1 /* $NetBSD: deflate.c,v 1.4 2017/01/10 01:27:41 christos Exp $ */ 2 3 /* deflate.c -- compress data using the deflation algorithm 4 * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler 5 * For conditions of distribution and use, see copyright notice in zlib.h 6 */ 7 8 /* 9 * ALGORITHM 10 * 11 * The "deflation" process depends on being able to identify portions 12 * of the input text which are identical to earlier input (within a 13 * sliding window trailing behind the input currently being processed). 14 * 15 * The most straightforward technique turns out to be the fastest for 16 * most input files: try all possible matches and select the longest. 17 * The key feature of this algorithm is that insertions into the string 18 * dictionary are very simple and thus fast, and deletions are avoided 19 * completely. Insertions are performed at each input character, whereas 20 * string matches are performed only when the previous match ends. So it 21 * is preferable to spend more time in matches to allow very fast string 22 * insertions and avoid deletions. The matching algorithm for small 23 * strings is inspired from that of Rabin & Karp. A brute force approach 24 * is used to find longer strings when a small match has been found. 25 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze 26 * (by Leonid Broukhis). 27 * A previous version of this file used a more sophisticated algorithm 28 * (by Fiala and Greene) which is guaranteed to run in linear amortized 29 * time, but has a larger average cost, uses more memory and is patented. 30 * However the F&G algorithm may be faster for some highly redundant 31 * files if the parameter max_chain_length (described below) is too large. 32 * 33 * ACKNOWLEDGEMENTS 34 * 35 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and 36 * I found it in 'freeze' written by Leonid Broukhis. 37 * Thanks to many people for bug reports and testing. 38 * 39 * REFERENCES 40 * 41 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification". 42 * Available in http://tools.ietf.org/html/rfc1951 43 * 44 * A description of the Rabin and Karp algorithm is given in the book 45 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252. 46 * 47 * Fiala,E.R., and Greene,D.H. 48 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595 49 * 50 */ 51 52 /* @(#) $Id: deflate.c,v 1.4 2017/01/10 01:27:41 christos Exp $ */ 53 54 #include "deflate.h" 55 56 const char deflate_copyright[] = 57 " deflate 1.2.10 Copyright 1995-2017 Jean-loup Gailly and Mark Adler "; 58 /* 59 If you use the zlib library in a product, an acknowledgment is welcome 60 in the documentation of your product. If for some reason you cannot 61 include such an acknowledgment, I would appreciate that you keep this 62 copyright string in the executable of your product. 63 */ 64 65 /* =========================================================================== 66 * Function prototypes. 67 */ 68 typedef enum { 69 need_more, /* block not completed, need more input or more output */ 70 block_done, /* block flush performed */ 71 finish_started, /* finish started, need only more output at next deflate */ 72 finish_done /* finish done, accept no more input or output */ 73 } block_state; 74 75 typedef block_state (*compress_func) OF((deflate_state *s, int flush)); 76 /* Compression function. Returns the block state after the call. */ 77 78 local int deflateStateCheck OF((z_streamp strm)); 79 local void slide_hash OF((deflate_state *s)); 80 local void fill_window OF((deflate_state *s)); 81 local block_state deflate_stored OF((deflate_state *s, int flush)); 82 local block_state deflate_fast OF((deflate_state *s, int flush)); 83 #ifndef FASTEST 84 local block_state deflate_slow OF((deflate_state *s, int flush)); 85 #endif 86 local block_state deflate_rle OF((deflate_state *s, int flush)); 87 local block_state deflate_huff OF((deflate_state *s, int flush)); 88 local void lm_init OF((deflate_state *s)); 89 local void putShortMSB OF((deflate_state *s, uInt b)); 90 local void flush_pending OF((z_streamp strm)); 91 local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); 92 #ifdef ASMV 93 # pragma message("Assembler code may have bugs -- use at your own risk") 94 void match_init OF((void)); /* asm code initialization */ 95 uInt longest_match OF((deflate_state *s, IPos cur_match)); 96 #else 97 local uInt longest_match OF((deflate_state *s, IPos cur_match)); 98 #endif 99 100 #ifdef ZLIB_DEBUG 101 local void check_match OF((deflate_state *s, IPos start, IPos match, 102 int length)); 103 #endif 104 105 /* =========================================================================== 106 * Local data 107 */ 108 109 #define NIL 0 110 /* Tail of hash chains */ 111 112 #ifndef TOO_FAR 113 # define TOO_FAR 4096 114 #endif 115 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ 116 117 /* Values for max_lazy_match, good_match and max_chain_length, depending on 118 * the desired pack level (0..9). The values given below have been tuned to 119 * exclude worst case performance for pathological files. Better values may be 120 * found for specific files. 121 */ 122 typedef struct config_s { 123 ush good_length; /* reduce lazy search above this match length */ 124 ush max_lazy; /* do not perform lazy search above this match length */ 125 ush nice_length; /* quit search above this match length */ 126 ush max_chain; 127 compress_func func; 128 } config; 129 130 #ifdef FASTEST 131 local const config configuration_table[2] = { 132 /* good lazy nice chain */ 133 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 134 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */ 135 #else 136 local const config configuration_table[10] = { 137 /* good lazy nice chain */ 138 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 139 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */ 140 /* 2 */ {4, 5, 16, 8, deflate_fast}, 141 /* 3 */ {4, 6, 32, 32, deflate_fast}, 142 143 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ 144 /* 5 */ {8, 16, 32, 32, deflate_slow}, 145 /* 6 */ {8, 16, 128, 128, deflate_slow}, 146 /* 7 */ {8, 32, 128, 256, deflate_slow}, 147 /* 8 */ {32, 128, 258, 1024, deflate_slow}, 148 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */ 149 #endif 150 151 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 152 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different 153 * meaning. 154 */ 155 156 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */ 157 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0)) 158 159 /* =========================================================================== 160 * Update a hash value with the given input byte 161 * IN assertion: all calls to UPDATE_HASH are made with consecutive input 162 * characters, so that a running hash key can be computed from the previous 163 * key instead of complete recalculation each time. 164 */ 165 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask) 166 167 168 /* =========================================================================== 169 * Insert string str in the dictionary and set match_head to the previous head 170 * of the hash chain (the most recent string with same hash key). Return 171 * the previous length of the hash chain. 172 * If this file is compiled with -DFASTEST, the compression level is forced 173 * to 1, and no hash chains are maintained. 174 * IN assertion: all calls to INSERT_STRING are made with consecutive input 175 * characters and the first MIN_MATCH bytes of str are valid (except for 176 * the last MIN_MATCH-1 bytes of the input file). 177 */ 178 #ifdef FASTEST 179 #define INSERT_STRING(s, str, match_head) \ 180 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 181 match_head = s->head[s->ins_h], \ 182 s->head[s->ins_h] = (Pos)(str)) 183 #else 184 #define INSERT_STRING(s, str, match_head) \ 185 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 186 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \ 187 s->head[s->ins_h] = (Pos)(str)) 188 #endif 189 190 /* =========================================================================== 191 * Initialize the hash table (avoiding 64K overflow for 16 bit systems). 192 * prev[] will be initialized on the fly. 193 */ 194 #define CLEAR_HASH(s) \ 195 s->head[s->hash_size-1] = NIL; \ 196 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head)); 197 198 /* =========================================================================== 199 * Slide the hash table when sliding the window down (could be avoided with 32 200 * bit values at the expense of memory usage). We slide even when level == 0 to 201 * keep the hash table consistent if we switch back to level > 0 later. 202 */ 203 local void slide_hash(s) 204 deflate_state *s; 205 { 206 unsigned n, m; 207 Posf *p; 208 uInt wsize = s->w_size; 209 210 n = s->hash_size; 211 p = &s->head[n]; 212 do { 213 m = *--p; 214 *p = (Pos)(m >= wsize ? m - wsize : NIL); 215 } while (--n); 216 n = wsize; 217 #ifndef FASTEST 218 p = &s->prev[n]; 219 do { 220 m = *--p; 221 *p = (Pos)(m >= wsize ? m - wsize : NIL); 222 /* If n is not on any hash chain, prev[n] is garbage but 223 * its value will never be used. 224 */ 225 } while (--n); 226 #endif 227 } 228 229 /* ========================================================================= */ 230 int ZEXPORT deflateInit_(strm, level, version, stream_size) 231 z_streamp strm; 232 int level; 233 const char *version; 234 int stream_size; 235 { 236 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, 237 Z_DEFAULT_STRATEGY, version, stream_size); 238 /* To do: ignore strm->next_in if we use it as window */ 239 } 240 241 /* ========================================================================= */ 242 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, 243 version, stream_size) 244 z_streamp strm; 245 int level; 246 int method; 247 int windowBits; 248 int memLevel; 249 int strategy; 250 const char *version; 251 int stream_size; 252 { 253 deflate_state *s; 254 int wrap = 1; 255 static const char my_version[] = ZLIB_VERSION; 256 257 ushf *overlay; 258 /* We overlay pending_buf and d_buf+l_buf. This works since the average 259 * output size for (length,distance) codes is <= 24 bits. 260 */ 261 262 if (version == Z_NULL || version[0] != my_version[0] || 263 stream_size != sizeof(z_stream)) { 264 return Z_VERSION_ERROR; 265 } 266 if (strm == Z_NULL) return Z_STREAM_ERROR; 267 268 strm->msg = Z_NULL; 269 if (strm->zalloc == (alloc_func)0) { 270 #ifdef Z_SOLO 271 return Z_STREAM_ERROR; 272 #else 273 strm->zalloc = zcalloc; 274 strm->opaque = (voidpf)0; 275 #endif 276 } 277 if (strm->zfree == (free_func)0) 278 #ifdef Z_SOLO 279 return Z_STREAM_ERROR; 280 #else 281 strm->zfree = zcfree; 282 #endif 283 284 #ifdef FASTEST 285 if (level != 0) level = 1; 286 #else 287 if (level == Z_DEFAULT_COMPRESSION) level = 6; 288 #endif 289 290 if (windowBits < 0) { /* suppress zlib wrapper */ 291 wrap = 0; 292 windowBits = -windowBits; 293 } 294 #ifdef GZIP 295 else if (windowBits > 15) { 296 wrap = 2; /* write gzip wrapper instead */ 297 windowBits -= 16; 298 } 299 #endif 300 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || 301 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || 302 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) { 303 return Z_STREAM_ERROR; 304 } 305 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ 306 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); 307 if (s == Z_NULL) return Z_MEM_ERROR; 308 strm->state = (struct internal_state FAR *)s; 309 s->strm = strm; 310 s->status = INIT_STATE; /* to pass state test in deflateReset() */ 311 312 s->wrap = wrap; 313 s->gzhead = Z_NULL; 314 s->w_bits = (uInt)windowBits; 315 s->w_size = 1 << s->w_bits; 316 s->w_mask = s->w_size - 1; 317 318 s->hash_bits = (uInt)memLevel + 7; 319 s->hash_size = 1 << s->hash_bits; 320 s->hash_mask = s->hash_size - 1; 321 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); 322 323 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); 324 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); 325 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); 326 327 s->high_water = 0; /* nothing written to s->window yet */ 328 329 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ 330 331 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2); 332 s->pending_buf = (uchf *) overlay; 333 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); 334 335 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || 336 s->pending_buf == Z_NULL) { 337 s->status = FINISH_STATE; 338 strm->msg = __UNCONST(ERR_MSG(Z_MEM_ERROR)); 339 deflateEnd (strm); 340 return Z_MEM_ERROR; 341 } 342 s->d_buf = overlay + s->lit_bufsize/sizeof(ush); 343 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; 344 345 s->level = level; 346 s->strategy = strategy; 347 s->method = (Byte)method; 348 349 return deflateReset(strm); 350 } 351 352 /* ========================================================================= 353 * Check for a valid deflate stream state. Return 0 if ok, 1 if not. 354 */ 355 local int deflateStateCheck (strm) 356 z_streamp strm; 357 { 358 deflate_state *s; 359 if (strm == Z_NULL || 360 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) 361 return 1; 362 s = strm->state; 363 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE && 364 #ifdef GZIP 365 s->status != GZIP_STATE && 366 #endif 367 s->status != EXTRA_STATE && 368 s->status != NAME_STATE && 369 s->status != COMMENT_STATE && 370 s->status != HCRC_STATE && 371 s->status != BUSY_STATE && 372 s->status != FINISH_STATE)) 373 return 1; 374 return 0; 375 } 376 377 /* ========================================================================= */ 378 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength) 379 z_streamp strm; 380 const Bytef *dictionary; 381 uInt dictLength; 382 { 383 deflate_state *s; 384 uInt str, n; 385 int wrap; 386 unsigned avail; 387 z_const unsigned char *next; 388 389 if (deflateStateCheck(strm) || dictionary == Z_NULL) 390 return Z_STREAM_ERROR; 391 s = strm->state; 392 wrap = s->wrap; 393 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead) 394 return Z_STREAM_ERROR; 395 396 /* when using zlib wrappers, compute Adler-32 for provided dictionary */ 397 if (wrap == 1) 398 strm->adler = adler32(strm->adler, dictionary, dictLength); 399 s->wrap = 0; /* avoid computing Adler-32 in read_buf */ 400 401 /* if dictionary would fill window, just replace the history */ 402 if (dictLength >= s->w_size) { 403 if (wrap == 0) { /* already empty otherwise */ 404 CLEAR_HASH(s); 405 s->strstart = 0; 406 s->block_start = 0L; 407 s->insert = 0; 408 } 409 dictionary += dictLength - s->w_size; /* use the tail */ 410 dictLength = s->w_size; 411 } 412 413 /* insert dictionary into window and hash */ 414 avail = strm->avail_in; 415 next = strm->next_in; 416 strm->avail_in = dictLength; 417 strm->next_in = __UNCONST(dictionary); 418 fill_window(s); 419 while (s->lookahead >= MIN_MATCH) { 420 str = s->strstart; 421 n = s->lookahead - (MIN_MATCH-1); 422 do { 423 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 424 #ifndef FASTEST 425 s->prev[str & s->w_mask] = s->head[s->ins_h]; 426 #endif 427 s->head[s->ins_h] = (Pos)str; 428 str++; 429 } while (--n); 430 s->strstart = str; 431 s->lookahead = MIN_MATCH-1; 432 fill_window(s); 433 } 434 s->strstart += s->lookahead; 435 s->block_start = (long)s->strstart; 436 s->insert = s->lookahead; 437 s->lookahead = 0; 438 s->match_length = s->prev_length = MIN_MATCH-1; 439 s->match_available = 0; 440 strm->next_in = next; 441 strm->avail_in = avail; 442 s->wrap = wrap; 443 return Z_OK; 444 } 445 446 /* ========================================================================= */ 447 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength) 448 z_streamp strm; 449 Bytef *dictionary; 450 uInt *dictLength; 451 { 452 deflate_state *s; 453 uInt len; 454 455 if (deflateStateCheck(strm)) 456 return Z_STREAM_ERROR; 457 s = strm->state; 458 len = s->strstart + s->lookahead; 459 if (len > s->w_size) 460 len = s->w_size; 461 if (dictionary != Z_NULL && len) 462 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len); 463 if (dictLength != Z_NULL) 464 *dictLength = len; 465 return Z_OK; 466 } 467 468 /* ========================================================================= */ 469 int ZEXPORT deflateResetKeep (strm) 470 z_streamp strm; 471 { 472 deflate_state *s; 473 474 if (deflateStateCheck(strm)) { 475 return Z_STREAM_ERROR; 476 } 477 478 strm->total_in = strm->total_out = 0; 479 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ 480 strm->data_type = Z_UNKNOWN; 481 482 s = (deflate_state *)strm->state; 483 s->pending = 0; 484 s->pending_out = s->pending_buf; 485 486 if (s->wrap < 0) { 487 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ 488 } 489 s->status = 490 #ifdef GZIP 491 s->wrap == 2 ? GZIP_STATE : 492 #endif 493 s->wrap ? INIT_STATE : BUSY_STATE; 494 strm->adler = 495 #ifdef GZIP 496 s->wrap == 2 ? crc32(0L, Z_NULL, 0) : 497 #endif 498 adler32(0L, Z_NULL, 0); 499 s->last_flush = Z_NO_FLUSH; 500 501 _tr_init(s); 502 503 return Z_OK; 504 } 505 506 /* ========================================================================= */ 507 int ZEXPORT deflateReset (strm) 508 z_streamp strm; 509 { 510 int ret; 511 512 ret = deflateResetKeep(strm); 513 if (ret == Z_OK) 514 lm_init(strm->state); 515 return ret; 516 } 517 518 /* ========================================================================= */ 519 int ZEXPORT deflateSetHeader (strm, head) 520 z_streamp strm; 521 gz_headerp head; 522 { 523 if (deflateStateCheck(strm) || strm->state->wrap != 2) 524 return Z_STREAM_ERROR; 525 strm->state->gzhead = head; 526 return Z_OK; 527 } 528 529 /* ========================================================================= */ 530 int ZEXPORT deflatePending (strm, pending, bits) 531 unsigned *pending; 532 int *bits; 533 z_streamp strm; 534 { 535 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 536 if (pending != Z_NULL) 537 *pending = strm->state->pending; 538 if (bits != Z_NULL) 539 *bits = strm->state->bi_valid; 540 return Z_OK; 541 } 542 543 /* ========================================================================= */ 544 int ZEXPORT deflatePrime (strm, bits, value) 545 z_streamp strm; 546 int bits; 547 int value; 548 { 549 deflate_state *s; 550 int put; 551 552 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 553 s = strm->state; 554 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3)) 555 return Z_BUF_ERROR; 556 do { 557 put = Buf_size - s->bi_valid; 558 if (put > bits) 559 put = bits; 560 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid); 561 s->bi_valid += put; 562 _tr_flush_bits(s); 563 value >>= put; 564 bits -= put; 565 } while (bits); 566 return Z_OK; 567 } 568 569 /* ========================================================================= */ 570 int ZEXPORT deflateParams(strm, level, strategy) 571 z_streamp strm; 572 int level; 573 int strategy; 574 { 575 deflate_state *s; 576 compress_func func; 577 578 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 579 s = strm->state; 580 581 #ifdef FASTEST 582 if (level != 0) level = 1; 583 #else 584 if (level == Z_DEFAULT_COMPRESSION) level = 6; 585 #endif 586 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { 587 return Z_STREAM_ERROR; 588 } 589 func = configuration_table[s->level].func; 590 591 if ((strategy != s->strategy || func != configuration_table[level].func)) { 592 /* Flush the last buffer: */ 593 int err = deflate(strm, Z_BLOCK); 594 if (err == Z_STREAM_ERROR) 595 return err; 596 if (strm->avail_out == 0) 597 return Z_BUF_ERROR; 598 } 599 if (s->level != level) { 600 if (s->level == 0 && s->matches != 0) { 601 if (s->matches == 1) 602 slide_hash(s); 603 else 604 CLEAR_HASH(s); 605 s->matches = 0; 606 } 607 s->level = level; 608 s->max_lazy_match = configuration_table[level].max_lazy; 609 s->good_match = configuration_table[level].good_length; 610 s->nice_match = configuration_table[level].nice_length; 611 s->max_chain_length = configuration_table[level].max_chain; 612 } 613 s->strategy = strategy; 614 return Z_OK; 615 } 616 617 /* ========================================================================= */ 618 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain) 619 z_streamp strm; 620 int good_length; 621 int max_lazy; 622 int nice_length; 623 int max_chain; 624 { 625 deflate_state *s; 626 627 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 628 s = strm->state; 629 s->good_match = (uInt)good_length; 630 s->max_lazy_match = (uInt)max_lazy; 631 s->nice_match = nice_length; 632 s->max_chain_length = (uInt)max_chain; 633 return Z_OK; 634 } 635 636 /* ========================================================================= 637 * For the default windowBits of 15 and memLevel of 8, this function returns 638 * a close to exact, as well as small, upper bound on the compressed size. 639 * They are coded as constants here for a reason--if the #define's are 640 * changed, then this function needs to be changed as well. The return 641 * value for 15 and 8 only works for those exact settings. 642 * 643 * For any setting other than those defaults for windowBits and memLevel, 644 * the value returned is a conservative worst case for the maximum expansion 645 * resulting from using fixed blocks instead of stored blocks, which deflate 646 * can emit on compressed data for some combinations of the parameters. 647 * 648 * This function could be more sophisticated to provide closer upper bounds for 649 * every combination of windowBits and memLevel. But even the conservative 650 * upper bound of about 14% expansion does not seem onerous for output buffer 651 * allocation. 652 */ 653 uLong ZEXPORT deflateBound(strm, sourceLen) 654 z_streamp strm; 655 uLong sourceLen; 656 { 657 deflate_state *s; 658 uLong complen, wraplen; 659 660 /* conservative upper bound for compressed data */ 661 complen = sourceLen + 662 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5; 663 664 /* if can't get parameters, return conservative bound plus zlib wrapper */ 665 if (deflateStateCheck(strm)) 666 return complen + 6; 667 668 /* compute wrapper length */ 669 s = strm->state; 670 switch (s->wrap) { 671 case 0: /* raw deflate */ 672 wraplen = 0; 673 break; 674 case 1: /* zlib wrapper */ 675 wraplen = 6 + (s->strstart ? 4 : 0); 676 break; 677 #ifdef GZIP 678 case 2: /* gzip wrapper */ 679 wraplen = 18; 680 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */ 681 Bytef *str; 682 if (s->gzhead->extra != Z_NULL) 683 wraplen += 2 + s->gzhead->extra_len; 684 str = s->gzhead->name; 685 if (str != Z_NULL) 686 do { 687 wraplen++; 688 } while (*str++); 689 str = s->gzhead->comment; 690 if (str != Z_NULL) 691 do { 692 wraplen++; 693 } while (*str++); 694 if (s->gzhead->hcrc) 695 wraplen += 2; 696 } 697 break; 698 #endif 699 default: /* for compiler happiness */ 700 wraplen = 6; 701 } 702 703 /* if not default parameters, return conservative bound */ 704 if (s->w_bits != 15 || s->hash_bits != 8 + 7) 705 return complen + wraplen; 706 707 /* default settings: return tight bound for that case */ 708 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) + 709 (sourceLen >> 25) + 13 - 6 + wraplen; 710 } 711 712 /* ========================================================================= 713 * Put a short in the pending buffer. The 16-bit value is put in MSB order. 714 * IN assertion: the stream state is correct and there is enough room in 715 * pending_buf. 716 */ 717 local void putShortMSB (s, b) 718 deflate_state *s; 719 uInt b; 720 { 721 put_byte(s, (Byte)(b >> 8)); 722 put_byte(s, (Byte)(b & 0xff)); 723 } 724 725 /* ========================================================================= 726 * Flush as much pending output as possible. All deflate() output, except for 727 * some deflate_stored() output, goes through this function so some 728 * applications may wish to modify it to avoid allocating a large 729 * strm->next_out buffer and copying into it. (See also read_buf()). 730 */ 731 local void flush_pending(strm) 732 z_streamp strm; 733 { 734 unsigned len; 735 deflate_state *s = strm->state; 736 737 _tr_flush_bits(s); 738 len = s->pending; 739 if (len > strm->avail_out) len = strm->avail_out; 740 if (len == 0) return; 741 742 zmemcpy(strm->next_out, s->pending_out, len); 743 strm->next_out += len; 744 s->pending_out += len; 745 strm->total_out += len; 746 strm->avail_out -= len; 747 s->pending -= len; 748 if (s->pending == 0) { 749 s->pending_out = s->pending_buf; 750 } 751 } 752 753 /* =========================================================================== 754 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1]. 755 */ 756 #define HCRC_UPDATE(beg) \ 757 do { \ 758 if (s->gzhead->hcrc && s->pending > (beg)) \ 759 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \ 760 s->pending - (beg)); \ 761 } while (0) 762 763 /* ========================================================================= */ 764 int ZEXPORT deflate (strm, flush) 765 z_streamp strm; 766 int flush; 767 { 768 int old_flush; /* value of flush param for previous deflate call */ 769 deflate_state *s; 770 771 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) { 772 return Z_STREAM_ERROR; 773 } 774 s = strm->state; 775 776 if (strm->next_out == Z_NULL || 777 (strm->avail_in != 0 && strm->next_in == Z_NULL) || 778 (s->status == FINISH_STATE && flush != Z_FINISH)) { 779 ERR_RETURN(strm, Z_STREAM_ERROR); 780 } 781 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); 782 783 old_flush = s->last_flush; 784 s->last_flush = flush; 785 786 /* Flush as much pending output as possible */ 787 if (s->pending != 0) { 788 flush_pending(strm); 789 if (strm->avail_out == 0) { 790 /* Since avail_out is 0, deflate will be called again with 791 * more output space, but possibly with both pending and 792 * avail_in equal to zero. There won't be anything to do, 793 * but this is not an error situation so make sure we 794 * return OK instead of BUF_ERROR at next call of deflate: 795 */ 796 s->last_flush = -1; 797 return Z_OK; 798 } 799 800 /* Make sure there is something to do and avoid duplicate consecutive 801 * flushes. For repeated and useless calls with Z_FINISH, we keep 802 * returning Z_STREAM_END instead of Z_BUF_ERROR. 803 */ 804 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) && 805 flush != Z_FINISH) { 806 ERR_RETURN(strm, Z_BUF_ERROR); 807 } 808 809 /* User must not provide more input after the first FINISH: */ 810 if (s->status == FINISH_STATE && strm->avail_in != 0) { 811 ERR_RETURN(strm, Z_BUF_ERROR); 812 } 813 814 /* Write the header */ 815 if (s->status == INIT_STATE) { 816 /* zlib header */ 817 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; 818 uInt level_flags; 819 820 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) 821 level_flags = 0; 822 else if (s->level < 6) 823 level_flags = 1; 824 else if (s->level == 6) 825 level_flags = 2; 826 else 827 level_flags = 3; 828 header |= (level_flags << 6); 829 if (s->strstart != 0) header |= PRESET_DICT; 830 header += 31 - (header % 31); 831 832 putShortMSB(s, header); 833 834 /* Save the adler32 of the preset dictionary: */ 835 if (s->strstart != 0) { 836 putShortMSB(s, (uInt)(strm->adler >> 16)); 837 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 838 } 839 strm->adler = adler32(0L, Z_NULL, 0); 840 s->status = BUSY_STATE; 841 842 /* Compression must start with an empty pending buffer */ 843 flush_pending(strm); 844 if (s->pending != 0) { 845 s->last_flush = -1; 846 return Z_OK; 847 } 848 } 849 #ifdef GZIP 850 if (s->status == GZIP_STATE) { 851 /* gzip header */ 852 strm->adler = crc32(0L, Z_NULL, 0); 853 put_byte(s, 31); 854 put_byte(s, 139); 855 put_byte(s, 8); 856 if (s->gzhead == Z_NULL) { 857 put_byte(s, 0); 858 put_byte(s, 0); 859 put_byte(s, 0); 860 put_byte(s, 0); 861 put_byte(s, 0); 862 put_byte(s, s->level == 9 ? 2 : 863 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 864 4 : 0)); 865 put_byte(s, OS_CODE); 866 s->status = BUSY_STATE; 867 868 /* Compression must start with an empty pending buffer */ 869 flush_pending(strm); 870 if (s->pending != 0) { 871 s->last_flush = -1; 872 return Z_OK; 873 } 874 } 875 else { 876 put_byte(s, (s->gzhead->text ? 1 : 0) + 877 (s->gzhead->hcrc ? 2 : 0) + 878 (s->gzhead->extra == Z_NULL ? 0 : 4) + 879 (s->gzhead->name == Z_NULL ? 0 : 8) + 880 (s->gzhead->comment == Z_NULL ? 0 : 16) 881 ); 882 put_byte(s, (Byte)(s->gzhead->time & 0xff)); 883 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); 884 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); 885 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); 886 put_byte(s, s->level == 9 ? 2 : 887 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 888 4 : 0)); 889 put_byte(s, s->gzhead->os & 0xff); 890 if (s->gzhead->extra != Z_NULL) { 891 put_byte(s, s->gzhead->extra_len & 0xff); 892 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); 893 } 894 if (s->gzhead->hcrc) 895 strm->adler = crc32(strm->adler, s->pending_buf, 896 s->pending); 897 s->gzindex = 0; 898 s->status = EXTRA_STATE; 899 } 900 } 901 if (s->status == EXTRA_STATE) { 902 if (s->gzhead->extra != Z_NULL) { 903 ulg beg = s->pending; /* start of bytes to update crc */ 904 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex; 905 while (s->pending + left > s->pending_buf_size) { 906 uInt copy = s->pending_buf_size - s->pending; 907 zmemcpy(s->pending_buf + s->pending, 908 s->gzhead->extra + s->gzindex, copy); 909 s->pending = s->pending_buf_size; 910 HCRC_UPDATE(beg); 911 s->gzindex += copy; 912 flush_pending(strm); 913 if (s->pending != 0) { 914 s->last_flush = -1; 915 return Z_OK; 916 } 917 beg = 0; 918 left -= copy; 919 } 920 zmemcpy(s->pending_buf + s->pending, 921 s->gzhead->extra + s->gzindex, left); 922 s->pending += left; 923 HCRC_UPDATE(beg); 924 s->gzindex = 0; 925 } 926 s->status = NAME_STATE; 927 } 928 if (s->status == NAME_STATE) { 929 if (s->gzhead->name != Z_NULL) { 930 ulg beg = s->pending; /* start of bytes to update crc */ 931 int val; 932 do { 933 if (s->pending == s->pending_buf_size) { 934 HCRC_UPDATE(beg); 935 flush_pending(strm); 936 if (s->pending != 0) { 937 s->last_flush = -1; 938 return Z_OK; 939 } 940 beg = 0; 941 } 942 val = s->gzhead->name[s->gzindex++]; 943 put_byte(s, val); 944 } while (val != 0); 945 HCRC_UPDATE(beg); 946 s->gzindex = 0; 947 } 948 s->status = COMMENT_STATE; 949 } 950 if (s->status == COMMENT_STATE) { 951 if (s->gzhead->comment != Z_NULL) { 952 ulg beg = s->pending; /* start of bytes to update crc */ 953 int val; 954 do { 955 if (s->pending == s->pending_buf_size) { 956 HCRC_UPDATE(beg); 957 flush_pending(strm); 958 if (s->pending != 0) { 959 s->last_flush = -1; 960 return Z_OK; 961 } 962 beg = 0; 963 } 964 val = s->gzhead->comment[s->gzindex++]; 965 put_byte(s, val); 966 } while (val != 0); 967 HCRC_UPDATE(beg); 968 } 969 s->status = HCRC_STATE; 970 } 971 if (s->status == HCRC_STATE) { 972 if (s->gzhead->hcrc) { 973 if (s->pending + 2 > s->pending_buf_size) { 974 flush_pending(strm); 975 if (s->pending != 0) { 976 s->last_flush = -1; 977 return Z_OK; 978 } 979 } 980 put_byte(s, (Byte)(strm->adler & 0xff)); 981 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 982 strm->adler = crc32(0L, Z_NULL, 0); 983 } 984 s->status = BUSY_STATE; 985 986 /* Compression must start with an empty pending buffer */ 987 flush_pending(strm); 988 if (s->pending != 0) { 989 s->last_flush = -1; 990 return Z_OK; 991 } 992 } 993 #endif 994 995 /* Start a new block or continue the current one. 996 */ 997 if (strm->avail_in != 0 || s->lookahead != 0 || 998 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { 999 block_state bstate; 1000 1001 bstate = s->level == 0 ? deflate_stored(s, flush) : 1002 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : 1003 s->strategy == Z_RLE ? deflate_rle(s, flush) : 1004 (*(configuration_table[s->level].func))(s, flush); 1005 1006 if (bstate == finish_started || bstate == finish_done) { 1007 s->status = FINISH_STATE; 1008 } 1009 if (bstate == need_more || bstate == finish_started) { 1010 if (strm->avail_out == 0) { 1011 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ 1012 } 1013 return Z_OK; 1014 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call 1015 * of deflate should use the same flush parameter to make sure 1016 * that the flush is complete. So we don't have to output an 1017 * empty block here, this will be done at next call. This also 1018 * ensures that for a very small output buffer, we emit at most 1019 * one empty block. 1020 */ 1021 } 1022 if (bstate == block_done) { 1023 if (flush == Z_PARTIAL_FLUSH) { 1024 _tr_align(s); 1025 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */ 1026 _tr_stored_block(s, (char*)0, 0L, 0); 1027 /* For a full flush, this empty block will be recognized 1028 * as a special marker by inflate_sync(). 1029 */ 1030 if (flush == Z_FULL_FLUSH) { 1031 CLEAR_HASH(s); /* forget history */ 1032 if (s->lookahead == 0) { 1033 s->strstart = 0; 1034 s->block_start = 0L; 1035 s->insert = 0; 1036 } 1037 } 1038 } 1039 flush_pending(strm); 1040 if (strm->avail_out == 0) { 1041 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ 1042 return Z_OK; 1043 } 1044 } 1045 } 1046 1047 if (flush != Z_FINISH) return Z_OK; 1048 if (s->wrap <= 0) return Z_STREAM_END; 1049 1050 /* Write the trailer */ 1051 #ifdef GZIP 1052 if (s->wrap == 2) { 1053 put_byte(s, (Byte)(strm->adler & 0xff)); 1054 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 1055 put_byte(s, (Byte)((strm->adler >> 16) & 0xff)); 1056 put_byte(s, (Byte)((strm->adler >> 24) & 0xff)); 1057 put_byte(s, (Byte)(strm->total_in & 0xff)); 1058 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff)); 1059 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff)); 1060 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff)); 1061 } 1062 else 1063 #endif 1064 { 1065 putShortMSB(s, (uInt)(strm->adler >> 16)); 1066 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 1067 } 1068 flush_pending(strm); 1069 /* If avail_out is zero, the application will call deflate again 1070 * to flush the rest. 1071 */ 1072 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ 1073 return s->pending != 0 ? Z_OK : Z_STREAM_END; 1074 } 1075 1076 /* ========================================================================= */ 1077 int ZEXPORT deflateEnd (strm) 1078 z_streamp strm; 1079 { 1080 int status; 1081 1082 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 1083 1084 status = strm->state->status; 1085 1086 /* Deallocate in reverse order of allocations: */ 1087 TRY_FREE(strm, strm->state->pending_buf); 1088 TRY_FREE(strm, strm->state->head); 1089 TRY_FREE(strm, strm->state->prev); 1090 TRY_FREE(strm, strm->state->window); 1091 1092 ZFREE(strm, strm->state); 1093 strm->state = Z_NULL; 1094 1095 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; 1096 } 1097 1098 /* ========================================================================= 1099 * Copy the source state to the destination state. 1100 * To simplify the source, this is not supported for 16-bit MSDOS (which 1101 * doesn't have enough memory anyway to duplicate compression states). 1102 */ 1103 int ZEXPORT deflateCopy (dest, source) 1104 z_streamp dest; 1105 z_streamp source; 1106 { 1107 #ifdef MAXSEG_64K 1108 return Z_STREAM_ERROR; 1109 #else 1110 deflate_state *ds; 1111 deflate_state *ss; 1112 ushf *overlay; 1113 1114 1115 if (deflateStateCheck(source) || dest == Z_NULL) { 1116 return Z_STREAM_ERROR; 1117 } 1118 1119 ss = source->state; 1120 1121 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream)); 1122 1123 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); 1124 if (ds == Z_NULL) return Z_MEM_ERROR; 1125 dest->state = (struct internal_state FAR *) ds; 1126 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state)); 1127 ds->strm = dest; 1128 1129 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); 1130 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); 1131 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); 1132 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2); 1133 ds->pending_buf = (uchf *) overlay; 1134 1135 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || 1136 ds->pending_buf == Z_NULL) { 1137 deflateEnd (dest); 1138 return Z_MEM_ERROR; 1139 } 1140 /* following zmemcpy do not work for 16-bit MSDOS */ 1141 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); 1142 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos)); 1143 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos)); 1144 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); 1145 1146 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); 1147 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); 1148 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; 1149 1150 ds->l_desc.dyn_tree = ds->dyn_ltree; 1151 ds->d_desc.dyn_tree = ds->dyn_dtree; 1152 ds->bl_desc.dyn_tree = ds->bl_tree; 1153 1154 return Z_OK; 1155 #endif /* MAXSEG_64K */ 1156 } 1157 1158 /* =========================================================================== 1159 * Read a new buffer from the current input stream, update the adler32 1160 * and total number of bytes read. All deflate() input goes through 1161 * this function so some applications may wish to modify it to avoid 1162 * allocating a large strm->next_in buffer and copying from it. 1163 * (See also flush_pending()). 1164 */ 1165 local unsigned read_buf(strm, buf, size) 1166 z_streamp strm; 1167 Bytef *buf; 1168 unsigned size; 1169 { 1170 unsigned len = strm->avail_in; 1171 1172 if (len > size) len = size; 1173 if (len == 0) return 0; 1174 1175 strm->avail_in -= len; 1176 1177 zmemcpy(buf, strm->next_in, len); 1178 if (strm->state->wrap == 1) { 1179 strm->adler = adler32(strm->adler, buf, len); 1180 } 1181 #ifdef GZIP 1182 else if (strm->state->wrap == 2) { 1183 strm->adler = crc32(strm->adler, buf, len); 1184 } 1185 #endif 1186 strm->next_in += len; 1187 strm->total_in += len; 1188 1189 return len; 1190 } 1191 1192 /* =========================================================================== 1193 * Initialize the "longest match" routines for a new zlib stream 1194 */ 1195 local void lm_init (s) 1196 deflate_state *s; 1197 { 1198 s->window_size = (ulg)2L*s->w_size; 1199 1200 CLEAR_HASH(s); 1201 1202 /* Set the default configuration parameters: 1203 */ 1204 s->max_lazy_match = configuration_table[s->level].max_lazy; 1205 s->good_match = configuration_table[s->level].good_length; 1206 s->nice_match = configuration_table[s->level].nice_length; 1207 s->max_chain_length = configuration_table[s->level].max_chain; 1208 1209 s->strstart = 0; 1210 s->block_start = 0L; 1211 s->lookahead = 0; 1212 s->insert = 0; 1213 s->match_length = s->prev_length = MIN_MATCH-1; 1214 s->match_available = 0; 1215 s->ins_h = 0; 1216 #ifndef FASTEST 1217 #ifdef ASMV 1218 match_init(); /* initialize the asm code */ 1219 #endif 1220 #endif 1221 } 1222 1223 #ifndef FASTEST 1224 /* =========================================================================== 1225 * Set match_start to the longest match starting at the given string and 1226 * return its length. Matches shorter or equal to prev_length are discarded, 1227 * in which case the result is equal to prev_length and match_start is 1228 * garbage. 1229 * IN assertions: cur_match is the head of the hash chain for the current 1230 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 1231 * OUT assertion: the match length is not greater than s->lookahead. 1232 */ 1233 #ifndef ASMV 1234 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or 1235 * match.S. The code will be functionally equivalent. 1236 */ 1237 local uInt longest_match(s, cur_match) 1238 deflate_state *s; 1239 IPos cur_match; /* current match */ 1240 { 1241 unsigned chain_length = s->max_chain_length;/* max hash chain length */ 1242 register Bytef *scan = s->window + s->strstart; /* current string */ 1243 register Bytef *match; /* matched string */ 1244 register int len; /* length of current match */ 1245 int best_len = (int)s->prev_length; /* best match length so far */ 1246 int nice_match = s->nice_match; /* stop if match long enough */ 1247 IPos limit = s->strstart > (IPos)MAX_DIST(s) ? 1248 s->strstart - (IPos)MAX_DIST(s) : NIL; 1249 /* Stop when cur_match becomes <= limit. To simplify the code, 1250 * we prevent matches with the string of window index 0. 1251 */ 1252 Posf *prev = s->prev; 1253 uInt wmask = s->w_mask; 1254 1255 #ifdef UNALIGNED_OK 1256 /* Compare two bytes at a time. Note: this is not always beneficial. 1257 * Try with and without -DUNALIGNED_OK to check. 1258 */ 1259 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; 1260 register ush scan_start = *(ushf*)scan; 1261 register ush scan_end = *(ushf*)(scan+best_len-1); 1262 #else 1263 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1264 register Byte scan_end1 = scan[best_len-1]; 1265 register Byte scan_end = scan[best_len]; 1266 #endif 1267 1268 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1269 * It is easy to get rid of this optimization if necessary. 1270 */ 1271 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1272 1273 /* Do not waste too much time if we already have a good match: */ 1274 if (s->prev_length >= s->good_match) { 1275 chain_length >>= 2; 1276 } 1277 /* Do not look for matches beyond the end of the input. This is necessary 1278 * to make deflate deterministic. 1279 */ 1280 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead; 1281 1282 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); 1283 1284 do { 1285 Assert(cur_match < s->strstart, "no future"); 1286 match = s->window + cur_match; 1287 1288 /* Skip to next match if the match length cannot increase 1289 * or if the match length is less than 2. Note that the checks below 1290 * for insufficient lookahead only occur occasionally for performance 1291 * reasons. Therefore uninitialized memory will be accessed, and 1292 * conditional jumps will be made that depend on those values. 1293 * However the length of the match is limited to the lookahead, so 1294 * the output of deflate is not affected by the uninitialized values. 1295 */ 1296 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) 1297 /* This code assumes sizeof(unsigned short) == 2. Do not use 1298 * UNALIGNED_OK if your compiler uses a different size. 1299 */ 1300 if (*(ushf*)(match+best_len-1) != scan_end || 1301 *(ushf*)match != scan_start) continue; 1302 1303 /* It is not necessary to compare scan[2] and match[2] since they are 1304 * always equal when the other bytes match, given that the hash keys 1305 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at 1306 * strstart+3, +5, ... up to strstart+257. We check for insufficient 1307 * lookahead only every 4th comparison; the 128th check will be made 1308 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is 1309 * necessary to put more guard bytes at the end of the window, or 1310 * to check more often for insufficient lookahead. 1311 */ 1312 Assert(scan[2] == match[2], "scan[2]?"); 1313 scan++, match++; 1314 do { 1315 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1316 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1317 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1318 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1319 scan < strend); 1320 /* The funny "do {}" generates better code on most compilers */ 1321 1322 /* Here, scan <= window+strstart+257 */ 1323 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1324 if (*scan == *match) scan++; 1325 1326 len = (MAX_MATCH - 1) - (int)(strend-scan); 1327 scan = strend - (MAX_MATCH-1); 1328 1329 #else /* UNALIGNED_OK */ 1330 1331 if (match[best_len] != scan_end || 1332 match[best_len-1] != scan_end1 || 1333 *match != *scan || 1334 *++match != scan[1]) continue; 1335 1336 /* The check at best_len-1 can be removed because it will be made 1337 * again later. (This heuristic is not always a win.) 1338 * It is not necessary to compare scan[2] and match[2] since they 1339 * are always equal when the other bytes match, given that 1340 * the hash keys are equal and that HASH_BITS >= 8. 1341 */ 1342 scan += 2, match++; 1343 Assert(*scan == *match, "match[2]?"); 1344 1345 /* We check for insufficient lookahead only every 8th comparison; 1346 * the 256th check will be made at strstart+258. 1347 */ 1348 do { 1349 } while (*++scan == *++match && *++scan == *++match && 1350 *++scan == *++match && *++scan == *++match && 1351 *++scan == *++match && *++scan == *++match && 1352 *++scan == *++match && *++scan == *++match && 1353 scan < strend); 1354 1355 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1356 1357 len = MAX_MATCH - (int)(strend - scan); 1358 scan = strend - MAX_MATCH; 1359 1360 #endif /* UNALIGNED_OK */ 1361 1362 if (len > best_len) { 1363 s->match_start = cur_match; 1364 best_len = len; 1365 if (len >= nice_match) break; 1366 #ifdef UNALIGNED_OK 1367 scan_end = *(ushf*)(scan+best_len-1); 1368 #else 1369 scan_end1 = scan[best_len-1]; 1370 scan_end = scan[best_len]; 1371 #endif 1372 } 1373 } while ((cur_match = prev[cur_match & wmask]) > limit 1374 && --chain_length != 0); 1375 1376 if ((uInt)best_len <= s->lookahead) return (uInt)best_len; 1377 return s->lookahead; 1378 } 1379 #endif /* ASMV */ 1380 1381 #else /* FASTEST */ 1382 1383 /* --------------------------------------------------------------------------- 1384 * Optimized version for FASTEST only 1385 */ 1386 local uInt longest_match(s, cur_match) 1387 deflate_state *s; 1388 IPos cur_match; /* current match */ 1389 { 1390 register Bytef *scan = s->window + s->strstart; /* current string */ 1391 register Bytef *match; /* matched string */ 1392 register int len; /* length of current match */ 1393 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1394 1395 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1396 * It is easy to get rid of this optimization if necessary. 1397 */ 1398 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1399 1400 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); 1401 1402 Assert(cur_match < s->strstart, "no future"); 1403 1404 match = s->window + cur_match; 1405 1406 /* Return failure if the match length is less than 2: 1407 */ 1408 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; 1409 1410 /* The check at best_len-1 can be removed because it will be made 1411 * again later. (This heuristic is not always a win.) 1412 * It is not necessary to compare scan[2] and match[2] since they 1413 * are always equal when the other bytes match, given that 1414 * the hash keys are equal and that HASH_BITS >= 8. 1415 */ 1416 scan += 2, match += 2; 1417 Assert(*scan == *match, "match[2]?"); 1418 1419 /* We check for insufficient lookahead only every 8th comparison; 1420 * the 256th check will be made at strstart+258. 1421 */ 1422 do { 1423 } while (*++scan == *++match && *++scan == *++match && 1424 *++scan == *++match && *++scan == *++match && 1425 *++scan == *++match && *++scan == *++match && 1426 *++scan == *++match && *++scan == *++match && 1427 scan < strend); 1428 1429 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1430 1431 len = MAX_MATCH - (int)(strend - scan); 1432 1433 if (len < MIN_MATCH) return MIN_MATCH - 1; 1434 1435 s->match_start = cur_match; 1436 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead; 1437 } 1438 1439 #endif /* FASTEST */ 1440 1441 #ifdef ZLIB_DEBUG 1442 1443 #define EQUAL 0 1444 /* result of memcmp for equal strings */ 1445 1446 /* =========================================================================== 1447 * Check that the match at match_start is indeed a match. 1448 */ 1449 local void check_match(s, start, match, length) 1450 deflate_state *s; 1451 IPos start, match; 1452 int length; 1453 { 1454 /* check that the match is indeed a match */ 1455 if (zmemcmp(s->window + match, 1456 s->window + start, length) != EQUAL) { 1457 fprintf(stderr, " start %u, match %u, length %d\n", 1458 start, match, length); 1459 do { 1460 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); 1461 } while (--length != 0); 1462 z_error("invalid match"); 1463 } 1464 if (z_verbose > 1) { 1465 fprintf(stderr,"\\[%d,%d]", start-match, length); 1466 do { putc(s->window[start++], stderr); } while (--length != 0); 1467 } 1468 } 1469 #else 1470 # define check_match(s, start, match, length) 1471 #endif /* ZLIB_DEBUG */ 1472 1473 /* =========================================================================== 1474 * Fill the window when the lookahead becomes insufficient. 1475 * Updates strstart and lookahead. 1476 * 1477 * IN assertion: lookahead < MIN_LOOKAHEAD 1478 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD 1479 * At least one byte has been read, or avail_in == 0; reads are 1480 * performed for at least two bytes (required for the zip translate_eol 1481 * option -- not supported here). 1482 */ 1483 local void fill_window(s) 1484 deflate_state *s; 1485 { 1486 unsigned n; 1487 unsigned more; /* Amount of free space at the end of the window. */ 1488 uInt wsize = s->w_size; 1489 1490 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead"); 1491 1492 do { 1493 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); 1494 1495 /* Deal with !@#$% 64K limit: */ 1496 if (sizeof(int) <= 2) { 1497 if (more == 0 && s->strstart == 0 && s->lookahead == 0) { 1498 more = wsize; 1499 1500 } else if (more == (unsigned)(-1)) { 1501 /* Very unlikely, but possible on 16 bit machine if 1502 * strstart == 0 && lookahead == 1 (input done a byte at time) 1503 */ 1504 more--; 1505 } 1506 } 1507 1508 /* If the window is almost full and there is insufficient lookahead, 1509 * move the upper half to the lower one to make room in the upper half. 1510 */ 1511 if (s->strstart >= wsize+MAX_DIST(s)) { 1512 1513 zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more); 1514 s->match_start -= wsize; 1515 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ 1516 s->block_start -= (long) wsize; 1517 slide_hash(s); 1518 more += wsize; 1519 } 1520 if (s->strm->avail_in == 0) break; 1521 1522 /* If there was no sliding: 1523 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && 1524 * more == window_size - lookahead - strstart 1525 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) 1526 * => more >= window_size - 2*WSIZE + 2 1527 * In the BIG_MEM or MMAP case (not yet supported), 1528 * window_size == input_size + MIN_LOOKAHEAD && 1529 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. 1530 * Otherwise, window_size == 2*WSIZE so more >= 2. 1531 * If there was sliding, more >= WSIZE. So in all cases, more >= 2. 1532 */ 1533 Assert(more >= 2, "more < 2"); 1534 1535 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); 1536 s->lookahead += n; 1537 1538 /* Initialize the hash value now that we have some input: */ 1539 if (s->lookahead + s->insert >= MIN_MATCH) { 1540 uInt str = s->strstart - s->insert; 1541 s->ins_h = s->window[str]; 1542 UPDATE_HASH(s, s->ins_h, s->window[str + 1]); 1543 #if MIN_MATCH != 3 1544 Call UPDATE_HASH() MIN_MATCH-3 more times 1545 #endif 1546 while (s->insert) { 1547 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 1548 #ifndef FASTEST 1549 s->prev[str & s->w_mask] = s->head[s->ins_h]; 1550 #endif 1551 s->head[s->ins_h] = (Pos)str; 1552 str++; 1553 s->insert--; 1554 if (s->lookahead + s->insert < MIN_MATCH) 1555 break; 1556 } 1557 } 1558 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, 1559 * but this is not important since only literal bytes will be emitted. 1560 */ 1561 1562 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); 1563 1564 /* If the WIN_INIT bytes after the end of the current data have never been 1565 * written, then zero those bytes in order to avoid memory check reports of 1566 * the use of uninitialized (or uninitialised as Julian writes) bytes by 1567 * the longest match routines. Update the high water mark for the next 1568 * time through here. WIN_INIT is set to MAX_MATCH since the longest match 1569 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead. 1570 */ 1571 if (s->high_water < s->window_size) { 1572 ulg curr = s->strstart + (ulg)(s->lookahead); 1573 ulg init; 1574 1575 if (s->high_water < curr) { 1576 /* Previous high water mark below current data -- zero WIN_INIT 1577 * bytes or up to end of window, whichever is less. 1578 */ 1579 init = s->window_size - curr; 1580 if (init > WIN_INIT) 1581 init = WIN_INIT; 1582 zmemzero(s->window + curr, (unsigned)init); 1583 s->high_water = curr + init; 1584 } 1585 else if (s->high_water < (ulg)curr + WIN_INIT) { 1586 /* High water mark at or above current data, but below current data 1587 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up 1588 * to end of window, whichever is less. 1589 */ 1590 init = (ulg)curr + WIN_INIT - s->high_water; 1591 if (init > s->window_size - s->high_water) 1592 init = s->window_size - s->high_water; 1593 zmemzero(s->window + s->high_water, (unsigned)init); 1594 s->high_water += init; 1595 } 1596 } 1597 1598 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 1599 "not enough room for search"); 1600 } 1601 1602 /* =========================================================================== 1603 * Flush the current block, with given end-of-file flag. 1604 * IN assertion: strstart is set to the end of the current match. 1605 */ 1606 #define FLUSH_BLOCK_ONLY(s, last) { \ 1607 _tr_flush_block(s, (s->block_start >= 0L ? \ 1608 (charf *)&s->window[(unsigned)s->block_start] : \ 1609 (charf *)Z_NULL), \ 1610 (ulg)((long)s->strstart - s->block_start), \ 1611 (last)); \ 1612 s->block_start = s->strstart; \ 1613 flush_pending(s->strm); \ 1614 Tracev((stderr,"[FLUSH]")); \ 1615 } 1616 1617 /* Same but force premature exit if necessary. */ 1618 #define FLUSH_BLOCK(s, last) { \ 1619 FLUSH_BLOCK_ONLY(s, last); \ 1620 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \ 1621 } 1622 1623 /* Maximum stored block length in deflate format (not including header). */ 1624 #define MAX_STORED 65535 1625 1626 /* Minimum of a and b. */ 1627 #define MIN(a, b) ((a) > (b) ? (b) : (a)) 1628 1629 /* =========================================================================== 1630 * Copy without compression as much as possible from the input stream, return 1631 * the current block state. 1632 * 1633 * In case deflateParams() is used to later switch to a non-zero compression 1634 * level, s->matches (otherwise unused when storing) keeps track of the number 1635 * of hash table slides to perform. If s->matches is 1, then one hash table 1636 * slide will be done when switching. If s->matches is 2, the maximum value 1637 * allowed here, then the hash table will be cleared, since two or more slides 1638 * is the same as a clear. 1639 * 1640 * deflate_stored() is written to minimize the number of times an input byte is 1641 * copied. It is most efficient with large input and output buffers, which 1642 * maximizes the opportunites to have a single copy from next_in to next_out. 1643 */ 1644 local block_state deflate_stored(s, flush) 1645 deflate_state *s; 1646 int flush; 1647 { 1648 /* Smallest worthy block size when not flushing or finishing. By default 1649 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For 1650 * large input and output buffers, the stored block size will be larger. 1651 */ 1652 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size); 1653 1654 /* Copy as many min_block or larger stored blocks directly to next_out as 1655 * possible. If flushing, copy the remaining available input to next_out as 1656 * stored blocks, if there is enough space. 1657 */ 1658 unsigned len, left, have, last = 0; 1659 unsigned used = s->strm->avail_in; 1660 do { 1661 /* Set len to the maximum size block that we can copy directly with the 1662 * available input data and output space. Set left to how much of that 1663 * would be copied from what's left in the window. 1664 */ 1665 len = MAX_STORED; /* maximum deflate stored block length */ 1666 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1667 if (s->strm->avail_out < have) /* need room for header */ 1668 break; 1669 /* maximum stored block length that will fit in avail_out: */ 1670 have = s->strm->avail_out - have; 1671 left = s->strstart - s->block_start; /* bytes left in window */ 1672 if (len > (ulg)left + s->strm->avail_in) 1673 len = left + s->strm->avail_in; /* limit len to the input */ 1674 if (len > have) 1675 len = have; /* limit len to the output */ 1676 if (left > len) 1677 left = len; /* limit window pull to len */ 1678 1679 /* If the stored block would be less than min_block in length, or if 1680 * unable to copy all of the available input when flushing, then try 1681 * copying to the window and the pending buffer instead. Also don't 1682 * write an empty block when flushing -- deflate() does that. 1683 */ 1684 if (len < min_block && ((len == 0 && flush != Z_FINISH) || 1685 flush == Z_NO_FLUSH || 1686 len - left != s->strm->avail_in)) 1687 break; 1688 1689 /* Make a dummy stored block in pending to get the header bytes, 1690 * including any pending bits. This also updates the debugging counts. 1691 */ 1692 last = flush == Z_FINISH && len - left == s->strm->avail_in ? 1 : 0; 1693 _tr_stored_block(s, (char *)0, 0L, last); 1694 1695 /* Replace the lengths in the dummy stored block with len. */ 1696 s->pending_buf[s->pending - 4] = len; 1697 s->pending_buf[s->pending - 3] = len >> 8; 1698 s->pending_buf[s->pending - 2] = ~len; 1699 s->pending_buf[s->pending - 1] = ~len >> 8; 1700 1701 /* Write the stored block header bytes. */ 1702 flush_pending(s->strm); 1703 1704 /* Update debugging counts for the data about to be copied. */ 1705 #ifdef ZLIB_DEBUG 1706 s->compressed_len += len << 3; 1707 s->bits_sent += len << 3; 1708 #endif 1709 1710 /* Copy uncompressed bytes from the window to next_out. */ 1711 if (left) { 1712 zmemcpy(s->strm->next_out, s->window + s->block_start, left); 1713 s->strm->next_out += left; 1714 s->strm->avail_out -= left; 1715 s->strm->total_out += left; 1716 s->block_start += left; 1717 len -= left; 1718 } 1719 1720 /* Copy uncompressed bytes directly from next_in to next_out, updating 1721 * the check value. 1722 */ 1723 if (len) { 1724 read_buf(s->strm, s->strm->next_out, len); 1725 s->strm->next_out += len; 1726 s->strm->avail_out -= len; 1727 s->strm->total_out += len; 1728 } 1729 } while (last == 0); 1730 1731 /* Update the sliding window with the last s->w_size bytes of the copied 1732 * data, or append all of the copied data to the existing window if less 1733 * than s->w_size bytes were copied. Also update the number of bytes to 1734 * insert in the hash tables, in the event that deflateParams() switches to 1735 * a non-zero compression level. 1736 */ 1737 used -= s->strm->avail_in; /* number of input bytes directly copied */ 1738 if (used) { 1739 /* If any input was used, then no unused input remains in the window, 1740 * therefore s->block_start == s->strstart. 1741 */ 1742 if (used >= s->w_size) { /* supplant the previous history */ 1743 s->matches = 2; /* clear hash */ 1744 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size); 1745 s->strstart = s->w_size; 1746 } 1747 else { 1748 if (s->window_size - s->strstart <= used) { 1749 /* Slide the window down. */ 1750 s->strstart -= s->w_size; 1751 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1752 if (s->matches < 2) 1753 s->matches++; /* add a pending slide_hash() */ 1754 } 1755 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used); 1756 s->strstart += used; 1757 } 1758 s->block_start = s->strstart; 1759 s->insert += MIN(used, s->w_size - s->insert); 1760 } 1761 1762 /* If the last block was written to next_out, then done. */ 1763 if (last) 1764 return finish_done; 1765 1766 /* If flushing and all input has been consumed, then done. */ 1767 if (flush != Z_NO_FLUSH && flush != Z_FINISH && 1768 s->strm->avail_in == 0 && (long)s->strstart == s->block_start) 1769 return block_done; 1770 1771 /* Fill the window with any remaining input. */ 1772 have = s->window_size - s->strstart - 1; 1773 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) { 1774 /* Slide the window down. */ 1775 s->block_start -= s->w_size; 1776 s->strstart -= s->w_size; 1777 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1778 if (s->matches < 2) 1779 s->matches++; /* add a pending slide_hash() */ 1780 have += s->w_size; /* more space now */ 1781 } 1782 if (have > s->strm->avail_in) 1783 have = s->strm->avail_in; 1784 if (have) { 1785 read_buf(s->strm, s->window + s->strstart, have); 1786 s->strstart += have; 1787 } 1788 1789 /* There was not enough avail_out to write a complete worthy or flushed 1790 * stored block to next_out. Write a stored block to pending instead, if we 1791 * have enough input for a worthy block, or if flushing and there is enough 1792 * room for the remaining input as a stored block in the pending buffer. 1793 */ 1794 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1795 /* maximum stored block length that will fit in pending: */ 1796 have = MIN(s->pending_buf_size - have, MAX_STORED); 1797 min_block = MIN(have, s->w_size); 1798 left = s->strstart - s->block_start; 1799 if (left >= min_block || 1800 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH && 1801 s->strm->avail_in == 0 && left <= have)) { 1802 len = MIN(left, have); 1803 last = flush == Z_FINISH && s->strm->avail_in == 0 && 1804 len == left ? 1 : 0; 1805 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last); 1806 s->block_start += len; 1807 flush_pending(s->strm); 1808 } 1809 1810 /* We've done all we can with the available input and output. */ 1811 return last ? finish_started : need_more; 1812 } 1813 1814 /* =========================================================================== 1815 * Compress as much as possible from the input stream, return the current 1816 * block state. 1817 * This function does not perform lazy evaluation of matches and inserts 1818 * new strings in the dictionary only for unmatched strings or for short 1819 * matches. It is used only for the fast compression options. 1820 */ 1821 local block_state deflate_fast(s, flush) 1822 deflate_state *s; 1823 int flush; 1824 { 1825 IPos hash_head; /* head of the hash chain */ 1826 int bflush; /* set if current block must be flushed */ 1827 1828 for (;;) { 1829 /* Make sure that we always have enough lookahead, except 1830 * at the end of the input file. We need MAX_MATCH bytes 1831 * for the next match, plus MIN_MATCH bytes to insert the 1832 * string following the next match. 1833 */ 1834 if (s->lookahead < MIN_LOOKAHEAD) { 1835 fill_window(s); 1836 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1837 return need_more; 1838 } 1839 if (s->lookahead == 0) break; /* flush the current block */ 1840 } 1841 1842 /* Insert the string window[strstart .. strstart+2] in the 1843 * dictionary, and set hash_head to the head of the hash chain: 1844 */ 1845 hash_head = NIL; 1846 if (s->lookahead >= MIN_MATCH) { 1847 INSERT_STRING(s, s->strstart, hash_head); 1848 } 1849 1850 /* Find the longest match, discarding those <= prev_length. 1851 * At this point we have always match_length < MIN_MATCH 1852 */ 1853 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { 1854 /* To simplify the code, we prevent matches with the string 1855 * of window index 0 (in particular we have to avoid a match 1856 * of the string with itself at the start of the input file). 1857 */ 1858 s->match_length = longest_match (s, hash_head); 1859 /* longest_match() sets match_start */ 1860 } 1861 if (s->match_length >= MIN_MATCH) { 1862 check_match(s, s->strstart, s->match_start, s->match_length); 1863 1864 _tr_tally_dist(s, s->strstart - s->match_start, 1865 s->match_length - MIN_MATCH, bflush); 1866 1867 s->lookahead -= s->match_length; 1868 1869 /* Insert new strings in the hash table only if the match length 1870 * is not too large. This saves time but degrades compression. 1871 */ 1872 #ifndef FASTEST 1873 if (s->match_length <= s->max_insert_length && 1874 s->lookahead >= MIN_MATCH) { 1875 s->match_length--; /* string at strstart already in table */ 1876 do { 1877 s->strstart++; 1878 INSERT_STRING(s, s->strstart, hash_head); 1879 /* strstart never exceeds WSIZE-MAX_MATCH, so there are 1880 * always MIN_MATCH bytes ahead. 1881 */ 1882 } while (--s->match_length != 0); 1883 s->strstart++; 1884 } else 1885 #endif 1886 { 1887 s->strstart += s->match_length; 1888 s->match_length = 0; 1889 s->ins_h = s->window[s->strstart]; 1890 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); 1891 #if MIN_MATCH != 3 1892 Call UPDATE_HASH() MIN_MATCH-3 more times 1893 #endif 1894 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not 1895 * matter since it will be recomputed at next deflate call. 1896 */ 1897 } 1898 } else { 1899 /* No match, output a literal byte */ 1900 Tracevv((stderr,"%c", s->window[s->strstart])); 1901 _tr_tally_lit (s, s->window[s->strstart], bflush); 1902 s->lookahead--; 1903 s->strstart++; 1904 } 1905 if (bflush) FLUSH_BLOCK(s, 0); 1906 } 1907 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 1908 if (flush == Z_FINISH) { 1909 FLUSH_BLOCK(s, 1); 1910 return finish_done; 1911 } 1912 if (s->last_lit) 1913 FLUSH_BLOCK(s, 0); 1914 return block_done; 1915 } 1916 1917 #ifndef FASTEST 1918 /* =========================================================================== 1919 * Same as above, but achieves better compression. We use a lazy 1920 * evaluation for matches: a match is finally adopted only if there is 1921 * no better match at the next window position. 1922 */ 1923 local block_state deflate_slow(s, flush) 1924 deflate_state *s; 1925 int flush; 1926 { 1927 IPos hash_head; /* head of hash chain */ 1928 int bflush; /* set if current block must be flushed */ 1929 1930 /* Process the input block. */ 1931 for (;;) { 1932 /* Make sure that we always have enough lookahead, except 1933 * at the end of the input file. We need MAX_MATCH bytes 1934 * for the next match, plus MIN_MATCH bytes to insert the 1935 * string following the next match. 1936 */ 1937 if (s->lookahead < MIN_LOOKAHEAD) { 1938 fill_window(s); 1939 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1940 return need_more; 1941 } 1942 if (s->lookahead == 0) break; /* flush the current block */ 1943 } 1944 1945 /* Insert the string window[strstart .. strstart+2] in the 1946 * dictionary, and set hash_head to the head of the hash chain: 1947 */ 1948 hash_head = NIL; 1949 if (s->lookahead >= MIN_MATCH) { 1950 INSERT_STRING(s, s->strstart, hash_head); 1951 } 1952 1953 /* Find the longest match, discarding those <= prev_length. 1954 */ 1955 s->prev_length = s->match_length, s->prev_match = s->match_start; 1956 s->match_length = MIN_MATCH-1; 1957 1958 if (hash_head != NIL && s->prev_length < s->max_lazy_match && 1959 s->strstart - hash_head <= MAX_DIST(s)) { 1960 /* To simplify the code, we prevent matches with the string 1961 * of window index 0 (in particular we have to avoid a match 1962 * of the string with itself at the start of the input file). 1963 */ 1964 s->match_length = longest_match (s, hash_head); 1965 /* longest_match() sets match_start */ 1966 1967 if (s->match_length <= 5 && (s->strategy == Z_FILTERED 1968 #if TOO_FAR <= 32767 1969 || (s->match_length == MIN_MATCH && 1970 s->strstart - s->match_start > TOO_FAR) 1971 #endif 1972 )) { 1973 1974 /* If prev_match is also MIN_MATCH, match_start is garbage 1975 * but we will ignore the current match anyway. 1976 */ 1977 s->match_length = MIN_MATCH-1; 1978 } 1979 } 1980 /* If there was a match at the previous step and the current 1981 * match is not better, output the previous match: 1982 */ 1983 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { 1984 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; 1985 /* Do not insert strings in hash table beyond this. */ 1986 1987 check_match(s, s->strstart-1, s->prev_match, s->prev_length); 1988 1989 _tr_tally_dist(s, s->strstart -1 - s->prev_match, 1990 s->prev_length - MIN_MATCH, bflush); 1991 1992 /* Insert in hash table all strings up to the end of the match. 1993 * strstart-1 and strstart are already inserted. If there is not 1994 * enough lookahead, the last two strings are not inserted in 1995 * the hash table. 1996 */ 1997 s->lookahead -= s->prev_length-1; 1998 s->prev_length -= 2; 1999 do { 2000 if (++s->strstart <= max_insert) { 2001 INSERT_STRING(s, s->strstart, hash_head); 2002 } 2003 } while (--s->prev_length != 0); 2004 s->match_available = 0; 2005 s->match_length = MIN_MATCH-1; 2006 s->strstart++; 2007 2008 if (bflush) FLUSH_BLOCK(s, 0); 2009 2010 } else if (s->match_available) { 2011 /* If there was no match at the previous position, output a 2012 * single literal. If there was a match but the current match 2013 * is longer, truncate the previous match to a single literal. 2014 */ 2015 Tracevv((stderr,"%c", s->window[s->strstart-1])); 2016 _tr_tally_lit(s, s->window[s->strstart-1], bflush); 2017 if (bflush) { 2018 FLUSH_BLOCK_ONLY(s, 0); 2019 } 2020 s->strstart++; 2021 s->lookahead--; 2022 if (s->strm->avail_out == 0) return need_more; 2023 } else { 2024 /* There is no previous match to compare with, wait for 2025 * the next step to decide. 2026 */ 2027 s->match_available = 1; 2028 s->strstart++; 2029 s->lookahead--; 2030 } 2031 } 2032 Assert (flush != Z_NO_FLUSH, "no flush?"); 2033 if (s->match_available) { 2034 Tracevv((stderr,"%c", s->window[s->strstart-1])); 2035 _tr_tally_lit(s, s->window[s->strstart-1], bflush); 2036 s->match_available = 0; 2037 } 2038 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 2039 if (flush == Z_FINISH) { 2040 FLUSH_BLOCK(s, 1); 2041 return finish_done; 2042 } 2043 if (s->last_lit) 2044 FLUSH_BLOCK(s, 0); 2045 return block_done; 2046 } 2047 #endif /* FASTEST */ 2048 2049 /* =========================================================================== 2050 * For Z_RLE, simply look for runs of bytes, generate matches only of distance 2051 * one. Do not maintain a hash table. (It will be regenerated if this run of 2052 * deflate switches away from Z_RLE.) 2053 */ 2054 local block_state deflate_rle(s, flush) 2055 deflate_state *s; 2056 int flush; 2057 { 2058 int bflush; /* set if current block must be flushed */ 2059 uInt prev; /* byte at distance one to match */ 2060 Bytef *scan, *strend; /* scan goes up to strend for length of run */ 2061 2062 for (;;) { 2063 /* Make sure that we always have enough lookahead, except 2064 * at the end of the input file. We need MAX_MATCH bytes 2065 * for the longest run, plus one for the unrolled loop. 2066 */ 2067 if (s->lookahead <= MAX_MATCH) { 2068 fill_window(s); 2069 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) { 2070 return need_more; 2071 } 2072 if (s->lookahead == 0) break; /* flush the current block */ 2073 } 2074 2075 /* See how many times the previous byte repeats */ 2076 s->match_length = 0; 2077 if (s->lookahead >= MIN_MATCH && s->strstart > 0) { 2078 scan = s->window + s->strstart - 1; 2079 prev = *scan; 2080 if (prev == *++scan && prev == *++scan && prev == *++scan) { 2081 strend = s->window + s->strstart + MAX_MATCH; 2082 do { 2083 } while (prev == *++scan && prev == *++scan && 2084 prev == *++scan && prev == *++scan && 2085 prev == *++scan && prev == *++scan && 2086 prev == *++scan && prev == *++scan && 2087 scan < strend); 2088 s->match_length = MAX_MATCH - (uInt)(strend - scan); 2089 if (s->match_length > s->lookahead) 2090 s->match_length = s->lookahead; 2091 } 2092 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan"); 2093 } 2094 2095 /* Emit match if have run of MIN_MATCH or longer, else emit literal */ 2096 if (s->match_length >= MIN_MATCH) { 2097 check_match(s, s->strstart, s->strstart - 1, s->match_length); 2098 2099 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush); 2100 2101 s->lookahead -= s->match_length; 2102 s->strstart += s->match_length; 2103 s->match_length = 0; 2104 } else { 2105 /* No match, output a literal byte */ 2106 Tracevv((stderr,"%c", s->window[s->strstart])); 2107 _tr_tally_lit (s, s->window[s->strstart], bflush); 2108 s->lookahead--; 2109 s->strstart++; 2110 } 2111 if (bflush) FLUSH_BLOCK(s, 0); 2112 } 2113 s->insert = 0; 2114 if (flush == Z_FINISH) { 2115 FLUSH_BLOCK(s, 1); 2116 return finish_done; 2117 } 2118 if (s->last_lit) 2119 FLUSH_BLOCK(s, 0); 2120 return block_done; 2121 } 2122 2123 /* =========================================================================== 2124 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table. 2125 * (It will be regenerated if this run of deflate switches away from Huffman.) 2126 */ 2127 local block_state deflate_huff(s, flush) 2128 deflate_state *s; 2129 int flush; 2130 { 2131 int bflush; /* set if current block must be flushed */ 2132 2133 for (;;) { 2134 /* Make sure that we have a literal to write. */ 2135 if (s->lookahead == 0) { 2136 fill_window(s); 2137 if (s->lookahead == 0) { 2138 if (flush == Z_NO_FLUSH) 2139 return need_more; 2140 break; /* flush the current block */ 2141 } 2142 } 2143 2144 /* Output a literal byte */ 2145 s->match_length = 0; 2146 Tracevv((stderr,"%c", s->window[s->strstart])); 2147 _tr_tally_lit (s, s->window[s->strstart], bflush); 2148 s->lookahead--; 2149 s->strstart++; 2150 if (bflush) FLUSH_BLOCK(s, 0); 2151 } 2152 s->insert = 0; 2153 if (flush == Z_FINISH) { 2154 FLUSH_BLOCK(s, 1); 2155 return finish_done; 2156 } 2157 if (s->last_lit) 2158 FLUSH_BLOCK(s, 0); 2159 return block_done; 2160 } 2161