| 1 | /* -*- mode: C++; c-basic-offset: 4; indent-tabs-mode: nil -*- */ |
| 2 | // vim: ft=cpp:expandtab:ts=8:sw=4:softtabstop=4: |
| 3 | #ident "$Id$" |
| 4 | /*====== |
| 5 | This file is part of TokuDB |
| 6 | |
| 7 | |
| 8 | Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved. |
| 9 | |
| 10 | TokuDBis is free software: you can redistribute it and/or modify |
| 11 | it under the terms of the GNU General Public License, version 2, |
| 12 | as published by the Free Software Foundation. |
| 13 | |
| 14 | TokuDB is distributed in the hope that it will be useful, |
| 15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | GNU General Public License for more details. |
| 18 | |
| 19 | You should have received a copy of the GNU General Public License |
| 20 | along with TokuDB. If not, see <http://www.gnu.org/licenses/>. |
| 21 | |
| 22 | ======= */ |
| 23 | |
| 24 | #ident "Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved." |
| 25 | |
| 26 | #include "hatoku_cmp.h" |
| 27 | |
| 28 | #ifdef WORDS_BIGENDIAN |
| 29 | #error "WORDS_BIGENDIAN not supported" |
| 30 | #endif |
| 31 | |
| 32 | // returns true if the field is a valid field to be used |
| 33 | // in a TokuDB table. The non-valid fields are those |
| 34 | // that have been deprecated since before 5.1, and can |
| 35 | // only exist through upgrades of old versions of MySQL |
| 36 | static bool field_valid_for_tokudb_table(Field* field) { |
| 37 | bool ret_val = false; |
| 38 | enum_field_types mysql_type = field->real_type(); |
| 39 | switch (mysql_type) { |
| 40 | case MYSQL_TYPE_LONG: |
| 41 | case MYSQL_TYPE_LONGLONG: |
| 42 | case MYSQL_TYPE_TINY: |
| 43 | case MYSQL_TYPE_SHORT: |
| 44 | case MYSQL_TYPE_INT24: |
| 45 | case MYSQL_TYPE_DATE: |
| 46 | case MYSQL_TYPE_YEAR: |
| 47 | case MYSQL_TYPE_NEWDATE: |
| 48 | case MYSQL_TYPE_ENUM: |
| 49 | case MYSQL_TYPE_SET: |
| 50 | case MYSQL_TYPE_TIME: |
| 51 | case MYSQL_TYPE_DATETIME: |
| 52 | case MYSQL_TYPE_TIMESTAMP: |
| 53 | case MYSQL_TYPE_DOUBLE: |
| 54 | case MYSQL_TYPE_FLOAT: |
| 55 | #if (50600 <= MYSQL_VERSION_ID && MYSQL_VERSION_ID <= 50699) || \ |
| 56 | (50700 <= MYSQL_VERSION_ID && MYSQL_VERSION_ID <= 50799) || \ |
| 57 | (100000 <= MYSQL_VERSION_ID) |
| 58 | case MYSQL_TYPE_DATETIME2: |
| 59 | case MYSQL_TYPE_TIMESTAMP2: |
| 60 | case MYSQL_TYPE_TIME2: |
| 61 | #endif |
| 62 | case MYSQL_TYPE_NEWDECIMAL: |
| 63 | case MYSQL_TYPE_BIT: |
| 64 | case MYSQL_TYPE_STRING: |
| 65 | case MYSQL_TYPE_VARCHAR: |
| 66 | case MYSQL_TYPE_TINY_BLOB: |
| 67 | case MYSQL_TYPE_MEDIUM_BLOB: |
| 68 | case MYSQL_TYPE_BLOB: |
| 69 | case MYSQL_TYPE_LONG_BLOB: |
| 70 | ret_val = true; |
| 71 | goto exit; |
| 72 | // |
| 73 | // I believe these are old types that are no longer |
| 74 | // in any 5.1 tables, so tokudb does not need |
| 75 | // to worry about them |
| 76 | // Putting in this assert in case I am wrong. |
| 77 | // Do not support geometry yet. |
| 78 | // |
| 79 | case MYSQL_TYPE_GEOMETRY: |
| 80 | case MYSQL_TYPE_DECIMAL: |
| 81 | case MYSQL_TYPE_VAR_STRING: |
| 82 | case MYSQL_TYPE_NULL: |
| 83 | case MYSQL_TYPE_VARCHAR_COMPRESSED: |
| 84 | case MYSQL_TYPE_BLOB_COMPRESSED: |
| 85 | ret_val = false; |
| 86 | } |
| 87 | exit: |
| 88 | return ret_val; |
| 89 | } |
| 90 | |
| 91 | static void get_var_field_info( |
| 92 | uint32_t* field_len, // output: length of field |
| 93 | uint32_t* start_offset, // output, length of offset where data starts |
| 94 | uint32_t var_field_index, //input, index of var field we want info on |
| 95 | const uchar* var_field_offset_ptr, //input, pointer to where offset information for all var fields begins |
| 96 | uint32_t num_offset_bytes //input, number of bytes used to store offsets starting at var_field_offset_ptr |
| 97 | ) |
| 98 | { |
| 99 | uint32_t data_start_offset = 0; |
| 100 | uint32_t data_end_offset = 0; |
| 101 | switch (num_offset_bytes) { |
| 102 | case (1): |
| 103 | data_end_offset = (var_field_offset_ptr + var_field_index)[0]; |
| 104 | break; |
| 105 | case (2): |
| 106 | data_end_offset = uint2korr(var_field_offset_ptr + 2*var_field_index); |
| 107 | break; |
| 108 | default: |
| 109 | assert_unreachable(); |
| 110 | } |
| 111 | |
| 112 | if (var_field_index) { |
| 113 | switch (num_offset_bytes) { |
| 114 | case (1): |
| 115 | data_start_offset = (var_field_offset_ptr + var_field_index - 1)[0]; |
| 116 | break; |
| 117 | case (2): |
| 118 | data_start_offset = uint2korr(var_field_offset_ptr + 2*(var_field_index-1)); |
| 119 | break; |
| 120 | default: |
| 121 | assert_unreachable(); |
| 122 | } |
| 123 | } |
| 124 | else { |
| 125 | data_start_offset = 0; |
| 126 | } |
| 127 | |
| 128 | *start_offset = data_start_offset; |
| 129 | assert_always(data_end_offset >= data_start_offset); |
| 130 | *field_len = data_end_offset - data_start_offset; |
| 131 | } |
| 132 | |
| 133 | static void get_blob_field_info( |
| 134 | uint32_t* start_offset, |
| 135 | uint32_t len_of_offsets, |
| 136 | const uchar* var_field_data_ptr, |
| 137 | uint32_t num_offset_bytes |
| 138 | ) |
| 139 | { |
| 140 | uint32_t data_end_offset; |
| 141 | // |
| 142 | // need to set var_field_data_ptr to point to beginning of blobs, which |
| 143 | // is at the end of the var stuff (if they exist), if var stuff does not exist |
| 144 | // then the bottom variable will be 0, and var_field_data_ptr is already |
| 145 | // set correctly |
| 146 | // |
| 147 | if (len_of_offsets) { |
| 148 | switch (num_offset_bytes) { |
| 149 | case (1): |
| 150 | data_end_offset = (var_field_data_ptr - 1)[0]; |
| 151 | break; |
| 152 | case (2): |
| 153 | data_end_offset = uint2korr(var_field_data_ptr - 2); |
| 154 | break; |
| 155 | default: |
| 156 | assert_unreachable(); |
| 157 | } |
| 158 | } |
| 159 | else { |
| 160 | data_end_offset = 0; |
| 161 | } |
| 162 | *start_offset = data_end_offset; |
| 163 | } |
| 164 | |
| 165 | |
| 166 | // this function is pattern matched from |
| 167 | // InnoDB's get_innobase_type_from_mysql_type |
| 168 | static TOKU_TYPE mysql_to_toku_type (Field* field) { |
| 169 | TOKU_TYPE ret_val = toku_type_unknown; |
| 170 | enum_field_types mysql_type = field->real_type(); |
| 171 | switch (mysql_type) { |
| 172 | case MYSQL_TYPE_LONG: |
| 173 | case MYSQL_TYPE_LONGLONG: |
| 174 | case MYSQL_TYPE_TINY: |
| 175 | case MYSQL_TYPE_SHORT: |
| 176 | case MYSQL_TYPE_INT24: |
| 177 | case MYSQL_TYPE_DATE: |
| 178 | case MYSQL_TYPE_YEAR: |
| 179 | case MYSQL_TYPE_NEWDATE: |
| 180 | case MYSQL_TYPE_ENUM: |
| 181 | case MYSQL_TYPE_SET: |
| 182 | ret_val = toku_type_int; |
| 183 | goto exit; |
| 184 | case MYSQL_TYPE_TIME: |
| 185 | case MYSQL_TYPE_DATETIME: |
| 186 | case MYSQL_TYPE_TIMESTAMP: |
| 187 | #ifdef MARIADB_BASE_VERSION |
| 188 | // case to handle fractional seconds in MariaDB |
| 189 | // |
| 190 | if (field->key_type() == HA_KEYTYPE_BINARY) { |
| 191 | ret_val = toku_type_fixbinary; |
| 192 | goto exit; |
| 193 | } |
| 194 | #endif |
| 195 | ret_val = toku_type_int; |
| 196 | goto exit; |
| 197 | case MYSQL_TYPE_DOUBLE: |
| 198 | ret_val = toku_type_double; |
| 199 | goto exit; |
| 200 | case MYSQL_TYPE_FLOAT: |
| 201 | ret_val = toku_type_float; |
| 202 | goto exit; |
| 203 | #if (50600 <= MYSQL_VERSION_ID && MYSQL_VERSION_ID <= 50699) || \ |
| 204 | (50700 <= MYSQL_VERSION_ID && MYSQL_VERSION_ID <= 50799) || \ |
| 205 | (100000 <= MYSQL_VERSION_ID) |
| 206 | case MYSQL_TYPE_DATETIME2: |
| 207 | case MYSQL_TYPE_TIMESTAMP2: |
| 208 | case MYSQL_TYPE_TIME2: |
| 209 | #endif |
| 210 | case MYSQL_TYPE_NEWDECIMAL: |
| 211 | case MYSQL_TYPE_BIT: |
| 212 | ret_val = toku_type_fixbinary; |
| 213 | goto exit; |
| 214 | case MYSQL_TYPE_STRING: |
| 215 | if (field->binary()) { |
| 216 | ret_val = toku_type_fixbinary; |
| 217 | } |
| 218 | else { |
| 219 | ret_val = toku_type_fixstring; |
| 220 | } |
| 221 | goto exit; |
| 222 | case MYSQL_TYPE_VARCHAR: |
| 223 | if (field->binary()) { |
| 224 | ret_val = toku_type_varbinary; |
| 225 | } |
| 226 | else { |
| 227 | ret_val = toku_type_varstring; |
| 228 | } |
| 229 | goto exit; |
| 230 | case MYSQL_TYPE_TINY_BLOB: |
| 231 | case MYSQL_TYPE_MEDIUM_BLOB: |
| 232 | case MYSQL_TYPE_BLOB: |
| 233 | case MYSQL_TYPE_LONG_BLOB: |
| 234 | ret_val = toku_type_blob; |
| 235 | goto exit; |
| 236 | // |
| 237 | // I believe these are old types that are no longer |
| 238 | // in any 5.1 tables, so tokudb does not need |
| 239 | // to worry about them |
| 240 | // Putting in this assert in case I am wrong. |
| 241 | // Do not support geometry yet. |
| 242 | // |
| 243 | case MYSQL_TYPE_GEOMETRY: |
| 244 | case MYSQL_TYPE_DECIMAL: |
| 245 | case MYSQL_TYPE_VAR_STRING: |
| 246 | case MYSQL_TYPE_NULL: |
| 247 | case MYSQL_TYPE_VARCHAR_COMPRESSED: |
| 248 | case MYSQL_TYPE_BLOB_COMPRESSED: |
| 249 | assert_unreachable(); |
| 250 | } |
| 251 | exit: |
| 252 | return ret_val; |
| 253 | } |
| 254 | |
| 255 | |
| 256 | static inline CHARSET_INFO* get_charset_from_num (uint32_t charset_number) { |
| 257 | // |
| 258 | // patternmatched off of InnoDB, due to MySQL bug 42649 |
| 259 | // |
| 260 | if (charset_number == default_charset_info->number) { |
| 261 | return default_charset_info; |
| 262 | } |
| 263 | else if (charset_number == my_charset_latin1.number) { |
| 264 | return &my_charset_latin1; |
| 265 | } |
| 266 | else { |
| 267 | return get_charset(charset_number, MYF(MY_WME)); |
| 268 | } |
| 269 | } |
| 270 | |
| 271 | |
| 272 | |
| 273 | // |
| 274 | // used to read the length of a variable sized field in a tokudb key (buf). |
| 275 | // |
| 276 | static inline uint32_t get_length_from_var_tokudata (uchar* buf, uint32_t length_bytes) { |
| 277 | uint32_t length = (uint32_t)(buf[0]); |
| 278 | if (length_bytes == 2) { |
| 279 | uint32_t rest_of_length = (uint32_t)buf[1]; |
| 280 | length += rest_of_length<<8; |
| 281 | } |
| 282 | return length; |
| 283 | } |
| 284 | |
| 285 | // |
| 286 | // used to deduce the number of bytes used to store the length of a varstring/varbinary |
| 287 | // in a key field stored in tokudb |
| 288 | // |
| 289 | static inline uint32_t get_length_bytes_from_max(uint32_t max_num_bytes) { |
| 290 | return (max_num_bytes > 255) ? 2 : 1; |
| 291 | } |
| 292 | |
| 293 | |
| 294 | |
| 295 | // |
| 296 | // assuming MySQL in little endian, and we are storing in little endian |
| 297 | // |
| 298 | static inline uchar* pack_toku_int (uchar* to_tokudb, uchar* from_mysql, uint32_t num_bytes) { |
| 299 | switch (num_bytes) { |
| 300 | case (1): |
| 301 | memcpy(to_tokudb, from_mysql, 1); |
| 302 | break; |
| 303 | case (2): |
| 304 | memcpy(to_tokudb, from_mysql, 2); |
| 305 | break; |
| 306 | case (3): |
| 307 | memcpy(to_tokudb, from_mysql, 3); |
| 308 | break; |
| 309 | case (4): |
| 310 | memcpy(to_tokudb, from_mysql, 4); |
| 311 | break; |
| 312 | case (8): |
| 313 | memcpy(to_tokudb, from_mysql, 8); |
| 314 | break; |
| 315 | default: |
| 316 | assert_unreachable(); |
| 317 | } |
| 318 | return to_tokudb+num_bytes; |
| 319 | } |
| 320 | |
| 321 | // |
| 322 | // assuming MySQL in little endian, and we are unpacking to little endian |
| 323 | // |
| 324 | static inline uchar* unpack_toku_int(uchar* to_mysql, uchar* from_tokudb, uint32_t num_bytes) { |
| 325 | switch (num_bytes) { |
| 326 | case (1): |
| 327 | memcpy(to_mysql, from_tokudb, 1); |
| 328 | break; |
| 329 | case (2): |
| 330 | memcpy(to_mysql, from_tokudb, 2); |
| 331 | break; |
| 332 | case (3): |
| 333 | memcpy(to_mysql, from_tokudb, 3); |
| 334 | break; |
| 335 | case (4): |
| 336 | memcpy(to_mysql, from_tokudb, 4); |
| 337 | break; |
| 338 | case (8): |
| 339 | memcpy(to_mysql, from_tokudb, 8); |
| 340 | break; |
| 341 | default: |
| 342 | assert_unreachable(); |
| 343 | } |
| 344 | return from_tokudb+num_bytes; |
| 345 | } |
| 346 | |
| 347 | static inline int cmp_toku_int (uchar* a_buf, uchar* b_buf, bool is_unsigned, uint32_t num_bytes) { |
| 348 | int ret_val = 0; |
| 349 | // |
| 350 | // case for unsigned integers |
| 351 | // |
| 352 | if (is_unsigned) { |
| 353 | uint32_t a_num, b_num = 0; |
| 354 | uint64_t a_big_num, b_big_num = 0; |
| 355 | switch (num_bytes) { |
| 356 | case (1): |
| 357 | a_num = *a_buf; |
| 358 | b_num = *b_buf; |
| 359 | ret_val = a_num-b_num; |
| 360 | goto exit; |
| 361 | case (2): |
| 362 | a_num = uint2korr(a_buf); |
| 363 | b_num = uint2korr(b_buf); |
| 364 | ret_val = a_num-b_num; |
| 365 | goto exit; |
| 366 | case (3): |
| 367 | a_num = tokudb_uint3korr(a_buf); |
| 368 | b_num = tokudb_uint3korr(b_buf); |
| 369 | ret_val = a_num-b_num; |
| 370 | goto exit; |
| 371 | case (4): |
| 372 | a_num = uint4korr(a_buf); |
| 373 | b_num = uint4korr(b_buf); |
| 374 | if (a_num < b_num) { |
| 375 | ret_val = -1; goto exit; |
| 376 | } |
| 377 | if (a_num > b_num) { |
| 378 | ret_val = 1; goto exit; |
| 379 | } |
| 380 | ret_val = 0; |
| 381 | goto exit; |
| 382 | case (8): |
| 383 | a_big_num = uint8korr(a_buf); |
| 384 | b_big_num = uint8korr(b_buf); |
| 385 | if (a_big_num < b_big_num) { |
| 386 | ret_val = -1; goto exit; |
| 387 | } |
| 388 | else if (a_big_num > b_big_num) { |
| 389 | ret_val = 1; goto exit; |
| 390 | } |
| 391 | ret_val = 0; |
| 392 | goto exit; |
| 393 | default: |
| 394 | assert_unreachable(); |
| 395 | } |
| 396 | } |
| 397 | // |
| 398 | // case for signed integers |
| 399 | // |
| 400 | else { |
| 401 | int32_t a_num, b_num = 0; |
| 402 | int64_t a_big_num, b_big_num = 0; |
| 403 | switch (num_bytes) { |
| 404 | case (1): |
| 405 | a_num = *(signed char *)a_buf; |
| 406 | b_num = *(signed char *)b_buf; |
| 407 | ret_val = a_num-b_num; |
| 408 | goto exit; |
| 409 | case (2): |
| 410 | a_num = sint2korr(a_buf); |
| 411 | b_num = sint2korr(b_buf); |
| 412 | ret_val = a_num-b_num; |
| 413 | goto exit; |
| 414 | case (3): |
| 415 | a_num = sint3korr(a_buf); |
| 416 | b_num = sint3korr(b_buf); |
| 417 | ret_val = a_num - b_num; |
| 418 | goto exit; |
| 419 | case (4): |
| 420 | a_num = sint4korr(a_buf); |
| 421 | b_num = sint4korr(b_buf); |
| 422 | if (a_num < b_num) { |
| 423 | ret_val = -1; goto exit; |
| 424 | } |
| 425 | if (a_num > b_num) { |
| 426 | ret_val = 1; goto exit; |
| 427 | } |
| 428 | ret_val = 0; |
| 429 | goto exit; |
| 430 | case (8): |
| 431 | a_big_num = sint8korr(a_buf); |
| 432 | b_big_num = sint8korr(b_buf); |
| 433 | if (a_big_num < b_big_num) { |
| 434 | ret_val = -1; goto exit; |
| 435 | } |
| 436 | else if (a_big_num > b_big_num) { |
| 437 | ret_val = 1; goto exit; |
| 438 | } |
| 439 | ret_val = 0; |
| 440 | goto exit; |
| 441 | default: |
| 442 | assert_unreachable(); |
| 443 | } |
| 444 | } |
| 445 | // |
| 446 | // if this is hit, indicates bug in writing of this function |
| 447 | // |
| 448 | assert_unreachable(); |
| 449 | exit: |
| 450 | return ret_val; |
| 451 | } |
| 452 | |
| 453 | static inline uchar* pack_toku_double (uchar* to_tokudb, uchar* from_mysql) { |
| 454 | memcpy(to_tokudb, from_mysql, sizeof(double)); |
| 455 | return to_tokudb + sizeof(double); |
| 456 | } |
| 457 | |
| 458 | |
| 459 | static inline uchar* unpack_toku_double(uchar* to_mysql, uchar* from_tokudb) { |
| 460 | memcpy(to_mysql, from_tokudb, sizeof(double)); |
| 461 | return from_tokudb + sizeof(double); |
| 462 | } |
| 463 | |
| 464 | static inline int cmp_toku_double(uchar* a_buf, uchar* b_buf) { |
| 465 | int ret_val; |
| 466 | double a_num; |
| 467 | double b_num; |
| 468 | doubleget(a_num, a_buf); |
| 469 | doubleget(b_num, b_buf); |
| 470 | if (a_num < b_num) { |
| 471 | ret_val = -1; |
| 472 | goto exit; |
| 473 | } |
| 474 | else if (a_num > b_num) { |
| 475 | ret_val = 1; |
| 476 | goto exit; |
| 477 | } |
| 478 | ret_val = 0; |
| 479 | exit: |
| 480 | return ret_val; |
| 481 | } |
| 482 | |
| 483 | |
| 484 | static inline uchar* pack_toku_float (uchar* to_tokudb, uchar* from_mysql) { |
| 485 | memcpy(to_tokudb, from_mysql, sizeof(float)); |
| 486 | return to_tokudb + sizeof(float); |
| 487 | } |
| 488 | |
| 489 | |
| 490 | static inline uchar* unpack_toku_float(uchar* to_mysql, uchar* from_tokudb) { |
| 491 | memcpy(to_mysql, from_tokudb, sizeof(float)); |
| 492 | return from_tokudb + sizeof(float); |
| 493 | } |
| 494 | |
| 495 | static inline int cmp_toku_float(uchar* a_buf, uchar* b_buf) { |
| 496 | int ret_val; |
| 497 | float a_num; |
| 498 | float b_num; |
| 499 | // |
| 500 | // This is the way Field_float::cmp gets the floats from the buffers |
| 501 | // |
| 502 | memcpy(&a_num, a_buf, sizeof(float)); |
| 503 | memcpy(&b_num, b_buf, sizeof(float)); |
| 504 | if (a_num < b_num) { |
| 505 | ret_val = -1; |
| 506 | goto exit; |
| 507 | } |
| 508 | else if (a_num > b_num) { |
| 509 | ret_val = 1; |
| 510 | goto exit; |
| 511 | } |
| 512 | ret_val = 0; |
| 513 | exit: |
| 514 | return ret_val; |
| 515 | } |
| 516 | |
| 517 | |
| 518 | static inline uchar* pack_toku_binary(uchar* to_tokudb, uchar* from_mysql, uint32_t num_bytes) { |
| 519 | memcpy(to_tokudb, from_mysql, num_bytes); |
| 520 | return to_tokudb + num_bytes; |
| 521 | } |
| 522 | |
| 523 | static inline uchar* unpack_toku_binary(uchar* to_mysql, uchar* from_tokudb, uint32_t num_bytes) { |
| 524 | memcpy(to_mysql, from_tokudb, num_bytes); |
| 525 | return from_tokudb + num_bytes; |
| 526 | } |
| 527 | |
| 528 | |
| 529 | static inline int cmp_toku_binary( |
| 530 | uchar* a_buf, |
| 531 | uint32_t a_num_bytes, |
| 532 | uchar* b_buf, |
| 533 | uint32_t b_num_bytes |
| 534 | ) |
| 535 | { |
| 536 | int ret_val = 0; |
| 537 | uint32_t num_bytes_to_cmp = (a_num_bytes < b_num_bytes) ? a_num_bytes : b_num_bytes; |
| 538 | ret_val = memcmp(a_buf, b_buf, num_bytes_to_cmp); |
| 539 | if ((ret_val != 0) || (a_num_bytes == b_num_bytes)) { |
| 540 | goto exit; |
| 541 | } |
| 542 | if (a_num_bytes < b_num_bytes) { |
| 543 | ret_val = -1; |
| 544 | goto exit; |
| 545 | } |
| 546 | else { |
| 547 | ret_val = 1; |
| 548 | goto exit; |
| 549 | } |
| 550 | exit: |
| 551 | return ret_val; |
| 552 | } |
| 553 | |
| 554 | // |
| 555 | // partially copied from below |
| 556 | // |
| 557 | static uchar* pack_toku_varbinary_from_desc( |
| 558 | uchar* to_tokudb, |
| 559 | const uchar* from_desc, |
| 560 | uint32_t key_part_length, //number of bytes to use to encode the length in to_tokudb |
| 561 | uint32_t field_length //length of field |
| 562 | ) |
| 563 | { |
| 564 | uint32_t length_bytes_in_tokudb = get_length_bytes_from_max(key_part_length); |
| 565 | uint32_t length = field_length; |
| 566 | set_if_smaller(length, key_part_length); |
| 567 | |
| 568 | // |
| 569 | // copy the length bytes, assuming both are in little endian |
| 570 | // |
| 571 | to_tokudb[0] = (uchar)length & 255; |
| 572 | if (length_bytes_in_tokudb > 1) { |
| 573 | to_tokudb[1] = (uchar) (length >> 8); |
| 574 | } |
| 575 | // |
| 576 | // copy the string |
| 577 | // |
| 578 | memcpy(to_tokudb + length_bytes_in_tokudb, from_desc, length); |
| 579 | return to_tokudb + length + length_bytes_in_tokudb; |
| 580 | } |
| 581 | |
| 582 | static inline uchar* pack_toku_varbinary( |
| 583 | uchar* to_tokudb, |
| 584 | uchar* from_mysql, |
| 585 | uint32_t length_bytes_in_mysql, //number of bytes used to encode the length in from_mysql |
| 586 | uint32_t max_num_bytes |
| 587 | ) |
| 588 | { |
| 589 | uint32_t length = 0; |
| 590 | uint32_t length_bytes_in_tokudb; |
| 591 | switch (length_bytes_in_mysql) { |
| 592 | case (0): |
| 593 | length = max_num_bytes; |
| 594 | break; |
| 595 | case (1): |
| 596 | length = (uint32_t)(*from_mysql); |
| 597 | break; |
| 598 | case (2): |
| 599 | length = uint2korr(from_mysql); |
| 600 | break; |
| 601 | case (3): |
| 602 | length = tokudb_uint3korr(from_mysql); |
| 603 | break; |
| 604 | case (4): |
| 605 | length = uint4korr(from_mysql); |
| 606 | break; |
| 607 | } |
| 608 | |
| 609 | // |
| 610 | // from this point on, functionality equivalent to pack_toku_varbinary_from_desc |
| 611 | // |
| 612 | set_if_smaller(length,max_num_bytes); |
| 613 | |
| 614 | length_bytes_in_tokudb = get_length_bytes_from_max(max_num_bytes); |
| 615 | // |
| 616 | // copy the length bytes, assuming both are in little endian |
| 617 | // |
| 618 | to_tokudb[0] = (uchar)length & 255; |
| 619 | if (length_bytes_in_tokudb > 1) { |
| 620 | to_tokudb[1] = (uchar) (length >> 8); |
| 621 | } |
| 622 | // |
| 623 | // copy the string |
| 624 | // |
| 625 | memcpy(to_tokudb + length_bytes_in_tokudb, from_mysql + length_bytes_in_mysql, length); |
| 626 | return to_tokudb + length + length_bytes_in_tokudb; |
| 627 | } |
| 628 | |
| 629 | static inline uchar* unpack_toku_varbinary( |
| 630 | uchar* to_mysql, |
| 631 | uchar* from_tokudb, |
| 632 | uint32_t length_bytes_in_tokudb, // number of bytes used to encode length in from_tokudb |
| 633 | uint32_t length_bytes_in_mysql // number of bytes used to encode length in to_mysql |
| 634 | ) |
| 635 | { |
| 636 | uint32_t length = get_length_from_var_tokudata(from_tokudb, length_bytes_in_tokudb); |
| 637 | |
| 638 | // |
| 639 | // copy the length into the mysql buffer |
| 640 | // |
| 641 | switch (length_bytes_in_mysql) { |
| 642 | case (0): |
| 643 | break; |
| 644 | case (1): |
| 645 | *to_mysql = (uchar) length; |
| 646 | break; |
| 647 | case (2): |
| 648 | int2store(to_mysql, length); |
| 649 | break; |
| 650 | case (3): |
| 651 | int3store(to_mysql, length); |
| 652 | break; |
| 653 | case (4): |
| 654 | int4store(to_mysql, length); |
| 655 | break; |
| 656 | default: |
| 657 | assert_unreachable(); |
| 658 | } |
| 659 | // |
| 660 | // copy the binary data |
| 661 | // |
| 662 | memcpy(to_mysql + length_bytes_in_mysql, from_tokudb + length_bytes_in_tokudb, length); |
| 663 | return from_tokudb + length_bytes_in_tokudb+ length; |
| 664 | } |
| 665 | |
| 666 | static inline int cmp_toku_varbinary( |
| 667 | uchar* a_buf, |
| 668 | uchar* b_buf, |
| 669 | uint32_t length_bytes, //number of bytes used to encode length in a_buf and b_buf |
| 670 | uint32_t* a_bytes_read, |
| 671 | uint32_t* b_bytes_read |
| 672 | ) |
| 673 | { |
| 674 | int ret_val = 0; |
| 675 | uint32_t a_len = get_length_from_var_tokudata(a_buf, length_bytes); |
| 676 | uint32_t b_len = get_length_from_var_tokudata(b_buf, length_bytes); |
| 677 | ret_val = cmp_toku_binary( |
| 678 | a_buf + length_bytes, |
| 679 | a_len, |
| 680 | b_buf + length_bytes, |
| 681 | b_len |
| 682 | ); |
| 683 | *a_bytes_read = a_len + length_bytes; |
| 684 | *b_bytes_read = b_len + length_bytes; |
| 685 | return ret_val; |
| 686 | } |
| 687 | |
| 688 | static inline uchar* pack_toku_blob( |
| 689 | uchar* to_tokudb, |
| 690 | uchar* from_mysql, |
| 691 | uint32_t length_bytes_in_tokudb, //number of bytes to use to encode the length in to_tokudb |
| 692 | uint32_t length_bytes_in_mysql, //number of bytes used to encode the length in from_mysql |
| 693 | uint32_t max_num_bytes, |
| 694 | #if MYSQL_VERSION_ID >= 50600 |
| 695 | const CHARSET_INFO* charset |
| 696 | #else |
| 697 | CHARSET_INFO* charset |
| 698 | #endif |
| 699 | ) |
| 700 | { |
| 701 | uint32_t length = 0; |
| 702 | uint32_t local_char_length = 0; |
| 703 | uchar* blob_buf = NULL; |
| 704 | |
| 705 | switch (length_bytes_in_mysql) { |
| 706 | case (0): |
| 707 | length = max_num_bytes; |
| 708 | break; |
| 709 | case (1): |
| 710 | length = (uint32_t)(*from_mysql); |
| 711 | break; |
| 712 | case (2): |
| 713 | length = uint2korr(from_mysql); |
| 714 | break; |
| 715 | case (3): |
| 716 | length = tokudb_uint3korr(from_mysql); |
| 717 | break; |
| 718 | case (4): |
| 719 | length = uint4korr(from_mysql); |
| 720 | break; |
| 721 | } |
| 722 | set_if_smaller(length,max_num_bytes); |
| 723 | |
| 724 | memcpy(&blob_buf,from_mysql+length_bytes_in_mysql,sizeof(uchar *)); |
| 725 | |
| 726 | local_char_length= ((charset->mbmaxlen > 1) ? |
| 727 | max_num_bytes/charset->mbmaxlen : max_num_bytes); |
| 728 | if (length > local_char_length) |
| 729 | { |
| 730 | local_char_length= my_charpos( |
| 731 | charset, |
| 732 | blob_buf, |
| 733 | blob_buf+length, |
| 734 | local_char_length |
| 735 | ); |
| 736 | set_if_smaller(length, local_char_length); |
| 737 | } |
| 738 | |
| 739 | |
| 740 | // |
| 741 | // copy the length bytes, assuming both are in little endian |
| 742 | // |
| 743 | to_tokudb[0] = (uchar)length & 255; |
| 744 | if (length_bytes_in_tokudb > 1) { |
| 745 | to_tokudb[1] = (uchar) (length >> 8); |
| 746 | } |
| 747 | // |
| 748 | // copy the string |
| 749 | // |
| 750 | memcpy(to_tokudb + length_bytes_in_tokudb, blob_buf, length); |
| 751 | return to_tokudb + length + length_bytes_in_tokudb; |
| 752 | } |
| 753 | |
| 754 | |
| 755 | static inline uchar* unpack_toku_blob( |
| 756 | uchar* to_mysql, |
| 757 | uchar* from_tokudb, |
| 758 | uint32_t length_bytes_in_tokudb, // number of bytes used to encode length in from_tokudb |
| 759 | uint32_t length_bytes_in_mysql // number of bytes used to encode length in to_mysql |
| 760 | ) |
| 761 | { |
| 762 | uint32_t length = get_length_from_var_tokudata(from_tokudb, length_bytes_in_tokudb); |
| 763 | uchar* blob_pos = NULL; |
| 764 | // |
| 765 | // copy the length into the mysql buffer |
| 766 | // |
| 767 | switch (length_bytes_in_mysql) { |
| 768 | case (0): |
| 769 | break; |
| 770 | case (1): |
| 771 | *to_mysql = (uchar) length; |
| 772 | break; |
| 773 | case (2): |
| 774 | int2store(to_mysql, length); |
| 775 | break; |
| 776 | case (3): |
| 777 | int3store(to_mysql, length); |
| 778 | break; |
| 779 | case (4): |
| 780 | int4store(to_mysql, length); |
| 781 | break; |
| 782 | default: |
| 783 | assert_unreachable(); |
| 784 | } |
| 785 | // |
| 786 | // copy the binary data |
| 787 | // |
| 788 | blob_pos = from_tokudb + length_bytes_in_tokudb; |
| 789 | memcpy(to_mysql + length_bytes_in_mysql, &blob_pos, sizeof(uchar *)); |
| 790 | return from_tokudb + length_bytes_in_tokudb+ length; |
| 791 | } |
| 792 | |
| 793 | |
| 794 | // |
| 795 | // partially copied from below |
| 796 | // |
| 797 | static uchar* pack_toku_varstring_from_desc( |
| 798 | uchar* to_tokudb, |
| 799 | const uchar* from_desc, |
| 800 | uint32_t key_part_length, //number of bytes to use to encode the length in to_tokudb |
| 801 | uint32_t field_length, |
| 802 | uint32_t charset_num//length of field |
| 803 | ) |
| 804 | { |
| 805 | CHARSET_INFO* charset = NULL; |
| 806 | uint32_t length_bytes_in_tokudb = get_length_bytes_from_max(key_part_length); |
| 807 | uint32_t length = field_length; |
| 808 | uint32_t local_char_length = 0; |
| 809 | set_if_smaller(length, key_part_length); |
| 810 | |
| 811 | charset = get_charset_from_num(charset_num); |
| 812 | |
| 813 | // |
| 814 | // copy the string |
| 815 | // |
| 816 | local_char_length= ((charset->mbmaxlen > 1) ? |
| 817 | key_part_length/charset->mbmaxlen : key_part_length); |
| 818 | if (length > local_char_length) |
| 819 | { |
| 820 | local_char_length= my_charpos( |
| 821 | charset, |
| 822 | from_desc, |
| 823 | from_desc+length, |
| 824 | local_char_length |
| 825 | ); |
| 826 | set_if_smaller(length, local_char_length); |
| 827 | } |
| 828 | |
| 829 | |
| 830 | // |
| 831 | // copy the length bytes, assuming both are in little endian |
| 832 | // |
| 833 | to_tokudb[0] = (uchar)length & 255; |
| 834 | if (length_bytes_in_tokudb > 1) { |
| 835 | to_tokudb[1] = (uchar) (length >> 8); |
| 836 | } |
| 837 | // |
| 838 | // copy the string |
| 839 | // |
| 840 | memcpy(to_tokudb + length_bytes_in_tokudb, from_desc, length); |
| 841 | return to_tokudb + length + length_bytes_in_tokudb; |
| 842 | } |
| 843 | |
| 844 | static inline uchar* pack_toku_varstring( |
| 845 | uchar* to_tokudb, |
| 846 | uchar* from_mysql, |
| 847 | uint32_t length_bytes_in_tokudb, //number of bytes to use to encode the length in to_tokudb |
| 848 | uint32_t length_bytes_in_mysql, //number of bytes used to encode the length in from_mysql |
| 849 | uint32_t max_num_bytes, |
| 850 | #if MYSQL_VERSION_ID >= 50600 |
| 851 | const CHARSET_INFO *charset |
| 852 | #else |
| 853 | CHARSET_INFO* charset |
| 854 | #endif |
| 855 | ) |
| 856 | { |
| 857 | uint32_t length = 0; |
| 858 | uint32_t local_char_length = 0; |
| 859 | |
| 860 | switch (length_bytes_in_mysql) { |
| 861 | case (0): |
| 862 | length = max_num_bytes; |
| 863 | break; |
| 864 | case (1): |
| 865 | length = (uint32_t)(*from_mysql); |
| 866 | break; |
| 867 | case (2): |
| 868 | length = uint2korr(from_mysql); |
| 869 | break; |
| 870 | case (3): |
| 871 | length = tokudb_uint3korr(from_mysql); |
| 872 | break; |
| 873 | case (4): |
| 874 | length = uint4korr(from_mysql); |
| 875 | break; |
| 876 | } |
| 877 | set_if_smaller(length,max_num_bytes); |
| 878 | |
| 879 | local_char_length= ((charset->mbmaxlen > 1) ? |
| 880 | max_num_bytes/charset->mbmaxlen : max_num_bytes); |
| 881 | if (length > local_char_length) |
| 882 | { |
| 883 | local_char_length= my_charpos( |
| 884 | charset, |
| 885 | from_mysql+length_bytes_in_mysql, |
| 886 | from_mysql+length_bytes_in_mysql+length, |
| 887 | local_char_length |
| 888 | ); |
| 889 | set_if_smaller(length, local_char_length); |
| 890 | } |
| 891 | |
| 892 | |
| 893 | // |
| 894 | // copy the length bytes, assuming both are in little endian |
| 895 | // |
| 896 | to_tokudb[0] = (uchar)length & 255; |
| 897 | if (length_bytes_in_tokudb > 1) { |
| 898 | to_tokudb[1] = (uchar) (length >> 8); |
| 899 | } |
| 900 | // |
| 901 | // copy the string |
| 902 | // |
| 903 | memcpy(to_tokudb + length_bytes_in_tokudb, from_mysql + length_bytes_in_mysql, length); |
| 904 | return to_tokudb + length + length_bytes_in_tokudb; |
| 905 | } |
| 906 | |
| 907 | static inline int cmp_toku_string( |
| 908 | uchar* a_buf, |
| 909 | uint32_t a_num_bytes, |
| 910 | uchar* b_buf, |
| 911 | uint32_t b_num_bytes, |
| 912 | uint32_t charset_number |
| 913 | ) |
| 914 | { |
| 915 | int ret_val = 0; |
| 916 | CHARSET_INFO* charset = NULL; |
| 917 | |
| 918 | charset = get_charset_from_num(charset_number); |
| 919 | |
| 920 | ret_val = charset->coll->strnncollsp( |
| 921 | charset, |
| 922 | a_buf, |
| 923 | a_num_bytes, |
| 924 | b_buf, |
| 925 | b_num_bytes |
| 926 | ); |
| 927 | return ret_val; |
| 928 | } |
| 929 | |
| 930 | static inline int cmp_toku_varstring( |
| 931 | uchar* a_buf, |
| 932 | uchar* b_buf, |
| 933 | uint32_t length_bytes, //number of bytes used to encode length in a_buf and b_buf |
| 934 | uint32_t charset_num, |
| 935 | uint32_t* a_bytes_read, |
| 936 | uint32_t* b_bytes_read |
| 937 | ) |
| 938 | { |
| 939 | int ret_val = 0; |
| 940 | uint32_t a_len = get_length_from_var_tokudata(a_buf, length_bytes); |
| 941 | uint32_t b_len = get_length_from_var_tokudata(b_buf, length_bytes); |
| 942 | ret_val = cmp_toku_string( |
| 943 | a_buf + length_bytes, |
| 944 | a_len, |
| 945 | b_buf + length_bytes, |
| 946 | b_len, |
| 947 | charset_num |
| 948 | ); |
| 949 | *a_bytes_read = a_len + length_bytes; |
| 950 | *b_bytes_read = b_len + length_bytes; |
| 951 | return ret_val; |
| 952 | } |
| 953 | |
| 954 | static inline int tokudb_compare_two_hidden_keys( |
| 955 | const void* new_key_data, |
| 956 | const uint32_t new_key_size, |
| 957 | const void* saved_key_data, |
| 958 | const uint32_t saved_key_size |
| 959 | ) { |
| 960 | assert_always( |
| 961 | (new_key_size >= TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH) && |
| 962 | (saved_key_size >= TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH)); |
| 963 | ulonglong a = hpk_char_to_num((uchar *) new_key_data); |
| 964 | ulonglong b = hpk_char_to_num((uchar *) saved_key_data); |
| 965 | return a < b ? -1 : (a > b ? 1 : 0); |
| 966 | } |
| 967 | |
| 968 | // |
| 969 | // Returns number of bytes used for a given TOKU_TYPE |
| 970 | // in a key descriptor. The number of bytes returned |
| 971 | // here MUST match the number of bytes used for the encoding |
| 972 | // in create_toku_key_descriptor_for_key |
| 973 | // Parameters: |
| 974 | // [in] row_desc - buffer that contains portion of descriptor |
| 975 | // created in create_toku_key_descriptor_for_key. The first |
| 976 | // byte points to the TOKU_TYPE. |
| 977 | // |
| 978 | static uint32_t skip_field_in_descriptor(uchar* row_desc) { |
| 979 | uchar* row_desc_pos = row_desc; |
| 980 | TOKU_TYPE toku_type = (TOKU_TYPE)row_desc_pos[0]; |
| 981 | row_desc_pos++; |
| 982 | |
| 983 | switch (toku_type) { |
| 984 | case (toku_type_hpk): |
| 985 | case (toku_type_double): |
| 986 | case (toku_type_float): |
| 987 | break; |
| 988 | case (toku_type_int): |
| 989 | row_desc_pos += 2; |
| 990 | break; |
| 991 | case (toku_type_fixbinary): |
| 992 | case (toku_type_varbinary): |
| 993 | row_desc_pos++; |
| 994 | break; |
| 995 | case (toku_type_fixstring): |
| 996 | case (toku_type_varstring): |
| 997 | case (toku_type_blob): |
| 998 | row_desc_pos++; |
| 999 | row_desc_pos += sizeof(uint32_t); |
| 1000 | break; |
| 1001 | default: |
| 1002 | assert_unreachable(); |
| 1003 | } |
| 1004 | return (uint32_t)(row_desc_pos - row_desc); |
| 1005 | } |
| 1006 | |
| 1007 | // |
| 1008 | // outputs a descriptor for key into buf. Returns number of bytes used in buf |
| 1009 | // to store the descriptor. Number of bytes used MUST match number of bytes |
| 1010 | // we would skip in skip_field_in_descriptor |
| 1011 | // |
| 1012 | static int create_toku_key_descriptor_for_key(KEY* key, uchar* buf) { |
| 1013 | uchar* pos = buf; |
| 1014 | uint32_t num_bytes_in_field = 0; |
| 1015 | uint32_t charset_num = 0; |
| 1016 | for (uint i = 0; i < key->user_defined_key_parts; i++) { |
| 1017 | Field* field = key->key_part[i].field; |
| 1018 | // |
| 1019 | // The first byte states if there is a null byte |
| 1020 | // 0 means no null byte, non-zer means there |
| 1021 | // is one |
| 1022 | // |
| 1023 | *pos = field->null_bit; |
| 1024 | pos++; |
| 1025 | |
| 1026 | // |
| 1027 | // The second byte for each field is the type |
| 1028 | // |
| 1029 | TOKU_TYPE type = mysql_to_toku_type(field); |
| 1030 | assert_always((int)type < 256); |
| 1031 | *pos = (uchar)(type & 255); |
| 1032 | pos++; |
| 1033 | |
| 1034 | // |
| 1035 | // based on the type, extra data follows afterwards |
| 1036 | // |
| 1037 | switch (type) { |
| 1038 | // |
| 1039 | // two bytes follow for ints, first one states how many |
| 1040 | // bytes the int is (1 , 2, 3, 4 or 8) |
| 1041 | // next one states if it is signed or not |
| 1042 | // |
| 1043 | case (toku_type_int): |
| 1044 | num_bytes_in_field = field->pack_length(); |
| 1045 | assert_always (num_bytes_in_field < 256); |
| 1046 | *pos = (uchar)(num_bytes_in_field & 255); |
| 1047 | pos++; |
| 1048 | *pos = (field->flags & UNSIGNED_FLAG) ? 1 : 0; |
| 1049 | pos++; |
| 1050 | break; |
| 1051 | // |
| 1052 | // nothing follows floats and doubles |
| 1053 | // |
| 1054 | case (toku_type_double): |
| 1055 | case (toku_type_float): |
| 1056 | break; |
| 1057 | // |
| 1058 | // one byte follow stating the length of the field |
| 1059 | // |
| 1060 | case (toku_type_fixbinary): |
| 1061 | num_bytes_in_field = field->pack_length(); |
| 1062 | set_if_smaller(num_bytes_in_field, key->key_part[i].length); |
| 1063 | assert_always(num_bytes_in_field < 256); |
| 1064 | pos[0] = (uchar)(num_bytes_in_field & 255); |
| 1065 | pos++; |
| 1066 | break; |
| 1067 | // |
| 1068 | // one byte follows: the number of bytes used to encode the length |
| 1069 | // |
| 1070 | case (toku_type_varbinary): |
| 1071 | *pos = (uchar)(get_length_bytes_from_max(key->key_part[i].length) & 255); |
| 1072 | pos++; |
| 1073 | break; |
| 1074 | // |
| 1075 | // five bytes follow: one for the number of bytes to encode the length, |
| 1076 | // four for the charset number |
| 1077 | // |
| 1078 | case (toku_type_fixstring): |
| 1079 | case (toku_type_varstring): |
| 1080 | case (toku_type_blob): |
| 1081 | *pos = (uchar)(get_length_bytes_from_max(key->key_part[i].length) & 255); |
| 1082 | pos++; |
| 1083 | charset_num = field->charset()->number; |
| 1084 | pos[0] = (uchar)(charset_num & 255); |
| 1085 | pos[1] = (uchar)((charset_num >> 8) & 255); |
| 1086 | pos[2] = (uchar)((charset_num >> 16) & 255); |
| 1087 | pos[3] = (uchar)((charset_num >> 24) & 255); |
| 1088 | pos += 4; |
| 1089 | break; |
| 1090 | default: |
| 1091 | assert_unreachable(); |
| 1092 | } |
| 1093 | } |
| 1094 | return pos - buf; |
| 1095 | } |
| 1096 | |
| 1097 | |
| 1098 | // |
| 1099 | // Creates a descriptor for a DB. That contains all information necessary |
| 1100 | // to do both key comparisons and data comparisons (for dup-sort databases). |
| 1101 | // |
| 1102 | // There are two types of descriptors we care about: |
| 1103 | // 1) Primary key, (in a no-dup database) |
| 1104 | // 2) secondary keys, which are a secondary key followed by a primary key, |
| 1105 | // but in a no-dup database. |
| 1106 | // |
| 1107 | // I realize this may be confusing, but here is how it works. |
| 1108 | // All DB's have a key compare. |
| 1109 | // The format of the descriptor must be able to handle both. |
| 1110 | // |
| 1111 | // The first four bytes store an offset into the descriptor to the second piece |
| 1112 | // used for data comparisons. So, if in the future we want to append something |
| 1113 | // to the descriptor, we can. |
| 1114 | // |
| 1115 | // |
| 1116 | static int create_toku_key_descriptor( |
| 1117 | uchar* buf, |
| 1118 | bool is_first_hpk, |
| 1119 | KEY* first_key, |
| 1120 | bool is_second_hpk, |
| 1121 | KEY* second_key |
| 1122 | ) |
| 1123 | { |
| 1124 | // |
| 1125 | // The first four bytes always contain the offset of where the first key |
| 1126 | // ends. |
| 1127 | // |
| 1128 | uchar* pos = buf + 4; |
| 1129 | uint32_t num_bytes = 0; |
| 1130 | uint32_t offset = 0; |
| 1131 | |
| 1132 | |
| 1133 | if (is_first_hpk) { |
| 1134 | pos[0] = 0; //say there is NO infinity byte |
| 1135 | pos[1] = 0; //field cannot be NULL, stating it |
| 1136 | pos[2] = toku_type_hpk; |
| 1137 | pos += 3; |
| 1138 | } |
| 1139 | else { |
| 1140 | // |
| 1141 | // first key is NOT a hidden primary key, so we now pack first_key |
| 1142 | // |
| 1143 | pos[0] = 1; //say there is an infinity byte |
| 1144 | pos++; |
| 1145 | num_bytes = create_toku_key_descriptor_for_key(first_key, pos); |
| 1146 | pos += num_bytes; |
| 1147 | } |
| 1148 | |
| 1149 | // |
| 1150 | // if we do not have a second key, we can jump to exit right now |
| 1151 | // we do not have a second key if it is not a hidden primary key |
| 1152 | // and if second_key is NULL |
| 1153 | // |
| 1154 | if (is_first_hpk || (!is_second_hpk && (second_key == NULL)) ) { |
| 1155 | goto exit; |
| 1156 | } |
| 1157 | |
| 1158 | // |
| 1159 | // if we have a second key, and it is an hpk, we need to pack it, and |
| 1160 | // write in the offset to this position in the first four bytes |
| 1161 | // |
| 1162 | if (is_second_hpk) { |
| 1163 | pos[0] = 0; //field cannot be NULL, stating it |
| 1164 | pos[1] = toku_type_hpk; |
| 1165 | pos += 2; |
| 1166 | } |
| 1167 | else { |
| 1168 | // |
| 1169 | // second key is NOT a hidden primary key, so we now pack second_key |
| 1170 | // |
| 1171 | num_bytes = create_toku_key_descriptor_for_key(second_key, pos); |
| 1172 | pos += num_bytes; |
| 1173 | } |
| 1174 | |
| 1175 | |
| 1176 | exit: |
| 1177 | offset = pos - buf; |
| 1178 | buf[0] = (uchar)(offset & 255); |
| 1179 | buf[1] = (uchar)((offset >> 8) & 255); |
| 1180 | buf[2] = (uchar)((offset >> 16) & 255); |
| 1181 | buf[3] = (uchar)((offset >> 24) & 255); |
| 1182 | |
| 1183 | return pos - buf; |
| 1184 | } |
| 1185 | |
| 1186 | |
| 1187 | static inline int compare_toku_field( |
| 1188 | uchar* a_buf, |
| 1189 | uchar* b_buf, |
| 1190 | uchar* row_desc, |
| 1191 | uint32_t* a_bytes_read, |
| 1192 | uint32_t* b_bytes_read, |
| 1193 | uint32_t* row_desc_bytes_read, |
| 1194 | bool* read_string |
| 1195 | ) |
| 1196 | { |
| 1197 | int ret_val = 0; |
| 1198 | uchar* row_desc_pos = row_desc; |
| 1199 | uint32_t num_bytes = 0; |
| 1200 | uint32_t length_bytes = 0; |
| 1201 | uint32_t charset_num = 0; |
| 1202 | bool is_unsigned = false; |
| 1203 | |
| 1204 | TOKU_TYPE toku_type = (TOKU_TYPE)row_desc_pos[0]; |
| 1205 | row_desc_pos++; |
| 1206 | |
| 1207 | switch (toku_type) { |
| 1208 | case (toku_type_hpk): |
| 1209 | ret_val = tokudb_compare_two_hidden_keys( |
| 1210 | a_buf, |
| 1211 | TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH, |
| 1212 | b_buf, |
| 1213 | TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH |
| 1214 | ); |
| 1215 | *a_bytes_read = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH; |
| 1216 | *b_bytes_read = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH; |
| 1217 | break; |
| 1218 | case (toku_type_int): |
| 1219 | num_bytes = row_desc_pos[0]; |
| 1220 | is_unsigned = row_desc_pos[1]; |
| 1221 | ret_val = cmp_toku_int( |
| 1222 | a_buf, |
| 1223 | b_buf, |
| 1224 | is_unsigned, |
| 1225 | num_bytes |
| 1226 | ); |
| 1227 | *a_bytes_read = num_bytes; |
| 1228 | *b_bytes_read = num_bytes; |
| 1229 | row_desc_pos += 2; |
| 1230 | break; |
| 1231 | case (toku_type_double): |
| 1232 | ret_val = cmp_toku_double(a_buf, b_buf); |
| 1233 | *a_bytes_read = sizeof(double); |
| 1234 | *b_bytes_read = sizeof(double); |
| 1235 | break; |
| 1236 | case (toku_type_float): |
| 1237 | ret_val = cmp_toku_float(a_buf, b_buf); |
| 1238 | *a_bytes_read = sizeof(float); |
| 1239 | *b_bytes_read = sizeof(float); |
| 1240 | break; |
| 1241 | case (toku_type_fixbinary): |
| 1242 | num_bytes = row_desc_pos[0]; |
| 1243 | ret_val = cmp_toku_binary(a_buf, num_bytes, b_buf,num_bytes); |
| 1244 | *a_bytes_read = num_bytes; |
| 1245 | *b_bytes_read = num_bytes; |
| 1246 | row_desc_pos++; |
| 1247 | break; |
| 1248 | case (toku_type_varbinary): |
| 1249 | length_bytes = row_desc_pos[0]; |
| 1250 | ret_val = cmp_toku_varbinary( |
| 1251 | a_buf, |
| 1252 | b_buf, |
| 1253 | length_bytes, |
| 1254 | a_bytes_read, |
| 1255 | b_bytes_read |
| 1256 | ); |
| 1257 | row_desc_pos++; |
| 1258 | break; |
| 1259 | case (toku_type_fixstring): |
| 1260 | case (toku_type_varstring): |
| 1261 | case (toku_type_blob): |
| 1262 | length_bytes = row_desc_pos[0]; |
| 1263 | row_desc_pos++; |
| 1264 | // |
| 1265 | // not sure we want to read charset_num like this |
| 1266 | // |
| 1267 | charset_num = *(uint32_t *)row_desc_pos; |
| 1268 | row_desc_pos += sizeof(uint32_t); |
| 1269 | ret_val = cmp_toku_varstring( |
| 1270 | a_buf, |
| 1271 | b_buf, |
| 1272 | length_bytes, |
| 1273 | charset_num, |
| 1274 | a_bytes_read, |
| 1275 | b_bytes_read |
| 1276 | ); |
| 1277 | *read_string = true; |
| 1278 | break; |
| 1279 | default: |
| 1280 | assert_unreachable(); |
| 1281 | } |
| 1282 | |
| 1283 | *row_desc_bytes_read = row_desc_pos - row_desc; |
| 1284 | return ret_val; |
| 1285 | } |
| 1286 | |
| 1287 | // |
| 1288 | // packs a field from a MySQL buffer into a tokudb buffer. |
| 1289 | // Used for inserts/updates |
| 1290 | // |
| 1291 | static uchar* pack_toku_key_field( |
| 1292 | uchar* to_tokudb, |
| 1293 | uchar* from_mysql, |
| 1294 | Field* field, |
| 1295 | uint32_t key_part_length //I really hope this is temporary as I phase out the pack_cmp stuff |
| 1296 | ) |
| 1297 | { |
| 1298 | uchar* new_pos = NULL; |
| 1299 | uint32_t num_bytes = 0; |
| 1300 | TOKU_TYPE toku_type = mysql_to_toku_type(field); |
| 1301 | switch(toku_type) { |
| 1302 | case (toku_type_int): |
| 1303 | assert_always(key_part_length == field->pack_length()); |
| 1304 | new_pos = pack_toku_int( |
| 1305 | to_tokudb, |
| 1306 | from_mysql, |
| 1307 | field->pack_length() |
| 1308 | ); |
| 1309 | goto exit; |
| 1310 | case (toku_type_double): |
| 1311 | assert_always(field->pack_length() == sizeof(double)); |
| 1312 | assert_always(key_part_length == sizeof(double)); |
| 1313 | new_pos = pack_toku_double(to_tokudb, from_mysql); |
| 1314 | goto exit; |
| 1315 | case (toku_type_float): |
| 1316 | assert_always(field->pack_length() == sizeof(float)); |
| 1317 | assert_always(key_part_length == sizeof(float)); |
| 1318 | new_pos = pack_toku_float(to_tokudb, from_mysql); |
| 1319 | goto exit; |
| 1320 | case (toku_type_fixbinary): |
| 1321 | num_bytes = field->pack_length(); |
| 1322 | set_if_smaller(num_bytes, key_part_length); |
| 1323 | new_pos = pack_toku_binary( |
| 1324 | to_tokudb, |
| 1325 | from_mysql, |
| 1326 | num_bytes |
| 1327 | ); |
| 1328 | goto exit; |
| 1329 | case (toku_type_fixstring): |
| 1330 | num_bytes = field->pack_length(); |
| 1331 | set_if_smaller(num_bytes, key_part_length); |
| 1332 | new_pos = pack_toku_varstring( |
| 1333 | to_tokudb, |
| 1334 | from_mysql, |
| 1335 | get_length_bytes_from_max(key_part_length), |
| 1336 | 0, |
| 1337 | num_bytes, |
| 1338 | field->charset() |
| 1339 | ); |
| 1340 | goto exit; |
| 1341 | case (toku_type_varbinary): |
| 1342 | new_pos = pack_toku_varbinary( |
| 1343 | to_tokudb, |
| 1344 | from_mysql, |
| 1345 | ((Field_varstring *)field)->length_bytes, |
| 1346 | key_part_length |
| 1347 | ); |
| 1348 | goto exit; |
| 1349 | case (toku_type_varstring): |
| 1350 | new_pos = pack_toku_varstring( |
| 1351 | to_tokudb, |
| 1352 | from_mysql, |
| 1353 | get_length_bytes_from_max(key_part_length), |
| 1354 | ((Field_varstring *)field)->length_bytes, |
| 1355 | key_part_length, |
| 1356 | field->charset() |
| 1357 | ); |
| 1358 | goto exit; |
| 1359 | case (toku_type_blob): |
| 1360 | new_pos = pack_toku_blob( |
| 1361 | to_tokudb, |
| 1362 | from_mysql, |
| 1363 | get_length_bytes_from_max(key_part_length), |
| 1364 | ((Field_blob *)field)->row_pack_length(), //only calling this because packlength is returned |
| 1365 | key_part_length, |
| 1366 | field->charset() |
| 1367 | ); |
| 1368 | goto exit; |
| 1369 | default: |
| 1370 | assert_unreachable(); |
| 1371 | } |
| 1372 | assert_unreachable(); |
| 1373 | exit: |
| 1374 | return new_pos; |
| 1375 | } |
| 1376 | |
| 1377 | // |
| 1378 | // packs a field from a MySQL buffer into a tokudb buffer. |
| 1379 | // Used for queries. The only difference between this function |
| 1380 | // and pack_toku_key_field is that all variable sized columns |
| 1381 | // use 2 bytes to encode the length, regardless of the field |
| 1382 | // So varchar(4) will still use 2 bytes to encode the field |
| 1383 | // |
| 1384 | static uchar* pack_key_toku_key_field( |
| 1385 | uchar* to_tokudb, |
| 1386 | uchar* from_mysql, |
| 1387 | Field* field, |
| 1388 | uint32_t key_part_length //I really hope this is temporary as I phase out the pack_cmp stuff |
| 1389 | ) |
| 1390 | { |
| 1391 | uchar* new_pos = NULL; |
| 1392 | TOKU_TYPE toku_type = mysql_to_toku_type(field); |
| 1393 | switch(toku_type) { |
| 1394 | case (toku_type_int): |
| 1395 | case (toku_type_double): |
| 1396 | case (toku_type_float): |
| 1397 | case (toku_type_fixbinary): |
| 1398 | case (toku_type_fixstring): |
| 1399 | new_pos = pack_toku_key_field(to_tokudb, from_mysql, field, key_part_length); |
| 1400 | goto exit; |
| 1401 | case (toku_type_varbinary): |
| 1402 | new_pos = pack_toku_varbinary( |
| 1403 | to_tokudb, |
| 1404 | from_mysql, |
| 1405 | 2, // for some idiotic reason, 2 bytes are always used here, regardless of length of field |
| 1406 | key_part_length |
| 1407 | ); |
| 1408 | goto exit; |
| 1409 | case (toku_type_varstring): |
| 1410 | case (toku_type_blob): |
| 1411 | new_pos = pack_toku_varstring( |
| 1412 | to_tokudb, |
| 1413 | from_mysql, |
| 1414 | get_length_bytes_from_max(key_part_length), |
| 1415 | 2, // for some idiotic reason, 2 bytes are always used here, regardless of length of field |
| 1416 | key_part_length, |
| 1417 | field->charset() |
| 1418 | ); |
| 1419 | goto exit; |
| 1420 | default: |
| 1421 | assert_unreachable(); |
| 1422 | } |
| 1423 | |
| 1424 | assert_unreachable(); |
| 1425 | exit: |
| 1426 | return new_pos; |
| 1427 | } |
| 1428 | |
| 1429 | |
| 1430 | uchar* unpack_toku_key_field( |
| 1431 | uchar* to_mysql, |
| 1432 | uchar* from_tokudb, |
| 1433 | Field* field, |
| 1434 | uint32_t key_part_length) { |
| 1435 | |
| 1436 | uchar* new_pos = NULL; |
| 1437 | uint32_t num_bytes = 0; |
| 1438 | uint32_t num_bytes_copied; |
| 1439 | TOKU_TYPE toku_type = mysql_to_toku_type(field); |
| 1440 | switch(toku_type) { |
| 1441 | case (toku_type_int): |
| 1442 | assert_always(key_part_length == field->pack_length()); |
| 1443 | new_pos = unpack_toku_int( |
| 1444 | to_mysql, |
| 1445 | from_tokudb, |
| 1446 | field->pack_length() |
| 1447 | ); |
| 1448 | goto exit; |
| 1449 | case (toku_type_double): |
| 1450 | assert_always(field->pack_length() == sizeof(double)); |
| 1451 | assert_always(key_part_length == sizeof(double)); |
| 1452 | new_pos = unpack_toku_double(to_mysql, from_tokudb); |
| 1453 | goto exit; |
| 1454 | case (toku_type_float): |
| 1455 | assert_always(field->pack_length() == sizeof(float)); |
| 1456 | assert_always(key_part_length == sizeof(float)); |
| 1457 | new_pos = unpack_toku_float(to_mysql, from_tokudb); |
| 1458 | goto exit; |
| 1459 | case (toku_type_fixbinary): |
| 1460 | num_bytes = field->pack_length(); |
| 1461 | set_if_smaller(num_bytes, key_part_length); |
| 1462 | new_pos = unpack_toku_binary( |
| 1463 | to_mysql, |
| 1464 | from_tokudb, |
| 1465 | num_bytes); |
| 1466 | goto exit; |
| 1467 | case (toku_type_fixstring): |
| 1468 | num_bytes = field->pack_length(); |
| 1469 | new_pos = unpack_toku_varbinary( |
| 1470 | to_mysql, |
| 1471 | from_tokudb, |
| 1472 | get_length_bytes_from_max(key_part_length), |
| 1473 | 0); |
| 1474 | num_bytes_copied = |
| 1475 | new_pos - |
| 1476 | (from_tokudb + get_length_bytes_from_max(key_part_length)); |
| 1477 | assert_always(num_bytes_copied <= num_bytes); |
| 1478 | memset( |
| 1479 | to_mysql + num_bytes_copied, |
| 1480 | field->charset()->pad_char, |
| 1481 | num_bytes - num_bytes_copied); |
| 1482 | goto exit; |
| 1483 | case (toku_type_varbinary): |
| 1484 | case (toku_type_varstring): |
| 1485 | new_pos = unpack_toku_varbinary( |
| 1486 | to_mysql, |
| 1487 | from_tokudb, |
| 1488 | get_length_bytes_from_max(key_part_length), |
| 1489 | ((Field_varstring*)field)->length_bytes); |
| 1490 | goto exit; |
| 1491 | case (toku_type_blob): |
| 1492 | new_pos = unpack_toku_blob( |
| 1493 | to_mysql, |
| 1494 | from_tokudb, |
| 1495 | get_length_bytes_from_max(key_part_length), |
| 1496 | //only calling this because packlength is returned |
| 1497 | ((Field_blob *)field)->row_pack_length()); |
| 1498 | goto exit; |
| 1499 | default: |
| 1500 | assert_unreachable(); |
| 1501 | } |
| 1502 | assert_unreachable(); |
| 1503 | exit: |
| 1504 | return new_pos; |
| 1505 | } |
| 1506 | |
| 1507 | |
| 1508 | static int tokudb_compare_two_keys( |
| 1509 | const void* new_key_data, |
| 1510 | const uint32_t new_key_size, |
| 1511 | const void* saved_key_data, |
| 1512 | const uint32_t saved_key_size, |
| 1513 | const void* row_desc, |
| 1514 | const uint32_t row_desc_size, |
| 1515 | bool cmp_prefix, |
| 1516 | bool* read_string) { |
| 1517 | |
| 1518 | int ret_val = 0; |
| 1519 | int8_t new_key_inf_val = COL_NEG_INF; |
| 1520 | int8_t saved_key_inf_val = COL_NEG_INF; |
| 1521 | |
| 1522 | uchar* row_desc_ptr = (uchar *)row_desc; |
| 1523 | uchar *new_key_ptr = (uchar *)new_key_data; |
| 1524 | uchar *saved_key_ptr = (uchar *)saved_key_data; |
| 1525 | |
| 1526 | uint32_t new_key_bytes_left = new_key_size; |
| 1527 | uint32_t saved_key_bytes_left = saved_key_size; |
| 1528 | |
| 1529 | // |
| 1530 | // if the keys have an infinity byte, set it |
| 1531 | // |
| 1532 | if (row_desc_ptr[0]) { |
| 1533 | new_key_inf_val = (int8_t)new_key_ptr[0]; |
| 1534 | saved_key_inf_val = (int8_t)saved_key_ptr[0]; |
| 1535 | new_key_ptr++; |
| 1536 | saved_key_ptr++; |
| 1537 | } |
| 1538 | row_desc_ptr++; |
| 1539 | |
| 1540 | while ((uint32_t)(new_key_ptr - (uchar*)new_key_data) < new_key_size && |
| 1541 | (uint32_t)(saved_key_ptr - (uchar*)saved_key_data) < saved_key_size && |
| 1542 | (uint32_t)(row_desc_ptr - (uchar*)row_desc) < row_desc_size) { |
| 1543 | uint32_t new_key_field_length; |
| 1544 | uint32_t saved_key_field_length; |
| 1545 | uint32_t row_desc_field_length; |
| 1546 | // |
| 1547 | // if there is a null byte at this point in the key |
| 1548 | // |
| 1549 | if (row_desc_ptr[0]) { |
| 1550 | // |
| 1551 | // compare null bytes. If different, return |
| 1552 | // |
| 1553 | if (new_key_ptr[0] != saved_key_ptr[0]) { |
| 1554 | ret_val = ((int) *new_key_ptr - (int) *saved_key_ptr); |
| 1555 | goto exit; |
| 1556 | } |
| 1557 | saved_key_ptr++; |
| 1558 | // |
| 1559 | // in case we just read the fact that new_key_ptr and saved_key_ptr |
| 1560 | // have NULL as their next field |
| 1561 | // |
| 1562 | if (!*new_key_ptr++) { |
| 1563 | // |
| 1564 | // skip row_desc_ptr[0] read in if clause |
| 1565 | // |
| 1566 | row_desc_ptr++; |
| 1567 | // |
| 1568 | // skip data that describes rest of field |
| 1569 | // |
| 1570 | row_desc_ptr += skip_field_in_descriptor(row_desc_ptr); |
| 1571 | continue; |
| 1572 | } |
| 1573 | } |
| 1574 | row_desc_ptr++; |
| 1575 | |
| 1576 | ret_val = compare_toku_field( |
| 1577 | new_key_ptr, |
| 1578 | saved_key_ptr, |
| 1579 | row_desc_ptr, |
| 1580 | &new_key_field_length, |
| 1581 | &saved_key_field_length, |
| 1582 | &row_desc_field_length, |
| 1583 | read_string); |
| 1584 | new_key_ptr += new_key_field_length; |
| 1585 | saved_key_ptr += saved_key_field_length; |
| 1586 | row_desc_ptr += row_desc_field_length; |
| 1587 | if (ret_val) { |
| 1588 | goto exit; |
| 1589 | } |
| 1590 | |
| 1591 | assert_always( |
| 1592 | (uint32_t)(new_key_ptr - (uchar*)new_key_data) <= new_key_size); |
| 1593 | assert_always( |
| 1594 | (uint32_t)(saved_key_ptr - (uchar*)saved_key_data) <= saved_key_size); |
| 1595 | assert_always( |
| 1596 | (uint32_t)(row_desc_ptr - (uchar*)row_desc) <= row_desc_size); |
| 1597 | } |
| 1598 | new_key_bytes_left = |
| 1599 | new_key_size - ((uint32_t)(new_key_ptr - (uchar*)new_key_data)); |
| 1600 | saved_key_bytes_left = |
| 1601 | saved_key_size - ((uint32_t)(saved_key_ptr - (uchar*)saved_key_data)); |
| 1602 | if (cmp_prefix) { |
| 1603 | ret_val = 0; |
| 1604 | } else if (new_key_bytes_left== 0 && saved_key_bytes_left== 0) { |
| 1605 | // in this case, read both keys to completion, now read infinity byte |
| 1606 | ret_val = new_key_inf_val - saved_key_inf_val; |
| 1607 | } else if (new_key_bytes_left == 0 && saved_key_bytes_left > 0) { |
| 1608 | // at this point, one SHOULD be 0 |
| 1609 | ret_val = (new_key_inf_val == COL_POS_INF ) ? 1 : -1; |
| 1610 | } else if (new_key_bytes_left > 0 && saved_key_bytes_left == 0) { |
| 1611 | ret_val = (saved_key_inf_val == COL_POS_INF ) ? -1 : 1; |
| 1612 | } else { |
| 1613 | // this should never happen, perhaps we should assert(false) |
| 1614 | assert_unreachable(); |
| 1615 | ret_val = new_key_bytes_left - saved_key_bytes_left; |
| 1616 | } |
| 1617 | exit: |
| 1618 | return ret_val; |
| 1619 | } |
| 1620 | |
| 1621 | static int simple_memcmp(const DBT *keya, const DBT *keyb) { |
| 1622 | int cmp; |
| 1623 | int num_bytes_cmp = keya->size < keyb->size ? |
| 1624 | keya->size : keyb->size; |
| 1625 | cmp = memcmp(keya->data,keyb->data,num_bytes_cmp); |
| 1626 | if (cmp == 0 && (keya->size != keyb->size)) { |
| 1627 | cmp = keya->size < keyb->size ? -1 : 1; |
| 1628 | } |
| 1629 | return cmp; |
| 1630 | } |
| 1631 | |
| 1632 | // comparison function to be used by the fractal trees. |
| 1633 | static int tokudb_cmp_dbt_key(DB* file, const DBT *keya, const DBT *keyb) { |
| 1634 | int cmp; |
| 1635 | if (file->cmp_descriptor->dbt.size == 0) { |
| 1636 | cmp = simple_memcmp(keya, keyb); |
| 1637 | } |
| 1638 | else { |
| 1639 | bool read_string = false; |
| 1640 | cmp = tokudb_compare_two_keys( |
| 1641 | keya->data, |
| 1642 | keya->size, |
| 1643 | keyb->data, |
| 1644 | keyb->size, |
| 1645 | (uchar *)file->cmp_descriptor->dbt.data + 4, |
| 1646 | (*(uint32_t *)file->cmp_descriptor->dbt.data) - 4, |
| 1647 | false, |
| 1648 | &read_string |
| 1649 | ); |
| 1650 | // comparison above may be case-insensitive, but fractal tree |
| 1651 | // needs to distinguish between different data, so we do this |
| 1652 | // additional check here |
| 1653 | if (read_string && (cmp == 0)) { |
| 1654 | cmp = simple_memcmp(keya, keyb); |
| 1655 | } |
| 1656 | } |
| 1657 | return cmp; |
| 1658 | } |
| 1659 | |
| 1660 | //TODO: QQQ Only do one direction for prefix. |
| 1661 | static int tokudb_prefix_cmp_dbt_key(DB *file, const DBT *keya, const DBT *keyb) { |
| 1662 | // calls to this function are done by the handlerton, and are |
| 1663 | // comparing just the keys as MySQL would compare them. |
| 1664 | bool read_string = false; |
| 1665 | int cmp = tokudb_compare_two_keys( |
| 1666 | keya->data, |
| 1667 | keya->size, |
| 1668 | keyb->data, |
| 1669 | keyb->size, |
| 1670 | (uchar *)file->cmp_descriptor->dbt.data + 4, |
| 1671 | *(uint32_t *)file->cmp_descriptor->dbt.data - 4, |
| 1672 | true, |
| 1673 | &read_string |
| 1674 | ); |
| 1675 | return cmp; |
| 1676 | } |
| 1677 | |
| 1678 | static int tokudb_compare_two_key_parts( |
| 1679 | const void* new_key_data, |
| 1680 | const uint32_t new_key_size, |
| 1681 | const void* saved_key_data, |
| 1682 | const uint32_t saved_key_size, |
| 1683 | const void* row_desc, |
| 1684 | const uint32_t row_desc_size, |
| 1685 | uint max_parts |
| 1686 | ) |
| 1687 | { |
| 1688 | int ret_val = 0; |
| 1689 | |
| 1690 | uchar* row_desc_ptr = (uchar *)row_desc; |
| 1691 | uchar *new_key_ptr = (uchar *)new_key_data; |
| 1692 | uchar *saved_key_ptr = (uchar *)saved_key_data; |
| 1693 | |
| 1694 | // |
| 1695 | // if the keys have an infinity byte, set it |
| 1696 | // |
| 1697 | if (row_desc_ptr[0]) { |
| 1698 | // new_key_inf_val = (int8_t)new_key_ptr[0]; |
| 1699 | // saved_key_inf_val = (int8_t)saved_key_ptr[0]; |
| 1700 | new_key_ptr++; |
| 1701 | saved_key_ptr++; |
| 1702 | } |
| 1703 | row_desc_ptr++; |
| 1704 | |
| 1705 | for (uint i = 0; i < max_parts; i++) { |
| 1706 | if (!((uint32_t)(new_key_ptr - (uchar *)new_key_data) < new_key_size && |
| 1707 | (uint32_t)(saved_key_ptr - (uchar *)saved_key_data) < saved_key_size && |
| 1708 | (uint32_t)(row_desc_ptr - (uchar *)row_desc) < row_desc_size)) |
| 1709 | break; |
| 1710 | uint32_t new_key_field_length; |
| 1711 | uint32_t saved_key_field_length; |
| 1712 | uint32_t row_desc_field_length; |
| 1713 | // |
| 1714 | // if there is a null byte at this point in the key |
| 1715 | // |
| 1716 | if (row_desc_ptr[0]) { |
| 1717 | // |
| 1718 | // compare null bytes. If different, return |
| 1719 | // |
| 1720 | if (new_key_ptr[0] != saved_key_ptr[0]) { |
| 1721 | ret_val = ((int) *new_key_ptr - (int) *saved_key_ptr); |
| 1722 | goto exit; |
| 1723 | } |
| 1724 | saved_key_ptr++; |
| 1725 | // |
| 1726 | // in case we just read the fact that new_key_ptr and saved_key_ptr |
| 1727 | // have NULL as their next field |
| 1728 | // |
| 1729 | if (!*new_key_ptr++) { |
| 1730 | // |
| 1731 | // skip row_desc_ptr[0] read in if clause |
| 1732 | // |
| 1733 | row_desc_ptr++; |
| 1734 | // |
| 1735 | // skip data that describes rest of field |
| 1736 | // |
| 1737 | row_desc_ptr += skip_field_in_descriptor(row_desc_ptr); |
| 1738 | continue; |
| 1739 | } |
| 1740 | } |
| 1741 | row_desc_ptr++; |
| 1742 | bool read_string = false; |
| 1743 | ret_val = compare_toku_field( |
| 1744 | new_key_ptr, |
| 1745 | saved_key_ptr, |
| 1746 | row_desc_ptr, |
| 1747 | &new_key_field_length, |
| 1748 | &saved_key_field_length, |
| 1749 | &row_desc_field_length, |
| 1750 | &read_string |
| 1751 | ); |
| 1752 | new_key_ptr += new_key_field_length; |
| 1753 | saved_key_ptr += saved_key_field_length; |
| 1754 | row_desc_ptr += row_desc_field_length; |
| 1755 | if (ret_val) { |
| 1756 | goto exit; |
| 1757 | } |
| 1758 | |
| 1759 | assert_always((uint32_t)(new_key_ptr - (uchar *)new_key_data) <= new_key_size); |
| 1760 | assert_always((uint32_t)(saved_key_ptr - (uchar *)saved_key_data) <= saved_key_size); |
| 1761 | assert_always((uint32_t)(row_desc_ptr - (uchar *)row_desc) <= row_desc_size); |
| 1762 | } |
| 1763 | |
| 1764 | ret_val = 0; |
| 1765 | exit: |
| 1766 | return ret_val; |
| 1767 | } |
| 1768 | |
| 1769 | static int tokudb_cmp_dbt_key_parts(DB *file, const DBT *keya, const DBT *keyb, uint max_parts) { |
| 1770 | assert_always(file->cmp_descriptor->dbt.size); |
| 1771 | return tokudb_compare_two_key_parts( |
| 1772 | keya->data, |
| 1773 | keya->size, |
| 1774 | keyb->data, |
| 1775 | keyb->size, |
| 1776 | (uchar *)file->cmp_descriptor->dbt.data + 4, |
| 1777 | (*(uint32_t *)file->cmp_descriptor->dbt.data) - 4, |
| 1778 | max_parts); |
| 1779 | } |
| 1780 | |
| 1781 | static uint32_t create_toku_main_key_pack_descriptor ( |
| 1782 | uchar* buf |
| 1783 | ) |
| 1784 | { |
| 1785 | // |
| 1786 | // The first four bytes always contain the offset of where the first key |
| 1787 | // ends. |
| 1788 | // |
| 1789 | uchar* pos = buf + 4; |
| 1790 | uint32_t offset = 0; |
| 1791 | // |
| 1792 | // one byte states if this is the main dictionary |
| 1793 | // |
| 1794 | pos[0] = 1; |
| 1795 | pos++; |
| 1796 | goto exit; |
| 1797 | |
| 1798 | |
| 1799 | exit: |
| 1800 | offset = pos - buf; |
| 1801 | buf[0] = (uchar)(offset & 255); |
| 1802 | buf[1] = (uchar)((offset >> 8) & 255); |
| 1803 | buf[2] = (uchar)((offset >> 16) & 255); |
| 1804 | buf[3] = (uchar)((offset >> 24) & 255); |
| 1805 | |
| 1806 | return pos - buf; |
| 1807 | } |
| 1808 | |
| 1809 | #define COL_HAS_NO_CHARSET 0x44 |
| 1810 | #define COL_HAS_CHARSET 0x55 |
| 1811 | |
| 1812 | #define COL_FIX_PK_OFFSET 0x66 |
| 1813 | #define COL_VAR_PK_OFFSET 0x77 |
| 1814 | |
| 1815 | #define CK_FIX_RANGE 0x88 |
| 1816 | #define CK_VAR_RANGE 0x99 |
| 1817 | |
| 1818 | #define COPY_OFFSET_TO_BUF memcpy ( \ |
| 1819 | pos, \ |
| 1820 | &kc_info->cp_info[pk_index][field_index].col_pack_val, \ |
| 1821 | sizeof(uint32_t) \ |
| 1822 | ); \ |
| 1823 | pos += sizeof(uint32_t); |
| 1824 | |
| 1825 | |
| 1826 | static uint32_t pack_desc_pk_info(uchar* buf, KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, KEY_PART_INFO* key_part) { |
| 1827 | uchar* pos = buf; |
| 1828 | uint16 field_index = key_part->field->field_index; |
| 1829 | Field* field = table_share->field[field_index]; |
| 1830 | TOKU_TYPE toku_type = mysql_to_toku_type(field); |
| 1831 | uint32_t key_part_length = key_part->length; |
| 1832 | uint32_t field_length; |
| 1833 | uchar len_bytes = 0; |
| 1834 | |
| 1835 | switch(toku_type) { |
| 1836 | case (toku_type_int): |
| 1837 | case (toku_type_double): |
| 1838 | case (toku_type_float): |
| 1839 | pos[0] = COL_FIX_FIELD; |
| 1840 | pos++; |
| 1841 | assert_always(kc_info->field_lengths[field_index] < 256); |
| 1842 | pos[0] = kc_info->field_lengths[field_index]; |
| 1843 | pos++; |
| 1844 | break; |
| 1845 | case (toku_type_fixbinary): |
| 1846 | pos[0] = COL_FIX_FIELD; |
| 1847 | pos++; |
| 1848 | field_length = field->pack_length(); |
| 1849 | set_if_smaller(key_part_length, field_length); |
| 1850 | assert_always(key_part_length < 256); |
| 1851 | pos[0] = (uchar)key_part_length; |
| 1852 | pos++; |
| 1853 | break; |
| 1854 | case (toku_type_fixstring): |
| 1855 | case (toku_type_varbinary): |
| 1856 | case (toku_type_varstring): |
| 1857 | case (toku_type_blob): |
| 1858 | pos[0] = COL_VAR_FIELD; |
| 1859 | pos++; |
| 1860 | len_bytes = (key_part_length > 255) ? 2 : 1; |
| 1861 | pos[0] = len_bytes; |
| 1862 | pos++; |
| 1863 | break; |
| 1864 | default: |
| 1865 | assert_unreachable(); |
| 1866 | } |
| 1867 | |
| 1868 | return pos - buf; |
| 1869 | } |
| 1870 | |
| 1871 | static uint32_t pack_desc_pk_offset_info( |
| 1872 | uchar* buf, |
| 1873 | KEY_AND_COL_INFO* kc_info, |
| 1874 | TABLE_SHARE* table_share, |
| 1875 | KEY_PART_INFO* key_part, |
| 1876 | KEY* prim_key, |
| 1877 | uchar* pk_info |
| 1878 | ) |
| 1879 | { |
| 1880 | uchar* pos = buf; |
| 1881 | uint16 field_index = key_part->field->field_index; |
| 1882 | bool found_col_in_pk = false; |
| 1883 | uint32_t index_in_pk; |
| 1884 | |
| 1885 | bool is_constant_offset = true; |
| 1886 | uint32_t offset = 0; |
| 1887 | for (uint i = 0; i < prim_key->user_defined_key_parts; i++) { |
| 1888 | KEY_PART_INFO curr = prim_key->key_part[i]; |
| 1889 | uint16 curr_field_index = curr.field->field_index; |
| 1890 | |
| 1891 | if (pk_info[2*i] == COL_VAR_FIELD) { |
| 1892 | is_constant_offset = false; |
| 1893 | } |
| 1894 | |
| 1895 | if (curr_field_index == field_index) { |
| 1896 | found_col_in_pk = true; |
| 1897 | index_in_pk = i; |
| 1898 | break; |
| 1899 | } |
| 1900 | offset += pk_info[2*i + 1]; |
| 1901 | } |
| 1902 | assert_always(found_col_in_pk); |
| 1903 | if (is_constant_offset) { |
| 1904 | pos[0] = COL_FIX_PK_OFFSET; |
| 1905 | pos++; |
| 1906 | |
| 1907 | memcpy (pos, &offset, sizeof(offset)); |
| 1908 | pos += sizeof(offset); |
| 1909 | } |
| 1910 | else { |
| 1911 | pos[0] = COL_VAR_PK_OFFSET; |
| 1912 | pos++; |
| 1913 | |
| 1914 | memcpy(pos, &index_in_pk, sizeof(index_in_pk)); |
| 1915 | pos += sizeof(index_in_pk); |
| 1916 | } |
| 1917 | return pos - buf; |
| 1918 | } |
| 1919 | |
| 1920 | static uint32_t pack_desc_offset_info(uchar* buf, KEY_AND_COL_INFO* kc_info, uint pk_index, TABLE_SHARE* table_share, KEY_PART_INFO* key_part) { |
| 1921 | uchar* pos = buf; |
| 1922 | uint16 field_index = key_part->field->field_index; |
| 1923 | Field* field = table_share->field[field_index]; |
| 1924 | TOKU_TYPE toku_type = mysql_to_toku_type(field); |
| 1925 | bool found_index = false; |
| 1926 | |
| 1927 | switch(toku_type) { |
| 1928 | case (toku_type_int): |
| 1929 | case (toku_type_double): |
| 1930 | case (toku_type_float): |
| 1931 | case (toku_type_fixbinary): |
| 1932 | case (toku_type_fixstring): |
| 1933 | pos[0] = COL_FIX_FIELD; |
| 1934 | pos++; |
| 1935 | |
| 1936 | // copy the offset |
| 1937 | COPY_OFFSET_TO_BUF; |
| 1938 | break; |
| 1939 | case (toku_type_varbinary): |
| 1940 | case (toku_type_varstring): |
| 1941 | pos[0] = COL_VAR_FIELD; |
| 1942 | pos++; |
| 1943 | |
| 1944 | // copy the offset |
| 1945 | COPY_OFFSET_TO_BUF; |
| 1946 | break; |
| 1947 | case (toku_type_blob): |
| 1948 | pos[0] = COL_BLOB_FIELD; |
| 1949 | pos++; |
| 1950 | for (uint32_t i = 0; i < kc_info->num_blobs; i++) { |
| 1951 | uint32_t blob_index = kc_info->blob_fields[i]; |
| 1952 | if (blob_index == field_index) { |
| 1953 | uint32_t val = i; |
| 1954 | memcpy(pos, &val, sizeof(uint32_t)); |
| 1955 | pos += sizeof(uint32_t); |
| 1956 | found_index = true; |
| 1957 | break; |
| 1958 | } |
| 1959 | } |
| 1960 | assert_always(found_index); |
| 1961 | break; |
| 1962 | default: |
| 1963 | assert_unreachable(); |
| 1964 | } |
| 1965 | |
| 1966 | return pos - buf; |
| 1967 | } |
| 1968 | |
| 1969 | static uint32_t pack_desc_key_length_info(uchar* buf, KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, KEY_PART_INFO* key_part) { |
| 1970 | uchar* pos = buf; |
| 1971 | uint16 field_index = key_part->field->field_index; |
| 1972 | Field* field = table_share->field[field_index]; |
| 1973 | TOKU_TYPE toku_type = mysql_to_toku_type(field); |
| 1974 | uint32_t key_part_length = key_part->length; |
| 1975 | uint32_t field_length; |
| 1976 | |
| 1977 | switch(toku_type) { |
| 1978 | case (toku_type_int): |
| 1979 | case (toku_type_double): |
| 1980 | case (toku_type_float): |
| 1981 | // copy the key_part length |
| 1982 | field_length = kc_info->field_lengths[field_index]; |
| 1983 | memcpy(pos, &field_length, sizeof(field_length)); |
| 1984 | pos += sizeof(key_part_length); |
| 1985 | break; |
| 1986 | case (toku_type_fixbinary): |
| 1987 | case (toku_type_fixstring): |
| 1988 | field_length = field->pack_length(); |
| 1989 | set_if_smaller(key_part_length, field_length); |
| 1990 | // fallthrough |
| 1991 | case (toku_type_varbinary): |
| 1992 | case (toku_type_varstring): |
| 1993 | case (toku_type_blob): |
| 1994 | // copy the key_part length |
| 1995 | memcpy(pos, &key_part_length, sizeof(key_part_length)); |
| 1996 | pos += sizeof(key_part_length); |
| 1997 | break; |
| 1998 | default: |
| 1999 | assert_unreachable(); |
| 2000 | } |
| 2001 | |
| 2002 | return pos - buf; |
| 2003 | } |
| 2004 | |
| 2005 | static uint32_t pack_desc_char_info(uchar* buf, KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, KEY_PART_INFO* key_part) { |
| 2006 | uchar* pos = buf; |
| 2007 | uint16 field_index = key_part->field->field_index; |
| 2008 | Field* field = table_share->field[field_index]; |
| 2009 | TOKU_TYPE toku_type = mysql_to_toku_type(field); |
| 2010 | uint32_t charset_num = 0; |
| 2011 | |
| 2012 | switch(toku_type) { |
| 2013 | case (toku_type_int): |
| 2014 | case (toku_type_double): |
| 2015 | case (toku_type_float): |
| 2016 | case (toku_type_fixbinary): |
| 2017 | case (toku_type_varbinary): |
| 2018 | pos[0] = COL_HAS_NO_CHARSET; |
| 2019 | pos++; |
| 2020 | break; |
| 2021 | case (toku_type_fixstring): |
| 2022 | case (toku_type_varstring): |
| 2023 | case (toku_type_blob): |
| 2024 | pos[0] = COL_HAS_CHARSET; |
| 2025 | pos++; |
| 2026 | |
| 2027 | // copy the charset |
| 2028 | charset_num = field->charset()->number; |
| 2029 | pos[0] = (uchar)(charset_num & 255); |
| 2030 | pos[1] = (uchar)((charset_num >> 8) & 255); |
| 2031 | pos[2] = (uchar)((charset_num >> 16) & 255); |
| 2032 | pos[3] = (uchar)((charset_num >> 24) & 255); |
| 2033 | pos += 4; |
| 2034 | break; |
| 2035 | default: |
| 2036 | assert_unreachable(); |
| 2037 | } |
| 2038 | |
| 2039 | return pos - buf; |
| 2040 | } |
| 2041 | |
| 2042 | static uint32_t pack_some_row_info ( |
| 2043 | uchar* buf, |
| 2044 | uint pk_index, |
| 2045 | TABLE_SHARE* table_share, |
| 2046 | KEY_AND_COL_INFO* kc_info |
| 2047 | ) |
| 2048 | { |
| 2049 | uchar* pos = buf; |
| 2050 | uint32_t num_null_bytes = 0; |
| 2051 | // |
| 2052 | // four bytes stating number of null bytes |
| 2053 | // |
| 2054 | num_null_bytes = table_share->null_bytes; |
| 2055 | memcpy(pos, &num_null_bytes, sizeof(num_null_bytes)); |
| 2056 | pos += sizeof(num_null_bytes); |
| 2057 | // |
| 2058 | // eight bytes stating mcp_info |
| 2059 | // |
| 2060 | memcpy(pos, &kc_info->mcp_info[pk_index], sizeof(MULTI_COL_PACK_INFO)); |
| 2061 | pos += sizeof(MULTI_COL_PACK_INFO); |
| 2062 | // |
| 2063 | // one byte for the number of offset bytes |
| 2064 | // |
| 2065 | pos[0] = (uchar)kc_info->num_offset_bytes; |
| 2066 | pos++; |
| 2067 | |
| 2068 | return pos - buf; |
| 2069 | } |
| 2070 | |
| 2071 | static uint32_t get_max_clustering_val_pack_desc_size( |
| 2072 | TABLE_SHARE* table_share |
| 2073 | ) |
| 2074 | { |
| 2075 | uint32_t ret_val = 0; |
| 2076 | // |
| 2077 | // the fixed stuff: |
| 2078 | // first the things in pack_some_row_info |
| 2079 | // second another mcp_info |
| 2080 | // third a byte that states if blobs exist |
| 2081 | ret_val += sizeof(uint32_t) + sizeof(MULTI_COL_PACK_INFO) + 1; |
| 2082 | ret_val += sizeof(MULTI_COL_PACK_INFO); |
| 2083 | ret_val++; |
| 2084 | // |
| 2085 | // now the variable stuff |
| 2086 | // an upper bound is, for each field, byte stating if it is fixed or var, followed |
| 2087 | // by 8 bytes for endpoints |
| 2088 | // |
| 2089 | ret_val += (table_share->fields)*(1 + 2*sizeof(uint32_t)); |
| 2090 | // |
| 2091 | // four bytes storing the length of this portion |
| 2092 | // |
| 2093 | ret_val += 4; |
| 2094 | |
| 2095 | return ret_val; |
| 2096 | } |
| 2097 | |
| 2098 | static uint32_t create_toku_clustering_val_pack_descriptor ( |
| 2099 | uchar* buf, |
| 2100 | uint pk_index, |
| 2101 | TABLE_SHARE* table_share, |
| 2102 | KEY_AND_COL_INFO* kc_info, |
| 2103 | uint32_t keynr, |
| 2104 | bool is_clustering |
| 2105 | ) |
| 2106 | { |
| 2107 | uchar* pos = buf + 4; |
| 2108 | uint32_t offset = 0; |
| 2109 | bool start_range_set = false; |
| 2110 | uint32_t last_col = 0; |
| 2111 | // |
| 2112 | // do not need to write anything if the key is not clustering |
| 2113 | // |
| 2114 | if (!is_clustering) { |
| 2115 | goto exit; |
| 2116 | } |
| 2117 | |
| 2118 | pos += pack_some_row_info( |
| 2119 | pos, |
| 2120 | pk_index, |
| 2121 | table_share, |
| 2122 | kc_info |
| 2123 | ); |
| 2124 | |
| 2125 | // |
| 2126 | // eight bytes stating mcp_info of clustering key |
| 2127 | // |
| 2128 | memcpy(pos, &kc_info->mcp_info[keynr], sizeof(MULTI_COL_PACK_INFO)); |
| 2129 | pos += sizeof(MULTI_COL_PACK_INFO); |
| 2130 | |
| 2131 | // |
| 2132 | // store bit that states if blobs exist |
| 2133 | // |
| 2134 | pos[0] = (kc_info->num_blobs) ? 1 : 0; |
| 2135 | pos++; |
| 2136 | |
| 2137 | // |
| 2138 | // descriptor assumes that all fields filtered from pk are |
| 2139 | // also filtered from clustering key val. Doing check here to |
| 2140 | // make sure something unexpected does not happen |
| 2141 | // |
| 2142 | for (uint i = 0; i < table_share->fields; i++) { |
| 2143 | bool col_filtered = bitmap_is_set(&kc_info->key_filters[keynr],i); |
| 2144 | bool col_filtered_in_pk = bitmap_is_set(&kc_info->key_filters[pk_index],i); |
| 2145 | if (col_filtered_in_pk) { |
| 2146 | assert_always(col_filtered); |
| 2147 | } |
| 2148 | } |
| 2149 | |
| 2150 | // |
| 2151 | // first handle the fixed fields |
| 2152 | // |
| 2153 | start_range_set = false; |
| 2154 | last_col = 0; |
| 2155 | for (uint i = 0; i < table_share->fields; i++) { |
| 2156 | bool col_filtered = bitmap_is_set(&kc_info->key_filters[keynr],i); |
| 2157 | if (!is_fixed_field(kc_info, i)) { |
| 2158 | // |
| 2159 | // not a fixed field, continue |
| 2160 | // |
| 2161 | continue; |
| 2162 | } |
| 2163 | if (col_filtered && start_range_set) { |
| 2164 | // |
| 2165 | // need to set the end range |
| 2166 | // |
| 2167 | start_range_set = false; |
| 2168 | uint32_t end_offset = kc_info->cp_info[pk_index][last_col].col_pack_val + kc_info->field_lengths[last_col]; |
| 2169 | memcpy(pos, &end_offset, sizeof(end_offset)); |
| 2170 | pos += sizeof(end_offset); |
| 2171 | } |
| 2172 | else if (!col_filtered) { |
| 2173 | if (!start_range_set) { |
| 2174 | pos[0] = CK_FIX_RANGE; |
| 2175 | pos++; |
| 2176 | start_range_set = true; |
| 2177 | uint32_t start_offset = kc_info->cp_info[pk_index][i].col_pack_val; |
| 2178 | memcpy(pos, &start_offset , sizeof(start_offset)); |
| 2179 | pos += sizeof(start_offset); |
| 2180 | } |
| 2181 | last_col = i; |
| 2182 | } |
| 2183 | else { |
| 2184 | continue; |
| 2185 | } |
| 2186 | } |
| 2187 | if (start_range_set) { |
| 2188 | // |
| 2189 | // need to set the end range |
| 2190 | // |
| 2191 | start_range_set = false; |
| 2192 | uint32_t end_offset = kc_info->cp_info[pk_index][last_col].col_pack_val+ kc_info->field_lengths[last_col]; |
| 2193 | memcpy(pos, &end_offset, sizeof(end_offset)); |
| 2194 | pos += sizeof(end_offset); |
| 2195 | } |
| 2196 | |
| 2197 | // |
| 2198 | // now handle the var fields |
| 2199 | // |
| 2200 | start_range_set = false; |
| 2201 | last_col = 0; |
| 2202 | for (uint i = 0; i < table_share->fields; i++) { |
| 2203 | bool col_filtered = bitmap_is_set(&kc_info->key_filters[keynr],i); |
| 2204 | if (!is_variable_field(kc_info, i)) { |
| 2205 | // |
| 2206 | // not a var field, continue |
| 2207 | // |
| 2208 | continue; |
| 2209 | } |
| 2210 | if (col_filtered && start_range_set) { |
| 2211 | // |
| 2212 | // need to set the end range |
| 2213 | // |
| 2214 | start_range_set = false; |
| 2215 | uint32_t end_offset = kc_info->cp_info[pk_index][last_col].col_pack_val; |
| 2216 | memcpy(pos, &end_offset, sizeof(end_offset)); |
| 2217 | pos += sizeof(end_offset); |
| 2218 | } |
| 2219 | else if (!col_filtered) { |
| 2220 | if (!start_range_set) { |
| 2221 | pos[0] = CK_VAR_RANGE; |
| 2222 | pos++; |
| 2223 | |
| 2224 | start_range_set = true; |
| 2225 | uint32_t start_offset = kc_info->cp_info[pk_index][i].col_pack_val; |
| 2226 | memcpy(pos, &start_offset , sizeof(start_offset)); |
| 2227 | pos += sizeof(start_offset); |
| 2228 | } |
| 2229 | last_col = i; |
| 2230 | } |
| 2231 | else { |
| 2232 | continue; |
| 2233 | } |
| 2234 | } |
| 2235 | if (start_range_set) { |
| 2236 | start_range_set = false; |
| 2237 | uint32_t end_offset = kc_info->cp_info[pk_index][last_col].col_pack_val; |
| 2238 | memcpy(pos, &end_offset, sizeof(end_offset)); |
| 2239 | pos += sizeof(end_offset); |
| 2240 | } |
| 2241 | |
| 2242 | exit: |
| 2243 | offset = pos - buf; |
| 2244 | buf[0] = (uchar)(offset & 255); |
| 2245 | buf[1] = (uchar)((offset >> 8) & 255); |
| 2246 | buf[2] = (uchar)((offset >> 16) & 255); |
| 2247 | buf[3] = (uchar)((offset >> 24) & 255); |
| 2248 | |
| 2249 | return pos - buf; |
| 2250 | } |
| 2251 | |
| 2252 | static uint32_t pack_clustering_val_from_desc( |
| 2253 | uchar* buf, |
| 2254 | void* row_desc, |
| 2255 | uint32_t row_desc_size, |
| 2256 | const DBT* pk_val |
| 2257 | ) |
| 2258 | { |
| 2259 | uchar* null_bytes_src_ptr = NULL; |
| 2260 | uchar* fixed_src_ptr = NULL; |
| 2261 | uchar* var_src_offset_ptr = NULL; |
| 2262 | uchar* var_src_data_ptr = NULL; |
| 2263 | uchar* fixed_dest_ptr = NULL; |
| 2264 | uchar* var_dest_offset_ptr = NULL; |
| 2265 | uchar* var_dest_data_ptr = NULL; |
| 2266 | uchar* orig_var_dest_data_ptr = NULL; |
| 2267 | uchar* desc_pos = (uchar *)row_desc; |
| 2268 | uint32_t num_null_bytes = 0; |
| 2269 | uint32_t num_offset_bytes; |
| 2270 | MULTI_COL_PACK_INFO src_mcp_info, dest_mcp_info; |
| 2271 | uchar has_blobs; |
| 2272 | |
| 2273 | memcpy(&num_null_bytes, desc_pos, sizeof(num_null_bytes)); |
| 2274 | desc_pos += sizeof(num_null_bytes); |
| 2275 | |
| 2276 | memcpy(&src_mcp_info, desc_pos, sizeof(src_mcp_info)); |
| 2277 | desc_pos += sizeof(src_mcp_info); |
| 2278 | |
| 2279 | num_offset_bytes = desc_pos[0]; |
| 2280 | desc_pos++; |
| 2281 | |
| 2282 | memcpy(&dest_mcp_info, desc_pos, sizeof(dest_mcp_info)); |
| 2283 | desc_pos += sizeof(dest_mcp_info); |
| 2284 | |
| 2285 | has_blobs = desc_pos[0]; |
| 2286 | desc_pos++; |
| 2287 | |
| 2288 | // |
| 2289 | //set the variables |
| 2290 | // |
| 2291 | null_bytes_src_ptr = (uchar *)pk_val->data; |
| 2292 | fixed_src_ptr = null_bytes_src_ptr + num_null_bytes; |
| 2293 | var_src_offset_ptr = fixed_src_ptr + src_mcp_info.fixed_field_size; |
| 2294 | var_src_data_ptr = var_src_offset_ptr + src_mcp_info.len_of_offsets; |
| 2295 | |
| 2296 | fixed_dest_ptr = buf + num_null_bytes; |
| 2297 | var_dest_offset_ptr = fixed_dest_ptr + dest_mcp_info.fixed_field_size; |
| 2298 | var_dest_data_ptr = var_dest_offset_ptr + dest_mcp_info.len_of_offsets; |
| 2299 | orig_var_dest_data_ptr = var_dest_data_ptr; |
| 2300 | |
| 2301 | // |
| 2302 | // copy the null bytes |
| 2303 | // |
| 2304 | memcpy(buf, null_bytes_src_ptr, num_null_bytes); |
| 2305 | while ( (uint32_t)(desc_pos - (uchar *)row_desc) < row_desc_size) { |
| 2306 | uint32_t start, end, length; |
| 2307 | uchar curr = desc_pos[0]; |
| 2308 | desc_pos++; |
| 2309 | |
| 2310 | memcpy(&start, desc_pos, sizeof(start)); |
| 2311 | desc_pos += sizeof(start); |
| 2312 | |
| 2313 | memcpy(&end, desc_pos, sizeof(end)); |
| 2314 | desc_pos += sizeof(end); |
| 2315 | |
| 2316 | assert_always (start <= end); |
| 2317 | |
| 2318 | if (curr == CK_FIX_RANGE) { |
| 2319 | length = end - start; |
| 2320 | |
| 2321 | memcpy(fixed_dest_ptr, fixed_src_ptr + start, length); |
| 2322 | fixed_dest_ptr += length; |
| 2323 | } |
| 2324 | else if (curr == CK_VAR_RANGE) { |
| 2325 | uint32_t start_data_size; |
| 2326 | uint32_t start_data_offset; |
| 2327 | uint32_t end_data_size; |
| 2328 | uint32_t end_data_offset; |
| 2329 | uint32_t offset_diffs; |
| 2330 | |
| 2331 | get_var_field_info( |
| 2332 | &start_data_size, |
| 2333 | &start_data_offset, |
| 2334 | start, |
| 2335 | var_src_offset_ptr, |
| 2336 | num_offset_bytes |
| 2337 | ); |
| 2338 | get_var_field_info( |
| 2339 | &end_data_size, |
| 2340 | &end_data_offset, |
| 2341 | end, |
| 2342 | var_src_offset_ptr, |
| 2343 | num_offset_bytes |
| 2344 | ); |
| 2345 | length = end_data_offset + end_data_size - start_data_offset; |
| 2346 | // |
| 2347 | // copy the data |
| 2348 | // |
| 2349 | memcpy( |
| 2350 | var_dest_data_ptr, |
| 2351 | var_src_data_ptr + start_data_offset, |
| 2352 | length |
| 2353 | ); |
| 2354 | var_dest_data_ptr += length; |
| 2355 | |
| 2356 | // |
| 2357 | // put in offset info |
| 2358 | // |
| 2359 | offset_diffs = (end_data_offset + end_data_size) - (uint32_t)(var_dest_data_ptr - orig_var_dest_data_ptr); |
| 2360 | for (uint32_t i = start; i <= end; i++) { |
| 2361 | if ( num_offset_bytes == 1 ) { |
| 2362 | assert_always(offset_diffs < 256); |
| 2363 | var_dest_offset_ptr[0] = var_src_offset_ptr[i] - (uchar)offset_diffs; |
| 2364 | var_dest_offset_ptr++; |
| 2365 | } else if ( num_offset_bytes == 2 ) { |
| 2366 | uint32_t tmp = uint2korr(var_src_offset_ptr + 2*i); |
| 2367 | uint32_t new_offset = tmp - offset_diffs; |
| 2368 | assert_always(new_offset < 1<<16); |
| 2369 | int2store(var_dest_offset_ptr,new_offset); |
| 2370 | var_dest_offset_ptr += 2; |
| 2371 | } else { |
| 2372 | assert_unreachable(); |
| 2373 | } |
| 2374 | } |
| 2375 | } else { |
| 2376 | assert_unreachable(); |
| 2377 | } |
| 2378 | } |
| 2379 | // |
| 2380 | // copy blobs |
| 2381 | // at this point, var_dest_data_ptr is pointing to the end, where blobs should be located |
| 2382 | // so, we put the blobs at var_dest_data_ptr |
| 2383 | // |
| 2384 | if (has_blobs) { |
| 2385 | uint32_t num_blob_bytes; |
| 2386 | uint32_t start_offset; |
| 2387 | uchar* src_blob_ptr = NULL; |
| 2388 | get_blob_field_info( |
| 2389 | &start_offset, |
| 2390 | src_mcp_info.len_of_offsets, |
| 2391 | var_src_data_ptr, |
| 2392 | num_offset_bytes |
| 2393 | ); |
| 2394 | src_blob_ptr = var_src_data_ptr + start_offset; |
| 2395 | num_blob_bytes = pk_val->size - (start_offset + (var_src_data_ptr - null_bytes_src_ptr)); |
| 2396 | memcpy(var_dest_data_ptr, src_blob_ptr, num_blob_bytes); |
| 2397 | var_dest_data_ptr += num_blob_bytes; |
| 2398 | } |
| 2399 | return var_dest_data_ptr - buf; |
| 2400 | } |
| 2401 | |
| 2402 | |
| 2403 | static uint32_t get_max_secondary_key_pack_desc_size( |
| 2404 | KEY_AND_COL_INFO* kc_info |
| 2405 | ) |
| 2406 | { |
| 2407 | uint32_t ret_val = 0; |
| 2408 | // |
| 2409 | // the fixed stuff: |
| 2410 | // byte that states if main dictionary |
| 2411 | // byte that states if hpk |
| 2412 | // the things in pack_some_row_info |
| 2413 | ret_val++; |
| 2414 | ret_val++; |
| 2415 | ret_val += sizeof(uint32_t) + sizeof(MULTI_COL_PACK_INFO) + 1; |
| 2416 | // |
| 2417 | // now variable sized stuff |
| 2418 | // |
| 2419 | |
| 2420 | // first the blobs |
| 2421 | ret_val += sizeof(kc_info->num_blobs); |
| 2422 | ret_val+= kc_info->num_blobs; |
| 2423 | |
| 2424 | // then the pk |
| 2425 | // one byte for num key parts |
| 2426 | // two bytes for each key part |
| 2427 | ret_val++; |
| 2428 | ret_val += MAX_REF_PARTS*2; |
| 2429 | |
| 2430 | // then the key |
| 2431 | // null bit, then null byte, |
| 2432 | // then 1 byte stating what it is, then 4 for offset, 4 for key length, |
| 2433 | // 1 for if charset exists, and 4 for charset |
| 2434 | ret_val += MAX_REF_PARTS*(1 + sizeof(uint32_t) + 1 + 3*sizeof(uint32_t) + 1); |
| 2435 | // |
| 2436 | // four bytes storing the length of this portion |
| 2437 | // |
| 2438 | ret_val += 4; |
| 2439 | return ret_val; |
| 2440 | } |
| 2441 | |
| 2442 | static uint32_t create_toku_secondary_key_pack_descriptor ( |
| 2443 | uchar* buf, |
| 2444 | bool has_hpk, |
| 2445 | uint pk_index, |
| 2446 | TABLE_SHARE* table_share, |
| 2447 | TABLE* table, |
| 2448 | KEY_AND_COL_INFO* kc_info, |
| 2449 | KEY* key_info, |
| 2450 | KEY* prim_key |
| 2451 | ) |
| 2452 | { |
| 2453 | // |
| 2454 | // The first four bytes always contain the offset of where the first key |
| 2455 | // ends. |
| 2456 | // |
| 2457 | uchar* pk_info = NULL; |
| 2458 | uchar* pos = buf + 4; |
| 2459 | uint32_t offset = 0; |
| 2460 | |
| 2461 | // |
| 2462 | // first byte states that it is NOT main dictionary |
| 2463 | // |
| 2464 | pos[0] = 0; |
| 2465 | pos++; |
| 2466 | |
| 2467 | // |
| 2468 | // one byte states if main dictionary has an hpk or not |
| 2469 | // |
| 2470 | if (has_hpk) { |
| 2471 | pos[0] = 1; |
| 2472 | } |
| 2473 | else { |
| 2474 | pos[0] = 0; |
| 2475 | } |
| 2476 | pos++; |
| 2477 | |
| 2478 | pos += pack_some_row_info( |
| 2479 | pos, |
| 2480 | pk_index, |
| 2481 | table_share, |
| 2482 | kc_info |
| 2483 | ); |
| 2484 | |
| 2485 | // |
| 2486 | // store blob information |
| 2487 | // |
| 2488 | memcpy(pos, &kc_info->num_blobs, sizeof(kc_info->num_blobs)); |
| 2489 | pos += sizeof(uint32_t); |
| 2490 | for (uint32_t i = 0; i < kc_info->num_blobs; i++) { |
| 2491 | // |
| 2492 | // store length bytes for each blob |
| 2493 | // |
| 2494 | Field* field = table_share->field[kc_info->blob_fields[i]]; |
| 2495 | pos[0] = (uchar)field->row_pack_length(); |
| 2496 | pos++; |
| 2497 | } |
| 2498 | |
| 2499 | // |
| 2500 | // store the pk information |
| 2501 | // |
| 2502 | if (has_hpk) { |
| 2503 | pos[0] = 0; |
| 2504 | pos++; |
| 2505 | } |
| 2506 | else { |
| 2507 | // |
| 2508 | // store number of parts |
| 2509 | // |
| 2510 | assert_always(prim_key->user_defined_key_parts < 128); |
| 2511 | pos[0] = 2 * prim_key->user_defined_key_parts; |
| 2512 | pos++; |
| 2513 | // |
| 2514 | // for each part, store if it is a fixed field or var field |
| 2515 | // if fixed, store number of bytes, if var, store |
| 2516 | // number of length bytes |
| 2517 | // total should be two bytes per key part stored |
| 2518 | // |
| 2519 | pk_info = pos; |
| 2520 | uchar* tmp = pos; |
| 2521 | for (uint i = 0; i < prim_key->user_defined_key_parts; i++) { |
| 2522 | tmp += pack_desc_pk_info( |
| 2523 | tmp, |
| 2524 | kc_info, |
| 2525 | table_share, |
| 2526 | &prim_key->key_part[i] |
| 2527 | ); |
| 2528 | } |
| 2529 | // |
| 2530 | // asserting that we moved forward as much as we think we have |
| 2531 | // |
| 2532 | assert_always(tmp - pos == (2 * prim_key->user_defined_key_parts)); |
| 2533 | pos = tmp; |
| 2534 | } |
| 2535 | |
| 2536 | for (uint i = 0; i < key_info->user_defined_key_parts; i++) { |
| 2537 | KEY_PART_INFO curr_kpi = key_info->key_part[i]; |
| 2538 | uint16 field_index = curr_kpi.field->field_index; |
| 2539 | Field* field = table_share->field[field_index]; |
| 2540 | bool is_col_in_pk = false; |
| 2541 | |
| 2542 | if (bitmap_is_set(&kc_info->key_filters[pk_index],field_index)) { |
| 2543 | assert_always(!has_hpk && prim_key != NULL); |
| 2544 | is_col_in_pk = true; |
| 2545 | } |
| 2546 | else { |
| 2547 | is_col_in_pk = false; |
| 2548 | } |
| 2549 | |
| 2550 | pos[0] = field->null_bit; |
| 2551 | pos++; |
| 2552 | |
| 2553 | if (is_col_in_pk) { |
| 2554 | // |
| 2555 | // assert that columns in pk do not have a null bit |
| 2556 | // because in MySQL, pk columns cannot be null |
| 2557 | // |
| 2558 | assert_always(!field->null_bit); |
| 2559 | } |
| 2560 | |
| 2561 | if (field->null_bit) { |
| 2562 | uint32_t null_offset = get_null_offset(table,table->field[field_index]); |
| 2563 | memcpy(pos, &null_offset, sizeof(uint32_t)); |
| 2564 | pos += sizeof(uint32_t); |
| 2565 | } |
| 2566 | if (is_col_in_pk) { |
| 2567 | pos += pack_desc_pk_offset_info( |
| 2568 | pos, |
| 2569 | kc_info, |
| 2570 | table_share, |
| 2571 | &curr_kpi, |
| 2572 | prim_key, |
| 2573 | pk_info |
| 2574 | ); |
| 2575 | } |
| 2576 | else { |
| 2577 | pos += pack_desc_offset_info( |
| 2578 | pos, |
| 2579 | kc_info, |
| 2580 | pk_index, |
| 2581 | table_share, |
| 2582 | &curr_kpi |
| 2583 | ); |
| 2584 | } |
| 2585 | pos += pack_desc_key_length_info( |
| 2586 | pos, |
| 2587 | kc_info, |
| 2588 | table_share, |
| 2589 | &curr_kpi |
| 2590 | ); |
| 2591 | pos += pack_desc_char_info( |
| 2592 | pos, |
| 2593 | kc_info, |
| 2594 | table_share, |
| 2595 | &curr_kpi |
| 2596 | ); |
| 2597 | } |
| 2598 | |
| 2599 | offset = pos - buf; |
| 2600 | buf[0] = (uchar)(offset & 255); |
| 2601 | buf[1] = (uchar)((offset >> 8) & 255); |
| 2602 | buf[2] = (uchar)((offset >> 16) & 255); |
| 2603 | buf[3] = (uchar)((offset >> 24) & 255); |
| 2604 | |
| 2605 | return pos - buf; |
| 2606 | } |
| 2607 | |
| 2608 | static uint32_t skip_key_in_desc( |
| 2609 | uchar* row_desc |
| 2610 | ) |
| 2611 | { |
| 2612 | uchar* pos = row_desc; |
| 2613 | uchar col_bin_or_char; |
| 2614 | // |
| 2615 | // skip the byte that states if it is a fix field or var field, we do not care |
| 2616 | // |
| 2617 | pos++; |
| 2618 | |
| 2619 | // |
| 2620 | // skip the offset information |
| 2621 | // |
| 2622 | pos += sizeof(uint32_t); |
| 2623 | |
| 2624 | // |
| 2625 | // skip the key_part_length info |
| 2626 | // |
| 2627 | pos += sizeof(uint32_t); |
| 2628 | col_bin_or_char = pos[0]; |
| 2629 | pos++; |
| 2630 | if (col_bin_or_char == COL_HAS_NO_CHARSET) { |
| 2631 | goto exit; |
| 2632 | } |
| 2633 | // |
| 2634 | // skip the charset info |
| 2635 | // |
| 2636 | pos += 4; |
| 2637 | |
| 2638 | |
| 2639 | exit: |
| 2640 | return (uint32_t)(pos-row_desc); |
| 2641 | } |
| 2642 | |
| 2643 | |
| 2644 | static uint32_t max_key_size_from_desc( |
| 2645 | void* row_desc, |
| 2646 | uint32_t row_desc_size |
| 2647 | ) |
| 2648 | { |
| 2649 | uchar* desc_pos = (uchar *)row_desc; |
| 2650 | uint32_t num_blobs; |
| 2651 | uint32_t num_pk_columns; |
| 2652 | // |
| 2653 | // start at 1 for the infinity byte |
| 2654 | // |
| 2655 | uint32_t max_size = 1; |
| 2656 | |
| 2657 | // skip byte that states if main dictionary |
| 2658 | bool is_main_dictionary = desc_pos[0]; |
| 2659 | desc_pos++; |
| 2660 | assert_always(!is_main_dictionary); |
| 2661 | |
| 2662 | // skip hpk byte |
| 2663 | desc_pos++; |
| 2664 | |
| 2665 | // skip num_null_bytes |
| 2666 | desc_pos += sizeof(uint32_t); |
| 2667 | |
| 2668 | // skip mcp_info |
| 2669 | desc_pos += sizeof(MULTI_COL_PACK_INFO); |
| 2670 | |
| 2671 | // skip offset_bytes |
| 2672 | desc_pos++; |
| 2673 | |
| 2674 | // skip over blobs |
| 2675 | memcpy(&num_blobs, desc_pos, sizeof(num_blobs)); |
| 2676 | desc_pos += sizeof(num_blobs); |
| 2677 | desc_pos += num_blobs; |
| 2678 | |
| 2679 | // skip over pk info |
| 2680 | num_pk_columns = desc_pos[0]/2; |
| 2681 | desc_pos++; |
| 2682 | desc_pos += 2*num_pk_columns; |
| 2683 | |
| 2684 | while ( (uint32_t)(desc_pos - (uchar *)row_desc) < row_desc_size) { |
| 2685 | uchar has_charset; |
| 2686 | uint32_t key_length = 0; |
| 2687 | |
| 2688 | uchar null_bit = desc_pos[0]; |
| 2689 | desc_pos++; |
| 2690 | |
| 2691 | if (null_bit) { |
| 2692 | // |
| 2693 | // column is NULLable, skip null_offset, and add a null byte |
| 2694 | // |
| 2695 | max_size++; |
| 2696 | desc_pos += sizeof(uint32_t); |
| 2697 | } |
| 2698 | // |
| 2699 | // skip over byte that states if fix or var |
| 2700 | // |
| 2701 | desc_pos++; |
| 2702 | |
| 2703 | // skip over offset |
| 2704 | desc_pos += sizeof(uint32_t); |
| 2705 | |
| 2706 | // |
| 2707 | // get the key length and add it to return value |
| 2708 | // |
| 2709 | memcpy(&key_length, desc_pos, sizeof(key_length)); |
| 2710 | desc_pos += sizeof(key_length); |
| 2711 | max_size += key_length; |
| 2712 | max_size += 2; // 2 bytes for a potential length bytes, we are upperbounding, does not need to be super tight |
| 2713 | |
| 2714 | has_charset = desc_pos[0]; |
| 2715 | desc_pos++; |
| 2716 | |
| 2717 | uint32_t charset_num; |
| 2718 | if (has_charset == COL_HAS_CHARSET) { |
| 2719 | // skip over charsent num |
| 2720 | desc_pos += sizeof(charset_num); |
| 2721 | } |
| 2722 | else { |
| 2723 | assert_always(has_charset == COL_HAS_NO_CHARSET); |
| 2724 | } |
| 2725 | } |
| 2726 | return max_size; |
| 2727 | } |
| 2728 | |
| 2729 | static uint32_t pack_key_from_desc( |
| 2730 | uchar* buf, |
| 2731 | void* row_desc, |
| 2732 | uint32_t row_desc_size, |
| 2733 | const DBT* pk_key, |
| 2734 | const DBT* pk_val) { |
| 2735 | |
| 2736 | MULTI_COL_PACK_INFO mcp_info; |
| 2737 | uint32_t num_null_bytes; |
| 2738 | uint32_t num_blobs; |
| 2739 | uint32_t num_pk_columns; |
| 2740 | uchar* blob_lengths = NULL; |
| 2741 | uchar* pk_info = NULL; |
| 2742 | uchar* pk_data_ptr = NULL; |
| 2743 | uchar* null_bytes_ptr = NULL; |
| 2744 | uchar* fixed_field_ptr = NULL; |
| 2745 | uchar* var_field_offset_ptr = NULL; |
| 2746 | const uchar* var_field_data_ptr = NULL; |
| 2747 | uint32_t num_offset_bytes; |
| 2748 | uchar* packed_key_pos = buf; |
| 2749 | uchar* desc_pos = (uchar *)row_desc; |
| 2750 | |
| 2751 | bool is_main_dictionary = desc_pos[0]; |
| 2752 | desc_pos++; |
| 2753 | assert_always(!is_main_dictionary); |
| 2754 | |
| 2755 | // |
| 2756 | // get the constant info out of descriptor |
| 2757 | // |
| 2758 | bool hpk = desc_pos[0]; |
| 2759 | desc_pos++; |
| 2760 | |
| 2761 | memcpy(&num_null_bytes, desc_pos, sizeof(num_null_bytes)); |
| 2762 | desc_pos += sizeof(num_null_bytes); |
| 2763 | |
| 2764 | memcpy(&mcp_info, desc_pos, sizeof(mcp_info)); |
| 2765 | desc_pos += sizeof(mcp_info); |
| 2766 | |
| 2767 | num_offset_bytes = desc_pos[0]; |
| 2768 | desc_pos++; |
| 2769 | |
| 2770 | memcpy(&num_blobs, desc_pos, sizeof(num_blobs)); |
| 2771 | desc_pos += sizeof(num_blobs); |
| 2772 | |
| 2773 | blob_lengths = desc_pos; |
| 2774 | desc_pos += num_blobs; |
| 2775 | |
| 2776 | num_pk_columns = desc_pos[0]/2; |
| 2777 | desc_pos++; |
| 2778 | pk_info = desc_pos; |
| 2779 | desc_pos += 2*num_pk_columns; |
| 2780 | |
| 2781 | // |
| 2782 | // now start packing the key |
| 2783 | // |
| 2784 | |
| 2785 | // |
| 2786 | // pack the infinity byte |
| 2787 | // |
| 2788 | packed_key_pos[0] = COL_ZERO; |
| 2789 | packed_key_pos++; |
| 2790 | // |
| 2791 | // now start packing each column of the key, as described in descriptor |
| 2792 | // |
| 2793 | if (!hpk) { |
| 2794 | // +1 for the infinity byte |
| 2795 | pk_data_ptr = (uchar *)pk_key->data + 1; |
| 2796 | } |
| 2797 | null_bytes_ptr = (uchar *)pk_val->data; |
| 2798 | fixed_field_ptr = null_bytes_ptr + num_null_bytes; |
| 2799 | var_field_offset_ptr = fixed_field_ptr + mcp_info.fixed_field_size; |
| 2800 | var_field_data_ptr = var_field_offset_ptr + mcp_info.len_of_offsets; |
| 2801 | while ((uint32_t)(desc_pos - (uchar*)row_desc) < row_desc_size) { |
| 2802 | uchar col_fix_val; |
| 2803 | uchar has_charset; |
| 2804 | uint32_t col_pack_val = 0; |
| 2805 | uint32_t key_length = 0; |
| 2806 | |
| 2807 | uchar null_bit = desc_pos[0]; |
| 2808 | desc_pos++; |
| 2809 | |
| 2810 | if (null_bit) { |
| 2811 | // |
| 2812 | // column is NULLable, need to check the null bytes to see if it is NULL |
| 2813 | // |
| 2814 | uint32_t null_offset = 0; |
| 2815 | bool is_field_null; |
| 2816 | memcpy(&null_offset, desc_pos, sizeof(null_offset)); |
| 2817 | desc_pos += sizeof(null_offset); |
| 2818 | |
| 2819 | is_field_null = (null_bytes_ptr[null_offset] & null_bit) ? true: false; |
| 2820 | if (is_field_null) { |
| 2821 | packed_key_pos[0] = NULL_COL_VAL; |
| 2822 | packed_key_pos++; |
| 2823 | desc_pos += skip_key_in_desc(desc_pos); |
| 2824 | continue; |
| 2825 | } else { |
| 2826 | packed_key_pos[0] = NONNULL_COL_VAL; |
| 2827 | packed_key_pos++; |
| 2828 | } |
| 2829 | } |
| 2830 | // |
| 2831 | // now pack the column (unless it was NULL, and we continued) |
| 2832 | // |
| 2833 | col_fix_val = desc_pos[0]; |
| 2834 | desc_pos++; |
| 2835 | |
| 2836 | memcpy(&col_pack_val, desc_pos, sizeof(col_pack_val)); |
| 2837 | desc_pos += sizeof(col_pack_val); |
| 2838 | |
| 2839 | memcpy(&key_length, desc_pos, sizeof(key_length)); |
| 2840 | desc_pos += sizeof(key_length); |
| 2841 | |
| 2842 | has_charset = desc_pos[0]; |
| 2843 | desc_pos++; |
| 2844 | |
| 2845 | uint32_t charset_num = 0; |
| 2846 | if (has_charset == COL_HAS_CHARSET) { |
| 2847 | memcpy(&charset_num, desc_pos, sizeof(charset_num)); |
| 2848 | desc_pos += sizeof(charset_num); |
| 2849 | } else { |
| 2850 | assert_always(has_charset == COL_HAS_NO_CHARSET); |
| 2851 | } |
| 2852 | // |
| 2853 | // case where column is in pk val |
| 2854 | // |
| 2855 | if (col_fix_val == COL_FIX_FIELD || |
| 2856 | col_fix_val == COL_VAR_FIELD || |
| 2857 | col_fix_val == COL_BLOB_FIELD) { |
| 2858 | if (col_fix_val == COL_FIX_FIELD && |
| 2859 | has_charset == COL_HAS_NO_CHARSET) { |
| 2860 | memcpy( |
| 2861 | packed_key_pos, |
| 2862 | &fixed_field_ptr[col_pack_val], |
| 2863 | key_length); |
| 2864 | packed_key_pos += key_length; |
| 2865 | } else if (col_fix_val == COL_VAR_FIELD && |
| 2866 | has_charset == COL_HAS_NO_CHARSET) { |
| 2867 | uint32_t data_start_offset = 0; |
| 2868 | |
| 2869 | uint32_t data_size = 0; |
| 2870 | get_var_field_info( |
| 2871 | &data_size, |
| 2872 | &data_start_offset, |
| 2873 | col_pack_val, |
| 2874 | var_field_offset_ptr, |
| 2875 | num_offset_bytes); |
| 2876 | |
| 2877 | // |
| 2878 | // length of this field in this row is data_size |
| 2879 | // data is located beginning at var_field_data_ptr + data_start_offset |
| 2880 | // |
| 2881 | packed_key_pos = pack_toku_varbinary_from_desc( |
| 2882 | packed_key_pos, |
| 2883 | var_field_data_ptr + data_start_offset, |
| 2884 | //number of bytes to use to encode the length in to_tokudb |
| 2885 | key_length, |
| 2886 | //length of field |
| 2887 | data_size); |
| 2888 | } else { |
| 2889 | const uchar* data_start = NULL; |
| 2890 | uint32_t data_start_offset = 0; |
| 2891 | uint32_t data_size = 0; |
| 2892 | |
| 2893 | if (col_fix_val == COL_FIX_FIELD) { |
| 2894 | data_start_offset = col_pack_val; |
| 2895 | data_size = key_length; |
| 2896 | data_start = fixed_field_ptr + data_start_offset; |
| 2897 | } else if (col_fix_val == COL_VAR_FIELD){ |
| 2898 | get_var_field_info( |
| 2899 | &data_size, |
| 2900 | &data_start_offset, |
| 2901 | col_pack_val, |
| 2902 | var_field_offset_ptr, |
| 2903 | num_offset_bytes); |
| 2904 | data_start = var_field_data_ptr + data_start_offset; |
| 2905 | } else if (col_fix_val == COL_BLOB_FIELD) { |
| 2906 | uint32_t blob_index = col_pack_val; |
| 2907 | uint32_t blob_offset; |
| 2908 | const uchar* blob_ptr = NULL; |
| 2909 | uint32_t field_len; |
| 2910 | uint32_t field_len_bytes = blob_lengths[blob_index]; |
| 2911 | get_blob_field_info( |
| 2912 | &blob_offset, |
| 2913 | mcp_info.len_of_offsets, |
| 2914 | var_field_data_ptr, |
| 2915 | num_offset_bytes); |
| 2916 | blob_ptr = var_field_data_ptr + blob_offset; |
| 2917 | assert_always(num_blobs > 0); |
| 2918 | |
| 2919 | // skip over other blobs to get to the one we want to |
| 2920 | // make a key out of |
| 2921 | for (uint32_t i = 0; i < blob_index; i++) { |
| 2922 | blob_ptr = unpack_toku_field_blob( |
| 2923 | NULL, |
| 2924 | blob_ptr, |
| 2925 | blob_lengths[i], |
| 2926 | true); |
| 2927 | } |
| 2928 | // at this point, blob_ptr is pointing to the blob we |
| 2929 | // want to make a key from |
| 2930 | field_len = get_blob_field_len(blob_ptr, field_len_bytes); |
| 2931 | // now we set the variables to make the key |
| 2932 | data_start = blob_ptr + field_len_bytes; |
| 2933 | data_size = field_len; |
| 2934 | } else { |
| 2935 | assert_unreachable(); |
| 2936 | } |
| 2937 | |
| 2938 | packed_key_pos = pack_toku_varstring_from_desc(packed_key_pos, |
| 2939 | data_start, |
| 2940 | key_length, |
| 2941 | data_size, |
| 2942 | charset_num); |
| 2943 | } |
| 2944 | } else { |
| 2945 | // case where column is in pk key |
| 2946 | if (col_fix_val == COL_FIX_PK_OFFSET) { |
| 2947 | memcpy(packed_key_pos, &pk_data_ptr[col_pack_val], key_length); |
| 2948 | packed_key_pos += key_length; |
| 2949 | } else if (col_fix_val == COL_VAR_PK_OFFSET) { |
| 2950 | uchar* tmp_pk_data_ptr = pk_data_ptr; |
| 2951 | uint32_t index_in_pk = col_pack_val; |
| 2952 | // |
| 2953 | // skip along in pk to the right column |
| 2954 | // |
| 2955 | for (uint32_t i = 0; i < index_in_pk; i++) { |
| 2956 | if (pk_info[2*i] == COL_FIX_FIELD) { |
| 2957 | tmp_pk_data_ptr += pk_info[2*i + 1]; |
| 2958 | } else if (pk_info[2*i] == COL_VAR_FIELD) { |
| 2959 | uint32_t len_bytes = pk_info[2*i + 1]; |
| 2960 | uint32_t len; |
| 2961 | if (len_bytes == 1) { |
| 2962 | len = tmp_pk_data_ptr[0]; |
| 2963 | tmp_pk_data_ptr++; |
| 2964 | } else if (len_bytes == 2) { |
| 2965 | len = uint2korr(tmp_pk_data_ptr); |
| 2966 | tmp_pk_data_ptr += 2; |
| 2967 | } else { |
| 2968 | assert_unreachable(); |
| 2969 | } |
| 2970 | tmp_pk_data_ptr += len; |
| 2971 | } else { |
| 2972 | assert_unreachable(); |
| 2973 | } |
| 2974 | } |
| 2975 | // |
| 2976 | // at this point, tmp_pk_data_ptr is pointing at the column |
| 2977 | // |
| 2978 | uint32_t is_fix_field = pk_info[2*index_in_pk]; |
| 2979 | if (is_fix_field == COL_FIX_FIELD) { |
| 2980 | memcpy(packed_key_pos, tmp_pk_data_ptr, key_length); |
| 2981 | packed_key_pos += key_length; |
| 2982 | } else if (is_fix_field == COL_VAR_FIELD) { |
| 2983 | const uchar* data_start = NULL; |
| 2984 | uint32_t data_size = 0; |
| 2985 | uint32_t len_bytes = pk_info[2*index_in_pk + 1]; |
| 2986 | if (len_bytes == 1) { |
| 2987 | data_size = tmp_pk_data_ptr[0]; |
| 2988 | tmp_pk_data_ptr++; |
| 2989 | } else if (len_bytes == 2) { |
| 2990 | data_size = uint2korr(tmp_pk_data_ptr); |
| 2991 | tmp_pk_data_ptr += 2; |
| 2992 | } else { |
| 2993 | assert_unreachable(); |
| 2994 | } |
| 2995 | data_start = tmp_pk_data_ptr; |
| 2996 | |
| 2997 | if (has_charset == COL_HAS_CHARSET) { |
| 2998 | packed_key_pos = pack_toku_varstring_from_desc( |
| 2999 | packed_key_pos, |
| 3000 | data_start, |
| 3001 | key_length, |
| 3002 | data_size, |
| 3003 | charset_num); |
| 3004 | } else if (has_charset == COL_HAS_NO_CHARSET) { |
| 3005 | packed_key_pos = pack_toku_varbinary_from_desc( |
| 3006 | packed_key_pos, |
| 3007 | data_start, |
| 3008 | key_length, |
| 3009 | data_size); |
| 3010 | } else { |
| 3011 | assert_unreachable(); |
| 3012 | } |
| 3013 | } else { |
| 3014 | assert_unreachable(); |
| 3015 | } |
| 3016 | } else { |
| 3017 | assert_unreachable(); |
| 3018 | } |
| 3019 | } |
| 3020 | |
| 3021 | } |
| 3022 | assert_always( (uint32_t)(desc_pos - (uchar *)row_desc) == row_desc_size); |
| 3023 | |
| 3024 | // |
| 3025 | // now append the primary key to the end of the key |
| 3026 | // |
| 3027 | if (hpk) { |
| 3028 | memcpy(packed_key_pos, pk_key->data, pk_key->size); |
| 3029 | packed_key_pos += pk_key->size; |
| 3030 | } else { |
| 3031 | memcpy(packed_key_pos, (uchar *)pk_key->data + 1, pk_key->size - 1); |
| 3032 | packed_key_pos += (pk_key->size - 1); |
| 3033 | } |
| 3034 | |
| 3035 | return (uint32_t)(packed_key_pos - buf); |
| 3036 | } |
| 3037 | |
| 3038 | static bool fields_have_same_name(Field* a, Field* b) { |
| 3039 | return strcmp(a->field_name.str, b->field_name.str) == 0; |
| 3040 | } |
| 3041 | |
| 3042 | static bool fields_are_same_type(Field* a, Field* b) { |
| 3043 | bool retval = true; |
| 3044 | enum_field_types a_mysql_type = a->real_type(); |
| 3045 | enum_field_types b_mysql_type = b->real_type(); |
| 3046 | TOKU_TYPE a_toku_type = mysql_to_toku_type(a); |
| 3047 | TOKU_TYPE b_toku_type = mysql_to_toku_type(b); |
| 3048 | // make sure have same names |
| 3049 | // make sure have same types |
| 3050 | if (a_mysql_type != b_mysql_type) { |
| 3051 | retval = false; |
| 3052 | goto cleanup; |
| 3053 | } |
| 3054 | // Thanks to MariaDB 5.5, we can have two fields |
| 3055 | // be the same MySQL type but not the same toku type, |
| 3056 | // This is an issue introduced with MariaDB's fractional time |
| 3057 | // implementation |
| 3058 | if (a_toku_type != b_toku_type) { |
| 3059 | retval = false; |
| 3060 | goto cleanup; |
| 3061 | } |
| 3062 | // make sure that either both are nullable, or both not nullable |
| 3063 | if ((a->null_bit && !b->null_bit) || (!a->null_bit && b->null_bit)) { |
| 3064 | retval = false; |
| 3065 | goto cleanup; |
| 3066 | } |
| 3067 | switch (a_mysql_type) { |
| 3068 | case MYSQL_TYPE_TINY: |
| 3069 | case MYSQL_TYPE_SHORT: |
| 3070 | case MYSQL_TYPE_INT24: |
| 3071 | case MYSQL_TYPE_LONG: |
| 3072 | case MYSQL_TYPE_LONGLONG: |
| 3073 | // length, unsigned, auto increment |
| 3074 | if (a->pack_length() != b->pack_length() || |
| 3075 | (a->flags & UNSIGNED_FLAG) != (b->flags & UNSIGNED_FLAG) || |
| 3076 | (a->flags & AUTO_INCREMENT_FLAG) != (b->flags & AUTO_INCREMENT_FLAG)) { |
| 3077 | retval = false; |
| 3078 | goto cleanup; |
| 3079 | } |
| 3080 | break; |
| 3081 | case MYSQL_TYPE_DOUBLE: |
| 3082 | case MYSQL_TYPE_FLOAT: |
| 3083 | // length, unsigned, auto increment |
| 3084 | if (a->pack_length() != b->pack_length() || |
| 3085 | (a->flags & UNSIGNED_FLAG) != (b->flags & UNSIGNED_FLAG) || |
| 3086 | (a->flags & AUTO_INCREMENT_FLAG) != (b->flags & AUTO_INCREMENT_FLAG)) { |
| 3087 | retval = false; |
| 3088 | goto cleanup; |
| 3089 | } |
| 3090 | break; |
| 3091 | case MYSQL_TYPE_NEWDECIMAL: |
| 3092 | // length, unsigned |
| 3093 | if (a->pack_length() != b->pack_length() || |
| 3094 | (a->flags & UNSIGNED_FLAG) != (b->flags & UNSIGNED_FLAG)) { |
| 3095 | retval = false; |
| 3096 | goto cleanup; |
| 3097 | } |
| 3098 | break; |
| 3099 | case MYSQL_TYPE_ENUM: { |
| 3100 | Field_enum *a_enum = static_cast<Field_enum *>(a); |
| 3101 | if (!a_enum->eq_def(b)) { |
| 3102 | retval = false; |
| 3103 | goto cleanup; |
| 3104 | } |
| 3105 | break; |
| 3106 | } |
| 3107 | case MYSQL_TYPE_SET: { |
| 3108 | Field_set *a_set = static_cast<Field_set *>(a); |
| 3109 | if (!a_set->eq_def(b)) { |
| 3110 | retval = false; |
| 3111 | goto cleanup; |
| 3112 | } |
| 3113 | break; |
| 3114 | } |
| 3115 | case MYSQL_TYPE_BIT: |
| 3116 | // length |
| 3117 | if (a->pack_length() != b->pack_length()) { |
| 3118 | retval = false; |
| 3119 | goto cleanup; |
| 3120 | } |
| 3121 | break; |
| 3122 | case MYSQL_TYPE_DATE: |
| 3123 | case MYSQL_TYPE_DATETIME: |
| 3124 | case MYSQL_TYPE_YEAR: |
| 3125 | case MYSQL_TYPE_NEWDATE: |
| 3126 | case MYSQL_TYPE_TIME: |
| 3127 | case MYSQL_TYPE_TIMESTAMP: |
| 3128 | #if (50600 <= MYSQL_VERSION_ID && MYSQL_VERSION_ID <= 50699) || \ |
| 3129 | (50700 <= MYSQL_VERSION_ID && MYSQL_VERSION_ID <= 50799) || \ |
| 3130 | (100000 <= MYSQL_VERSION_ID) |
| 3131 | case MYSQL_TYPE_DATETIME2: |
| 3132 | case MYSQL_TYPE_TIMESTAMP2: |
| 3133 | case MYSQL_TYPE_TIME2: |
| 3134 | #endif |
| 3135 | // length |
| 3136 | if (a->pack_length() != b->pack_length()) { |
| 3137 | retval = false; |
| 3138 | goto cleanup; |
| 3139 | } |
| 3140 | break; |
| 3141 | case MYSQL_TYPE_TINY_BLOB: |
| 3142 | case MYSQL_TYPE_MEDIUM_BLOB: |
| 3143 | case MYSQL_TYPE_BLOB: |
| 3144 | case MYSQL_TYPE_LONG_BLOB: |
| 3145 | // test the charset |
| 3146 | if (a->charset()->number != b->charset()->number) { |
| 3147 | retval = false; |
| 3148 | goto cleanup; |
| 3149 | } |
| 3150 | if (a->row_pack_length() != b->row_pack_length()) { |
| 3151 | retval = false; |
| 3152 | goto cleanup; |
| 3153 | } |
| 3154 | break; |
| 3155 | case MYSQL_TYPE_STRING: |
| 3156 | if (a->pack_length() != b->pack_length()) { |
| 3157 | retval = false; |
| 3158 | goto cleanup; |
| 3159 | } |
| 3160 | // if both are binary, we know have same pack lengths, |
| 3161 | // so we can goto end |
| 3162 | if (a->binary() && b->binary()) { |
| 3163 | // nothing to do, we are good |
| 3164 | } |
| 3165 | else if (!a->binary() && !b->binary()) { |
| 3166 | // test the charset |
| 3167 | if (a->charset()->number != b->charset()->number) { |
| 3168 | retval = false; |
| 3169 | goto cleanup; |
| 3170 | } |
| 3171 | } |
| 3172 | else { |
| 3173 | // one is binary and the other is not, so not the same |
| 3174 | retval = false; |
| 3175 | goto cleanup; |
| 3176 | } |
| 3177 | break; |
| 3178 | case MYSQL_TYPE_VARCHAR: |
| 3179 | if (a->field_length != b->field_length) { |
| 3180 | retval = false; |
| 3181 | goto cleanup; |
| 3182 | } |
| 3183 | // if both are binary, we know have same pack lengths, |
| 3184 | // so we can goto end |
| 3185 | if (a->binary() && b->binary()) { |
| 3186 | // nothing to do, we are good |
| 3187 | } |
| 3188 | else if (!a->binary() && !b->binary()) { |
| 3189 | // test the charset |
| 3190 | if (a->charset()->number != b->charset()->number) { |
| 3191 | retval = false; |
| 3192 | goto cleanup; |
| 3193 | } |
| 3194 | } |
| 3195 | else { |
| 3196 | // one is binary and the other is not, so not the same |
| 3197 | retval = false; |
| 3198 | goto cleanup; |
| 3199 | } |
| 3200 | break; |
| 3201 | // |
| 3202 | // I believe these are old types that are no longer |
| 3203 | // in any 5.1 tables, so tokudb does not need |
| 3204 | // to worry about them |
| 3205 | // Putting in this assert in case I am wrong. |
| 3206 | // Do not support geometry yet. |
| 3207 | // |
| 3208 | case MYSQL_TYPE_GEOMETRY: |
| 3209 | case MYSQL_TYPE_DECIMAL: |
| 3210 | case MYSQL_TYPE_VAR_STRING: |
| 3211 | case MYSQL_TYPE_NULL: |
| 3212 | case MYSQL_TYPE_VARCHAR_COMPRESSED: |
| 3213 | case MYSQL_TYPE_BLOB_COMPRESSED: |
| 3214 | assert_unreachable(); |
| 3215 | } |
| 3216 | |
| 3217 | cleanup: |
| 3218 | return retval; |
| 3219 | } |
| 3220 | |
| 3221 | static bool are_two_fields_same(Field* a, Field* b) { |
| 3222 | return fields_have_same_name(a, b) && fields_are_same_type(a, b); |
| 3223 | } |
| 3224 | |
| 3225 | |
| 3226 | |