1 /* deflate.c -- compress data using the deflation algorithm 2 * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler 3 * For conditions of distribution and use, see copyright notice in zlib.h 4 */ 5 6 /* 7 * ALGORITHM 8 * 9 * The "deflation" process depends on being able to identify portions 10 * of the input text which are identical to earlier input (within a 11 * sliding window trailing behind the input currently being processed). 12 * 13 * The most straightforward technique turns out to be the fastest for 14 * most input files: try all possible matches and select the longest. 15 * The key feature of this algorithm is that insertions into the string 16 * dictionary are very simple and thus fast, and deletions are avoided 17 * completely. Insertions are performed at each input character, whereas 18 * string matches are performed only when the previous match ends. So it 19 * is preferable to spend more time in matches to allow very fast string 20 * insertions and avoid deletions. The matching algorithm for small 21 * strings is inspired from that of Rabin & Karp. A brute force approach 22 * is used to find longer strings when a small match has been found. 23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze 24 * (by Leonid Broukhis). 25 * A previous version of this file used a more sophisticated algorithm 26 * (by Fiala and Greene) which is guaranteed to run in linear amortized 27 * time, but has a larger average cost, uses more memory and is patented. 28 * However the F&G algorithm may be faster for some highly redundant 29 * files if the parameter max_chain_length (described below) is too large. 30 * 31 * ACKNOWLEDGEMENTS 32 * 33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and 34 * I found it in 'freeze' written by Leonid Broukhis. 35 * Thanks to many people for bug reports and testing. 36 * 37 * REFERENCES 38 * 39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification". 40 * Available in http://tools.ietf.org/html/rfc1951 41 * 42 * A description of the Rabin and Karp algorithm is given in the book 43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252. 44 * 45 * Fiala,E.R., and Greene,D.H. 46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595 47 * 48 */ 49 50 /* @(#) $Id$ */ 51 52 #include "deflate.h" 53 54 const char deflate_copyright[] = 55 " deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler "; 56 /* 57 If you use the zlib library in a product, an acknowledgment is welcome 58 in the documentation of your product. If for some reason you cannot 59 include such an acknowledgment, I would appreciate that you keep this 60 copyright string in the executable of your product. 61 */ 62 63 /* =========================================================================== 64 * Function prototypes. 65 */ 66 typedef enum { 67 need_more, /* block not completed, need more input or more output */ 68 block_done, /* block flush performed */ 69 finish_started, /* finish started, need only more output at next deflate */ 70 finish_done /* finish done, accept no more input or output */ 71 } block_state; 72 73 typedef block_state (*compress_func) OF((deflate_state *s, int flush)); 74 /* Compression function. Returns the block state after the call. */ 75 76 local int deflateStateCheck OF((z_streamp strm)); 77 local void slide_hash OF((deflate_state *s)); 78 local void fill_window OF((deflate_state *s)); 79 local block_state deflate_stored OF((deflate_state *s, int flush)); 80 local block_state deflate_fast OF((deflate_state *s, int flush)); 81 #ifndef FASTEST 82 local block_state deflate_slow OF((deflate_state *s, int flush)); 83 #endif 84 local block_state deflate_rle OF((deflate_state *s, int flush)); 85 local block_state deflate_huff OF((deflate_state *s, int flush)); 86 local void lm_init OF((deflate_state *s)); 87 local void putShortMSB OF((deflate_state *s, uInt b)); 88 local void flush_pending OF((z_streamp strm)); 89 local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); 90 #ifdef ASMV 91 # pragma message("Assembler code may have bugs -- use at your own risk") 92 void match_init OF((void)); /* asm code initialization */ 93 uInt longest_match OF((deflate_state *s, IPos cur_match)); 94 #else 95 local uInt longest_match OF((deflate_state *s, IPos cur_match)); 96 #endif 97 98 #ifdef ZLIB_DEBUG 99 local void check_match OF((deflate_state *s, IPos start, IPos match, 100 int length)); 101 #endif 102 103 /* =========================================================================== 104 * Local data 105 */ 106 107 #define NIL 0 108 /* Tail of hash chains */ 109 110 #ifndef TOO_FAR 111 # define TOO_FAR 4096 112 #endif 113 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ 114 115 /* Values for max_lazy_match, good_match and max_chain_length, depending on 116 * the desired pack level (0..9). The values given below have been tuned to 117 * exclude worst case performance for pathological files. Better values may be 118 * found for specific files. 119 */ 120 typedef struct config_s { 121 ush good_length; /* reduce lazy search above this match length */ 122 ush max_lazy; /* do not perform lazy search above this match length */ 123 ush nice_length; /* quit search above this match length */ 124 ush max_chain; 125 compress_func func; 126 } config; 127 128 #ifdef FASTEST 129 local const config configuration_table[2] = { 130 /* good lazy nice chain */ 131 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 132 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */ 133 #else 134 local const config configuration_table[10] = { 135 /* good lazy nice chain */ 136 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 137 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */ 138 /* 2 */ {4, 5, 16, 8, deflate_fast}, 139 /* 3 */ {4, 6, 32, 32, deflate_fast}, 140 141 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ 142 /* 5 */ {8, 16, 32, 32, deflate_slow}, 143 /* 6 */ {8, 16, 128, 128, deflate_slow}, 144 /* 7 */ {8, 32, 128, 256, deflate_slow}, 145 /* 8 */ {32, 128, 258, 1024, deflate_slow}, 146 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */ 147 #endif 148 149 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 150 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different 151 * meaning. 152 */ 153 154 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */ 155 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0)) 156 157 /* =========================================================================== 158 * Update a hash value with the given input byte 159 * IN assertion: all calls to UPDATE_HASH are made with consecutive input 160 * characters, so that a running hash key can be computed from the previous 161 * key instead of complete recalculation each time. 162 */ 163 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask) 164 165 166 /* =========================================================================== 167 * Insert string str in the dictionary and set match_head to the previous head 168 * of the hash chain (the most recent string with same hash key). Return 169 * the previous length of the hash chain. 170 * If this file is compiled with -DFASTEST, the compression level is forced 171 * to 1, and no hash chains are maintained. 172 * IN assertion: all calls to INSERT_STRING are made with consecutive input 173 * characters and the first MIN_MATCH bytes of str are valid (except for 174 * the last MIN_MATCH-1 bytes of the input file). 175 */ 176 #ifdef FASTEST 177 #define INSERT_STRING(s, str, match_head) \ 178 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 179 match_head = s->head[s->ins_h], \ 180 s->head[s->ins_h] = (Pos)(str)) 181 #else 182 #define INSERT_STRING(s, str, match_head) \ 183 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 184 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \ 185 s->head[s->ins_h] = (Pos)(str)) 186 #endif 187 188 /* =========================================================================== 189 * Initialize the hash table (avoiding 64K overflow for 16 bit systems). 190 * prev[] will be initialized on the fly. 191 */ 192 #define CLEAR_HASH(s) \ 193 do { \ 194 s->head[s->hash_size-1] = NIL; \ 195 zmemzero((Bytef *)s->head, \ 196 (unsigned)(s->hash_size-1)*sizeof(*s->head)); \ 197 } while (0) 198 199 /* =========================================================================== 200 * Slide the hash table when sliding the window down (could be avoided with 32 201 * bit values at the expense of memory usage). We slide even when level == 0 to 202 * keep the hash table consistent if we switch back to level > 0 later. 203 */ 204 local void slide_hash(s) 205 deflate_state *s; 206 { 207 unsigned n, m; 208 Posf *p; 209 uInt wsize = s->w_size; 210 211 n = s->hash_size; 212 p = &s->head[n]; 213 do { 214 m = *--p; 215 *p = (Pos)(m >= wsize ? m - wsize : NIL); 216 } while (--n); 217 n = wsize; 218 #ifndef FASTEST 219 p = &s->prev[n]; 220 do { 221 m = *--p; 222 *p = (Pos)(m >= wsize ? m - wsize : NIL); 223 /* If n is not on any hash chain, prev[n] is garbage but 224 * its value will never be used. 225 */ 226 } while (--n); 227 #endif 228 } 229 230 /* ========================================================================= */ 231 int ZEXPORT deflateInit_(strm, level, version, stream_size) 232 z_streamp strm; 233 int level; 234 const char *version; 235 int stream_size; 236 { 237 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, 238 Z_DEFAULT_STRATEGY, version, stream_size); 239 /* To do: ignore strm->next_in if we use it as window */ 240 } 241 242 /* ========================================================================= */ 243 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, 244 version, stream_size) 245 z_streamp strm; 246 int level; 247 int method; 248 int windowBits; 249 int memLevel; 250 int strategy; 251 const char *version; 252 int stream_size; 253 { 254 deflate_state *s; 255 int wrap = 1; 256 static const char my_version[] = ZLIB_VERSION; 257 258 ushf *overlay; 259 /* We overlay pending_buf and d_buf+l_buf. This works since the average 260 * output size for (length,distance) codes is <= 24 bits. 261 */ 262 263 if (version == Z_NULL || version[0] != my_version[0] || 264 stream_size != sizeof(z_stream)) { 265 return Z_VERSION_ERROR; 266 } 267 if (strm == Z_NULL) return Z_STREAM_ERROR; 268 269 strm->msg = Z_NULL; 270 if (strm->zalloc == (alloc_func)0) { 271 #if defined(Z_SOLO) && !defined(_KERNEL) 272 return Z_STREAM_ERROR; 273 #else 274 strm->zalloc = zcalloc; 275 strm->opaque = (voidpf)0; 276 #endif 277 } 278 if (strm->zfree == (free_func)0) 279 #if defined(Z_SOLO) && !defined(_KERNEL) 280 return Z_STREAM_ERROR; 281 #else 282 strm->zfree = zcfree; 283 #endif 284 285 #ifdef FASTEST 286 if (level != 0) level = 1; 287 #else 288 if (level == Z_DEFAULT_COMPRESSION) level = 6; 289 #endif 290 291 if (windowBits < 0) { /* suppress zlib wrapper */ 292 wrap = 0; 293 windowBits = -windowBits; 294 } 295 #ifdef GZIP 296 else if (windowBits > 15) { 297 wrap = 2; /* write gzip wrapper instead */ 298 windowBits -= 16; 299 } 300 #endif 301 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || 302 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || 303 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) { 304 return Z_STREAM_ERROR; 305 } 306 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ 307 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); 308 if (s == Z_NULL) return Z_MEM_ERROR; 309 strm->state = (struct internal_state FAR *)s; 310 s->strm = strm; 311 s->status = INIT_STATE; /* to pass state test in deflateReset() */ 312 313 s->wrap = wrap; 314 s->gzhead = Z_NULL; 315 s->w_bits = (uInt)windowBits; 316 s->w_size = 1 << s->w_bits; 317 s->w_mask = s->w_size - 1; 318 319 s->hash_bits = (uInt)memLevel + 7; 320 s->hash_size = 1 << s->hash_bits; 321 s->hash_mask = s->hash_size - 1; 322 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); 323 324 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); 325 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); 326 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); 327 328 s->high_water = 0; /* nothing written to s->window yet */ 329 330 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ 331 332 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2); 333 s->pending_buf = (uchf *) overlay; 334 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); 335 336 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || 337 s->pending_buf == Z_NULL) { 338 s->status = FINISH_STATE; 339 strm->msg = ERR_MSG(Z_MEM_ERROR); 340 deflateEnd (strm); 341 return Z_MEM_ERROR; 342 } 343 s->d_buf = overlay + s->lit_bufsize/sizeof(ush); 344 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; 345 346 s->level = level; 347 s->strategy = strategy; 348 s->method = (Byte)method; 349 350 return deflateReset(strm); 351 } 352 353 /* ========================================================================= 354 * Check for a valid deflate stream state. Return 0 if ok, 1 if not. 355 */ 356 local int deflateStateCheck (strm) 357 z_streamp strm; 358 { 359 deflate_state *s; 360 if (strm == Z_NULL || 361 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) 362 return 1; 363 s = strm->state; 364 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE && 365 #ifdef GZIP 366 s->status != GZIP_STATE && 367 #endif 368 s->status != EXTRA_STATE && 369 s->status != NAME_STATE && 370 s->status != COMMENT_STATE && 371 s->status != HCRC_STATE && 372 s->status != BUSY_STATE && 373 s->status != FINISH_STATE)) 374 return 1; 375 return 0; 376 } 377 378 /* ========================================================================= */ 379 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength) 380 z_streamp strm; 381 const Bytef *dictionary; 382 uInt dictLength; 383 { 384 deflate_state *s; 385 uInt str, n; 386 int wrap; 387 unsigned avail; 388 z_const unsigned char *next; 389 390 if (deflateStateCheck(strm) || dictionary == Z_NULL) 391 return Z_STREAM_ERROR; 392 s = strm->state; 393 wrap = s->wrap; 394 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead) 395 return Z_STREAM_ERROR; 396 397 /* when using zlib wrappers, compute Adler-32 for provided dictionary */ 398 if (wrap == 1) 399 strm->adler = adler32(strm->adler, dictionary, dictLength); 400 s->wrap = 0; /* avoid computing Adler-32 in read_buf */ 401 402 /* if dictionary would fill window, just replace the history */ 403 if (dictLength >= s->w_size) { 404 if (wrap == 0) { /* already empty otherwise */ 405 CLEAR_HASH(s); 406 s->strstart = 0; 407 s->block_start = 0L; 408 s->insert = 0; 409 } 410 dictionary += dictLength - s->w_size; /* use the tail */ 411 dictLength = s->w_size; 412 } 413 414 /* insert dictionary into window and hash */ 415 avail = strm->avail_in; 416 next = strm->next_in; 417 strm->avail_in = dictLength; 418 strm->next_in = (z_const Bytef *)dictionary; 419 fill_window(s); 420 while (s->lookahead >= MIN_MATCH) { 421 str = s->strstart; 422 n = s->lookahead - (MIN_MATCH-1); 423 do { 424 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 425 #ifndef FASTEST 426 s->prev[str & s->w_mask] = s->head[s->ins_h]; 427 #endif 428 s->head[s->ins_h] = (Pos)str; 429 str++; 430 } while (--n); 431 s->strstart = str; 432 s->lookahead = MIN_MATCH-1; 433 fill_window(s); 434 } 435 s->strstart += s->lookahead; 436 s->block_start = (long)s->strstart; 437 s->insert = s->lookahead; 438 s->lookahead = 0; 439 s->match_length = s->prev_length = MIN_MATCH-1; 440 s->match_available = 0; 441 strm->next_in = next; 442 strm->avail_in = avail; 443 s->wrap = wrap; 444 return Z_OK; 445 } 446 447 /* ========================================================================= */ 448 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength) 449 z_streamp strm; 450 Bytef *dictionary; 451 uInt *dictLength; 452 { 453 deflate_state *s; 454 uInt len; 455 456 if (deflateStateCheck(strm)) 457 return Z_STREAM_ERROR; 458 s = strm->state; 459 len = s->strstart + s->lookahead; 460 if (len > s->w_size) 461 len = s->w_size; 462 if (dictionary != Z_NULL && len) 463 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len); 464 if (dictLength != Z_NULL) 465 *dictLength = len; 466 return Z_OK; 467 } 468 469 /* ========================================================================= */ 470 int ZEXPORT deflateResetKeep (strm) 471 z_streamp strm; 472 { 473 deflate_state *s; 474 475 if (deflateStateCheck(strm)) { 476 return Z_STREAM_ERROR; 477 } 478 479 strm->total_in = strm->total_out = 0; 480 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ 481 strm->data_type = Z_UNKNOWN; 482 483 s = (deflate_state *)strm->state; 484 s->pending = 0; 485 s->pending_out = s->pending_buf; 486 487 if (s->wrap < 0) { 488 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ 489 } 490 s->status = 491 #ifdef GZIP 492 s->wrap == 2 ? GZIP_STATE : 493 #endif 494 s->wrap ? INIT_STATE : BUSY_STATE; 495 strm->adler = 496 #ifdef GZIP 497 s->wrap == 2 ? crc32(0L, Z_NULL, 0) : 498 #endif 499 adler32(0L, Z_NULL, 0); 500 s->last_flush = -2; 501 502 _tr_init(s); 503 504 return Z_OK; 505 } 506 507 /* ========================================================================= */ 508 int ZEXPORT deflateReset (strm) 509 z_streamp strm; 510 { 511 int ret; 512 513 ret = deflateResetKeep(strm); 514 if (ret == Z_OK) 515 lm_init(strm->state); 516 return ret; 517 } 518 519 /* ========================================================================= */ 520 int ZEXPORT deflateSetHeader (strm, head) 521 z_streamp strm; 522 gz_headerp head; 523 { 524 if (deflateStateCheck(strm) || strm->state->wrap != 2) 525 return Z_STREAM_ERROR; 526 strm->state->gzhead = head; 527 return Z_OK; 528 } 529 530 /* ========================================================================= */ 531 int ZEXPORT deflatePending (strm, pending, bits) 532 unsigned *pending; 533 int *bits; 534 z_streamp strm; 535 { 536 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 537 if (pending != Z_NULL) 538 *pending = strm->state->pending; 539 if (bits != Z_NULL) 540 *bits = strm->state->bi_valid; 541 return Z_OK; 542 } 543 544 /* ========================================================================= */ 545 int ZEXPORT deflatePrime (strm, bits, value) 546 z_streamp strm; 547 int bits; 548 int value; 549 { 550 deflate_state *s; 551 int put; 552 553 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 554 s = strm->state; 555 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3)) 556 return Z_BUF_ERROR; 557 do { 558 put = Buf_size - s->bi_valid; 559 if (put > bits) 560 put = bits; 561 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid); 562 s->bi_valid += put; 563 _tr_flush_bits(s); 564 value >>= put; 565 bits -= put; 566 } while (bits); 567 return Z_OK; 568 } 569 570 /* ========================================================================= */ 571 int ZEXPORT deflateParams(strm, level, strategy) 572 z_streamp strm; 573 int level; 574 int strategy; 575 { 576 deflate_state *s; 577 compress_func func; 578 579 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 580 s = strm->state; 581 582 #ifdef FASTEST 583 if (level != 0) level = 1; 584 #else 585 if (level == Z_DEFAULT_COMPRESSION) level = 6; 586 #endif 587 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { 588 return Z_STREAM_ERROR; 589 } 590 func = configuration_table[s->level].func; 591 592 if ((strategy != s->strategy || func != configuration_table[level].func) && 593 s->last_flush != -2) { 594 /* Flush the last buffer: */ 595 int err = deflate(strm, Z_BLOCK); 596 if (err == Z_STREAM_ERROR) 597 return err; 598 if (strm->avail_in || (s->strstart - s->block_start) + s->lookahead) 599 return Z_BUF_ERROR; 600 } 601 if (s->level != level) { 602 if (s->level == 0 && s->matches != 0) { 603 if (s->matches == 1) 604 slide_hash(s); 605 else 606 CLEAR_HASH(s); 607 s->matches = 0; 608 } 609 s->level = level; 610 s->max_lazy_match = configuration_table[level].max_lazy; 611 s->good_match = configuration_table[level].good_length; 612 s->nice_match = configuration_table[level].nice_length; 613 s->max_chain_length = configuration_table[level].max_chain; 614 } 615 s->strategy = strategy; 616 return Z_OK; 617 } 618 619 /* ========================================================================= */ 620 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain) 621 z_streamp strm; 622 int good_length; 623 int max_lazy; 624 int nice_length; 625 int max_chain; 626 { 627 deflate_state *s; 628 629 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 630 s = strm->state; 631 s->good_match = (uInt)good_length; 632 s->max_lazy_match = (uInt)max_lazy; 633 s->nice_match = nice_length; 634 s->max_chain_length = (uInt)max_chain; 635 return Z_OK; 636 } 637 638 /* ========================================================================= 639 * For the default windowBits of 15 and memLevel of 8, this function returns 640 * a close to exact, as well as small, upper bound on the compressed size. 641 * They are coded as constants here for a reason--if the #define's are 642 * changed, then this function needs to be changed as well. The return 643 * value for 15 and 8 only works for those exact settings. 644 * 645 * For any setting other than those defaults for windowBits and memLevel, 646 * the value returned is a conservative worst case for the maximum expansion 647 * resulting from using fixed blocks instead of stored blocks, which deflate 648 * can emit on compressed data for some combinations of the parameters. 649 * 650 * This function could be more sophisticated to provide closer upper bounds for 651 * every combination of windowBits and memLevel. But even the conservative 652 * upper bound of about 14% expansion does not seem onerous for output buffer 653 * allocation. 654 */ 655 uLong ZEXPORT deflateBound(strm, sourceLen) 656 z_streamp strm; 657 uLong sourceLen; 658 { 659 deflate_state *s; 660 uLong complen, wraplen; 661 662 /* conservative upper bound for compressed data */ 663 complen = sourceLen + 664 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5; 665 666 /* if can't get parameters, return conservative bound plus zlib wrapper */ 667 if (deflateStateCheck(strm)) 668 return complen + 6; 669 670 /* compute wrapper length */ 671 s = strm->state; 672 switch (s->wrap) { 673 case 0: /* raw deflate */ 674 wraplen = 0; 675 break; 676 case 1: /* zlib wrapper */ 677 wraplen = 6 + (s->strstart ? 4 : 0); 678 break; 679 #ifdef GZIP 680 case 2: /* gzip wrapper */ 681 wraplen = 18; 682 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */ 683 Bytef *str; 684 if (s->gzhead->extra != Z_NULL) 685 wraplen += 2 + s->gzhead->extra_len; 686 str = s->gzhead->name; 687 if (str != Z_NULL) 688 do { 689 wraplen++; 690 } while (*str++); 691 str = s->gzhead->comment; 692 if (str != Z_NULL) 693 do { 694 wraplen++; 695 } while (*str++); 696 if (s->gzhead->hcrc) 697 wraplen += 2; 698 } 699 break; 700 #endif 701 default: /* for compiler happiness */ 702 wraplen = 6; 703 } 704 705 /* if not default parameters, return conservative bound */ 706 if (s->w_bits != 15 || s->hash_bits != 8 + 7) 707 return complen + wraplen; 708 709 /* default settings: return tight bound for that case */ 710 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) + 711 (sourceLen >> 25) + 13 - 6 + wraplen; 712 } 713 714 /* ========================================================================= 715 * Put a short in the pending buffer. The 16-bit value is put in MSB order. 716 * IN assertion: the stream state is correct and there is enough room in 717 * pending_buf. 718 */ 719 local void putShortMSB (s, b) 720 deflate_state *s; 721 uInt b; 722 { 723 put_byte(s, (Byte)(b >> 8)); 724 put_byte(s, (Byte)(b & 0xff)); 725 } 726 727 /* ========================================================================= 728 * Flush as much pending output as possible. All deflate() output, except for 729 * some deflate_stored() output, goes through this function so some 730 * applications may wish to modify it to avoid allocating a large 731 * strm->next_out buffer and copying into it. (See also read_buf()). 732 */ 733 local void flush_pending(strm) 734 z_streamp strm; 735 { 736 unsigned len; 737 deflate_state *s = strm->state; 738 739 _tr_flush_bits(s); 740 len = s->pending; 741 if (len > strm->avail_out) len = strm->avail_out; 742 if (len == 0) return; 743 744 zmemcpy(strm->next_out, s->pending_out, len); 745 strm->next_out += len; 746 s->pending_out += len; 747 strm->total_out += len; 748 strm->avail_out -= len; 749 s->pending -= len; 750 if (s->pending == 0) { 751 s->pending_out = s->pending_buf; 752 } 753 } 754 755 /* =========================================================================== 756 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1]. 757 */ 758 #define HCRC_UPDATE(beg) \ 759 do { \ 760 if (s->gzhead->hcrc && s->pending > (beg)) \ 761 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \ 762 s->pending - (beg)); \ 763 } while (0) 764 765 /* ========================================================================= */ 766 int ZEXPORT deflate (strm, flush) 767 z_streamp strm; 768 int flush; 769 { 770 int old_flush; /* value of flush param for previous deflate call */ 771 deflate_state *s; 772 773 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) { 774 return Z_STREAM_ERROR; 775 } 776 s = strm->state; 777 778 if (strm->next_out == Z_NULL || 779 (strm->avail_in != 0 && strm->next_in == Z_NULL) || 780 (s->status == FINISH_STATE && flush != Z_FINISH)) { 781 ERR_RETURN(strm, Z_STREAM_ERROR); 782 } 783 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); 784 785 old_flush = s->last_flush; 786 s->last_flush = flush; 787 788 /* Flush as much pending output as possible */ 789 if (s->pending != 0) { 790 flush_pending(strm); 791 if (strm->avail_out == 0) { 792 /* Since avail_out is 0, deflate will be called again with 793 * more output space, but possibly with both pending and 794 * avail_in equal to zero. There won't be anything to do, 795 * but this is not an error situation so make sure we 796 * return OK instead of BUF_ERROR at next call of deflate: 797 */ 798 s->last_flush = -1; 799 return Z_OK; 800 } 801 802 /* Make sure there is something to do and avoid duplicate consecutive 803 * flushes. For repeated and useless calls with Z_FINISH, we keep 804 * returning Z_STREAM_END instead of Z_BUF_ERROR. 805 */ 806 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) && 807 flush != Z_FINISH) { 808 ERR_RETURN(strm, Z_BUF_ERROR); 809 } 810 811 /* User must not provide more input after the first FINISH: */ 812 if (s->status == FINISH_STATE && strm->avail_in != 0) { 813 ERR_RETURN(strm, Z_BUF_ERROR); 814 } 815 816 /* Write the header */ 817 if (s->status == INIT_STATE) { 818 /* zlib header */ 819 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; 820 uInt level_flags; 821 822 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) 823 level_flags = 0; 824 else if (s->level < 6) 825 level_flags = 1; 826 else if (s->level == 6) 827 level_flags = 2; 828 else 829 level_flags = 3; 830 header |= (level_flags << 6); 831 if (s->strstart != 0) header |= PRESET_DICT; 832 header += 31 - (header % 31); 833 834 putShortMSB(s, header); 835 836 /* Save the adler32 of the preset dictionary: */ 837 if (s->strstart != 0) { 838 putShortMSB(s, (uInt)(strm->adler >> 16)); 839 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 840 } 841 strm->adler = adler32(0L, Z_NULL, 0); 842 s->status = BUSY_STATE; 843 844 /* Compression must start with an empty pending buffer */ 845 flush_pending(strm); 846 if (s->pending != 0) { 847 s->last_flush = -1; 848 return Z_OK; 849 } 850 } 851 #ifdef GZIP 852 if (s->status == GZIP_STATE) { 853 /* gzip header */ 854 strm->adler = crc32(0L, Z_NULL, 0); 855 put_byte(s, 31); 856 put_byte(s, 139); 857 put_byte(s, 8); 858 if (s->gzhead == Z_NULL) { 859 put_byte(s, 0); 860 put_byte(s, 0); 861 put_byte(s, 0); 862 put_byte(s, 0); 863 put_byte(s, 0); 864 put_byte(s, s->level == 9 ? 2 : 865 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 866 4 : 0)); 867 put_byte(s, OS_CODE); 868 s->status = BUSY_STATE; 869 870 /* Compression must start with an empty pending buffer */ 871 flush_pending(strm); 872 if (s->pending != 0) { 873 s->last_flush = -1; 874 return Z_OK; 875 } 876 } 877 else { 878 put_byte(s, (s->gzhead->text ? 1 : 0) + 879 (s->gzhead->hcrc ? 2 : 0) + 880 (s->gzhead->extra == Z_NULL ? 0 : 4) + 881 (s->gzhead->name == Z_NULL ? 0 : 8) + 882 (s->gzhead->comment == Z_NULL ? 0 : 16) 883 ); 884 put_byte(s, (Byte)(s->gzhead->time & 0xff)); 885 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); 886 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); 887 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); 888 put_byte(s, s->level == 9 ? 2 : 889 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 890 4 : 0)); 891 put_byte(s, s->gzhead->os & 0xff); 892 if (s->gzhead->extra != Z_NULL) { 893 put_byte(s, s->gzhead->extra_len & 0xff); 894 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); 895 } 896 if (s->gzhead->hcrc) 897 strm->adler = crc32(strm->adler, s->pending_buf, 898 s->pending); 899 s->gzindex = 0; 900 s->status = EXTRA_STATE; 901 } 902 } 903 if (s->status == EXTRA_STATE) { 904 if (s->gzhead->extra != Z_NULL) { 905 ulg beg = s->pending; /* start of bytes to update crc */ 906 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex; 907 while (s->pending + left > s->pending_buf_size) { 908 uInt copy = s->pending_buf_size - s->pending; 909 zmemcpy(s->pending_buf + s->pending, 910 s->gzhead->extra + s->gzindex, copy); 911 s->pending = s->pending_buf_size; 912 HCRC_UPDATE(beg); 913 s->gzindex += copy; 914 flush_pending(strm); 915 if (s->pending != 0) { 916 s->last_flush = -1; 917 return Z_OK; 918 } 919 beg = 0; 920 left -= copy; 921 } 922 zmemcpy(s->pending_buf + s->pending, 923 s->gzhead->extra + s->gzindex, left); 924 s->pending += left; 925 HCRC_UPDATE(beg); 926 s->gzindex = 0; 927 } 928 s->status = NAME_STATE; 929 } 930 if (s->status == NAME_STATE) { 931 if (s->gzhead->name != Z_NULL) { 932 ulg beg = s->pending; /* start of bytes to update crc */ 933 int val; 934 do { 935 if (s->pending == s->pending_buf_size) { 936 HCRC_UPDATE(beg); 937 flush_pending(strm); 938 if (s->pending != 0) { 939 s->last_flush = -1; 940 return Z_OK; 941 } 942 beg = 0; 943 } 944 val = s->gzhead->name[s->gzindex++]; 945 put_byte(s, val); 946 } while (val != 0); 947 HCRC_UPDATE(beg); 948 s->gzindex = 0; 949 } 950 s->status = COMMENT_STATE; 951 } 952 if (s->status == COMMENT_STATE) { 953 if (s->gzhead->comment != Z_NULL) { 954 ulg beg = s->pending; /* start of bytes to update crc */ 955 int val; 956 do { 957 if (s->pending == s->pending_buf_size) { 958 HCRC_UPDATE(beg); 959 flush_pending(strm); 960 if (s->pending != 0) { 961 s->last_flush = -1; 962 return Z_OK; 963 } 964 beg = 0; 965 } 966 val = s->gzhead->comment[s->gzindex++]; 967 put_byte(s, val); 968 } while (val != 0); 969 HCRC_UPDATE(beg); 970 } 971 s->status = HCRC_STATE; 972 } 973 if (s->status == HCRC_STATE) { 974 if (s->gzhead->hcrc) { 975 if (s->pending + 2 > s->pending_buf_size) { 976 flush_pending(strm); 977 if (s->pending != 0) { 978 s->last_flush = -1; 979 return Z_OK; 980 } 981 } 982 put_byte(s, (Byte)(strm->adler & 0xff)); 983 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 984 strm->adler = crc32(0L, Z_NULL, 0); 985 } 986 s->status = BUSY_STATE; 987 988 /* Compression must start with an empty pending buffer */ 989 flush_pending(strm); 990 if (s->pending != 0) { 991 s->last_flush = -1; 992 return Z_OK; 993 } 994 } 995 #endif 996 997 /* Start a new block or continue the current one. 998 */ 999 if (strm->avail_in != 0 || s->lookahead != 0 || 1000 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { 1001 block_state bstate; 1002 1003 bstate = s->level == 0 ? deflate_stored(s, flush) : 1004 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : 1005 s->strategy == Z_RLE ? deflate_rle(s, flush) : 1006 (*(configuration_table[s->level].func))(s, flush); 1007 1008 if (bstate == finish_started || bstate == finish_done) { 1009 s->status = FINISH_STATE; 1010 } 1011 if (bstate == need_more || bstate == finish_started) { 1012 if (strm->avail_out == 0) { 1013 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ 1014 } 1015 return Z_OK; 1016 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call 1017 * of deflate should use the same flush parameter to make sure 1018 * that the flush is complete. So we don't have to output an 1019 * empty block here, this will be done at next call. This also 1020 * ensures that for a very small output buffer, we emit at most 1021 * one empty block. 1022 */ 1023 } 1024 if (bstate == block_done) { 1025 if (flush == Z_PARTIAL_FLUSH) { 1026 _tr_align(s); 1027 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */ 1028 _tr_stored_block(s, (char*)0, 0L, 0); 1029 /* For a full flush, this empty block will be recognized 1030 * as a special marker by inflate_sync(). 1031 */ 1032 if (flush == Z_FULL_FLUSH) { 1033 CLEAR_HASH(s); /* forget history */ 1034 if (s->lookahead == 0) { 1035 s->strstart = 0; 1036 s->block_start = 0L; 1037 s->insert = 0; 1038 } 1039 } 1040 } 1041 flush_pending(strm); 1042 if (strm->avail_out == 0) { 1043 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ 1044 return Z_OK; 1045 } 1046 } 1047 } 1048 1049 if (flush != Z_FINISH) return Z_OK; 1050 if (s->wrap <= 0) return Z_STREAM_END; 1051 1052 /* Write the trailer */ 1053 #ifdef GZIP 1054 if (s->wrap == 2) { 1055 put_byte(s, (Byte)(strm->adler & 0xff)); 1056 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 1057 put_byte(s, (Byte)((strm->adler >> 16) & 0xff)); 1058 put_byte(s, (Byte)((strm->adler >> 24) & 0xff)); 1059 put_byte(s, (Byte)(strm->total_in & 0xff)); 1060 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff)); 1061 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff)); 1062 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff)); 1063 } 1064 else 1065 #endif 1066 { 1067 putShortMSB(s, (uInt)(strm->adler >> 16)); 1068 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 1069 } 1070 flush_pending(strm); 1071 /* If avail_out is zero, the application will call deflate again 1072 * to flush the rest. 1073 */ 1074 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ 1075 return s->pending != 0 ? Z_OK : Z_STREAM_END; 1076 } 1077 1078 /* ========================================================================= */ 1079 int ZEXPORT deflateEnd (strm) 1080 z_streamp strm; 1081 { 1082 int status; 1083 1084 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 1085 1086 status = strm->state->status; 1087 1088 /* Deallocate in reverse order of allocations: */ 1089 TRY_FREE(strm, strm->state->pending_buf); 1090 TRY_FREE(strm, strm->state->head); 1091 TRY_FREE(strm, strm->state->prev); 1092 TRY_FREE(strm, strm->state->window); 1093 1094 ZFREE(strm, strm->state); 1095 strm->state = Z_NULL; 1096 1097 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; 1098 } 1099 1100 /* ========================================================================= 1101 * Copy the source state to the destination state. 1102 * To simplify the source, this is not supported for 16-bit MSDOS (which 1103 * doesn't have enough memory anyway to duplicate compression states). 1104 */ 1105 int ZEXPORT deflateCopy (dest, source) 1106 z_streamp dest; 1107 z_streamp source; 1108 { 1109 #ifdef MAXSEG_64K 1110 return Z_STREAM_ERROR; 1111 #else 1112 deflate_state *ds; 1113 deflate_state *ss; 1114 ushf *overlay; 1115 1116 1117 if (deflateStateCheck(source) || dest == Z_NULL) { 1118 return Z_STREAM_ERROR; 1119 } 1120 1121 ss = source->state; 1122 1123 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream)); 1124 1125 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); 1126 if (ds == Z_NULL) return Z_MEM_ERROR; 1127 dest->state = (struct internal_state FAR *) ds; 1128 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state)); 1129 ds->strm = dest; 1130 1131 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); 1132 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); 1133 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); 1134 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2); 1135 ds->pending_buf = (uchf *) overlay; 1136 1137 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || 1138 ds->pending_buf == Z_NULL) { 1139 deflateEnd (dest); 1140 return Z_MEM_ERROR; 1141 } 1142 /* following zmemcpy do not work for 16-bit MSDOS */ 1143 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); 1144 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos)); 1145 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos)); 1146 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); 1147 1148 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); 1149 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); 1150 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; 1151 1152 ds->l_desc.dyn_tree = ds->dyn_ltree; 1153 ds->d_desc.dyn_tree = ds->dyn_dtree; 1154 ds->bl_desc.dyn_tree = ds->bl_tree; 1155 1156 return Z_OK; 1157 #endif /* MAXSEG_64K */ 1158 } 1159 1160 /* =========================================================================== 1161 * Read a new buffer from the current input stream, update the adler32 1162 * and total number of bytes read. All deflate() input goes through 1163 * this function so some applications may wish to modify it to avoid 1164 * allocating a large strm->next_in buffer and copying from it. 1165 * (See also flush_pending()). 1166 */ 1167 local unsigned read_buf(strm, buf, size) 1168 z_streamp strm; 1169 Bytef *buf; 1170 unsigned size; 1171 { 1172 unsigned len = strm->avail_in; 1173 1174 if (len > size) len = size; 1175 if (len == 0) return 0; 1176 1177 strm->avail_in -= len; 1178 1179 zmemcpy(buf, strm->next_in, len); 1180 if (strm->state->wrap == 1) { 1181 strm->adler = adler32(strm->adler, buf, len); 1182 } 1183 #ifdef GZIP 1184 else if (strm->state->wrap == 2) { 1185 strm->adler = crc32(strm->adler, buf, len); 1186 } 1187 #endif 1188 strm->next_in += len; 1189 strm->total_in += len; 1190 1191 return len; 1192 } 1193 1194 /* =========================================================================== 1195 * Initialize the "longest match" routines for a new zlib stream 1196 */ 1197 local void lm_init (s) 1198 deflate_state *s; 1199 { 1200 s->window_size = (ulg)2L*s->w_size; 1201 1202 CLEAR_HASH(s); 1203 1204 /* Set the default configuration parameters: 1205 */ 1206 s->max_lazy_match = configuration_table[s->level].max_lazy; 1207 s->good_match = configuration_table[s->level].good_length; 1208 s->nice_match = configuration_table[s->level].nice_length; 1209 s->max_chain_length = configuration_table[s->level].max_chain; 1210 1211 s->strstart = 0; 1212 s->block_start = 0L; 1213 s->lookahead = 0; 1214 s->insert = 0; 1215 s->match_length = s->prev_length = MIN_MATCH-1; 1216 s->match_available = 0; 1217 s->ins_h = 0; 1218 #ifndef FASTEST 1219 #ifdef ASMV 1220 match_init(); /* initialize the asm code */ 1221 #endif 1222 #endif 1223 } 1224 1225 #ifndef FASTEST 1226 /* =========================================================================== 1227 * Set match_start to the longest match starting at the given string and 1228 * return its length. Matches shorter or equal to prev_length are discarded, 1229 * in which case the result is equal to prev_length and match_start is 1230 * garbage. 1231 * IN assertions: cur_match is the head of the hash chain for the current 1232 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 1233 * OUT assertion: the match length is not greater than s->lookahead. 1234 */ 1235 #ifndef ASMV 1236 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or 1237 * match.S. The code will be functionally equivalent. 1238 */ 1239 local uInt longest_match(s, cur_match) 1240 deflate_state *s; 1241 IPos cur_match; /* current match */ 1242 { 1243 unsigned chain_length = s->max_chain_length;/* max hash chain length */ 1244 register Bytef *scan = s->window + s->strstart; /* current string */ 1245 register Bytef *match; /* matched string */ 1246 register int len; /* length of current match */ 1247 int best_len = (int)s->prev_length; /* best match length so far */ 1248 int nice_match = s->nice_match; /* stop if match long enough */ 1249 IPos limit = s->strstart > (IPos)MAX_DIST(s) ? 1250 s->strstart - (IPos)MAX_DIST(s) : NIL; 1251 /* Stop when cur_match becomes <= limit. To simplify the code, 1252 * we prevent matches with the string of window index 0. 1253 */ 1254 Posf *prev = s->prev; 1255 uInt wmask = s->w_mask; 1256 1257 #ifdef UNALIGNED_OK 1258 /* Compare two bytes at a time. Note: this is not always beneficial. 1259 * Try with and without -DUNALIGNED_OK to check. 1260 */ 1261 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; 1262 register ush scan_start = *(ushf*)scan; 1263 register ush scan_end = *(ushf*)(scan+best_len-1); 1264 #else 1265 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1266 register Byte scan_end1 = scan[best_len-1]; 1267 register Byte scan_end = scan[best_len]; 1268 #endif 1269 1270 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1271 * It is easy to get rid of this optimization if necessary. 1272 */ 1273 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1274 1275 /* Do not waste too much time if we already have a good match: */ 1276 if (s->prev_length >= s->good_match) { 1277 chain_length >>= 2; 1278 } 1279 /* Do not look for matches beyond the end of the input. This is necessary 1280 * to make deflate deterministic. 1281 */ 1282 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead; 1283 1284 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); 1285 1286 do { 1287 Assert(cur_match < s->strstart, "no future"); 1288 match = s->window + cur_match; 1289 1290 /* Skip to next match if the match length cannot increase 1291 * or if the match length is less than 2. Note that the checks below 1292 * for insufficient lookahead only occur occasionally for performance 1293 * reasons. Therefore uninitialized memory will be accessed, and 1294 * conditional jumps will be made that depend on those values. 1295 * However the length of the match is limited to the lookahead, so 1296 * the output of deflate is not affected by the uninitialized values. 1297 */ 1298 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) 1299 /* This code assumes sizeof(unsigned short) == 2. Do not use 1300 * UNALIGNED_OK if your compiler uses a different size. 1301 */ 1302 if (*(ushf*)(match+best_len-1) != scan_end || 1303 *(ushf*)match != scan_start) continue; 1304 1305 /* It is not necessary to compare scan[2] and match[2] since they are 1306 * always equal when the other bytes match, given that the hash keys 1307 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at 1308 * strstart+3, +5, ... up to strstart+257. We check for insufficient 1309 * lookahead only every 4th comparison; the 128th check will be made 1310 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is 1311 * necessary to put more guard bytes at the end of the window, or 1312 * to check more often for insufficient lookahead. 1313 */ 1314 Assert(scan[2] == match[2], "scan[2]?"); 1315 scan++, match++; 1316 do { 1317 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1318 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1319 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1320 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1321 scan < strend); 1322 /* The funny "do {}" generates better code on most compilers */ 1323 1324 /* Here, scan <= window+strstart+257 */ 1325 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1326 if (*scan == *match) scan++; 1327 1328 len = (MAX_MATCH - 1) - (int)(strend-scan); 1329 scan = strend - (MAX_MATCH-1); 1330 1331 #else /* UNALIGNED_OK */ 1332 1333 if (match[best_len] != scan_end || 1334 match[best_len-1] != scan_end1 || 1335 *match != *scan || 1336 *++match != scan[1]) continue; 1337 1338 /* The check at best_len-1 can be removed because it will be made 1339 * again later. (This heuristic is not always a win.) 1340 * It is not necessary to compare scan[2] and match[2] since they 1341 * are always equal when the other bytes match, given that 1342 * the hash keys are equal and that HASH_BITS >= 8. 1343 */ 1344 scan += 2, match++; 1345 Assert(*scan == *match, "match[2]?"); 1346 1347 /* We check for insufficient lookahead only every 8th comparison; 1348 * the 256th check will be made at strstart+258. 1349 */ 1350 do { 1351 } while (*++scan == *++match && *++scan == *++match && 1352 *++scan == *++match && *++scan == *++match && 1353 *++scan == *++match && *++scan == *++match && 1354 *++scan == *++match && *++scan == *++match && 1355 scan < strend); 1356 1357 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1358 1359 len = MAX_MATCH - (int)(strend - scan); 1360 scan = strend - MAX_MATCH; 1361 1362 #endif /* UNALIGNED_OK */ 1363 1364 if (len > best_len) { 1365 s->match_start = cur_match; 1366 best_len = len; 1367 if (len >= nice_match) break; 1368 #ifdef UNALIGNED_OK 1369 scan_end = *(ushf*)(scan+best_len-1); 1370 #else 1371 scan_end1 = scan[best_len-1]; 1372 scan_end = scan[best_len]; 1373 #endif 1374 } 1375 } while ((cur_match = prev[cur_match & wmask]) > limit 1376 && --chain_length != 0); 1377 1378 if ((uInt)best_len <= s->lookahead) return (uInt)best_len; 1379 return s->lookahead; 1380 } 1381 #endif /* ASMV */ 1382 1383 #else /* FASTEST */ 1384 1385 /* --------------------------------------------------------------------------- 1386 * Optimized version for FASTEST only 1387 */ 1388 local uInt longest_match(s, cur_match) 1389 deflate_state *s; 1390 IPos cur_match; /* current match */ 1391 { 1392 register Bytef *scan = s->window + s->strstart; /* current string */ 1393 register Bytef *match; /* matched string */ 1394 register int len; /* length of current match */ 1395 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1396 1397 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1398 * It is easy to get rid of this optimization if necessary. 1399 */ 1400 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1401 1402 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); 1403 1404 Assert(cur_match < s->strstart, "no future"); 1405 1406 match = s->window + cur_match; 1407 1408 /* Return failure if the match length is less than 2: 1409 */ 1410 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; 1411 1412 /* The check at best_len-1 can be removed because it will be made 1413 * again later. (This heuristic is not always a win.) 1414 * It is not necessary to compare scan[2] and match[2] since they 1415 * are always equal when the other bytes match, given that 1416 * the hash keys are equal and that HASH_BITS >= 8. 1417 */ 1418 scan += 2, match += 2; 1419 Assert(*scan == *match, "match[2]?"); 1420 1421 /* We check for insufficient lookahead only every 8th comparison; 1422 * the 256th check will be made at strstart+258. 1423 */ 1424 do { 1425 } while (*++scan == *++match && *++scan == *++match && 1426 *++scan == *++match && *++scan == *++match && 1427 *++scan == *++match && *++scan == *++match && 1428 *++scan == *++match && *++scan == *++match && 1429 scan < strend); 1430 1431 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1432 1433 len = MAX_MATCH - (int)(strend - scan); 1434 1435 if (len < MIN_MATCH) return MIN_MATCH - 1; 1436 1437 s->match_start = cur_match; 1438 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead; 1439 } 1440 1441 #endif /* FASTEST */ 1442 1443 #ifdef ZLIB_DEBUG 1444 1445 #define EQUAL 0 1446 /* result of memcmp for equal strings */ 1447 1448 /* =========================================================================== 1449 * Check that the match at match_start is indeed a match. 1450 */ 1451 local void check_match(s, start, match, length) 1452 deflate_state *s; 1453 IPos start, match; 1454 int length; 1455 { 1456 /* check that the match is indeed a match */ 1457 if (zmemcmp(s->window + match, 1458 s->window + start, length) != EQUAL) { 1459 fprintf(stderr, " start %u, match %u, length %d\n", 1460 start, match, length); 1461 do { 1462 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); 1463 } while (--length != 0); 1464 z_error("invalid match"); 1465 } 1466 if (z_verbose > 1) { 1467 fprintf(stderr,"\\[%d,%d]", start-match, length); 1468 do { putc(s->window[start++], stderr); } while (--length != 0); 1469 } 1470 } 1471 #else 1472 # define check_match(s, start, match, length) 1473 #endif /* ZLIB_DEBUG */ 1474 1475 /* =========================================================================== 1476 * Fill the window when the lookahead becomes insufficient. 1477 * Updates strstart and lookahead. 1478 * 1479 * IN assertion: lookahead < MIN_LOOKAHEAD 1480 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD 1481 * At least one byte has been read, or avail_in == 0; reads are 1482 * performed for at least two bytes (required for the zip translate_eol 1483 * option -- not supported here). 1484 */ 1485 local void fill_window(s) 1486 deflate_state *s; 1487 { 1488 unsigned n; 1489 unsigned more; /* Amount of free space at the end of the window. */ 1490 uInt wsize = s->w_size; 1491 1492 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead"); 1493 1494 do { 1495 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); 1496 1497 /* Deal with !@#$% 64K limit: */ 1498 if (sizeof(int) <= 2) { 1499 if (more == 0 && s->strstart == 0 && s->lookahead == 0) { 1500 more = wsize; 1501 1502 } else if (more == (unsigned)(-1)) { 1503 /* Very unlikely, but possible on 16 bit machine if 1504 * strstart == 0 && lookahead == 1 (input done a byte at time) 1505 */ 1506 more--; 1507 } 1508 } 1509 1510 /* If the window is almost full and there is insufficient lookahead, 1511 * move the upper half to the lower one to make room in the upper half. 1512 */ 1513 if (s->strstart >= wsize+MAX_DIST(s)) { 1514 1515 zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more); 1516 s->match_start -= wsize; 1517 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ 1518 s->block_start -= (long) wsize; 1519 slide_hash(s); 1520 more += wsize; 1521 } 1522 if (s->strm->avail_in == 0) break; 1523 1524 /* If there was no sliding: 1525 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && 1526 * more == window_size - lookahead - strstart 1527 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) 1528 * => more >= window_size - 2*WSIZE + 2 1529 * In the BIG_MEM or MMAP case (not yet supported), 1530 * window_size == input_size + MIN_LOOKAHEAD && 1531 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. 1532 * Otherwise, window_size == 2*WSIZE so more >= 2. 1533 * If there was sliding, more >= WSIZE. So in all cases, more >= 2. 1534 */ 1535 Assert(more >= 2, "more < 2"); 1536 1537 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); 1538 s->lookahead += n; 1539 1540 /* Initialize the hash value now that we have some input: */ 1541 if (s->lookahead + s->insert >= MIN_MATCH) { 1542 uInt str = s->strstart - s->insert; 1543 s->ins_h = s->window[str]; 1544 UPDATE_HASH(s, s->ins_h, s->window[str + 1]); 1545 #if MIN_MATCH != 3 1546 Call UPDATE_HASH() MIN_MATCH-3 more times 1547 #endif 1548 while (s->insert) { 1549 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 1550 #ifndef FASTEST 1551 s->prev[str & s->w_mask] = s->head[s->ins_h]; 1552 #endif 1553 s->head[s->ins_h] = (Pos)str; 1554 str++; 1555 s->insert--; 1556 if (s->lookahead + s->insert < MIN_MATCH) 1557 break; 1558 } 1559 } 1560 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, 1561 * but this is not important since only literal bytes will be emitted. 1562 */ 1563 1564 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); 1565 1566 /* If the WIN_INIT bytes after the end of the current data have never been 1567 * written, then zero those bytes in order to avoid memory check reports of 1568 * the use of uninitialized (or uninitialised as Julian writes) bytes by 1569 * the longest match routines. Update the high water mark for the next 1570 * time through here. WIN_INIT is set to MAX_MATCH since the longest match 1571 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead. 1572 */ 1573 if (s->high_water < s->window_size) { 1574 ulg curr = s->strstart + (ulg)(s->lookahead); 1575 ulg init; 1576 1577 if (s->high_water < curr) { 1578 /* Previous high water mark below current data -- zero WIN_INIT 1579 * bytes or up to end of window, whichever is less. 1580 */ 1581 init = s->window_size - curr; 1582 if (init > WIN_INIT) 1583 init = WIN_INIT; 1584 zmemzero(s->window + curr, (unsigned)init); 1585 s->high_water = curr + init; 1586 } 1587 else if (s->high_water < (ulg)curr + WIN_INIT) { 1588 /* High water mark at or above current data, but below current data 1589 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up 1590 * to end of window, whichever is less. 1591 */ 1592 init = (ulg)curr + WIN_INIT - s->high_water; 1593 if (init > s->window_size - s->high_water) 1594 init = s->window_size - s->high_water; 1595 zmemzero(s->window + s->high_water, (unsigned)init); 1596 s->high_water += init; 1597 } 1598 } 1599 1600 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 1601 "not enough room for search"); 1602 } 1603 1604 /* =========================================================================== 1605 * Flush the current block, with given end-of-file flag. 1606 * IN assertion: strstart is set to the end of the current match. 1607 */ 1608 #define FLUSH_BLOCK_ONLY(s, last) { \ 1609 _tr_flush_block(s, (s->block_start >= 0L ? \ 1610 (charf *)&s->window[(unsigned)s->block_start] : \ 1611 (charf *)Z_NULL), \ 1612 (ulg)((long)s->strstart - s->block_start), \ 1613 (last)); \ 1614 s->block_start = s->strstart; \ 1615 flush_pending(s->strm); \ 1616 Tracev((stderr,"[FLUSH]")); \ 1617 } 1618 1619 /* Same but force premature exit if necessary. */ 1620 #define FLUSH_BLOCK(s, last) { \ 1621 FLUSH_BLOCK_ONLY(s, last); \ 1622 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \ 1623 } 1624 1625 /* Maximum stored block length in deflate format (not including header). */ 1626 #define MAX_STORED 65535 1627 1628 #if !defined(MIN) 1629 /* Minimum of a and b. */ 1630 #define MIN(a, b) ((a) > (b) ? (b) : (a)) 1631 #endif 1632 1633 /* =========================================================================== 1634 * Copy without compression as much as possible from the input stream, return 1635 * the current block state. 1636 * 1637 * In case deflateParams() is used to later switch to a non-zero compression 1638 * level, s->matches (otherwise unused when storing) keeps track of the number 1639 * of hash table slides to perform. If s->matches is 1, then one hash table 1640 * slide will be done when switching. If s->matches is 2, the maximum value 1641 * allowed here, then the hash table will be cleared, since two or more slides 1642 * is the same as a clear. 1643 * 1644 * deflate_stored() is written to minimize the number of times an input byte is 1645 * copied. It is most efficient with large input and output buffers, which 1646 * maximizes the opportunites to have a single copy from next_in to next_out. 1647 */ 1648 local block_state deflate_stored(s, flush) 1649 deflate_state *s; 1650 int flush; 1651 { 1652 /* Smallest worthy block size when not flushing or finishing. By default 1653 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For 1654 * large input and output buffers, the stored block size will be larger. 1655 */ 1656 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size); 1657 1658 /* Copy as many min_block or larger stored blocks directly to next_out as 1659 * possible. If flushing, copy the remaining available input to next_out as 1660 * stored blocks, if there is enough space. 1661 */ 1662 unsigned len, left, have, last = 0; 1663 unsigned used = s->strm->avail_in; 1664 do { 1665 /* Set len to the maximum size block that we can copy directly with the 1666 * available input data and output space. Set left to how much of that 1667 * would be copied from what's left in the window. 1668 */ 1669 len = MAX_STORED; /* maximum deflate stored block length */ 1670 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1671 if (s->strm->avail_out < have) /* need room for header */ 1672 break; 1673 /* maximum stored block length that will fit in avail_out: */ 1674 have = s->strm->avail_out - have; 1675 left = s->strstart - s->block_start; /* bytes left in window */ 1676 if (len > (ulg)left + s->strm->avail_in) 1677 len = left + s->strm->avail_in; /* limit len to the input */ 1678 if (len > have) 1679 len = have; /* limit len to the output */ 1680 1681 /* If the stored block would be less than min_block in length, or if 1682 * unable to copy all of the available input when flushing, then try 1683 * copying to the window and the pending buffer instead. Also don't 1684 * write an empty block when flushing -- deflate() does that. 1685 */ 1686 if (len < min_block && ((len == 0 && flush != Z_FINISH) || 1687 flush == Z_NO_FLUSH || 1688 len != left + s->strm->avail_in)) 1689 break; 1690 1691 /* Make a dummy stored block in pending to get the header bytes, 1692 * including any pending bits. This also updates the debugging counts. 1693 */ 1694 last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0; 1695 _tr_stored_block(s, (char *)0, 0L, last); 1696 1697 /* Replace the lengths in the dummy stored block with len. */ 1698 s->pending_buf[s->pending - 4] = len; 1699 s->pending_buf[s->pending - 3] = len >> 8; 1700 s->pending_buf[s->pending - 2] = ~len; 1701 s->pending_buf[s->pending - 1] = ~len >> 8; 1702 1703 /* Write the stored block header bytes. */ 1704 flush_pending(s->strm); 1705 1706 #ifdef ZLIB_DEBUG 1707 /* Update debugging counts for the data about to be copied. */ 1708 s->compressed_len += len << 3; 1709 s->bits_sent += len << 3; 1710 #endif 1711 1712 /* Copy uncompressed bytes from the window to next_out. */ 1713 if (left) { 1714 if (left > len) 1715 left = len; 1716 zmemcpy(s->strm->next_out, s->window + s->block_start, left); 1717 s->strm->next_out += left; 1718 s->strm->avail_out -= left; 1719 s->strm->total_out += left; 1720 s->block_start += left; 1721 len -= left; 1722 } 1723 1724 /* Copy uncompressed bytes directly from next_in to next_out, updating 1725 * the check value. 1726 */ 1727 if (len) { 1728 read_buf(s->strm, s->strm->next_out, len); 1729 s->strm->next_out += len; 1730 s->strm->avail_out -= len; 1731 s->strm->total_out += len; 1732 } 1733 } while (last == 0); 1734 1735 /* Update the sliding window with the last s->w_size bytes of the copied 1736 * data, or append all of the copied data to the existing window if less 1737 * than s->w_size bytes were copied. Also update the number of bytes to 1738 * insert in the hash tables, in the event that deflateParams() switches to 1739 * a non-zero compression level. 1740 */ 1741 used -= s->strm->avail_in; /* number of input bytes directly copied */ 1742 if (used) { 1743 /* If any input was used, then no unused input remains in the window, 1744 * therefore s->block_start == s->strstart. 1745 */ 1746 if (used >= s->w_size) { /* supplant the previous history */ 1747 s->matches = 2; /* clear hash */ 1748 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size); 1749 s->strstart = s->w_size; 1750 } 1751 else { 1752 if (s->window_size - s->strstart <= used) { 1753 /* Slide the window down. */ 1754 s->strstart -= s->w_size; 1755 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1756 if (s->matches < 2) 1757 s->matches++; /* add a pending slide_hash() */ 1758 } 1759 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used); 1760 s->strstart += used; 1761 } 1762 s->block_start = s->strstart; 1763 s->insert += MIN(used, s->w_size - s->insert); 1764 } 1765 if (s->high_water < s->strstart) 1766 s->high_water = s->strstart; 1767 1768 /* If the last block was written to next_out, then done. */ 1769 if (last) 1770 return finish_done; 1771 1772 /* If flushing and all input has been consumed, then done. */ 1773 if (flush != Z_NO_FLUSH && flush != Z_FINISH && 1774 s->strm->avail_in == 0 && (long)s->strstart == s->block_start) 1775 return block_done; 1776 1777 /* Fill the window with any remaining input. */ 1778 have = s->window_size - s->strstart - 1; 1779 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) { 1780 /* Slide the window down. */ 1781 s->block_start -= s->w_size; 1782 s->strstart -= s->w_size; 1783 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1784 if (s->matches < 2) 1785 s->matches++; /* add a pending slide_hash() */ 1786 have += s->w_size; /* more space now */ 1787 } 1788 if (have > s->strm->avail_in) 1789 have = s->strm->avail_in; 1790 if (have) { 1791 read_buf(s->strm, s->window + s->strstart, have); 1792 s->strstart += have; 1793 } 1794 if (s->high_water < s->strstart) 1795 s->high_water = s->strstart; 1796 1797 /* There was not enough avail_out to write a complete worthy or flushed 1798 * stored block to next_out. Write a stored block to pending instead, if we 1799 * have enough input for a worthy block, or if flushing and there is enough 1800 * room for the remaining input as a stored block in the pending buffer. 1801 */ 1802 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1803 /* maximum stored block length that will fit in pending: */ 1804 have = MIN(s->pending_buf_size - have, MAX_STORED); 1805 min_block = MIN(have, s->w_size); 1806 left = s->strstart - s->block_start; 1807 if (left >= min_block || 1808 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH && 1809 s->strm->avail_in == 0 && left <= have)) { 1810 len = MIN(left, have); 1811 last = flush == Z_FINISH && s->strm->avail_in == 0 && 1812 len == left ? 1 : 0; 1813 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last); 1814 s->block_start += len; 1815 flush_pending(s->strm); 1816 } 1817 1818 /* We've done all we can with the available input and output. */ 1819 return last ? finish_started : need_more; 1820 } 1821 1822 /* =========================================================================== 1823 * Compress as much as possible from the input stream, return the current 1824 * block state. 1825 * This function does not perform lazy evaluation of matches and inserts 1826 * new strings in the dictionary only for unmatched strings or for short 1827 * matches. It is used only for the fast compression options. 1828 */ 1829 local block_state deflate_fast(s, flush) 1830 deflate_state *s; 1831 int flush; 1832 { 1833 IPos hash_head; /* head of the hash chain */ 1834 int bflush; /* set if current block must be flushed */ 1835 1836 for (;;) { 1837 /* Make sure that we always have enough lookahead, except 1838 * at the end of the input file. We need MAX_MATCH bytes 1839 * for the next match, plus MIN_MATCH bytes to insert the 1840 * string following the next match. 1841 */ 1842 if (s->lookahead < MIN_LOOKAHEAD) { 1843 fill_window(s); 1844 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1845 return need_more; 1846 } 1847 if (s->lookahead == 0) break; /* flush the current block */ 1848 } 1849 1850 /* Insert the string window[strstart .. strstart+2] in the 1851 * dictionary, and set hash_head to the head of the hash chain: 1852 */ 1853 hash_head = NIL; 1854 if (s->lookahead >= MIN_MATCH) { 1855 INSERT_STRING(s, s->strstart, hash_head); 1856 } 1857 1858 /* Find the longest match, discarding those <= prev_length. 1859 * At this point we have always match_length < MIN_MATCH 1860 */ 1861 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { 1862 /* To simplify the code, we prevent matches with the string 1863 * of window index 0 (in particular we have to avoid a match 1864 * of the string with itself at the start of the input file). 1865 */ 1866 s->match_length = longest_match (s, hash_head); 1867 /* longest_match() sets match_start */ 1868 } 1869 if (s->match_length >= MIN_MATCH) { 1870 check_match(s, s->strstart, s->match_start, s->match_length); 1871 1872 _tr_tally_dist(s, s->strstart - s->match_start, 1873 s->match_length - MIN_MATCH, bflush); 1874 1875 s->lookahead -= s->match_length; 1876 1877 /* Insert new strings in the hash table only if the match length 1878 * is not too large. This saves time but degrades compression. 1879 */ 1880 #ifndef FASTEST 1881 if (s->match_length <= s->max_insert_length && 1882 s->lookahead >= MIN_MATCH) { 1883 s->match_length--; /* string at strstart already in table */ 1884 do { 1885 s->strstart++; 1886 INSERT_STRING(s, s->strstart, hash_head); 1887 /* strstart never exceeds WSIZE-MAX_MATCH, so there are 1888 * always MIN_MATCH bytes ahead. 1889 */ 1890 } while (--s->match_length != 0); 1891 s->strstart++; 1892 } else 1893 #endif 1894 { 1895 s->strstart += s->match_length; 1896 s->match_length = 0; 1897 s->ins_h = s->window[s->strstart]; 1898 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); 1899 #if MIN_MATCH != 3 1900 Call UPDATE_HASH() MIN_MATCH-3 more times 1901 #endif 1902 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not 1903 * matter since it will be recomputed at next deflate call. 1904 */ 1905 } 1906 } else { 1907 /* No match, output a literal byte */ 1908 Tracevv((stderr,"%c", s->window[s->strstart])); 1909 _tr_tally_lit (s, s->window[s->strstart], bflush); 1910 s->lookahead--; 1911 s->strstart++; 1912 } 1913 if (bflush) FLUSH_BLOCK(s, 0); 1914 } 1915 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 1916 if (flush == Z_FINISH) { 1917 FLUSH_BLOCK(s, 1); 1918 return finish_done; 1919 } 1920 if (s->last_lit) 1921 FLUSH_BLOCK(s, 0); 1922 return block_done; 1923 } 1924 1925 #ifndef FASTEST 1926 /* =========================================================================== 1927 * Same as above, but achieves better compression. We use a lazy 1928 * evaluation for matches: a match is finally adopted only if there is 1929 * no better match at the next window position. 1930 */ 1931 local block_state deflate_slow(s, flush) 1932 deflate_state *s; 1933 int flush; 1934 { 1935 IPos hash_head; /* head of hash chain */ 1936 int bflush; /* set if current block must be flushed */ 1937 1938 /* Process the input block. */ 1939 for (;;) { 1940 /* Make sure that we always have enough lookahead, except 1941 * at the end of the input file. We need MAX_MATCH bytes 1942 * for the next match, plus MIN_MATCH bytes to insert the 1943 * string following the next match. 1944 */ 1945 if (s->lookahead < MIN_LOOKAHEAD) { 1946 fill_window(s); 1947 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1948 return need_more; 1949 } 1950 if (s->lookahead == 0) break; /* flush the current block */ 1951 } 1952 1953 /* Insert the string window[strstart .. strstart+2] in the 1954 * dictionary, and set hash_head to the head of the hash chain: 1955 */ 1956 hash_head = NIL; 1957 if (s->lookahead >= MIN_MATCH) { 1958 INSERT_STRING(s, s->strstart, hash_head); 1959 } 1960 1961 /* Find the longest match, discarding those <= prev_length. 1962 */ 1963 s->prev_length = s->match_length, s->prev_match = s->match_start; 1964 s->match_length = MIN_MATCH-1; 1965 1966 if (hash_head != NIL && s->prev_length < s->max_lazy_match && 1967 s->strstart - hash_head <= MAX_DIST(s)) { 1968 /* To simplify the code, we prevent matches with the string 1969 * of window index 0 (in particular we have to avoid a match 1970 * of the string with itself at the start of the input file). 1971 */ 1972 s->match_length = longest_match (s, hash_head); 1973 /* longest_match() sets match_start */ 1974 1975 if (s->match_length <= 5 && (s->strategy == Z_FILTERED 1976 #if TOO_FAR <= 32767 1977 || (s->match_length == MIN_MATCH && 1978 s->strstart - s->match_start > TOO_FAR) 1979 #endif 1980 )) { 1981 1982 /* If prev_match is also MIN_MATCH, match_start is garbage 1983 * but we will ignore the current match anyway. 1984 */ 1985 s->match_length = MIN_MATCH-1; 1986 } 1987 } 1988 /* If there was a match at the previous step and the current 1989 * match is not better, output the previous match: 1990 */ 1991 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { 1992 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; 1993 /* Do not insert strings in hash table beyond this. */ 1994 1995 check_match(s, s->strstart-1, s->prev_match, s->prev_length); 1996 1997 _tr_tally_dist(s, s->strstart -1 - s->prev_match, 1998 s->prev_length - MIN_MATCH, bflush); 1999 2000 /* Insert in hash table all strings up to the end of the match. 2001 * strstart-1 and strstart are already inserted. If there is not 2002 * enough lookahead, the last two strings are not inserted in 2003 * the hash table. 2004 */ 2005 s->lookahead -= s->prev_length-1; 2006 s->prev_length -= 2; 2007 do { 2008 if (++s->strstart <= max_insert) { 2009 INSERT_STRING(s, s->strstart, hash_head); 2010 } 2011 } while (--s->prev_length != 0); 2012 s->match_available = 0; 2013 s->match_length = MIN_MATCH-1; 2014 s->strstart++; 2015 2016 if (bflush) FLUSH_BLOCK(s, 0); 2017 2018 } else if (s->match_available) { 2019 /* If there was no match at the previous position, output a 2020 * single literal. If there was a match but the current match 2021 * is longer, truncate the previous match to a single literal. 2022 */ 2023 Tracevv((stderr,"%c", s->window[s->strstart-1])); 2024 _tr_tally_lit(s, s->window[s->strstart-1], bflush); 2025 if (bflush) { 2026 FLUSH_BLOCK_ONLY(s, 0); 2027 } 2028 s->strstart++; 2029 s->lookahead--; 2030 if (s->strm->avail_out == 0) return need_more; 2031 } else { 2032 /* There is no previous match to compare with, wait for 2033 * the next step to decide. 2034 */ 2035 s->match_available = 1; 2036 s->strstart++; 2037 s->lookahead--; 2038 } 2039 } 2040 Assert (flush != Z_NO_FLUSH, "no flush?"); 2041 if (s->match_available) { 2042 Tracevv((stderr,"%c", s->window[s->strstart-1])); 2043 _tr_tally_lit(s, s->window[s->strstart-1], bflush); 2044 s->match_available = 0; 2045 } 2046 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 2047 if (flush == Z_FINISH) { 2048 FLUSH_BLOCK(s, 1); 2049 return finish_done; 2050 } 2051 if (s->last_lit) 2052 FLUSH_BLOCK(s, 0); 2053 return block_done; 2054 } 2055 #endif /* FASTEST */ 2056 2057 /* =========================================================================== 2058 * For Z_RLE, simply look for runs of bytes, generate matches only of distance 2059 * one. Do not maintain a hash table. (It will be regenerated if this run of 2060 * deflate switches away from Z_RLE.) 2061 */ 2062 local block_state deflate_rle(s, flush) 2063 deflate_state *s; 2064 int flush; 2065 { 2066 int bflush; /* set if current block must be flushed */ 2067 uInt prev; /* byte at distance one to match */ 2068 Bytef *scan, *strend; /* scan goes up to strend for length of run */ 2069 2070 for (;;) { 2071 /* Make sure that we always have enough lookahead, except 2072 * at the end of the input file. We need MAX_MATCH bytes 2073 * for the longest run, plus one for the unrolled loop. 2074 */ 2075 if (s->lookahead <= MAX_MATCH) { 2076 fill_window(s); 2077 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) { 2078 return need_more; 2079 } 2080 if (s->lookahead == 0) break; /* flush the current block */ 2081 } 2082 2083 /* See how many times the previous byte repeats */ 2084 s->match_length = 0; 2085 if (s->lookahead >= MIN_MATCH && s->strstart > 0) { 2086 scan = s->window + s->strstart - 1; 2087 prev = *scan; 2088 if (prev == *++scan && prev == *++scan && prev == *++scan) { 2089 strend = s->window + s->strstart + MAX_MATCH; 2090 do { 2091 } while (prev == *++scan && prev == *++scan && 2092 prev == *++scan && prev == *++scan && 2093 prev == *++scan && prev == *++scan && 2094 prev == *++scan && prev == *++scan && 2095 scan < strend); 2096 s->match_length = MAX_MATCH - (uInt)(strend - scan); 2097 if (s->match_length > s->lookahead) 2098 s->match_length = s->lookahead; 2099 } 2100 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan"); 2101 } 2102 2103 /* Emit match if have run of MIN_MATCH or longer, else emit literal */ 2104 if (s->match_length >= MIN_MATCH) { 2105 check_match(s, s->strstart, s->strstart - 1, s->match_length); 2106 2107 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush); 2108 2109 s->lookahead -= s->match_length; 2110 s->strstart += s->match_length; 2111 s->match_length = 0; 2112 } else { 2113 /* No match, output a literal byte */ 2114 Tracevv((stderr,"%c", s->window[s->strstart])); 2115 _tr_tally_lit (s, s->window[s->strstart], bflush); 2116 s->lookahead--; 2117 s->strstart++; 2118 } 2119 if (bflush) FLUSH_BLOCK(s, 0); 2120 } 2121 s->insert = 0; 2122 if (flush == Z_FINISH) { 2123 FLUSH_BLOCK(s, 1); 2124 return finish_done; 2125 } 2126 if (s->last_lit) 2127 FLUSH_BLOCK(s, 0); 2128 return block_done; 2129 } 2130 2131 /* =========================================================================== 2132 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table. 2133 * (It will be regenerated if this run of deflate switches away from Huffman.) 2134 */ 2135 local block_state deflate_huff(s, flush) 2136 deflate_state *s; 2137 int flush; 2138 { 2139 int bflush; /* set if current block must be flushed */ 2140 2141 for (;;) { 2142 /* Make sure that we have a literal to write. */ 2143 if (s->lookahead == 0) { 2144 fill_window(s); 2145 if (s->lookahead == 0) { 2146 if (flush == Z_NO_FLUSH) 2147 return need_more; 2148 break; /* flush the current block */ 2149 } 2150 } 2151 2152 /* Output a literal byte */ 2153 s->match_length = 0; 2154 Tracevv((stderr,"%c", s->window[s->strstart])); 2155 _tr_tally_lit (s, s->window[s->strstart], bflush); 2156 s->lookahead--; 2157 s->strstart++; 2158 if (bflush) FLUSH_BLOCK(s, 0); 2159 } 2160 s->insert = 0; 2161 if (flush == Z_FINISH) { 2162 FLUSH_BLOCK(s, 1); 2163 return finish_done; 2164 } 2165 if (s->last_lit) 2166 FLUSH_BLOCK(s, 0); 2167 return block_done; 2168 } 2169