | extern const int BAD_GET; |
| 25 | extern const int NOT_IMPLEMENTED; |
| 26 | extern const int LOGICAL_ERROR; |
| 27 | extern const int ILLEGAL_TYPE_OF_ARGUMENT; |
| 28 | } |
| 29 | |
| 30 | template <typename T, typename SFINAE = void> |
| 31 | struct NearestFieldTypeImpl; |
| 32 | |
| 33 | template <typename T> |
| 34 | using NearestFieldType = typename NearestFieldTypeImpl<T>::Type; |
| 35 | |
| 36 | class Field; |
| 37 | using FieldVector = std::vector<Field>; |
| 38 | |
| 39 | /// Array and Tuple use the same storage type -- FieldVector, but we declare |
| 40 | /// distinct types for them, so that the caller can choose whether it wants to |
| 41 | /// construct a Field of Array or a Tuple type. An alternative approach would be |
| 42 | /// to construct both of these types from FieldVector, and have the caller |
| 43 | /// specify the desired Field type explicitly. |
| 44 | #define DEFINE_FIELD_VECTOR(X) \ |
| 45 | struct X : public FieldVector \ |
| 46 | { \ |
| 47 | using FieldVector::FieldVector; \ |
| 48 | } |
| 49 | |
| 50 | DEFINE_FIELD_VECTOR(Array); |
| 51 | DEFINE_FIELD_VECTOR(Tuple); |
| 52 | |
| 53 | #undef DEFINE_FIELD_VECTOR |
| 54 | |
| 55 | struct AggregateFunctionStateData |
| 56 | { |
| 57 | String name; /// Name with arguments. |
| 58 | String data; |
| 59 | |
| 60 | bool operator < (const AggregateFunctionStateData &) const |
| 61 | { |
| 62 | throw Exception("Operator < is not implemented for AggregateFunctionStateData." , ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT); |
| 63 | } |
| 64 | |
| 65 | bool operator <= (const AggregateFunctionStateData &) const |
| 66 | { |
| 67 | throw Exception("Operator <= is not implemented for AggregateFunctionStateData." , ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT); |
| 68 | } |
| 69 | |
| 70 | bool operator > (const AggregateFunctionStateData &) const |
| 71 | { |
| 72 | throw Exception("Operator > is not implemented for AggregateFunctionStateData." , ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT); |
| 73 | } |
| 74 | |
| 75 | bool operator >= (const AggregateFunctionStateData &) const |
| 76 | { |
| 77 | throw Exception("Operator >= is not implemented for AggregateFunctionStateData." , ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT); |
| 78 | } |
| 79 | |
| 80 | bool operator == (const AggregateFunctionStateData & rhs) const |
| 81 | { |
| 82 | if (name != rhs.name) |
| 83 | throw Exception("Comparing aggregate functions with different types: " + name + " and " + rhs.name, |
| 84 | ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT); |
| 85 | |
| 86 | return data == rhs.data; |
| 87 | } |
| 88 | }; |
| 89 | |
| 90 | template <typename T> bool decimalEqual(T x, T y, UInt32 x_scale, UInt32 y_scale); |
| 91 | template <typename T> bool decimalLess(T x, T y, UInt32 x_scale, UInt32 y_scale); |
| 92 | template <typename T> bool decimalLessOrEqual(T x, T y, UInt32 x_scale, UInt32 y_scale); |
| 93 | |
| 94 | template <typename T> |
| 95 | class DecimalField |
| 96 | { |
| 97 | public: |
| 98 | DecimalField(T value, UInt32 scale_) |
| 99 | : dec(value), |
| 100 | scale(scale_) |
| 101 | {} |
| 102 | |
| 103 | operator T() const { return dec; } |
| 104 | T getValue() const { return dec; } |
| 105 | T getScaleMultiplier() const { return T::getScaleMultiplier(scale); } |
| 106 | UInt32 getScale() const { return scale; } |
| 107 | |
| 108 | template <typename U> |
| 109 | bool operator < (const DecimalField<U> & r) const |
| 110 | { |
| 111 | using MaxType = std::conditional_t<(sizeof(T) > sizeof(U)), T, U>; |
| 112 | return decimalLess<MaxType>(dec, r.getValue(), scale, r.getScale()); |
| 113 | } |
| 114 | |
| 115 | template <typename U> |
| 116 | bool operator <= (const DecimalField<U> & r) const |
| 117 | { |
| 118 | using MaxType = std::conditional_t<(sizeof(T) > sizeof(U)), T, U>; |
| 119 | return decimalLessOrEqual<MaxType>(dec, r.getValue(), scale, r.getScale()); |
| 120 | } |
| 121 | |
| 122 | template <typename U> |
| 123 | bool operator == (const DecimalField<U> & r) const |
| 124 | { |
| 125 | using MaxType = std::conditional_t<(sizeof(T) > sizeof(U)), T, U>; |
| 126 | return decimalEqual<MaxType>(dec, r.getValue(), scale, r.getScale()); |
| 127 | } |
| 128 | |
| 129 | template <typename U> bool operator > (const DecimalField<U> & r) const { return r < *this; } |
| 130 | template <typename U> bool operator >= (const DecimalField<U> & r) const { return r <= * this; } |
| 131 | template <typename U> bool operator != (const DecimalField<U> & r) const { return !(*this == r); } |
| 132 | |
| 133 | const DecimalField<T> & operator += (const DecimalField<T> & r) |
| 134 | { |
| 135 | if (scale != r.getScale()) |
| 136 | throw Exception("Add different decimal fields" , ErrorCodes::LOGICAL_ERROR); |
| 137 | dec += r.getValue(); |
| 138 | return *this; |
| 139 | } |
| 140 | |
| 141 | const DecimalField<T> & operator -= (const DecimalField<T> & r) |
| 142 | { |
| 143 | if (scale != r.getScale()) |
| 144 | throw Exception("Sub different decimal fields" , ErrorCodes::LOGICAL_ERROR); |
| 145 | dec -= r.getValue(); |
| 146 | return *this; |
| 147 | } |
| 148 | |
| 149 | private: |
| 150 | T dec; |
| 151 | UInt32 scale; |
| 152 | }; |
| 153 | |
| 154 | /// char may be signed or unsigned, and behave identically to signed char or unsigned char, |
| 155 | /// but they are always three different types. |
| 156 | /// signedness of char is different in Linux on x86 and Linux on ARM. |
| 157 | template <> struct NearestFieldTypeImpl<char> { using Type = std::conditional_t<is_signed_v<char>, Int64, UInt64>; }; |
| 158 | template <> struct NearestFieldTypeImpl<signed char> { using Type = Int64; }; |
| 159 | template <> struct NearestFieldTypeImpl<unsigned char> { using Type = UInt64; }; |
| 160 | |
| 161 | template <> struct NearestFieldTypeImpl<UInt16> { using Type = UInt64; }; |
| 162 | template <> struct NearestFieldTypeImpl<UInt32> { using Type = UInt64; }; |
| 163 | |
| 164 | template <> struct NearestFieldTypeImpl<DayNum> { using Type = UInt64; }; |
| 165 | template <> struct NearestFieldTypeImpl<UInt128> { using Type = UInt128; }; |
| 166 | template <> struct NearestFieldTypeImpl<UUID> { using Type = UInt128; }; |
| 167 | template <> struct NearestFieldTypeImpl<Int16> { using Type = Int64; }; |
| 168 | template <> struct NearestFieldTypeImpl<Int32> { using Type = Int64; }; |
| 169 | |
| 170 | /// long and long long are always different types that may behave identically or not. |
| 171 | /// This is different on Linux and Mac. |
| 172 | template <> struct NearestFieldTypeImpl<long> { using Type = Int64; }; |
| 173 | template <> struct NearestFieldTypeImpl<long long> { using Type = Int64; }; |
| 174 | template <> struct NearestFieldTypeImpl<unsigned long> { using Type = UInt64; }; |
| 175 | template <> struct NearestFieldTypeImpl<unsigned long long> { using Type = UInt64; }; |
| 176 | |
| 177 | template <> struct NearestFieldTypeImpl<Int128> { using Type = Int128; }; |
| 178 | template <> struct NearestFieldTypeImpl<Decimal32> { using Type = DecimalField<Decimal32>; }; |
| 179 | template <> struct NearestFieldTypeImpl<Decimal64> { using Type = DecimalField<Decimal64>; }; |
| 180 | template <> struct NearestFieldTypeImpl<Decimal128> { using Type = DecimalField<Decimal128>; }; |
| 181 | template <> struct NearestFieldTypeImpl<DecimalField<Decimal32>> { using Type = DecimalField<Decimal32>; }; |
| 182 | template <> struct NearestFieldTypeImpl<DecimalField<Decimal64>> { using Type = DecimalField<Decimal64>; }; |
| 183 | template <> struct NearestFieldTypeImpl<DecimalField<Decimal128>> { using Type = DecimalField<Decimal128>; }; |
| 184 | template <> struct NearestFieldTypeImpl<Float32> { using Type = Float64; }; |
| 185 | template <> struct NearestFieldTypeImpl<Float64> { using Type = Float64; }; |
| 186 | template <> struct NearestFieldTypeImpl<const char *> { using Type = String; }; |
| 187 | template <> struct NearestFieldTypeImpl<String> { using Type = String; }; |
| 188 | template <> struct NearestFieldTypeImpl<Array> { using Type = Array; }; |
| 189 | template <> struct NearestFieldTypeImpl<Tuple> { using Type = Tuple; }; |
| 190 | template <> struct NearestFieldTypeImpl<bool> { using Type = UInt64; }; |
| 191 | template <> struct NearestFieldTypeImpl<Null> { using Type = Null; }; |
| 192 | |
| 193 | template <> struct NearestFieldTypeImpl<AggregateFunctionStateData> { using Type = AggregateFunctionStateData; }; |
| 194 | |
| 195 | // For enum types, use the field type that corresponds to their underlying type. |
| 196 | template <typename T> |
| 197 | struct NearestFieldTypeImpl<T, std::enable_if_t<std::is_enum_v<T>>> |
| 198 | { |
| 199 | using Type = NearestFieldType<std::underlying_type_t<T>>; |
| 200 | }; |
| 201 | |
| 202 | /** 32 is enough. Round number is used for alignment and for better arithmetic inside std::vector. |
| 203 | * NOTE: Actually, sizeof(std::string) is 32 when using libc++, so Field is 40 bytes. |
| 204 | */ |
| 205 | #define DBMS_MIN_FIELD_SIZE 32 |
| 206 | |
| 207 | |
| 208 | /** Discriminated union of several types. |
| 209 | * Made for replacement of `boost::variant` |
| 210 | * is not generalized, |
| 211 | * but somewhat more efficient, and simpler. |
| 212 | * |
| 213 | * Used to represent a single value of one of several types in memory. |
| 214 | * Warning! Prefer to use chunks of columns instead of single values. See Column.h |
| 215 | */ |
| 216 | class Field |
| 217 | { |
| 218 | public: |
| 219 | struct Types |
| 220 | { |
| 221 | /// Type tag. |
| 222 | enum Which |
| 223 | { |
| 224 | Null = 0, |
| 225 | UInt64 = 1, |
| 226 | Int64 = 2, |
| 227 | Float64 = 3, |
| 228 | UInt128 = 4, |
| 229 | Int128 = 5, |
| 230 | |
| 231 | /// Non-POD types. |
| 232 | |
| 233 | String = 16, |
| 234 | Array = 17, |
| 235 | Tuple = 18, |
| 236 | Decimal32 = 19, |
| 237 | Decimal64 = 20, |
| 238 | Decimal128 = 21, |
| 239 | AggregateFunctionState = 22, |
| 240 | }; |
| 241 | |
| 242 | static const int MIN_NON_POD = 16; |
| 243 | |
| 244 | static const char * toString(Which which) |
| 245 | { |
| 246 | switch (which) |
| 247 | { |
| 248 | case Null: return "Null" ; |
| 249 | case UInt64: return "UInt64" ; |
| 250 | case UInt128: return "UInt128" ; |
| 251 | case Int64: return "Int64" ; |
| 252 | case Int128: return "Int128" ; |
| 253 | case Float64: return "Float64" ; |
| 254 | case String: return "String" ; |
| 255 | case Array: return "Array" ; |
| 256 | case Tuple: return "Tuple" ; |
| 257 | case Decimal32: return "Decimal32" ; |
| 258 | case Decimal64: return "Decimal64" ; |
| 259 | case Decimal128: return "Decimal128" ; |
| 260 | case AggregateFunctionState: return "AggregateFunctionState" ; |
| 261 | } |
| 262 | |
| 263 | throw Exception("Bad type of Field" , ErrorCodes::BAD_TYPE_OF_FIELD); |
| 264 | } |
| 265 | }; |
| 266 | |
| 267 | |
| 268 | /// Returns an identifier for the type or vice versa. |
| 269 | template <typename T> struct TypeToEnum; |
| 270 | template <Types::Which which> struct EnumToType; |
| 271 | |
| 272 | static bool IsDecimal(Types::Which which) { return which >= Types::Decimal32 && which <= Types::Decimal128; } |
| 273 | |
| 274 | Field() |
| 275 | : which(Types::Null) |
| 276 | { |
| 277 | } |
| 278 | |
| 279 | /** Despite the presence of a template constructor, this constructor is still needed, |
| 280 | * since, in its absence, the compiler will still generate the default constructor. |
| 281 | */ |
| 282 | Field(const Field & rhs) |
| 283 | { |
| 284 | create(rhs); |
| 285 | } |
| 286 | |
| 287 | Field(Field && rhs) |
| 288 | { |
| 289 | create(std::move(rhs)); |
| 290 | } |
| 291 | |
| 292 | template <typename T> |
| 293 | Field(T && rhs, std::enable_if_t<!std::is_same_v<std::decay_t<T>, Field>, void *> = nullptr); |
| 294 | |
| 295 | /// Create a string inplace. |
| 296 | Field(const char * data, size_t size) |
| 297 | { |
| 298 | create(data, size); |
| 299 | } |
| 300 | |
| 301 | Field(const unsigned char * data, size_t size) |
| 302 | { |
| 303 | create(data, size); |
| 304 | } |
| 305 | |
| 306 | /// NOTE In case when field already has string type, more direct assign is possible. |
| 307 | void assignString(const char * data, size_t size) |
| 308 | { |
| 309 | destroy(); |
| 310 | create(data, size); |
| 311 | } |
| 312 | |
| 313 | void assignString(const unsigned char * data, size_t size) |
| 314 | { |
| 315 | destroy(); |
| 316 | create(data, size); |
| 317 | } |
| 318 | |
| 319 | Field & operator= (const Field & rhs) |
| 320 | { |
| 321 | if (this != &rhs) |
| 322 | { |
| 323 | if (which != rhs.which) |
| 324 | { |
| 325 | destroy(); |
| 326 | create(rhs); |
| 327 | } |
| 328 | else |
| 329 | assign(rhs); /// This assigns string or vector without deallocation of existing buffer. |
| 330 | } |
| 331 | return *this; |
| 332 | } |
| 333 | |
| 334 | Field & operator= (Field && rhs) |
| 335 | { |
| 336 | if (this != &rhs) |
| 337 | { |
| 338 | if (which != rhs.which) |
| 339 | { |
| 340 | destroy(); |
| 341 | create(std::move(rhs)); |
| 342 | } |
| 343 | else |
| 344 | assign(std::move(rhs)); |
| 345 | } |
| 346 | return *this; |
| 347 | } |
| 348 | |
| 349 | template <typename T> |
| 350 | std::enable_if_t<!std::is_same_v<std::decay_t<T>, Field>, Field &> |
| 351 | operator= (T && rhs); |
| 352 | |
| 353 | ~Field() |
| 354 | { |
| 355 | destroy(); |
| 356 | } |
| 357 | |
| 358 | |
| 359 | Types::Which getType() const { return which; } |
| 360 | const char * getTypeName() const { return Types::toString(which); } |
| 361 | |
| 362 | bool isNull() const { return which == Types::Null; } |
| 363 | |
| 364 | |
| 365 | template <typename T> |
| 366 | T & get(); |
| 367 | |
| 368 | template <typename T> |
| 369 | const T & get() const |
| 370 | { |
| 371 | auto mutable_this = const_cast<std::decay_t<decltype(*this)> *>(this); |
| 372 | return mutable_this->get<T>(); |
| 373 | } |
| 374 | |
| 375 | template <typename T> |
| 376 | T & reinterpret(); |
| 377 | |
| 378 | template <typename T> |
| 379 | const T & reinterpret() const |
| 380 | { |
| 381 | auto mutable_this = const_cast<std::decay_t<decltype(*this)> *>(this); |
| 382 | return mutable_this->reinterpret<T>(); |
| 383 | } |
| 384 | |
| 385 | template <typename T> bool tryGet(T & result) |
| 386 | { |
| 387 | const Types::Which requested = TypeToEnum<std::decay_t<T>>::value; |
| 388 | if (which != requested) |
| 389 | return false; |
| 390 | result = get<T>(); |
| 391 | return true; |
| 392 | } |
| 393 | |
| 394 | template <typename T> bool tryGet(T & result) const |
| 395 | { |
| 396 | const Types::Which requested = TypeToEnum<std::decay_t<T>>::value; |
| 397 | if (which != requested) |
| 398 | return false; |
| 399 | result = get<T>(); |
| 400 | return true; |
| 401 | } |
| 402 | |
| 403 | template <typename T> T & safeGet() |
| 404 | { |
| 405 | const Types::Which requested = TypeToEnum<std::decay_t<T>>::value; |
| 406 | if (which != requested) |
| 407 | throw Exception("Bad get: has " + std::string(getTypeName()) + ", requested " + std::string(Types::toString(requested)), ErrorCodes::BAD_GET); |
| 408 | return get<T>(); |
| 409 | } |
| 410 | |
| 411 | template <typename T> const T & safeGet() const |
| 412 | { |
| 413 | const Types::Which requested = TypeToEnum<std::decay_t<T>>::value; |
| 414 | if (which != requested) |
| 415 | throw Exception("Bad get: has " + std::string(getTypeName()) + ", requested " + std::string(Types::toString(requested)), ErrorCodes::BAD_GET); |
| 416 | return get<T>(); |
| 417 | } |
| 418 | |
| 419 | |
| 420 | bool operator< (const Field & rhs) const |
| 421 | { |
| 422 | if (which < rhs.which) |
| 423 | return true; |
| 424 | if (which > rhs.which) |
| 425 | return false; |
| 426 | |
| 427 | switch (which) |
| 428 | { |
| 429 | case Types::Null: return false; |
| 430 | case Types::UInt64: return get<UInt64>() < rhs.get<UInt64>(); |
| 431 | case Types::UInt128: return get<UInt128>() < rhs.get<UInt128>(); |
| 432 | case Types::Int64: return get<Int64>() < rhs.get<Int64>(); |
| 433 | case Types::Int128: return get<Int128>() < rhs.get<Int128>(); |
| 434 | case Types::Float64: return get<Float64>() < rhs.get<Float64>(); |
| 435 | case Types::String: return get<String>() < rhs.get<String>(); |
| 436 | case Types::Array: return get<Array>() < rhs.get<Array>(); |
| 437 | case Types::Tuple: return get<Tuple>() < rhs.get<Tuple>(); |
| 438 | case Types::Decimal32: return get<DecimalField<Decimal32>>() < rhs.get<DecimalField<Decimal32>>(); |
| 439 | case Types::Decimal64: return get<DecimalField<Decimal64>>() < rhs.get<DecimalField<Decimal64>>(); |
| 440 | case Types::Decimal128: return get<DecimalField<Decimal128>>() < rhs.get<DecimalField<Decimal128>>(); |
| 441 | case Types::AggregateFunctionState: return get<AggregateFunctionStateData>() < rhs.get<AggregateFunctionStateData>(); |
| 442 | } |
| 443 | |
| 444 | throw Exception("Bad type of Field" , ErrorCodes::BAD_TYPE_OF_FIELD); |
| 445 | } |
| 446 | |
| 447 | bool operator> (const Field & rhs) const |
| 448 | { |
| 449 | return rhs < *this; |
| 450 | } |
| 451 | |
| 452 | bool operator<= (const Field & rhs) const |
| 453 | { |
| 454 | if (which < rhs.which) |
| 455 | return true; |
| 456 | if (which > rhs.which) |
| 457 | return false; |
| 458 | |
| 459 | switch (which) |
| 460 | { |
| 461 | case Types::Null: return true; |
| 462 | case Types::UInt64: return get<UInt64>() <= rhs.get<UInt64>(); |
| 463 | case Types::UInt128: return get<UInt128>() <= rhs.get<UInt128>(); |
| 464 | case Types::Int64: return get<Int64>() <= rhs.get<Int64>(); |
| 465 | case Types::Int128: return get<Int128>() <= rhs.get<Int128>(); |
| 466 | case Types::Float64: return get<Float64>() <= rhs.get<Float64>(); |
| 467 | case Types::String: return get<String>() <= rhs.get<String>(); |
| 468 | case Types::Array: return get<Array>() <= rhs.get<Array>(); |
| 469 | case Types::Tuple: return get<Tuple>() <= rhs.get<Tuple>(); |
| 470 | case Types::Decimal32: return get<DecimalField<Decimal32>>() <= rhs.get<DecimalField<Decimal32>>(); |
| 471 | case Types::Decimal64: return get<DecimalField<Decimal64>>() <= rhs.get<DecimalField<Decimal64>>(); |
| 472 | case Types::Decimal128: return get<DecimalField<Decimal128>>() <= rhs.get<DecimalField<Decimal128>>(); |
| 473 | case Types::AggregateFunctionState: return get<AggregateFunctionStateData>() <= rhs.get<AggregateFunctionStateData>(); |
| 474 | } |
| 475 | |
| 476 | throw Exception("Bad type of Field" , ErrorCodes::BAD_TYPE_OF_FIELD); |
| 477 | } |
| 478 | |
| 479 | bool operator>= (const Field & rhs) const |
| 480 | { |
| 481 | return rhs <= *this; |
| 482 | } |
| 483 | |
| 484 | // More like bitwise equality as opposed to semantic equality: |
| 485 | // Null equals Null and NaN equals NaN. |
| 486 | bool operator== (const Field & rhs) const |
| 487 | { |
| 488 | if (which != rhs.which) |
| 489 | return false; |
| 490 | |
| 491 | switch (which) |
| 492 | { |
| 493 | case Types::Null: return true; |
| 494 | case Types::UInt64: return get<UInt64>() == rhs.get<UInt64>(); |
| 495 | case Types::Int64: return get<Int64>() == rhs.get<Int64>(); |
| 496 | case Types::Float64: |
| 497 | { |
| 498 | // Compare as UInt64 so that NaNs compare as equal. |
| 499 | return reinterpret<UInt64>() == rhs.reinterpret<UInt64>(); |
| 500 | } |
| 501 | case Types::String: return get<String>() == rhs.get<String>(); |
| 502 | case Types::Array: return get<Array>() == rhs.get<Array>(); |
| 503 | case Types::Tuple: return get<Tuple>() == rhs.get<Tuple>(); |
| 504 | case Types::UInt128: return get<UInt128>() == rhs.get<UInt128>(); |
| 505 | case Types::Int128: return get<Int128>() == rhs.get<Int128>(); |
| 506 | case Types::Decimal32: return get<DecimalField<Decimal32>>() == rhs.get<DecimalField<Decimal32>>(); |
| 507 | case Types::Decimal64: return get<DecimalField<Decimal64>>() == rhs.get<DecimalField<Decimal64>>(); |
| 508 | case Types::Decimal128: return get<DecimalField<Decimal128>>() == rhs.get<DecimalField<Decimal128>>(); |
| 509 | case Types::AggregateFunctionState: return get<AggregateFunctionStateData>() == rhs.get<AggregateFunctionStateData>(); |
| 510 | } |
| 511 | |
| 512 | throw Exception("Bad type of Field" , ErrorCodes::BAD_TYPE_OF_FIELD); |
| 513 | } |
| 514 | |
| 515 | bool operator!= (const Field & rhs) const |
| 516 | { |
| 517 | return !(*this == rhs); |
| 518 | } |
| 519 | |
| 520 | /// Field is template parameter, to allow universal reference for field, |
| 521 | /// that is useful for const and non-const . |
| 522 | template <typename F, typename FieldRef> |
| 523 | static auto dispatch(F && f, FieldRef && field) |
| 524 | { |
| 525 | switch (field.which) |
| 526 | { |
| 527 | case Types::Null: return f(field.template get<Null>()); |
| 528 | // gcc 8.2.1 |
| 529 | #if !__clang__ |
| 530 | #pragma GCC diagnostic push |
| 531 | #pragma GCC diagnostic ignored "-Wmaybe-uninitialized" |
| 532 | #endif |
| 533 | case Types::UInt64: return f(field.template get<UInt64>()); |
| 534 | case Types::UInt128: return f(field.template get<UInt128>()); |
| 535 | case Types::Int64: return f(field.template get<Int64>()); |
| 536 | case Types::Float64: return f(field.template get<Float64>()); |
| 537 | case Types::String: return f(field.template get<String>()); |
| 538 | case Types::Array: return f(field.template get<Array>()); |
| 539 | case Types::Tuple: return f(field.template get<Tuple>()); |
| 540 | #if !__clang__ |
| 541 | #pragma GCC diagnostic pop |
| 542 | #endif |
| 543 | case Types::Decimal32: return f(field.template get<DecimalField<Decimal32>>()); |
| 544 | case Types::Decimal64: return f(field.template get<DecimalField<Decimal64>>()); |
| 545 | case Types::Decimal128: return f(field.template get<DecimalField<Decimal128>>()); |
| 546 | case Types::AggregateFunctionState: return f(field.template get<AggregateFunctionStateData>()); |
| 547 | case Types::Int128: |
| 548 | // TODO: investigate where we need Int128 Fields. There are no |
| 549 | // field visitors that support them, and they only arise indirectly |
| 550 | // in some functions that use Decimal columns: they get the |
| 551 | // underlying Field value with get<Int128>(). Probably should be |
| 552 | // switched to DecimalField, but this is a whole endeavor in itself. |
| 553 | throw Exception("Unexpected Int128 in Field::dispatch()" , ErrorCodes::LOGICAL_ERROR); |
| 554 | } |
| 555 | |
| 556 | // GCC 9 complains that control reaches the end, despite that we handle |
| 557 | // all the cases above (maybe because of throw?). Return something to |
| 558 | // silence it. |
| 559 | Null null{}; |
| 560 | return f(null); |
| 561 | } |
| 562 | |
| 563 | |
| 564 | private: |
| 565 | std::aligned_union_t<DBMS_MIN_FIELD_SIZE - sizeof(Types::Which), |
| 566 | Null, UInt64, UInt128, Int64, Int128, Float64, String, Array, Tuple, |
| 567 | DecimalField<Decimal32>, DecimalField<Decimal64>, DecimalField<Decimal128>, AggregateFunctionStateData |
| 568 | > storage; |
| 569 | |
| 570 | Types::Which which; |
| 571 | |
| 572 | |
| 573 | /// Assuming there was no allocated state or it was deallocated (see destroy). |
| 574 | template <typename T> |
| 575 | void createConcrete(T && x) |
| 576 | { |
| 577 | using UnqualifiedType = std::decay_t<T>; |
| 578 | |
| 579 | // In both Field and PODArray, small types may be stored as wider types, |
| 580 | // e.g. char is stored as UInt64. Field can return this extended value |
| 581 | // with get<StorageType>(). To avoid uninitialized results from get(), |
| 582 | // we must initialize the entire wide stored type, and not just the |
| 583 | // nominal type. |
| 584 | using StorageType = NearestFieldType<UnqualifiedType>; |
| 585 | new (&storage) StorageType(std::forward<T>(x)); |
| 586 | which = TypeToEnum<UnqualifiedType>::value; |
| 587 | } |
| 588 | |
| 589 | /// Assuming same types. |
| 590 | template <typename T> |
| 591 | void assignConcrete(T && x) |
| 592 | { |
| 593 | using JustT = std::decay_t<T>; |
| 594 | assert(which == TypeToEnum<JustT>::value); |
| 595 | JustT * MAY_ALIAS ptr = reinterpret_cast<JustT *>(&storage); |
| 596 | *ptr = std::forward<T>(x); |
| 597 | } |
| 598 | |
| 599 | |
| 600 | void create(const Field & x) |
| 601 | { |
| 602 | dispatch([this] (auto & value) { createConcrete(value); }, x); |
| 603 | } |
| 604 | |
| 605 | void create(Field && x) |
| 606 | { |
| 607 | dispatch([this] (auto & value) { createConcrete(std::move(value)); }, x); |
| 608 | } |
| 609 | |
| 610 | void assign(const Field & x) |
| 611 | { |
| 612 | dispatch([this] (auto & value) { assignConcrete(value); }, x); |
| 613 | } |
| 614 | |
| 615 | void assign(Field && x) |
| 616 | { |
| 617 | dispatch([this] (auto & value) { assignConcrete(std::move(value)); }, x); |
| 618 | } |
| 619 | |
| 620 | |
| 621 | void create(const char * data, size_t size) |
| 622 | { |
| 623 | new (&storage) String(data, size); |
| 624 | which = Types::String; |
| 625 | } |
| 626 | |
| 627 | void create(const unsigned char * data, size_t size) |
| 628 | { |
| 629 | create(reinterpret_cast<const char *>(data), size); |
| 630 | } |
| 631 | |
| 632 | ALWAYS_INLINE void destroy() |
| 633 | { |
| 634 | if (which < Types::MIN_NON_POD) |
| 635 | return; |
| 636 | |
| 637 | switch (which) |
| 638 | { |
| 639 | case Types::String: |
| 640 | destroy<String>(); |
| 641 | break; |
| 642 | case Types::Array: |
| 643 | destroy<Array>(); |
| 644 | break; |
| 645 | case Types::Tuple: |
| 646 | destroy<Tuple>(); |
| 647 | break; |
| 648 | case Types::AggregateFunctionState: |
| 649 | destroy<AggregateFunctionStateData>(); |
| 650 | break; |
| 651 | default: |
| 652 | break; |
| 653 | } |
| 654 | |
| 655 | which = Types::Null; /// for exception safety in subsequent calls to destroy and create, when create fails. |
| 656 | } |
| 657 | |
| 658 | template <typename T> |
| 659 | void destroy() |
| 660 | { |
| 661 | T * MAY_ALIAS ptr = reinterpret_cast<T*>(&storage); |
| 662 | ptr->~T(); |
| 663 | } |
| 664 | }; |
| 665 | |
| 666 | #undef DBMS_MIN_FIELD_SIZE |
| 667 | |
| 668 | |
| 669 | template <> struct Field::TypeToEnum<Null> { static const Types::Which value = Types::Null; }; |
| 670 | template <> struct Field::TypeToEnum<UInt64> { static const Types::Which value = Types::UInt64; }; |
| 671 | template <> struct Field::TypeToEnum<UInt128> { static const Types::Which value = Types::UInt128; }; |
| 672 | template <> struct Field::TypeToEnum<Int64> { static const Types::Which value = Types::Int64; }; |
| 673 | template <> struct Field::TypeToEnum<Int128> { static const Types::Which value = Types::Int128; }; |
| 674 | template <> struct Field::TypeToEnum<Float64> { static const Types::Which value = Types::Float64; }; |
| 675 | template <> struct Field::TypeToEnum<String> { static const Types::Which value = Types::String; }; |
| 676 | template <> struct Field::TypeToEnum<Array> { static const Types::Which value = Types::Array; }; |
| 677 | template <> struct Field::TypeToEnum<Tuple> { static const Types::Which value = Types::Tuple; }; |
| 678 | template <> struct Field::TypeToEnum<DecimalField<Decimal32>>{ static const Types::Which value = Types::Decimal32; }; |
| 679 | template <> struct Field::TypeToEnum<DecimalField<Decimal64>>{ static const Types::Which value = Types::Decimal64; }; |
| 680 | template <> struct Field::TypeToEnum<DecimalField<Decimal128>>{ static const Types::Which value = Types::Decimal128; }; |
| 681 | template <> struct Field::TypeToEnum<AggregateFunctionStateData>{ static const Types::Which value = Types::AggregateFunctionState; }; |
| 682 | |
| 683 | template <> struct Field::EnumToType<Field::Types::Null> { using Type = Null; }; |
| 684 | template <> struct Field::EnumToType<Field::Types::UInt64> { using Type = UInt64; }; |
| 685 | template <> struct Field::EnumToType<Field::Types::UInt128> { using Type = UInt128; }; |
| 686 | template <> struct Field::EnumToType<Field::Types::Int64> { using Type = Int64; }; |
| 687 | template <> struct Field::EnumToType<Field::Types::Int128> { using Type = Int128; }; |
| 688 | template <> struct Field::EnumToType<Field::Types::Float64> { using Type = Float64; }; |
| 689 | template <> struct Field::EnumToType<Field::Types::String> { using Type = String; }; |
| 690 | template <> struct Field::EnumToType<Field::Types::Array> { using Type = Array; }; |
| 691 | template <> struct Field::EnumToType<Field::Types::Tuple> { using Type = Tuple; }; |
| 692 | template <> struct Field::EnumToType<Field::Types::Decimal32> { using Type = DecimalField<Decimal32>; }; |
| 693 | template <> struct Field::EnumToType<Field::Types::Decimal64> { using Type = DecimalField<Decimal64>; }; |
| 694 | template <> struct Field::EnumToType<Field::Types::Decimal128> { using Type = DecimalField<Decimal128>; }; |
| 695 | template <> struct Field::EnumToType<Field::Types::AggregateFunctionState> { using Type = DecimalField<AggregateFunctionStateData>; }; |
| 696 | |
| 697 | inline constexpr bool isInt64FieldType(Field::Types::Which t) |
| 698 | { |
| 699 | return t == Field::Types::Int64 |
| 700 | || t == Field::Types::UInt64; |
| 701 | } |
| 702 | |
| 703 | // Field value getter with type checking in debug builds. |
| 704 | template <typename T> |
| 705 | T & Field::get() |
| 706 | { |
| 707 | using ValueType = std::decay_t<T>; |
| 708 | |
| 709 | #ifndef NDEBUG |
| 710 | // Disregard signedness when converting between int64 types. |
| 711 | constexpr Field::Types::Which target = TypeToEnum<NearestFieldType<ValueType>>::value; |
| 712 | assert(target == which |
| 713 | || (isInt64FieldType(target) && isInt64FieldType(which))); |
| 714 | #endif |
| 715 | |
| 716 | ValueType * MAY_ALIAS ptr = reinterpret_cast<ValueType *>(&storage); |
| 717 | return *ptr; |
| 718 | } |
| 719 | |
| 720 | template <typename T> |
| 721 | T & Field::reinterpret() |
| 722 | { |
| 723 | using ValueType = std::decay_t<T>; |
| 724 | ValueType * MAY_ALIAS ptr = reinterpret_cast<ValueType *>(&storage); |
| 725 | return *ptr; |
| 726 | } |
| 727 | |
| 728 | template <typename T> |
| 729 | T get(const Field & field) |
| 730 | { |
| 731 | return field.template get<T>(); |
| 732 | } |
| 733 | |
| 734 | template <typename T> |
| 735 | T get(Field & field) |
| 736 | { |
| 737 | return field.template get<T>(); |
| 738 | } |
| 739 | |
| 740 | template <typename T> |
| 741 | T safeGet(const Field & field) |
| 742 | { |
| 743 | return field.template safeGet<T>(); |
| 744 | } |
| 745 | |
| 746 | template <typename T> |
| 747 | T safeGet(Field & field) |
| 748 | { |
| 749 | return field.template safeGet<T>(); |
| 750 | } |
| 751 | |
| 752 | |
| 753 | template <> struct TypeName<Array> { static std::string get() { return "Array" ; } }; |
| 754 | template <> struct TypeName<Tuple> { static std::string get() { return "Tuple" ; } }; |
| 755 | template <> struct TypeName<AggregateFunctionStateData> { static std::string get() { return "AggregateFunctionState" ; } }; |
| 756 | |
| 757 | template <typename T> |
| 758 | decltype(auto) castToNearestFieldType(T && x) |
| 759 | { |
| 760 | using U = NearestFieldType<std::decay_t<T>>; |
| 761 | if constexpr (std::is_same_v<std::decay_t<T>, U>) |
| 762 | return std::forward<T>(x); |
| 763 | else |
| 764 | return U(x); |
| 765 | } |
| 766 | |
| 767 | /// This (rather tricky) code is to avoid ambiguity in expressions like |
| 768 | /// Field f = 1; |
| 769 | /// instead of |
| 770 | /// Field f = Int64(1); |
| 771 | /// Things to note: |
| 772 | /// 1. float <--> int needs explicit cast |
| 773 | /// 2. customized types needs explicit cast |
| 774 | template <typename T> |
| 775 | Field::Field(T && rhs, std::enable_if_t<!std::is_same_v<std::decay_t<T>, Field>, void *>) |
| 776 | { |
| 777 | auto && val = castToNearestFieldType(std::forward<T>(rhs)); |
| 778 | createConcrete(std::forward<decltype(val)>(val)); |
| 779 | } |
| 780 | |
| 781 | template <typename T> |
| 782 | std::enable_if_t<!std::is_same_v<std::decay_t<T>, Field>, Field &> |
| 783 | Field::operator= (T && rhs) |
| 784 | { |
| 785 | auto && val = castToNearestFieldType(std::forward<T>(rhs)); |
| 786 | using U = decltype(val); |
| 787 | if (which != TypeToEnum<std::decay_t<U>>::value) |
| 788 | { |
| 789 | destroy(); |
| 790 | createConcrete(std::forward<U>(val)); |
| 791 | } |
| 792 | else |
| 793 | assignConcrete(std::forward<U>(val)); |
| 794 | |
| 795 | return *this; |
| 796 | } |
| 797 | |
| 798 | |
| 799 | class ReadBuffer; |
| 800 | class WriteBuffer; |
| 801 | |
| 802 | /// It is assumed that all elements of the array have the same type. |
| 803 | void readBinary(Array & x, ReadBuffer & buf); |
| 804 | |
| 805 | [[noreturn]] inline void readText(Array &, ReadBuffer &) { throw Exception("Cannot read Array." , ErrorCodes::NOT_IMPLEMENTED); } |
| 806 | [[noreturn]] inline void readQuoted(Array &, ReadBuffer &) { throw Exception("Cannot read Array." , ErrorCodes::NOT_IMPLEMENTED); } |
| 807 | |
| 808 | /// It is assumed that all elements of the array have the same type. |
| 809 | /// Also write size and type into buf. UInt64 and Int64 is written in variadic size form |
| 810 | void writeBinary(const Array & x, WriteBuffer & buf); |
| 811 | |
| 812 | void writeText(const Array & x, WriteBuffer & buf); |
| 813 | |
| 814 | [[noreturn]] inline void writeQuoted(const Array &, WriteBuffer &) { throw Exception("Cannot write Array quoted." , ErrorCodes::NOT_IMPLEMENTED); } |
| 815 | |
| 816 | void readBinary(Tuple & x, ReadBuffer & buf); |
| 817 | |
| 818 | [[noreturn]] inline void readText(Tuple &, ReadBuffer &) { throw Exception("Cannot read Tuple." , ErrorCodes::NOT_IMPLEMENTED); } |
| 819 | [[noreturn]] inline void readQuoted(Tuple &, ReadBuffer &) { throw Exception("Cannot read Tuple." , ErrorCodes::NOT_IMPLEMENTED); } |
| 820 | |
| 821 | void writeBinary(const Tuple & x, WriteBuffer & buf); |
| 822 | |
| 823 | void writeText(const Tuple & x, WriteBuffer & buf); |
| 824 | |
| 825 | void writeFieldText(const Field & x, WriteBuffer & buf); |
| 826 | |
| 827 | [[noreturn]] inline void writeQuoted(const Tuple &, WriteBuffer &) { throw Exception("Cannot write Tuple quoted." , ErrorCodes::NOT_IMPLEMENTED); } |
| 828 | } |
| 829 | |