| 1 | // @(#) $Revision: 4.1 $ $Source: /judy/src/JudyHS/JudyHS.c |
| 2 | //======================================================================= |
| 3 | // Author Douglas L. Baskins, Dec 2003. |
| 4 | // Permission to use this code is freely granted, provided that this |
| 5 | // statement is retained. |
| 6 | // email - doug@sourcejudy.com -or- dougbaskins@yahoo.com |
| 7 | //======================================================================= |
| 8 | |
| 9 | #include <string.h> // for memcmp(), memcpy() |
| 10 | |
| 11 | #include <Judy.h> // for JudyL* routines/macros |
| 12 | |
| 13 | /* |
| 14 | This routine is a very fast "string" version of an ADT that stores |
| 15 | (JudyHSIns()), retrieves (JudyHSGet()), deletes (JudyHSDel()) and |
| 16 | frees the entire ADT (JudyHSFreeArray()) strings. It uses the "Judy |
| 17 | arrays" JudyL() API as the main workhorse. The length of the string |
| 18 | is included in the calling parameters so that strings with embedded |
| 19 | \0s can be used. The string lengths can be from 0 bytes to whatever |
| 20 | malloc() can handle (~2GB). |
| 21 | |
| 22 | Compile: |
| 23 | |
| 24 | cc -O JudyHS.c -c needs to link with -lJudy (libJudy.a) |
| 25 | |
| 26 | Note: in gcc version 3.3.1, -O2 generates faster code than -O |
| 27 | Note: in gcc version 3.3.2, -O3 generates faster code than -O2 |
| 28 | |
| 29 | NOTES: |
| 30 | |
| 31 | 1) There may be some performance issues with 64 bit machines, because I |
| 32 | have not characterized that it yet. |
| 33 | |
| 34 | 2) It appears that a modern CPU (>2Ghz) that the instruction times are |
| 35 | much faster that a RAM access, so building up a word from bytes takes |
| 36 | no longer that a whole word access. I am taking advantage of this to |
| 37 | make this code endian neutral. A side effect of this is strings do |
| 38 | not need to be aligned, nor tested to be on to a word boundry. In |
| 39 | older and in slow (RISC) machines, this may be a performance issue. |
| 40 | I have given up trying to optimize for machines that have very slow |
| 41 | mpy, mod, variable shifts and call returns. |
| 42 | |
| 43 | 3) JudyHS is very scalable from 1 string to billions (with enough RAM). |
| 44 | The memory usage is also scales with population. I have attempted to |
| 45 | combine the best characteristics of JudyL arrays with Hashing methods |
| 46 | and well designed modern processors (such as the 1.3Ghz Intel |
| 47 | Centrino this is being written on). |
| 48 | |
| 49 | HOW JudyHS WORKS: ( 4[8] means 4 bytes in 32 bit machine and 8 in 64) |
| 50 | |
| 51 | A) A JudyL array is used to separate strings of equal lengths into |
| 52 | their own structures (a different hash table is used for each length |
| 53 | of string). The additional time overhead is very near zero because |
| 54 | of the CPU cache. The space efficiency is improved because the |
| 55 | length need not be stored with the string (ls_t). The "JLHash" ADT |
| 56 | in the test program "StringCompare" is verification of both these |
| 57 | assumptions. |
| 58 | |
| 59 | B) A 32 bit hash value is produced from the string. Many thanks to |
| 60 | the Internet and the author (Bob Jenkins) for coming up with a very |
| 61 | good and fast universal string hash. Next the 32 bit hash number is |
| 62 | used as an Index to another JudyL array. Notice that one (1) JudyL |
| 63 | array is used as a hash table per each string length. If there are |
| 64 | no hash collisions (normally) then the string is copied to a |
| 65 | structure (ls_t) along with room for storing a Value. A flag is |
| 66 | added to the pointer to note it is pointing to a ls_t structure. |
| 67 | Since the lengths of the strings are the same, there is no need to |
| 68 | stored length of string in the ls_t structure. This saves about a |
| 69 | word per string of memory. |
| 70 | |
| 71 | C) When there is a hashing collision (very rare), a JudyL array is |
| 72 | used to decode the next 4[8] bytes of the string. That is, the next |
| 73 | 4[8] bytes of the string are used as the Index. This process is |
| 74 | repeated until the remaining string is unique. The remaining string |
| 75 | (if any) is stored in a (now smaller) ls_t structure. If the |
| 76 | remaining string is less or equal to 4[8] bytes, then the ls_t |
| 77 | structure is not needed and the Value area in the JudyL array is |
| 78 | used. A compile option -DDONOTUSEHASH is available to test this |
| 79 | structure without using hashing (only the JudyL tree is used). This |
| 80 | is equivalent to having all strings hashed to the same bucket. The |
| 81 | speed is still better than all other tree based ADTs I have tested. |
| 82 | An added benefit of this is a very fast "hash collision" resolving. |
| 83 | It could foil hackers that exploit the slow synonym (linked-list) |
| 84 | collision handling property used with most hashing algorithms. If |
| 85 | this is not a necessary property, then a simpler ADT "JLHash" that is |
| 86 | documented the the test program "StringCompare.c" may be used with a |
| 87 | little loss of memory efficiency (because it includes the string |
| 88 | length with the ls_t structure). JudyHS was written to be the |
| 89 | fastest, very scalable, memory efficient, general purpose string ADT |
| 90 | possible. (However, I would like to eat those words someday). (dlb) |
| 91 | |
| 92 | */ |
| 93 | |
| 94 | #ifdef EXAMPLE_CODE |
| 95 | #include <stdio.h> |
| 96 | #include <unistd.h> |
| 97 | #include <string.h> |
| 98 | |
| 99 | #include <Judy.h> |
| 100 | |
| 101 | //#include "JudyHS.h" // for Judy.h without JudyHS*() |
| 102 | |
| 103 | // By Doug Baskins Apr 2004 - for JudyHS man page |
| 104 | |
| 105 | #define MAXLINE 1000000 /* max length of line */ |
| 106 | char Index[MAXLINE]; // string to check |
| 107 | |
| 108 | int // Usage: CheckDupLines < file |
| 109 | main() |
| 110 | { |
| 111 | Pvoid_t PJArray = (PWord_t)NULL; // Judy array. |
| 112 | PWord_t PValue; // ^ Judy array element. |
| 113 | Word_t Bytes; // size of JudyHS array. |
| 114 | Word_t LineNumb = 0; // current line number |
| 115 | Word_t Dups = 0; // number of duplicate lines |
| 116 | |
| 117 | while (fgets(Index, MAXLINE, stdin) != (char *)NULL) |
| 118 | { |
| 119 | LineNumb++; // line number |
| 120 | |
| 121 | // store string into array |
| 122 | JHSI(PValue, PJArray, Index, strlen(Index)); |
| 123 | if (*PValue) // check if duplicate |
| 124 | { |
| 125 | Dups++; // count duplicates |
| 126 | printf("Duplicate lines %lu:%lu:%s", *PValue, LineNumb, Index); |
| 127 | } |
| 128 | else |
| 129 | { |
| 130 | *PValue = LineNumb; // store Line number |
| 131 | } |
| 132 | } |
| 133 | printf("%lu Duplicates, free JudyHS array of %lu Lines\n", |
| 134 | Dups, LineNumb - Dups); |
| 135 | JHSFA(Bytes, PJArray); // free array |
| 136 | printf("The JudyHS array allocated %lu bytes of memory\n", Bytes); |
| 137 | return (0); |
| 138 | } |
| 139 | #endif // EXAMPLE_CODE |
| 140 | |
| 141 | // Note: Use JLAP_INVALID, which is non-zero, to mark pointers to a ls_t |
| 142 | // This makes it compatable with previous versions of JudyL() |
| 143 | |
| 144 | #define IS_PLS(PLS) (((Word_t) (PLS)) & JLAP_INVALID) |
| 145 | #define CLEAR_PLS(PLS) (((Word_t) (PLS)) & (~JLAP_INVALID)) |
| 146 | #define SET_PLS(PLS) (((Word_t) (PLS)) | JLAP_INVALID) |
| 147 | |
| 148 | #define WORDSIZE (sizeof(Word_t)) |
| 149 | |
| 150 | // this is the struct used for "leaf" strings. Note that |
| 151 | // the Value is followed by a "variable" length ls_String array. |
| 152 | // |
| 153 | typedef struct L_EAFSTRING |
| 154 | { |
| 155 | Word_t ls_Value; // Value area (cannot change size) |
| 156 | uint8_t ls_String[WORDSIZE]; // to fill out to a Word_t size |
| 157 | } ls_t , *Pls_t; |
| 158 | |
| 159 | #define LS_STRUCTOVD (sizeof(ls_t) - WORDSIZE) |
| 160 | |
| 161 | // Calculate size of ls_t including the string of length of LEN. |
| 162 | // |
| 163 | #define LS_WORDLEN(LEN) (((LEN) + LS_STRUCTOVD + WORDSIZE - 1) / WORDSIZE) |
| 164 | |
| 165 | // Copy from 0..4[8] bytes from string to a Word_t |
| 166 | // NOTE: the copy in in little-endian order to take advantage of improved |
| 167 | // memory efficiency of JudyLIns() with smaller numbers |
| 168 | // |
| 169 | #define COPYSTRING4toWORD(WORD,STR,LEN) \ |
| 170 | { \ |
| 171 | WORD = 0; \ |
| 172 | switch(LEN) \ |
| 173 | { \ |
| 174 | default: /* four and greater */ \ |
| 175 | case 4: \ |
| 176 | WORD += (Word_t)(((uint8_t *)(STR))[3] << 24); \ |
| 177 | case 3: \ |
| 178 | WORD += (Word_t)(((uint8_t *)(STR))[2] << 16); \ |
| 179 | case 2: \ |
| 180 | WORD += (Word_t)(((uint8_t *)(STR))[1] << 8); \ |
| 181 | case 1: \ |
| 182 | WORD += (Word_t)(((uint8_t *)(STR))[0]); \ |
| 183 | case 0: break; \ |
| 184 | } \ |
| 185 | } |
| 186 | |
| 187 | #ifdef JU_64BIT |
| 188 | |
| 189 | // copy from 0..8 bytes from string to Word_t |
| 190 | // |
| 191 | #define COPYSTRING8toWORD(WORD,STR,LEN) \ |
| 192 | { \ |
| 193 | WORD = 0UL; \ |
| 194 | switch(LEN) \ |
| 195 | { \ |
| 196 | default: /* eight and greater */ \ |
| 197 | case 8: \ |
| 198 | WORD += ((Word_t)((uint8_t *)(STR))[7] << 56); \ |
| 199 | case 7: \ |
| 200 | WORD += ((Word_t)((uint8_t *)(STR))[6] << 48); \ |
| 201 | case 6: \ |
| 202 | WORD += ((Word_t)((uint8_t *)(STR))[5] << 40); \ |
| 203 | case 5: \ |
| 204 | WORD += ((Word_t)((uint8_t *)(STR))[4] << 32); \ |
| 205 | case 4: \ |
| 206 | WORD += ((Word_t)((uint8_t *)(STR))[3] << 24); \ |
| 207 | case 3: \ |
| 208 | WORD += ((Word_t)((uint8_t *)(STR))[2] << 16); \ |
| 209 | case 2: \ |
| 210 | WORD += ((Word_t)((uint8_t *)(STR))[1] << 8); \ |
| 211 | case 1: \ |
| 212 | WORD += ((Word_t)((uint8_t *)(STR))[0]); \ |
| 213 | case 0: break; \ |
| 214 | } \ |
| 215 | } |
| 216 | |
| 217 | #define COPYSTRINGtoWORD COPYSTRING8toWORD |
| 218 | |
| 219 | #else // JU_32BIT |
| 220 | |
| 221 | #define COPYSTRINGtoWORD COPYSTRING4toWORD |
| 222 | |
| 223 | #endif // JU_32BIT |
| 224 | |
| 225 | // set JError_t locally |
| 226 | |
| 227 | #define JU_SET_ERRNO(PJERROR, JERRNO) \ |
| 228 | { \ |
| 229 | if (PJERROR != (PJError_t) NULL) \ |
| 230 | { \ |
| 231 | if (JERRNO) \ |
| 232 | JU_ERRNO(PJError) = (JERRNO); \ |
| 233 | JU_ERRID(PJERROR) = __LINE__; \ |
| 234 | } \ |
| 235 | } |
| 236 | |
| 237 | //======================================================================= |
| 238 | // This routine must hash string to 24..32 bits. The "goodness" of |
| 239 | // the hash is not as important as its speed. |
| 240 | //======================================================================= |
| 241 | |
| 242 | // hash to no more than 32 bits |
| 243 | |
| 244 | // extern Word_t gHmask; for hash bits experiments |
| 245 | |
| 246 | #define JUDYHASHSTR(HVALUE,STRING,LENGTH) \ |
| 247 | { \ |
| 248 | uint8_t *p_ = (uint8_t *)(STRING); \ |
| 249 | uint8_t *q_ = p_ + (LENGTH); \ |
| 250 | uint32_t c_ = 0; \ |
| 251 | for (; p_ != q_; ++p_) \ |
| 252 | { \ |
| 253 | c_ = (c_ * 31) + *p_; \ |
| 254 | } \ |
| 255 | /* c_ &= gHmask; see above */ \ |
| 256 | (HVALUE) = c_; \ |
| 257 | } |
| 258 | |
| 259 | // Find String of Len in JudyHS structure, return pointer to associated Value |
| 260 | |
| 261 | PPvoid_t |
| 262 | JudyHSGet(Pcvoid_t PArray, // pointer (^) to structure |
| 263 | void * Str, // pointer to string |
| 264 | Word_t Len // length of string |
| 265 | ) |
| 266 | { |
| 267 | uint8_t *String = (uint8_t *)Str; |
| 268 | PPvoid_t PPValue; // pointer to Value |
| 269 | Word_t Index; // 4[8] bytes of String |
| 270 | |
| 271 | JLG(PPValue, PArray, Len); // find hash table for strings of Len |
| 272 | if (PPValue == (PPvoid_t) NULL) |
| 273 | return ((PPvoid_t) NULL); // no strings of this Len |
| 274 | |
| 275 | // check for caller error (null pointer) |
| 276 | // |
| 277 | if ((String == (void *) NULL) && (Len != 0)) |
| 278 | return ((PPvoid_t) NULL); // avoid null-pointer dereference |
| 279 | |
| 280 | #ifndef DONOTUSEHASH |
| 281 | if (Len > WORDSIZE) // Hash table not necessary with short |
| 282 | { |
| 283 | uint32_t HValue; // hash of input string |
| 284 | JUDYHASHSTR(HValue, String, Len); // hash to no more than 32 bits |
| 285 | JLG(PPValue, *PPValue, (Word_t)HValue); // get ^ to hash bucket |
| 286 | if (PPValue == (PPvoid_t) NULL) |
| 287 | return ((PPvoid_t) NULL); // no entry in Hash table |
| 288 | } |
| 289 | #endif // DONOTUSEHASH |
| 290 | |
| 291 | /* |
| 292 | Each JudyL array decodes 4[8] bytes of the string. Since the hash |
| 293 | collisions occur very infrequently, the performance is not important. |
| 294 | However, even if the Hash code is not used this method still is |
| 295 | significantly faster than common tree methods (AVL, Red-Black, Splay, |
| 296 | b-tree, etc..). You can compare it yourself with #define DONOTUSEHASH |
| 297 | 1 or putting -DDONOTUSEHASH in the cc line. Use the "StringCompare.c" |
| 298 | code to compare (9Dec2003 dlb). |
| 299 | */ |
| 300 | while (Len > WORDSIZE) // traverse tree of JudyL arrays |
| 301 | { |
| 302 | if (IS_PLS(*PPValue)) // ^ to JudyL array or ls_t struct? |
| 303 | { |
| 304 | Pls_t Pls; // ls_t struct, termination of tree |
| 305 | Pls = (Pls_t) CLEAR_PLS(*PPValue); // remove flag from ^ |
| 306 | |
| 307 | // if remaining string matches, return ^ to Value, else NULL |
| 308 | |
| 309 | if (memcmp(String, Pls->ls_String, Len) == 0) |
| 310 | return ((PPvoid_t) (&(Pls->ls_Value))); |
| 311 | else |
| 312 | return ((PPvoid_t) NULL); // string does not match |
| 313 | } |
| 314 | else |
| 315 | { |
| 316 | COPYSTRINGtoWORD(Index, String, WORDSIZE); |
| 317 | |
| 318 | JLG(PPValue, *PPValue, Index); // decode next 4[8] bytes |
| 319 | if (PPValue == (PPvoid_t) NULL) // if NULL array, bail out |
| 320 | return ((PPvoid_t) NULL); // string does not match |
| 321 | |
| 322 | String += WORDSIZE; // advance |
| 323 | Len -= WORDSIZE; |
| 324 | } |
| 325 | } |
| 326 | |
| 327 | // Get remaining 1..4[8] bytes left in string |
| 328 | |
| 329 | COPYSTRINGtoWORD(Index, String, Len); |
| 330 | JLG(PPValue, *PPValue, Index); // decode last 1-4[8] bytes |
| 331 | return (PPValue); |
| 332 | } |
| 333 | |
| 334 | // Add string to a tree of JudyL arrays (all lengths must be same) |
| 335 | |
| 336 | static PPvoid_t |
| 337 | insStrJudyLTree(uint8_t * String, // string to add to tree of JudyL arrays |
| 338 | Word_t Len, // length of string |
| 339 | PPvoid_t PPValue, // pointer to root pointer |
| 340 | PJError_t PJError // for returning error info |
| 341 | ) |
| 342 | { |
| 343 | Word_t Index; // next 4[8] bytes of String |
| 344 | |
| 345 | while (Len > WORDSIZE) // add to JudyL tree |
| 346 | { |
| 347 | // CASE 1, pointer is to a NULL, make a new ls_t leaf |
| 348 | |
| 349 | if (*PPValue == (Pvoid_t)NULL) |
| 350 | { |
| 351 | Pls_t Pls; // memory for a ls_t |
| 352 | Pls = (Pls_t) JudyMalloc(LS_WORDLEN(Len)); |
| 353 | if (Pls == NULL) |
| 354 | { |
| 355 | JU_SET_ERRNO(PJError, JU_ERRNO_NOMEM); |
| 356 | return (PPJERR); |
| 357 | } |
| 358 | Pls->ls_Value = 0; // clear Value word |
| 359 | memcpy(Pls->ls_String, String, Len); // copy to new struct |
| 360 | *PPValue = (Pvoid_t)SET_PLS(Pls); // mark pointer |
| 361 | return ((PPvoid_t) (&Pls->ls_Value)); // return ^ to Value |
| 362 | } // no exit here |
| 363 | // CASE 2: is a ls_t, free (and shorten), then decode into JudyL tree |
| 364 | |
| 365 | if (IS_PLS(*PPValue)) // pointer to a ls_t? (leaf) |
| 366 | { |
| 367 | Pls_t Pls; // ^ to ls_t |
| 368 | uint8_t *String0; // ^ to string in ls_t |
| 369 | Word_t Index0; // 4[8] bytes in string |
| 370 | Word_t FreeLen; // length of ls_t |
| 371 | PPvoid_t PPsplit; |
| 372 | |
| 373 | FreeLen = LS_WORDLEN(Len); // length of ls_t |
| 374 | |
| 375 | Pls = (Pls_t) CLEAR_PLS(*PPValue); // demangle ^ to ls_t |
| 376 | String0 = Pls->ls_String; |
| 377 | if (memcmp(String, String0, Len) == 0) // check if match? |
| 378 | { |
| 379 | return ((PPvoid_t) (&Pls->ls_Value)); // yes, duplicate |
| 380 | } |
| 381 | |
| 382 | *PPValue = NULL; // clear ^ to ls_t and make JudyL |
| 383 | |
| 384 | // This do loop is technically not required, saves multiple JudyFree() |
| 385 | // when storing already sorted strings into structure |
| 386 | |
| 387 | do // decode next 4[8] bytes of string |
| 388 | { // with a JudyL array |
| 389 | // Note: string0 is always aligned |
| 390 | |
| 391 | COPYSTRINGtoWORD(Index0, String0, WORDSIZE); |
| 392 | String0 += WORDSIZE; |
| 393 | COPYSTRINGtoWORD(Index, String, WORDSIZE); |
| 394 | String += WORDSIZE; |
| 395 | Len -= WORDSIZE; |
| 396 | PPsplit = PPValue; // save for split below |
| 397 | PPValue = JudyLIns(PPValue, Index0, PJError); |
| 398 | if (PPValue == PPJERR) |
| 399 | { |
| 400 | JU_SET_ERRNO(PJError, 0); |
| 401 | return (PPJERR); |
| 402 | } |
| 403 | |
| 404 | } while ((Index0 == Index) && (Len > WORDSIZE)); |
| 405 | |
| 406 | // finish storing remainder of string that was in the ls_t |
| 407 | |
| 408 | PPValue = insStrJudyLTree(String0, Len, PPValue, PJError); |
| 409 | if (PPValue == PPJERR) |
| 410 | { |
| 411 | return (PPJERR); |
| 412 | } |
| 413 | // copy old Value to Value in new struct |
| 414 | |
| 415 | *(PWord_t)PPValue = Pls->ls_Value; |
| 416 | |
| 417 | // free the string buffer (ls_t) |
| 418 | |
| 419 | JudyFree((Pvoid_t)Pls, FreeLen); |
| 420 | PPValue = JudyLIns(PPsplit, Index, PJError); |
| 421 | if (PPValue == PPJERR) |
| 422 | { |
| 423 | JU_SET_ERRNO(PJError, 0); |
| 424 | return (PPValue); |
| 425 | } |
| 426 | |
| 427 | // finish remainder of newly inserted string |
| 428 | |
| 429 | PPValue = insStrJudyLTree(String, Len, PPValue, PJError); |
| 430 | return (PPValue); |
| 431 | } // no exit here |
| 432 | // CASE 3, more JudyL arrays, decode to next tree |
| 433 | |
| 434 | COPYSTRINGtoWORD(Index, String, WORDSIZE); |
| 435 | Len -= WORDSIZE; |
| 436 | String += WORDSIZE; |
| 437 | |
| 438 | PPValue = JudyLIns(PPValue, Index, PJError); // next 4[8] bytes |
| 439 | if (PPValue == PPJERR) |
| 440 | { |
| 441 | JU_SET_ERRNO(PJError, 0); |
| 442 | return (PPValue); |
| 443 | } |
| 444 | } |
| 445 | // this is done outside of loop so "Len" can be an unsigned number |
| 446 | |
| 447 | COPYSTRINGtoWORD(Index, String, Len); |
| 448 | PPValue = JudyLIns(PPValue, Index, PJError); // remaining 4[8] bytes |
| 449 | |
| 450 | return (PPValue); |
| 451 | } |
| 452 | |
| 453 | |
| 454 | // Insert string to JudyHS structure, return pointer to associated Value |
| 455 | |
| 456 | PPvoid_t |
| 457 | JudyHSIns(PPvoid_t PPArray, // ^ to JudyHashArray name |
| 458 | void * Str, // pointer to string |
| 459 | Word_t Len, // length of string |
| 460 | PJError_t PJError // optional, for returning error info |
| 461 | ) |
| 462 | { |
| 463 | uint8_t * String = (uint8_t *)Str; |
| 464 | PPvoid_t PPValue; |
| 465 | |
| 466 | // string can only be NULL if Len is 0. |
| 467 | |
| 468 | if ((String == (uint8_t *) NULL) && (Len != 0UL)) |
| 469 | { |
| 470 | JU_SET_ERRNO(PJError, JU_ERRNO_NULLPINDEX); |
| 471 | return (PPJERR); |
| 472 | } |
| 473 | JLG(PPValue, *PPArray, Len); // JudyL hash table for strings of Len |
| 474 | if (PPValue == (PPvoid_t) NULL) // make new if missing, (very rare) |
| 475 | { |
| 476 | PPValue = JudyLIns(PPArray, Len, PJError); |
| 477 | if (PPValue == PPJERR) |
| 478 | { |
| 479 | JU_SET_ERRNO(PJError, 0); |
| 480 | return (PPJERR); |
| 481 | } |
| 482 | } |
| 483 | #ifndef DONOTUSEHASH |
| 484 | if (Len > WORDSIZE) |
| 485 | { |
| 486 | uint32_t HValue; // hash of input string |
| 487 | JUDYHASHSTR(HValue, String, Len); // hash to no more than 32 bits |
| 488 | PPValue = JudyLIns(PPValue, (Word_t)HValue, PJError); |
| 489 | if (PPValue == PPJERR) |
| 490 | { |
| 491 | JU_SET_ERRNO(PJError, 0); |
| 492 | return (PPJERR); |
| 493 | } |
| 494 | } |
| 495 | #endif // DONOTUSEHASH |
| 496 | |
| 497 | PPValue = insStrJudyLTree(String, Len, PPValue, PJError); // add string |
| 498 | return (PPValue); // ^ to Value |
| 499 | } |
| 500 | |
| 501 | // Delete string from tree of JudyL arrays (all Lens must be same) |
| 502 | |
| 503 | static int |
| 504 | delStrJudyLTree(uint8_t * String, // delete from tree of JudyL arrays |
| 505 | Word_t Len, // length of string |
| 506 | PPvoid_t PPValue, // ^ to hash bucket |
| 507 | PJError_t PJError // for returning error info |
| 508 | ) |
| 509 | { |
| 510 | PPvoid_t PPValueN; // next pointer |
| 511 | Word_t Index; |
| 512 | int Ret; // -1=failed, 1=success, 2=quit del |
| 513 | |
| 514 | if (IS_PLS(*PPValue)) // is pointer to ls_t? |
| 515 | { |
| 516 | Pls_t Pls; |
| 517 | Pls = (Pls_t) CLEAR_PLS(*PPValue); // demangle pointer |
| 518 | JudyFree((Pvoid_t)Pls, LS_WORDLEN(Len)); // free the ls_t |
| 519 | |
| 520 | *PPValue = (Pvoid_t)NULL; // clean pointer |
| 521 | return (1); // successfully deleted |
| 522 | } |
| 523 | |
| 524 | if (Len > WORDSIZE) // delete from JudyL tree, not leaf |
| 525 | { |
| 526 | COPYSTRINGtoWORD(Index, String, WORDSIZE); // get Index |
| 527 | JLG(PPValueN, *PPValue, Index); // get pointer to next JudyL array |
| 528 | |
| 529 | String += WORDSIZE; // advance to next 4[8] bytes |
| 530 | Len -= WORDSIZE; |
| 531 | |
| 532 | Ret = delStrJudyLTree(String, Len, PPValueN, PJError); |
| 533 | if (Ret != 1) return(Ret); |
| 534 | |
| 535 | if (*PPValueN == (PPvoid_t) NULL) |
| 536 | { |
| 537 | // delete JudyL element from tree |
| 538 | |
| 539 | Ret = JudyLDel(PPValue, Index, PJError); |
| 540 | } |
| 541 | } |
| 542 | else |
| 543 | { |
| 544 | COPYSTRINGtoWORD(Index, String, Len); // get leaf element |
| 545 | |
| 546 | // delete last 1-4[8] bytes from leaf element |
| 547 | |
| 548 | Ret = JudyLDel(PPValue, Index, PJError); |
| 549 | } |
| 550 | return (Ret); |
| 551 | } |
| 552 | |
| 553 | // Delete string from JHS structure |
| 554 | |
| 555 | int |
| 556 | JudyHSDel(PPvoid_t PPArray, // ^ to JudyHashArray struct |
| 557 | void * Str, // pointer to string |
| 558 | Word_t Len, // length of string |
| 559 | PJError_t PJError // optional, for returning error info |
| 560 | ) |
| 561 | { |
| 562 | uint8_t * String = (uint8_t *)Str; |
| 563 | PPvoid_t PPBucket, PPHtble; |
| 564 | int Ret; // return bool from Delete routine |
| 565 | #ifndef DONOTUSEHASH |
| 566 | uint32_t HValue = 0; // hash value of input string |
| 567 | #endif // DONOTUSEHASH |
| 568 | |
| 569 | if (PPArray == NULL) |
| 570 | return (0); // no pointer, return not found |
| 571 | |
| 572 | // This is a little slower than optimum method, but not much in new CPU |
| 573 | // Verify that string is in the structure -- simplifies future assumptions |
| 574 | |
| 575 | if (JudyHSGet(*PPArray, String, Len) == (PPvoid_t) NULL) |
| 576 | return (0); // string not found, return |
| 577 | |
| 578 | // string is in structure, so testing for absence is not necessary |
| 579 | |
| 580 | JLG(PPHtble, *PPArray, Len); // JudyL hash table for strings of Len |
| 581 | |
| 582 | #ifdef DONOTUSEHASH |
| 583 | PPBucket = PPHtble; // simulate below code |
| 584 | #else // USEHASH |
| 585 | if (Len > WORDSIZE) |
| 586 | { |
| 587 | JUDYHASHSTR(HValue, String, Len); // hash to no more than 32 bits |
| 588 | |
| 589 | // get pointer to hash bucket |
| 590 | |
| 591 | JLG(PPBucket, *PPHtble, (Word_t)HValue); |
| 592 | } |
| 593 | else |
| 594 | { |
| 595 | PPBucket = PPHtble; // no bucket to JLGet |
| 596 | } |
| 597 | #endif // USEHASH |
| 598 | |
| 599 | // delete from JudyL tree |
| 600 | // |
| 601 | Ret = delStrJudyLTree(String, Len, PPBucket, PJError); |
| 602 | if (Ret != 1) |
| 603 | { |
| 604 | JU_SET_ERRNO(PJError, 0); |
| 605 | return(-1); |
| 606 | } |
| 607 | // handle case of missing JudyL array from hash table and length table |
| 608 | |
| 609 | if (*PPBucket == (Pvoid_t)NULL) // if JudyL tree gone |
| 610 | { |
| 611 | #ifndef DONOTUSEHASH |
| 612 | if (Len > WORDSIZE) |
| 613 | { |
| 614 | // delete entry in Hash table |
| 615 | |
| 616 | Ret = JudyLDel(PPHtble, (Word_t)HValue, PJError); |
| 617 | if (Ret != 1) |
| 618 | { |
| 619 | JU_SET_ERRNO(PJError, 0); |
| 620 | return(-1); |
| 621 | } |
| 622 | } |
| 623 | #endif // USEHASH |
| 624 | if (*PPHtble == (PPvoid_t) NULL) // if Hash table gone |
| 625 | { |
| 626 | // delete entry from the String length table |
| 627 | |
| 628 | Ret = JudyLDel(PPArray, Len, PJError); |
| 629 | if (Ret != 1) |
| 630 | { |
| 631 | JU_SET_ERRNO(PJError, 0); |
| 632 | return(-1); |
| 633 | } |
| 634 | } |
| 635 | } |
| 636 | return (1); // success |
| 637 | } |
| 638 | |
| 639 | static Word_t |
| 640 | delJudyLTree(PPvoid_t PPValue, // ^ to JudyL root pointer |
| 641 | Word_t Len, // length of string |
| 642 | PJError_t PJError) // for returning error info |
| 643 | { |
| 644 | Word_t bytes_freed = 0; // bytes freed at point |
| 645 | Word_t bytes_total = 0; // accumulated bytes freed |
| 646 | PPvoid_t PPValueN; |
| 647 | |
| 648 | // Pointer is to another tree of JudyL arrays or ls_t struct |
| 649 | |
| 650 | if (Len > WORDSIZE) // more depth to tree |
| 651 | { |
| 652 | Word_t NEntry; |
| 653 | |
| 654 | // Pointer is to a ls_t struct |
| 655 | |
| 656 | if (IS_PLS(*PPValue)) |
| 657 | { |
| 658 | Pls_t Pls; |
| 659 | Word_t freewords; |
| 660 | |
| 661 | freewords = LS_WORDLEN(Len); // calculate length |
| 662 | Pls = (Pls_t)CLEAR_PLS(*PPValue); // demangle pointer |
| 663 | |
| 664 | // *PPValue = (Pvoid_t)NULL; // clean pointer |
| 665 | JudyFree((Pvoid_t)Pls, freewords); // free the ls_t |
| 666 | |
| 667 | return(freewords * WORDSIZE); |
| 668 | } |
| 669 | // else |
| 670 | // Walk all the entrys in the JudyL array |
| 671 | |
| 672 | NEntry = 0; // start at beginning |
| 673 | for (PPValueN = JudyLFirst(*PPValue, &NEntry, PJError); |
| 674 | (PPValueN != (PPvoid_t) NULL) && (PPValueN != PPJERR); |
| 675 | PPValueN = JudyLNext(*PPValue, &NEntry, PJError)) |
| 676 | { |
| 677 | // recurse to the next level in the tree of arrays |
| 678 | |
| 679 | bytes_freed = delJudyLTree(PPValueN, Len - WORDSIZE, PJError); |
| 680 | if (bytes_freed == JERR) return(JERR); |
| 681 | bytes_total += bytes_freed; |
| 682 | } |
| 683 | if (PPValueN == PPJERR) return(JERR); |
| 684 | |
| 685 | // now free this JudyL array |
| 686 | |
| 687 | bytes_freed = JudyLFreeArray(PPValue, PJError); |
| 688 | if (bytes_freed == JERR) return(JERR); |
| 689 | bytes_total += bytes_freed; |
| 690 | |
| 691 | return(bytes_total); // return amount freed |
| 692 | } |
| 693 | // else |
| 694 | |
| 695 | // Pointer to simple JudyL array |
| 696 | |
| 697 | bytes_freed = JudyLFreeArray(PPValue, PJError); |
| 698 | |
| 699 | return(bytes_freed); |
| 700 | } |
| 701 | |
| 702 | |
| 703 | Word_t // bytes freed |
| 704 | JudyHSFreeArray(PPvoid_t PPArray, // ^ to JudyHashArray struct |
| 705 | PJError_t PJError // optional, for returning error info |
| 706 | ) |
| 707 | { |
| 708 | Word_t Len; // start at beginning |
| 709 | Word_t bytes_freed; // bytes freed at this level. |
| 710 | Word_t bytes_total; // bytes total at all levels. |
| 711 | PPvoid_t PPHtble; |
| 712 | |
| 713 | if (PPArray == NULL) |
| 714 | return (0); // no pointer, return none |
| 715 | |
| 716 | // Walk the string length table for subsidary hash structs |
| 717 | // NOTE: This is necessary to determine the depth of the tree |
| 718 | |
| 719 | bytes_freed = 0; |
| 720 | bytes_total = 0; |
| 721 | Len = 0; // walk to length table |
| 722 | |
| 723 | for (PPHtble = JudyLFirst(*PPArray, &Len, PJError); |
| 724 | (PPHtble != (PPvoid_t) NULL) && (PPHtble != PPJERR); |
| 725 | PPHtble = JudyLNext(*PPArray, &Len, PJError)) |
| 726 | { |
| 727 | PPvoid_t PPValueH; |
| 728 | |
| 729 | #ifndef DONOTUSEHASH |
| 730 | if (Len > WORDSIZE) |
| 731 | { |
| 732 | Word_t HEntry = 0; // walk the hash tables |
| 733 | |
| 734 | for (PPValueH = JudyLFirst(*PPHtble, &HEntry, PJError); |
| 735 | (PPValueH != (PPvoid_t) NULL) && (PPValueH != PPJERR); |
| 736 | PPValueH = JudyLNext(*PPHtble, &HEntry, PJError)) |
| 737 | { |
| 738 | bytes_freed = delJudyLTree(PPValueH, Len, PJError); |
| 739 | if (bytes_freed == JERR) return(JERR); |
| 740 | bytes_total += bytes_freed; |
| 741 | } |
| 742 | |
| 743 | if (PPValueH == PPJERR) return(JERR); |
| 744 | |
| 745 | // free the Hash table for this length of string |
| 746 | |
| 747 | bytes_freed = JudyLFreeArray(PPHtble, PJError); |
| 748 | if (bytes_freed == JERR) return(JERR); |
| 749 | bytes_total += bytes_freed; |
| 750 | } |
| 751 | else |
| 752 | #endif // DONOTUSEHASH |
| 753 | { |
| 754 | PPValueH = PPHtble; // simulate hash table |
| 755 | |
| 756 | bytes_freed = delJudyLTree(PPValueH, Len, PJError); |
| 757 | if (bytes_freed == JERR) return(JERR); |
| 758 | bytes_total += bytes_freed; |
| 759 | } |
| 760 | } |
| 761 | if (PPHtble == PPJERR) return(JERR); |
| 762 | |
| 763 | // free the length table |
| 764 | |
| 765 | bytes_freed = JudyLFreeArray(PPArray, PJError); |
| 766 | if (bytes_freed == JERR) return(JERR); |
| 767 | |
| 768 | bytes_total += bytes_freed; |
| 769 | |
| 770 | return(bytes_total); // return bytes freed |
| 771 | } |