1 /* deflate.c -- compress data using the deflation algorithm 2 * Copyright (C) 1995-2022 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 1.2.13 Copyright 1995-2022 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 local uInt longest_match OF((deflate_state *s, IPos cur_match)); 91 92 #ifdef ZLIB_DEBUG 93 local void check_match OF((deflate_state *s, IPos start, IPos match, 94 int length)); 95 #endif 96 97 /* =========================================================================== 98 * Local data 99 */ 100 101 #define NIL 0 102 /* Tail of hash chains */ 103 104 #ifndef TOO_FAR 105 # define TOO_FAR 4096 106 #endif 107 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ 108 109 /* Values for max_lazy_match, good_match and max_chain_length, depending on 110 * the desired pack level (0..9). The values given below have been tuned to 111 * exclude worst case performance for pathological files. Better values may be 112 * found for specific files. 113 */ 114 typedef struct config_s { 115 ush good_length; /* reduce lazy search above this match length */ 116 ush max_lazy; /* do not perform lazy search above this match length */ 117 ush nice_length; /* quit search above this match length */ 118 ush max_chain; 119 compress_func func; 120 } config; 121 122 #ifdef FASTEST 123 local const config configuration_table[2] = { 124 /* good lazy nice chain */ 125 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 126 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */ 127 #else 128 local const config configuration_table[10] = { 129 /* good lazy nice chain */ 130 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 131 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */ 132 /* 2 */ {4, 5, 16, 8, deflate_fast}, 133 /* 3 */ {4, 6, 32, 32, deflate_fast}, 134 135 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ 136 /* 5 */ {8, 16, 32, 32, deflate_slow}, 137 /* 6 */ {8, 16, 128, 128, deflate_slow}, 138 /* 7 */ {8, 32, 128, 256, deflate_slow}, 139 /* 8 */ {32, 128, 258, 1024, deflate_slow}, 140 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */ 141 #endif 142 143 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 144 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different 145 * meaning. 146 */ 147 148 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */ 149 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0)) 150 151 /* =========================================================================== 152 * Update a hash value with the given input byte 153 * IN assertion: all calls to UPDATE_HASH are made with consecutive input 154 * characters, so that a running hash key can be computed from the previous 155 * key instead of complete recalculation each time. 156 */ 157 #define UPDATE_HASH(s,h,c) (h = (((h) << s->hash_shift) ^ (c)) & s->hash_mask) 158 159 160 /* =========================================================================== 161 * Insert string str in the dictionary and set match_head to the previous head 162 * of the hash chain (the most recent string with same hash key). Return 163 * the previous length of the hash chain. 164 * If this file is compiled with -DFASTEST, the compression level is forced 165 * to 1, and no hash chains are maintained. 166 * IN assertion: all calls to INSERT_STRING are made with consecutive input 167 * characters and the first MIN_MATCH bytes of str are valid (except for 168 * the last MIN_MATCH-1 bytes of the input file). 169 */ 170 #ifdef FASTEST 171 #define INSERT_STRING(s, str, match_head) \ 172 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 173 match_head = s->head[s->ins_h], \ 174 s->head[s->ins_h] = (Pos)(str)) 175 #else 176 #define INSERT_STRING(s, str, match_head) \ 177 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 178 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \ 179 s->head[s->ins_h] = (Pos)(str)) 180 #endif 181 182 /* =========================================================================== 183 * Initialize the hash table (avoiding 64K overflow for 16 bit systems). 184 * prev[] will be initialized on the fly. 185 */ 186 #define CLEAR_HASH(s) \ 187 do { \ 188 s->head[s->hash_size - 1] = NIL; \ 189 zmemzero((Bytef *)s->head, \ 190 (unsigned)(s->hash_size - 1)*sizeof(*s->head)); \ 191 } while (0) 192 193 /* =========================================================================== 194 * Slide the hash table when sliding the window down (could be avoided with 32 195 * bit values at the expense of memory usage). We slide even when level == 0 to 196 * keep the hash table consistent if we switch back to level > 0 later. 197 */ 198 local void slide_hash(s) 199 deflate_state *s; 200 { 201 unsigned n, m; 202 Posf *p; 203 uInt wsize = s->w_size; 204 205 n = s->hash_size; 206 p = &s->head[n]; 207 do { 208 m = *--p; 209 *p = (Pos)(m >= wsize ? m - wsize : NIL); 210 } while (--n); 211 n = wsize; 212 #ifndef FASTEST 213 p = &s->prev[n]; 214 do { 215 m = *--p; 216 *p = (Pos)(m >= wsize ? m - wsize : NIL); 217 /* If n is not on any hash chain, prev[n] is garbage but 218 * its value will never be used. 219 */ 220 } while (--n); 221 #endif 222 } 223 224 /* ========================================================================= */ 225 int ZEXPORT deflateInit_(strm, level, version, stream_size) 226 z_streamp strm; 227 int level; 228 const char *version; 229 int stream_size; 230 { 231 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, 232 Z_DEFAULT_STRATEGY, version, stream_size); 233 /* To do: ignore strm->next_in if we use it as window */ 234 } 235 236 /* ========================================================================= */ 237 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, 238 version, stream_size) 239 z_streamp strm; 240 int level; 241 int method; 242 int windowBits; 243 int memLevel; 244 int strategy; 245 const char *version; 246 int stream_size; 247 { 248 deflate_state *s; 249 int wrap = 1; 250 static const char my_version[] = ZLIB_VERSION; 251 252 if (version == Z_NULL || version[0] != my_version[0] || 253 stream_size != sizeof(z_stream)) { 254 return Z_VERSION_ERROR; 255 } 256 if (strm == Z_NULL) return Z_STREAM_ERROR; 257 258 strm->msg = Z_NULL; 259 if (strm->zalloc == (alloc_func)0) { 260 #ifdef Z_SOLO 261 return Z_STREAM_ERROR; 262 #else 263 strm->zalloc = zcalloc; 264 strm->opaque = (voidpf)0; 265 #endif 266 } 267 if (strm->zfree == (free_func)0) 268 #ifdef Z_SOLO 269 return Z_STREAM_ERROR; 270 #else 271 strm->zfree = zcfree; 272 #endif 273 274 #ifdef FASTEST 275 if (level != 0) level = 1; 276 #else 277 if (level == Z_DEFAULT_COMPRESSION) level = 6; 278 #endif 279 280 if (windowBits < 0) { /* suppress zlib wrapper */ 281 wrap = 0; 282 if (windowBits < -15) 283 return Z_STREAM_ERROR; 284 windowBits = -windowBits; 285 } 286 #ifdef GZIP 287 else if (windowBits > 15) { 288 wrap = 2; /* write gzip wrapper instead */ 289 windowBits -= 16; 290 } 291 #endif 292 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || 293 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || 294 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) { 295 return Z_STREAM_ERROR; 296 } 297 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ 298 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); 299 if (s == Z_NULL) return Z_MEM_ERROR; 300 strm->state = (struct internal_state FAR *)s; 301 s->strm = strm; 302 s->status = INIT_STATE; /* to pass state test in deflateReset() */ 303 304 s->wrap = wrap; 305 s->gzhead = Z_NULL; 306 s->w_bits = (uInt)windowBits; 307 s->w_size = 1 << s->w_bits; 308 s->w_mask = s->w_size - 1; 309 310 s->hash_bits = (uInt)memLevel + 7; 311 s->hash_size = 1 << s->hash_bits; 312 s->hash_mask = s->hash_size - 1; 313 s->hash_shift = ((s->hash_bits + MIN_MATCH-1) / MIN_MATCH); 314 315 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); 316 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); 317 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); 318 319 s->high_water = 0; /* nothing written to s->window yet */ 320 321 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ 322 323 /* We overlay pending_buf and sym_buf. This works since the average size 324 * for length/distance pairs over any compressed block is assured to be 31 325 * bits or less. 326 * 327 * Analysis: The longest fixed codes are a length code of 8 bits plus 5 328 * extra bits, for lengths 131 to 257. The longest fixed distance codes are 329 * 5 bits plus 13 extra bits, for distances 16385 to 32768. The longest 330 * possible fixed-codes length/distance pair is then 31 bits total. 331 * 332 * sym_buf starts one-fourth of the way into pending_buf. So there are 333 * three bytes in sym_buf for every four bytes in pending_buf. Each symbol 334 * in sym_buf is three bytes -- two for the distance and one for the 335 * literal/length. As each symbol is consumed, the pointer to the next 336 * sym_buf value to read moves forward three bytes. From that symbol, up to 337 * 31 bits are written to pending_buf. The closest the written pending_buf 338 * bits gets to the next sym_buf symbol to read is just before the last 339 * code is written. At that time, 31*(n - 2) bits have been written, just 340 * after 24*(n - 2) bits have been consumed from sym_buf. sym_buf starts at 341 * 8*n bits into pending_buf. (Note that the symbol buffer fills when n - 1 342 * symbols are written.) The closest the writing gets to what is unread is 343 * then n + 14 bits. Here n is lit_bufsize, which is 16384 by default, and 344 * can range from 128 to 32768. 345 * 346 * Therefore, at a minimum, there are 142 bits of space between what is 347 * written and what is read in the overlain buffers, so the symbols cannot 348 * be overwritten by the compressed data. That space is actually 139 bits, 349 * due to the three-bit fixed-code block header. 350 * 351 * That covers the case where either Z_FIXED is specified, forcing fixed 352 * codes, or when the use of fixed codes is chosen, because that choice 353 * results in a smaller compressed block than dynamic codes. That latter 354 * condition then assures that the above analysis also covers all dynamic 355 * blocks. A dynamic-code block will only be chosen to be emitted if it has 356 * fewer bits than a fixed-code block would for the same set of symbols. 357 * Therefore its average symbol length is assured to be less than 31. So 358 * the compressed data for a dynamic block also cannot overwrite the 359 * symbols from which it is being constructed. 360 */ 361 362 s->pending_buf = (uchf *) ZALLOC(strm, s->lit_bufsize, 4); 363 s->pending_buf_size = (ulg)s->lit_bufsize * 4; 364 365 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || 366 s->pending_buf == Z_NULL) { 367 s->status = FINISH_STATE; 368 strm->msg = ERR_MSG(Z_MEM_ERROR); 369 deflateEnd (strm); 370 return Z_MEM_ERROR; 371 } 372 s->sym_buf = s->pending_buf + s->lit_bufsize; 373 s->sym_end = (s->lit_bufsize - 1) * 3; 374 /* We avoid equality with lit_bufsize*3 because of wraparound at 64K 375 * on 16 bit machines and because stored blocks are restricted to 376 * 64K-1 bytes. 377 */ 378 379 s->level = level; 380 s->strategy = strategy; 381 s->method = (Byte)method; 382 383 return deflateReset(strm); 384 } 385 386 /* ========================================================================= 387 * Check for a valid deflate stream state. Return 0 if ok, 1 if not. 388 */ 389 local int deflateStateCheck(strm) 390 z_streamp strm; 391 { 392 deflate_state *s; 393 if (strm == Z_NULL || 394 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) 395 return 1; 396 s = strm->state; 397 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE && 398 #ifdef GZIP 399 s->status != GZIP_STATE && 400 #endif 401 s->status != EXTRA_STATE && 402 s->status != NAME_STATE && 403 s->status != COMMENT_STATE && 404 s->status != HCRC_STATE && 405 s->status != BUSY_STATE && 406 s->status != FINISH_STATE)) 407 return 1; 408 return 0; 409 } 410 411 /* ========================================================================= */ 412 int ZEXPORT deflateSetDictionary(strm, dictionary, dictLength) 413 z_streamp strm; 414 const Bytef *dictionary; 415 uInt dictLength; 416 { 417 deflate_state *s; 418 uInt str, n; 419 int wrap; 420 unsigned avail; 421 z_const unsigned char *next; 422 423 if (deflateStateCheck(strm) || dictionary == Z_NULL) 424 return Z_STREAM_ERROR; 425 s = strm->state; 426 wrap = s->wrap; 427 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead) 428 return Z_STREAM_ERROR; 429 430 /* when using zlib wrappers, compute Adler-32 for provided dictionary */ 431 if (wrap == 1) 432 strm->adler = adler32(strm->adler, dictionary, dictLength); 433 s->wrap = 0; /* avoid computing Adler-32 in read_buf */ 434 435 /* if dictionary would fill window, just replace the history */ 436 if (dictLength >= s->w_size) { 437 if (wrap == 0) { /* already empty otherwise */ 438 CLEAR_HASH(s); 439 s->strstart = 0; 440 s->block_start = 0L; 441 s->insert = 0; 442 } 443 dictionary += dictLength - s->w_size; /* use the tail */ 444 dictLength = s->w_size; 445 } 446 447 /* insert dictionary into window and hash */ 448 avail = strm->avail_in; 449 next = strm->next_in; 450 strm->avail_in = dictLength; 451 strm->next_in = (z_const Bytef *)dictionary; 452 fill_window(s); 453 while (s->lookahead >= MIN_MATCH) { 454 str = s->strstart; 455 n = s->lookahead - (MIN_MATCH-1); 456 do { 457 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 458 #ifndef FASTEST 459 s->prev[str & s->w_mask] = s->head[s->ins_h]; 460 #endif 461 s->head[s->ins_h] = (Pos)str; 462 str++; 463 } while (--n); 464 s->strstart = str; 465 s->lookahead = MIN_MATCH-1; 466 fill_window(s); 467 } 468 s->strstart += s->lookahead; 469 s->block_start = (long)s->strstart; 470 s->insert = s->lookahead; 471 s->lookahead = 0; 472 s->match_length = s->prev_length = MIN_MATCH-1; 473 s->match_available = 0; 474 strm->next_in = next; 475 strm->avail_in = avail; 476 s->wrap = wrap; 477 return Z_OK; 478 } 479 480 /* ========================================================================= */ 481 int ZEXPORT deflateGetDictionary(strm, dictionary, dictLength) 482 z_streamp strm; 483 Bytef *dictionary; 484 uInt *dictLength; 485 { 486 deflate_state *s; 487 uInt len; 488 489 if (deflateStateCheck(strm)) 490 return Z_STREAM_ERROR; 491 s = strm->state; 492 len = s->strstart + s->lookahead; 493 if (len > s->w_size) 494 len = s->w_size; 495 if (dictionary != Z_NULL && len) 496 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len); 497 if (dictLength != Z_NULL) 498 *dictLength = len; 499 return Z_OK; 500 } 501 502 /* ========================================================================= */ 503 int ZEXPORT deflateResetKeep(strm) 504 z_streamp strm; 505 { 506 deflate_state *s; 507 508 if (deflateStateCheck(strm)) { 509 return Z_STREAM_ERROR; 510 } 511 512 strm->total_in = strm->total_out = 0; 513 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ 514 strm->data_type = Z_UNKNOWN; 515 516 s = (deflate_state *)strm->state; 517 s->pending = 0; 518 s->pending_out = s->pending_buf; 519 520 if (s->wrap < 0) { 521 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ 522 } 523 s->status = 524 #ifdef GZIP 525 s->wrap == 2 ? GZIP_STATE : 526 #endif 527 INIT_STATE; 528 strm->adler = 529 #ifdef GZIP 530 s->wrap == 2 ? crc32(0L, Z_NULL, 0) : 531 #endif 532 adler32(0L, Z_NULL, 0); 533 s->last_flush = -2; 534 535 _tr_init(s); 536 537 return Z_OK; 538 } 539 540 /* ========================================================================= */ 541 int ZEXPORT deflateReset(strm) 542 z_streamp strm; 543 { 544 int ret; 545 546 ret = deflateResetKeep(strm); 547 if (ret == Z_OK) 548 lm_init(strm->state); 549 return ret; 550 } 551 552 /* ========================================================================= */ 553 int ZEXPORT deflateSetHeader(strm, head) 554 z_streamp strm; 555 gz_headerp head; 556 { 557 if (deflateStateCheck(strm) || strm->state->wrap != 2) 558 return Z_STREAM_ERROR; 559 strm->state->gzhead = head; 560 return Z_OK; 561 } 562 563 /* ========================================================================= */ 564 int ZEXPORT deflatePending(strm, pending, bits) 565 unsigned *pending; 566 int *bits; 567 z_streamp strm; 568 { 569 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 570 if (pending != Z_NULL) 571 *pending = strm->state->pending; 572 if (bits != Z_NULL) 573 *bits = strm->state->bi_valid; 574 return Z_OK; 575 } 576 577 /* ========================================================================= */ 578 int ZEXPORT deflatePrime(strm, bits, value) 579 z_streamp strm; 580 int bits; 581 int value; 582 { 583 deflate_state *s; 584 int put; 585 586 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 587 s = strm->state; 588 if (bits < 0 || bits > 16 || 589 s->sym_buf < s->pending_out + ((Buf_size + 7) >> 3)) 590 return Z_BUF_ERROR; 591 do { 592 put = Buf_size - s->bi_valid; 593 if (put > bits) 594 put = bits; 595 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid); 596 s->bi_valid += put; 597 _tr_flush_bits(s); 598 value >>= put; 599 bits -= put; 600 } while (bits); 601 return Z_OK; 602 } 603 604 /* ========================================================================= */ 605 int ZEXPORT deflateParams(strm, level, strategy) 606 z_streamp strm; 607 int level; 608 int strategy; 609 { 610 deflate_state *s; 611 compress_func func; 612 613 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 614 s = strm->state; 615 616 #ifdef FASTEST 617 if (level != 0) level = 1; 618 #else 619 if (level == Z_DEFAULT_COMPRESSION) level = 6; 620 #endif 621 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { 622 return Z_STREAM_ERROR; 623 } 624 func = configuration_table[s->level].func; 625 626 if ((strategy != s->strategy || func != configuration_table[level].func) && 627 s->last_flush != -2) { 628 /* Flush the last buffer: */ 629 int err = deflate(strm, Z_BLOCK); 630 if (err == Z_STREAM_ERROR) 631 return err; 632 if (strm->avail_in || (s->strstart - s->block_start) + s->lookahead) 633 return Z_BUF_ERROR; 634 } 635 if (s->level != level) { 636 if (s->level == 0 && s->matches != 0) { 637 if (s->matches == 1) 638 slide_hash(s); 639 else 640 CLEAR_HASH(s); 641 s->matches = 0; 642 } 643 s->level = level; 644 s->max_lazy_match = configuration_table[level].max_lazy; 645 s->good_match = configuration_table[level].good_length; 646 s->nice_match = configuration_table[level].nice_length; 647 s->max_chain_length = configuration_table[level].max_chain; 648 } 649 s->strategy = strategy; 650 return Z_OK; 651 } 652 653 /* ========================================================================= */ 654 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain) 655 z_streamp strm; 656 int good_length; 657 int max_lazy; 658 int nice_length; 659 int max_chain; 660 { 661 deflate_state *s; 662 663 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 664 s = strm->state; 665 s->good_match = (uInt)good_length; 666 s->max_lazy_match = (uInt)max_lazy; 667 s->nice_match = nice_length; 668 s->max_chain_length = (uInt)max_chain; 669 return Z_OK; 670 } 671 672 /* ========================================================================= 673 * For the default windowBits of 15 and memLevel of 8, this function returns a 674 * close to exact, as well as small, upper bound on the compressed size. This 675 * is an expansion of ~0.03%, plus a small constant. 676 * 677 * For any setting other than those defaults for windowBits and memLevel, one 678 * of two worst case bounds is returned. This is at most an expansion of ~4% or 679 * ~13%, plus a small constant. 680 * 681 * Both the 0.03% and 4% derive from the overhead of stored blocks. The first 682 * one is for stored blocks of 16383 bytes (memLevel == 8), whereas the second 683 * is for stored blocks of 127 bytes (the worst case memLevel == 1). The 684 * expansion results from five bytes of header for each stored block. 685 * 686 * The larger expansion of 13% results from a window size less than or equal to 687 * the symbols buffer size (windowBits <= memLevel + 7). In that case some of 688 * the data being compressed may have slid out of the sliding window, impeding 689 * a stored block from being emitted. Then the only choice is a fixed or 690 * dynamic block, where a fixed block limits the maximum expansion to 9 bits 691 * per 8-bit byte, plus 10 bits for every block. The smallest block size for 692 * which this can occur is 255 (memLevel == 2). 693 * 694 * Shifts are used to approximate divisions, for speed. 695 */ 696 uLong ZEXPORT deflateBound(strm, sourceLen) 697 z_streamp strm; 698 uLong sourceLen; 699 { 700 deflate_state *s; 701 uLong fixedlen, storelen, wraplen; 702 703 /* upper bound for fixed blocks with 9-bit literals and length 255 704 (memLevel == 2, which is the lowest that may not use stored blocks) -- 705 ~13% overhead plus a small constant */ 706 fixedlen = sourceLen + (sourceLen >> 3) + (sourceLen >> 8) + 707 (sourceLen >> 9) + 4; 708 709 /* upper bound for stored blocks with length 127 (memLevel == 1) -- 710 ~4% overhead plus a small constant */ 711 storelen = sourceLen + (sourceLen >> 5) + (sourceLen >> 7) + 712 (sourceLen >> 11) + 7; 713 714 /* if can't get parameters, return larger bound plus a zlib wrapper */ 715 if (deflateStateCheck(strm)) 716 return (fixedlen > storelen ? fixedlen : storelen) + 6; 717 718 /* compute wrapper length */ 719 s = strm->state; 720 switch (s->wrap) { 721 case 0: /* raw deflate */ 722 wraplen = 0; 723 break; 724 case 1: /* zlib wrapper */ 725 wraplen = 6 + (s->strstart ? 4 : 0); 726 break; 727 #ifdef GZIP 728 case 2: /* gzip wrapper */ 729 wraplen = 18; 730 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */ 731 Bytef *str; 732 if (s->gzhead->extra != Z_NULL) 733 wraplen += 2 + s->gzhead->extra_len; 734 str = s->gzhead->name; 735 if (str != Z_NULL) 736 do { 737 wraplen++; 738 } while (*str++); 739 str = s->gzhead->comment; 740 if (str != Z_NULL) 741 do { 742 wraplen++; 743 } while (*str++); 744 if (s->gzhead->hcrc) 745 wraplen += 2; 746 } 747 break; 748 #endif 749 default: /* for compiler happiness */ 750 wraplen = 6; 751 } 752 753 /* if not default parameters, return one of the conservative bounds */ 754 if (s->w_bits != 15 || s->hash_bits != 8 + 7) 755 return (s->w_bits <= s->hash_bits ? fixedlen : storelen) + wraplen; 756 757 /* default settings: return tight bound for that case -- ~0.03% overhead 758 plus a small constant */ 759 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) + 760 (sourceLen >> 25) + 13 - 6 + wraplen; 761 } 762 763 /* ========================================================================= 764 * Put a short in the pending buffer. The 16-bit value is put in MSB order. 765 * IN assertion: the stream state is correct and there is enough room in 766 * pending_buf. 767 */ 768 local void putShortMSB(s, b) 769 deflate_state *s; 770 uInt b; 771 { 772 put_byte(s, (Byte)(b >> 8)); 773 put_byte(s, (Byte)(b & 0xff)); 774 } 775 776 /* ========================================================================= 777 * Flush as much pending output as possible. All deflate() output, except for 778 * some deflate_stored() output, goes through this function so some 779 * applications may wish to modify it to avoid allocating a large 780 * strm->next_out buffer and copying into it. (See also read_buf()). 781 */ 782 local void flush_pending(strm) 783 z_streamp strm; 784 { 785 unsigned len; 786 deflate_state *s = strm->state; 787 788 _tr_flush_bits(s); 789 len = s->pending; 790 if (len > strm->avail_out) len = strm->avail_out; 791 if (len == 0) return; 792 793 zmemcpy(strm->next_out, s->pending_out, len); 794 strm->next_out += len; 795 s->pending_out += len; 796 strm->total_out += len; 797 strm->avail_out -= len; 798 s->pending -= len; 799 if (s->pending == 0) { 800 s->pending_out = s->pending_buf; 801 } 802 } 803 804 /* =========================================================================== 805 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1]. 806 */ 807 #define HCRC_UPDATE(beg) \ 808 do { \ 809 if (s->gzhead->hcrc && s->pending > (beg)) \ 810 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \ 811 s->pending - (beg)); \ 812 } while (0) 813 814 /* ========================================================================= */ 815 int ZEXPORT deflate(strm, flush) 816 z_streamp strm; 817 int flush; 818 { 819 int old_flush; /* value of flush param for previous deflate call */ 820 deflate_state *s; 821 822 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) { 823 return Z_STREAM_ERROR; 824 } 825 s = strm->state; 826 827 if (strm->next_out == Z_NULL || 828 (strm->avail_in != 0 && strm->next_in == Z_NULL) || 829 (s->status == FINISH_STATE && flush != Z_FINISH)) { 830 ERR_RETURN(strm, Z_STREAM_ERROR); 831 } 832 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); 833 834 old_flush = s->last_flush; 835 s->last_flush = flush; 836 837 /* Flush as much pending output as possible */ 838 if (s->pending != 0) { 839 flush_pending(strm); 840 if (strm->avail_out == 0) { 841 /* Since avail_out is 0, deflate will be called again with 842 * more output space, but possibly with both pending and 843 * avail_in equal to zero. There won't be anything to do, 844 * but this is not an error situation so make sure we 845 * return OK instead of BUF_ERROR at next call of deflate: 846 */ 847 s->last_flush = -1; 848 return Z_OK; 849 } 850 851 /* Make sure there is something to do and avoid duplicate consecutive 852 * flushes. For repeated and useless calls with Z_FINISH, we keep 853 * returning Z_STREAM_END instead of Z_BUF_ERROR. 854 */ 855 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) && 856 flush != Z_FINISH) { 857 ERR_RETURN(strm, Z_BUF_ERROR); 858 } 859 860 /* User must not provide more input after the first FINISH: */ 861 if (s->status == FINISH_STATE && strm->avail_in != 0) { 862 ERR_RETURN(strm, Z_BUF_ERROR); 863 } 864 865 /* Write the header */ 866 if (s->status == INIT_STATE && s->wrap == 0) 867 s->status = BUSY_STATE; 868 if (s->status == INIT_STATE) { 869 /* zlib header */ 870 uInt header = (Z_DEFLATED + ((s->w_bits - 8) << 4)) << 8; 871 uInt level_flags; 872 873 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) 874 level_flags = 0; 875 else if (s->level < 6) 876 level_flags = 1; 877 else if (s->level == 6) 878 level_flags = 2; 879 else 880 level_flags = 3; 881 header |= (level_flags << 6); 882 if (s->strstart != 0) header |= PRESET_DICT; 883 header += 31 - (header % 31); 884 885 putShortMSB(s, header); 886 887 /* Save the adler32 of the preset dictionary: */ 888 if (s->strstart != 0) { 889 putShortMSB(s, (uInt)(strm->adler >> 16)); 890 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 891 } 892 strm->adler = adler32(0L, Z_NULL, 0); 893 s->status = BUSY_STATE; 894 895 /* Compression must start with an empty pending buffer */ 896 flush_pending(strm); 897 if (s->pending != 0) { 898 s->last_flush = -1; 899 return Z_OK; 900 } 901 } 902 #ifdef GZIP 903 if (s->status == GZIP_STATE) { 904 /* gzip header */ 905 strm->adler = crc32(0L, Z_NULL, 0); 906 put_byte(s, 31); 907 put_byte(s, 139); 908 put_byte(s, 8); 909 if (s->gzhead == Z_NULL) { 910 put_byte(s, 0); 911 put_byte(s, 0); 912 put_byte(s, 0); 913 put_byte(s, 0); 914 put_byte(s, 0); 915 put_byte(s, s->level == 9 ? 2 : 916 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 917 4 : 0)); 918 put_byte(s, OS_CODE); 919 s->status = BUSY_STATE; 920 921 /* Compression must start with an empty pending buffer */ 922 flush_pending(strm); 923 if (s->pending != 0) { 924 s->last_flush = -1; 925 return Z_OK; 926 } 927 } 928 else { 929 put_byte(s, (s->gzhead->text ? 1 : 0) + 930 (s->gzhead->hcrc ? 2 : 0) + 931 (s->gzhead->extra == Z_NULL ? 0 : 4) + 932 (s->gzhead->name == Z_NULL ? 0 : 8) + 933 (s->gzhead->comment == Z_NULL ? 0 : 16) 934 ); 935 put_byte(s, (Byte)(s->gzhead->time & 0xff)); 936 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); 937 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); 938 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); 939 put_byte(s, s->level == 9 ? 2 : 940 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 941 4 : 0)); 942 put_byte(s, s->gzhead->os & 0xff); 943 if (s->gzhead->extra != Z_NULL) { 944 put_byte(s, s->gzhead->extra_len & 0xff); 945 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); 946 } 947 if (s->gzhead->hcrc) 948 strm->adler = crc32(strm->adler, s->pending_buf, 949 s->pending); 950 s->gzindex = 0; 951 s->status = EXTRA_STATE; 952 } 953 } 954 if (s->status == EXTRA_STATE) { 955 if (s->gzhead->extra != Z_NULL) { 956 ulg beg = s->pending; /* start of bytes to update crc */ 957 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex; 958 while (s->pending + left > s->pending_buf_size) { 959 uInt copy = s->pending_buf_size - s->pending; 960 zmemcpy(s->pending_buf + s->pending, 961 s->gzhead->extra + s->gzindex, copy); 962 s->pending = s->pending_buf_size; 963 HCRC_UPDATE(beg); 964 s->gzindex += copy; 965 flush_pending(strm); 966 if (s->pending != 0) { 967 s->last_flush = -1; 968 return Z_OK; 969 } 970 beg = 0; 971 left -= copy; 972 } 973 zmemcpy(s->pending_buf + s->pending, 974 s->gzhead->extra + s->gzindex, left); 975 s->pending += left; 976 HCRC_UPDATE(beg); 977 s->gzindex = 0; 978 } 979 s->status = NAME_STATE; 980 } 981 if (s->status == NAME_STATE) { 982 if (s->gzhead->name != Z_NULL) { 983 ulg beg = s->pending; /* start of bytes to update crc */ 984 int val; 985 do { 986 if (s->pending == s->pending_buf_size) { 987 HCRC_UPDATE(beg); 988 flush_pending(strm); 989 if (s->pending != 0) { 990 s->last_flush = -1; 991 return Z_OK; 992 } 993 beg = 0; 994 } 995 val = s->gzhead->name[s->gzindex++]; 996 put_byte(s, val); 997 } while (val != 0); 998 HCRC_UPDATE(beg); 999 s->gzindex = 0; 1000 } 1001 s->status = COMMENT_STATE; 1002 } 1003 if (s->status == COMMENT_STATE) { 1004 if (s->gzhead->comment != Z_NULL) { 1005 ulg beg = s->pending; /* start of bytes to update crc */ 1006 int val; 1007 do { 1008 if (s->pending == s->pending_buf_size) { 1009 HCRC_UPDATE(beg); 1010 flush_pending(strm); 1011 if (s->pending != 0) { 1012 s->last_flush = -1; 1013 return Z_OK; 1014 } 1015 beg = 0; 1016 } 1017 val = s->gzhead->comment[s->gzindex++]; 1018 put_byte(s, val); 1019 } while (val != 0); 1020 HCRC_UPDATE(beg); 1021 } 1022 s->status = HCRC_STATE; 1023 } 1024 if (s->status == HCRC_STATE) { 1025 if (s->gzhead->hcrc) { 1026 if (s->pending + 2 > s->pending_buf_size) { 1027 flush_pending(strm); 1028 if (s->pending != 0) { 1029 s->last_flush = -1; 1030 return Z_OK; 1031 } 1032 } 1033 put_byte(s, (Byte)(strm->adler & 0xff)); 1034 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 1035 strm->adler = crc32(0L, Z_NULL, 0); 1036 } 1037 s->status = BUSY_STATE; 1038 1039 /* Compression must start with an empty pending buffer */ 1040 flush_pending(strm); 1041 if (s->pending != 0) { 1042 s->last_flush = -1; 1043 return Z_OK; 1044 } 1045 } 1046 #endif 1047 1048 /* Start a new block or continue the current one. 1049 */ 1050 if (strm->avail_in != 0 || s->lookahead != 0 || 1051 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { 1052 block_state bstate; 1053 1054 bstate = s->level == 0 ? deflate_stored(s, flush) : 1055 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : 1056 s->strategy == Z_RLE ? deflate_rle(s, flush) : 1057 (*(configuration_table[s->level].func))(s, flush); 1058 1059 if (bstate == finish_started || bstate == finish_done) { 1060 s->status = FINISH_STATE; 1061 } 1062 if (bstate == need_more || bstate == finish_started) { 1063 if (strm->avail_out == 0) { 1064 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ 1065 } 1066 return Z_OK; 1067 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call 1068 * of deflate should use the same flush parameter to make sure 1069 * that the flush is complete. So we don't have to output an 1070 * empty block here, this will be done at next call. This also 1071 * ensures that for a very small output buffer, we emit at most 1072 * one empty block. 1073 */ 1074 } 1075 if (bstate == block_done) { 1076 if (flush == Z_PARTIAL_FLUSH) { 1077 _tr_align(s); 1078 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */ 1079 _tr_stored_block(s, (char*)0, 0L, 0); 1080 /* For a full flush, this empty block will be recognized 1081 * as a special marker by inflate_sync(). 1082 */ 1083 if (flush == Z_FULL_FLUSH) { 1084 CLEAR_HASH(s); /* forget history */ 1085 if (s->lookahead == 0) { 1086 s->strstart = 0; 1087 s->block_start = 0L; 1088 s->insert = 0; 1089 } 1090 } 1091 } 1092 flush_pending(strm); 1093 if (strm->avail_out == 0) { 1094 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ 1095 return Z_OK; 1096 } 1097 } 1098 } 1099 1100 if (flush != Z_FINISH) return Z_OK; 1101 if (s->wrap <= 0) return Z_STREAM_END; 1102 1103 /* Write the trailer */ 1104 #ifdef GZIP 1105 if (s->wrap == 2) { 1106 put_byte(s, (Byte)(strm->adler & 0xff)); 1107 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 1108 put_byte(s, (Byte)((strm->adler >> 16) & 0xff)); 1109 put_byte(s, (Byte)((strm->adler >> 24) & 0xff)); 1110 put_byte(s, (Byte)(strm->total_in & 0xff)); 1111 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff)); 1112 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff)); 1113 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff)); 1114 } 1115 else 1116 #endif 1117 { 1118 putShortMSB(s, (uInt)(strm->adler >> 16)); 1119 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 1120 } 1121 flush_pending(strm); 1122 /* If avail_out is zero, the application will call deflate again 1123 * to flush the rest. 1124 */ 1125 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ 1126 return s->pending != 0 ? Z_OK : Z_STREAM_END; 1127 } 1128 1129 /* ========================================================================= */ 1130 int ZEXPORT deflateEnd(strm) 1131 z_streamp strm; 1132 { 1133 int status; 1134 1135 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 1136 1137 status = strm->state->status; 1138 1139 /* Deallocate in reverse order of allocations: */ 1140 TRY_FREE(strm, strm->state->pending_buf); 1141 TRY_FREE(strm, strm->state->head); 1142 TRY_FREE(strm, strm->state->prev); 1143 TRY_FREE(strm, strm->state->window); 1144 1145 ZFREE(strm, strm->state); 1146 strm->state = Z_NULL; 1147 1148 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; 1149 } 1150 1151 /* ========================================================================= 1152 * Copy the source state to the destination state. 1153 * To simplify the source, this is not supported for 16-bit MSDOS (which 1154 * doesn't have enough memory anyway to duplicate compression states). 1155 */ 1156 int ZEXPORT deflateCopy(dest, source) 1157 z_streamp dest; 1158 z_streamp source; 1159 { 1160 #ifdef MAXSEG_64K 1161 return Z_STREAM_ERROR; 1162 #else 1163 deflate_state *ds; 1164 deflate_state *ss; 1165 1166 1167 if (deflateStateCheck(source) || dest == Z_NULL) { 1168 return Z_STREAM_ERROR; 1169 } 1170 1171 ss = source->state; 1172 1173 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream)); 1174 1175 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); 1176 if (ds == Z_NULL) return Z_MEM_ERROR; 1177 dest->state = (struct internal_state FAR *) ds; 1178 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state)); 1179 ds->strm = dest; 1180 1181 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); 1182 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); 1183 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); 1184 ds->pending_buf = (uchf *) ZALLOC(dest, ds->lit_bufsize, 4); 1185 1186 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || 1187 ds->pending_buf == Z_NULL) { 1188 deflateEnd (dest); 1189 return Z_MEM_ERROR; 1190 } 1191 /* following zmemcpy do not work for 16-bit MSDOS */ 1192 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); 1193 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos)); 1194 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos)); 1195 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); 1196 1197 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); 1198 ds->sym_buf = ds->pending_buf + ds->lit_bufsize; 1199 1200 ds->l_desc.dyn_tree = ds->dyn_ltree; 1201 ds->d_desc.dyn_tree = ds->dyn_dtree; 1202 ds->bl_desc.dyn_tree = ds->bl_tree; 1203 1204 return Z_OK; 1205 #endif /* MAXSEG_64K */ 1206 } 1207 1208 /* =========================================================================== 1209 * Read a new buffer from the current input stream, update the adler32 1210 * and total number of bytes read. All deflate() input goes through 1211 * this function so some applications may wish to modify it to avoid 1212 * allocating a large strm->next_in buffer and copying from it. 1213 * (See also flush_pending()). 1214 */ 1215 local unsigned read_buf(strm, buf, size) 1216 z_streamp strm; 1217 Bytef *buf; 1218 unsigned size; 1219 { 1220 unsigned len = strm->avail_in; 1221 1222 if (len > size) len = size; 1223 if (len == 0) return 0; 1224 1225 strm->avail_in -= len; 1226 1227 zmemcpy(buf, strm->next_in, len); 1228 if (strm->state->wrap == 1) { 1229 strm->adler = adler32(strm->adler, buf, len); 1230 } 1231 #ifdef GZIP 1232 else if (strm->state->wrap == 2) { 1233 strm->adler = crc32(strm->adler, buf, len); 1234 } 1235 #endif 1236 strm->next_in += len; 1237 strm->total_in += len; 1238 1239 return len; 1240 } 1241 1242 /* =========================================================================== 1243 * Initialize the "longest match" routines for a new zlib stream 1244 */ 1245 local void lm_init(s) 1246 deflate_state *s; 1247 { 1248 s->window_size = (ulg)2L*s->w_size; 1249 1250 CLEAR_HASH(s); 1251 1252 /* Set the default configuration parameters: 1253 */ 1254 s->max_lazy_match = configuration_table[s->level].max_lazy; 1255 s->good_match = configuration_table[s->level].good_length; 1256 s->nice_match = configuration_table[s->level].nice_length; 1257 s->max_chain_length = configuration_table[s->level].max_chain; 1258 1259 s->strstart = 0; 1260 s->block_start = 0L; 1261 s->lookahead = 0; 1262 s->insert = 0; 1263 s->match_length = s->prev_length = MIN_MATCH-1; 1264 s->match_available = 0; 1265 s->ins_h = 0; 1266 } 1267 1268 #ifndef FASTEST 1269 /* =========================================================================== 1270 * Set match_start to the longest match starting at the given string and 1271 * return its length. Matches shorter or equal to prev_length are discarded, 1272 * in which case the result is equal to prev_length and match_start is 1273 * garbage. 1274 * IN assertions: cur_match is the head of the hash chain for the current 1275 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 1276 * OUT assertion: the match length is not greater than s->lookahead. 1277 */ 1278 local uInt longest_match(s, cur_match) 1279 deflate_state *s; 1280 IPos cur_match; /* current match */ 1281 { 1282 unsigned chain_length = s->max_chain_length;/* max hash chain length */ 1283 register Bytef *scan = s->window + s->strstart; /* current string */ 1284 register Bytef *match; /* matched string */ 1285 register int len; /* length of current match */ 1286 int best_len = (int)s->prev_length; /* best match length so far */ 1287 int nice_match = s->nice_match; /* stop if match long enough */ 1288 IPos limit = s->strstart > (IPos)MAX_DIST(s) ? 1289 s->strstart - (IPos)MAX_DIST(s) : NIL; 1290 /* Stop when cur_match becomes <= limit. To simplify the code, 1291 * we prevent matches with the string of window index 0. 1292 */ 1293 Posf *prev = s->prev; 1294 uInt wmask = s->w_mask; 1295 1296 #ifdef UNALIGNED_OK 1297 /* Compare two bytes at a time. Note: this is not always beneficial. 1298 * Try with and without -DUNALIGNED_OK to check. 1299 */ 1300 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; 1301 register ush scan_start = *(ushf*)scan; 1302 register ush scan_end = *(ushf*)(scan + best_len - 1); 1303 #else 1304 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1305 register Byte scan_end1 = scan[best_len - 1]; 1306 register Byte scan_end = scan[best_len]; 1307 #endif 1308 1309 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1310 * It is easy to get rid of this optimization if necessary. 1311 */ 1312 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1313 1314 /* Do not waste too much time if we already have a good match: */ 1315 if (s->prev_length >= s->good_match) { 1316 chain_length >>= 2; 1317 } 1318 /* Do not look for matches beyond the end of the input. This is necessary 1319 * to make deflate deterministic. 1320 */ 1321 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead; 1322 1323 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 1324 "need lookahead"); 1325 1326 do { 1327 Assert(cur_match < s->strstart, "no future"); 1328 match = s->window + cur_match; 1329 1330 /* Skip to next match if the match length cannot increase 1331 * or if the match length is less than 2. Note that the checks below 1332 * for insufficient lookahead only occur occasionally for performance 1333 * reasons. Therefore uninitialized memory will be accessed, and 1334 * conditional jumps will be made that depend on those values. 1335 * However the length of the match is limited to the lookahead, so 1336 * the output of deflate is not affected by the uninitialized values. 1337 */ 1338 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) 1339 /* This code assumes sizeof(unsigned short) == 2. Do not use 1340 * UNALIGNED_OK if your compiler uses a different size. 1341 */ 1342 if (*(ushf*)(match + best_len - 1) != scan_end || 1343 *(ushf*)match != scan_start) continue; 1344 1345 /* It is not necessary to compare scan[2] and match[2] since they are 1346 * always equal when the other bytes match, given that the hash keys 1347 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at 1348 * strstart + 3, + 5, up to strstart + 257. We check for insufficient 1349 * lookahead only every 4th comparison; the 128th check will be made 1350 * at strstart + 257. If MAX_MATCH-2 is not a multiple of 8, it is 1351 * necessary to put more guard bytes at the end of the window, or 1352 * to check more often for insufficient lookahead. 1353 */ 1354 Assert(scan[2] == match[2], "scan[2]?"); 1355 scan++, match++; 1356 do { 1357 } while (*(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1358 *(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1359 *(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1360 *(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1361 scan < strend); 1362 /* The funny "do {}" generates better code on most compilers */ 1363 1364 /* Here, scan <= window + strstart + 257 */ 1365 Assert(scan <= s->window + (unsigned)(s->window_size - 1), 1366 "wild scan"); 1367 if (*scan == *match) scan++; 1368 1369 len = (MAX_MATCH - 1) - (int)(strend - scan); 1370 scan = strend - (MAX_MATCH-1); 1371 1372 #else /* UNALIGNED_OK */ 1373 1374 if (match[best_len] != scan_end || 1375 match[best_len - 1] != scan_end1 || 1376 *match != *scan || 1377 *++match != scan[1]) continue; 1378 1379 /* The check at best_len - 1 can be removed because it will be made 1380 * again later. (This heuristic is not always a win.) 1381 * It is not necessary to compare scan[2] and match[2] since they 1382 * are always equal when the other bytes match, given that 1383 * the hash keys are equal and that HASH_BITS >= 8. 1384 */ 1385 scan += 2, match++; 1386 Assert(*scan == *match, "match[2]?"); 1387 1388 /* We check for insufficient lookahead only every 8th comparison; 1389 * the 256th check will be made at strstart + 258. 1390 */ 1391 do { 1392 } while (*++scan == *++match && *++scan == *++match && 1393 *++scan == *++match && *++scan == *++match && 1394 *++scan == *++match && *++scan == *++match && 1395 *++scan == *++match && *++scan == *++match && 1396 scan < strend); 1397 1398 Assert(scan <= s->window + (unsigned)(s->window_size - 1), 1399 "wild scan"); 1400 1401 len = MAX_MATCH - (int)(strend - scan); 1402 scan = strend - MAX_MATCH; 1403 1404 #endif /* UNALIGNED_OK */ 1405 1406 if (len > best_len) { 1407 s->match_start = cur_match; 1408 best_len = len; 1409 if (len >= nice_match) break; 1410 #ifdef UNALIGNED_OK 1411 scan_end = *(ushf*)(scan + best_len - 1); 1412 #else 1413 scan_end1 = scan[best_len - 1]; 1414 scan_end = scan[best_len]; 1415 #endif 1416 } 1417 } while ((cur_match = prev[cur_match & wmask]) > limit 1418 && --chain_length != 0); 1419 1420 if ((uInt)best_len <= s->lookahead) return (uInt)best_len; 1421 return s->lookahead; 1422 } 1423 1424 #else /* FASTEST */ 1425 1426 /* --------------------------------------------------------------------------- 1427 * Optimized version for FASTEST only 1428 */ 1429 local uInt longest_match(s, cur_match) 1430 deflate_state *s; 1431 IPos cur_match; /* current match */ 1432 { 1433 register Bytef *scan = s->window + s->strstart; /* current string */ 1434 register Bytef *match; /* matched string */ 1435 register int len; /* length of current match */ 1436 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1437 1438 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1439 * It is easy to get rid of this optimization if necessary. 1440 */ 1441 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1442 1443 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 1444 "need lookahead"); 1445 1446 Assert(cur_match < s->strstart, "no future"); 1447 1448 match = s->window + cur_match; 1449 1450 /* Return failure if the match length is less than 2: 1451 */ 1452 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; 1453 1454 /* The check at best_len - 1 can be removed because it will be made 1455 * again later. (This heuristic is not always a win.) 1456 * It is not necessary to compare scan[2] and match[2] since they 1457 * are always equal when the other bytes match, given that 1458 * the hash keys are equal and that HASH_BITS >= 8. 1459 */ 1460 scan += 2, match += 2; 1461 Assert(*scan == *match, "match[2]?"); 1462 1463 /* We check for insufficient lookahead only every 8th comparison; 1464 * the 256th check will be made at strstart + 258. 1465 */ 1466 do { 1467 } while (*++scan == *++match && *++scan == *++match && 1468 *++scan == *++match && *++scan == *++match && 1469 *++scan == *++match && *++scan == *++match && 1470 *++scan == *++match && *++scan == *++match && 1471 scan < strend); 1472 1473 Assert(scan <= s->window + (unsigned)(s->window_size - 1), "wild scan"); 1474 1475 len = MAX_MATCH - (int)(strend - scan); 1476 1477 if (len < MIN_MATCH) return MIN_MATCH - 1; 1478 1479 s->match_start = cur_match; 1480 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead; 1481 } 1482 1483 #endif /* FASTEST */ 1484 1485 #ifdef ZLIB_DEBUG 1486 1487 #define EQUAL 0 1488 /* result of memcmp for equal strings */ 1489 1490 /* =========================================================================== 1491 * Check that the match at match_start is indeed a match. 1492 */ 1493 local void check_match(s, start, match, length) 1494 deflate_state *s; 1495 IPos start, match; 1496 int length; 1497 { 1498 /* check that the match is indeed a match */ 1499 if (zmemcmp(s->window + match, 1500 s->window + start, length) != EQUAL) { 1501 fprintf(stderr, " start %u, match %u, length %d\n", 1502 start, match, length); 1503 do { 1504 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); 1505 } while (--length != 0); 1506 z_error("invalid match"); 1507 } 1508 if (z_verbose > 1) { 1509 fprintf(stderr,"\\[%d,%d]", start - match, length); 1510 do { putc(s->window[start++], stderr); } while (--length != 0); 1511 } 1512 } 1513 #else 1514 # define check_match(s, start, match, length) 1515 #endif /* ZLIB_DEBUG */ 1516 1517 /* =========================================================================== 1518 * Fill the window when the lookahead becomes insufficient. 1519 * Updates strstart and lookahead. 1520 * 1521 * IN assertion: lookahead < MIN_LOOKAHEAD 1522 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD 1523 * At least one byte has been read, or avail_in == 0; reads are 1524 * performed for at least two bytes (required for the zip translate_eol 1525 * option -- not supported here). 1526 */ 1527 local void fill_window(s) 1528 deflate_state *s; 1529 { 1530 unsigned n; 1531 unsigned more; /* Amount of free space at the end of the window. */ 1532 uInt wsize = s->w_size; 1533 1534 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead"); 1535 1536 do { 1537 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); 1538 1539 /* Deal with !@#$% 64K limit: */ 1540 if (sizeof(int) <= 2) { 1541 if (more == 0 && s->strstart == 0 && s->lookahead == 0) { 1542 more = wsize; 1543 1544 } else if (more == (unsigned)(-1)) { 1545 /* Very unlikely, but possible on 16 bit machine if 1546 * strstart == 0 && lookahead == 1 (input done a byte at time) 1547 */ 1548 more--; 1549 } 1550 } 1551 1552 /* If the window is almost full and there is insufficient lookahead, 1553 * move the upper half to the lower one to make room in the upper half. 1554 */ 1555 if (s->strstart >= wsize + MAX_DIST(s)) { 1556 1557 zmemcpy(s->window, s->window + wsize, (unsigned)wsize - more); 1558 s->match_start -= wsize; 1559 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ 1560 s->block_start -= (long) wsize; 1561 if (s->insert > s->strstart) 1562 s->insert = s->strstart; 1563 slide_hash(s); 1564 more += wsize; 1565 } 1566 if (s->strm->avail_in == 0) break; 1567 1568 /* If there was no sliding: 1569 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && 1570 * more == window_size - lookahead - strstart 1571 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) 1572 * => more >= window_size - 2*WSIZE + 2 1573 * In the BIG_MEM or MMAP case (not yet supported), 1574 * window_size == input_size + MIN_LOOKAHEAD && 1575 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. 1576 * Otherwise, window_size == 2*WSIZE so more >= 2. 1577 * If there was sliding, more >= WSIZE. So in all cases, more >= 2. 1578 */ 1579 Assert(more >= 2, "more < 2"); 1580 1581 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); 1582 s->lookahead += n; 1583 1584 /* Initialize the hash value now that we have some input: */ 1585 if (s->lookahead + s->insert >= MIN_MATCH) { 1586 uInt str = s->strstart - s->insert; 1587 s->ins_h = s->window[str]; 1588 UPDATE_HASH(s, s->ins_h, s->window[str + 1]); 1589 #if MIN_MATCH != 3 1590 Call UPDATE_HASH() MIN_MATCH-3 more times 1591 #endif 1592 while (s->insert) { 1593 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 1594 #ifndef FASTEST 1595 s->prev[str & s->w_mask] = s->head[s->ins_h]; 1596 #endif 1597 s->head[s->ins_h] = (Pos)str; 1598 str++; 1599 s->insert--; 1600 if (s->lookahead + s->insert < MIN_MATCH) 1601 break; 1602 } 1603 } 1604 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, 1605 * but this is not important since only literal bytes will be emitted. 1606 */ 1607 1608 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); 1609 1610 /* If the WIN_INIT bytes after the end of the current data have never been 1611 * written, then zero those bytes in order to avoid memory check reports of 1612 * the use of uninitialized (or uninitialised as Julian writes) bytes by 1613 * the longest match routines. Update the high water mark for the next 1614 * time through here. WIN_INIT is set to MAX_MATCH since the longest match 1615 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead. 1616 */ 1617 if (s->high_water < s->window_size) { 1618 ulg curr = s->strstart + (ulg)(s->lookahead); 1619 ulg init; 1620 1621 if (s->high_water < curr) { 1622 /* Previous high water mark below current data -- zero WIN_INIT 1623 * bytes or up to end of window, whichever is less. 1624 */ 1625 init = s->window_size - curr; 1626 if (init > WIN_INIT) 1627 init = WIN_INIT; 1628 zmemzero(s->window + curr, (unsigned)init); 1629 s->high_water = curr + init; 1630 } 1631 else if (s->high_water < (ulg)curr + WIN_INIT) { 1632 /* High water mark at or above current data, but below current data 1633 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up 1634 * to end of window, whichever is less. 1635 */ 1636 init = (ulg)curr + WIN_INIT - s->high_water; 1637 if (init > s->window_size - s->high_water) 1638 init = s->window_size - s->high_water; 1639 zmemzero(s->window + s->high_water, (unsigned)init); 1640 s->high_water += init; 1641 } 1642 } 1643 1644 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 1645 "not enough room for search"); 1646 } 1647 1648 /* =========================================================================== 1649 * Flush the current block, with given end-of-file flag. 1650 * IN assertion: strstart is set to the end of the current match. 1651 */ 1652 #define FLUSH_BLOCK_ONLY(s, last) { \ 1653 _tr_flush_block(s, (s->block_start >= 0L ? \ 1654 (charf *)&s->window[(unsigned)s->block_start] : \ 1655 (charf *)Z_NULL), \ 1656 (ulg)((long)s->strstart - s->block_start), \ 1657 (last)); \ 1658 s->block_start = s->strstart; \ 1659 flush_pending(s->strm); \ 1660 Tracev((stderr,"[FLUSH]")); \ 1661 } 1662 1663 /* Same but force premature exit if necessary. */ 1664 #define FLUSH_BLOCK(s, last) { \ 1665 FLUSH_BLOCK_ONLY(s, last); \ 1666 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \ 1667 } 1668 1669 /* Maximum stored block length in deflate format (not including header). */ 1670 #define MAX_STORED 65535 1671 1672 /* Minimum of a and b. */ 1673 #define MIN(a, b) ((a) > (b) ? (b) : (a)) 1674 1675 /* =========================================================================== 1676 * Copy without compression as much as possible from the input stream, return 1677 * the current block state. 1678 * 1679 * In case deflateParams() is used to later switch to a non-zero compression 1680 * level, s->matches (otherwise unused when storing) keeps track of the number 1681 * of hash table slides to perform. If s->matches is 1, then one hash table 1682 * slide will be done when switching. If s->matches is 2, the maximum value 1683 * allowed here, then the hash table will be cleared, since two or more slides 1684 * is the same as a clear. 1685 * 1686 * deflate_stored() is written to minimize the number of times an input byte is 1687 * copied. It is most efficient with large input and output buffers, which 1688 * maximizes the opportunities to have a single copy from next_in to next_out. 1689 */ 1690 local block_state deflate_stored(s, flush) 1691 deflate_state *s; 1692 int flush; 1693 { 1694 /* Smallest worthy block size when not flushing or finishing. By default 1695 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For 1696 * large input and output buffers, the stored block size will be larger. 1697 */ 1698 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size); 1699 1700 /* Copy as many min_block or larger stored blocks directly to next_out as 1701 * possible. If flushing, copy the remaining available input to next_out as 1702 * stored blocks, if there is enough space. 1703 */ 1704 unsigned len, left, have, last = 0; 1705 unsigned used = s->strm->avail_in; 1706 do { 1707 /* Set len to the maximum size block that we can copy directly with the 1708 * available input data and output space. Set left to how much of that 1709 * would be copied from what's left in the window. 1710 */ 1711 len = MAX_STORED; /* maximum deflate stored block length */ 1712 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1713 if (s->strm->avail_out < have) /* need room for header */ 1714 break; 1715 /* maximum stored block length that will fit in avail_out: */ 1716 have = s->strm->avail_out - have; 1717 left = s->strstart - s->block_start; /* bytes left in window */ 1718 if (len > (ulg)left + s->strm->avail_in) 1719 len = left + s->strm->avail_in; /* limit len to the input */ 1720 if (len > have) 1721 len = have; /* limit len to the output */ 1722 1723 /* If the stored block would be less than min_block in length, or if 1724 * unable to copy all of the available input when flushing, then try 1725 * copying to the window and the pending buffer instead. Also don't 1726 * write an empty block when flushing -- deflate() does that. 1727 */ 1728 if (len < min_block && ((len == 0 && flush != Z_FINISH) || 1729 flush == Z_NO_FLUSH || 1730 len != left + s->strm->avail_in)) 1731 break; 1732 1733 /* Make a dummy stored block in pending to get the header bytes, 1734 * including any pending bits. This also updates the debugging counts. 1735 */ 1736 last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0; 1737 _tr_stored_block(s, (char *)0, 0L, last); 1738 1739 /* Replace the lengths in the dummy stored block with len. */ 1740 s->pending_buf[s->pending - 4] = len; 1741 s->pending_buf[s->pending - 3] = len >> 8; 1742 s->pending_buf[s->pending - 2] = ~len; 1743 s->pending_buf[s->pending - 1] = ~len >> 8; 1744 1745 /* Write the stored block header bytes. */ 1746 flush_pending(s->strm); 1747 1748 #ifdef ZLIB_DEBUG 1749 /* Update debugging counts for the data about to be copied. */ 1750 s->compressed_len += len << 3; 1751 s->bits_sent += len << 3; 1752 #endif 1753 1754 /* Copy uncompressed bytes from the window to next_out. */ 1755 if (left) { 1756 if (left > len) 1757 left = len; 1758 zmemcpy(s->strm->next_out, s->window + s->block_start, left); 1759 s->strm->next_out += left; 1760 s->strm->avail_out -= left; 1761 s->strm->total_out += left; 1762 s->block_start += left; 1763 len -= left; 1764 } 1765 1766 /* Copy uncompressed bytes directly from next_in to next_out, updating 1767 * the check value. 1768 */ 1769 if (len) { 1770 read_buf(s->strm, s->strm->next_out, len); 1771 s->strm->next_out += len; 1772 s->strm->avail_out -= len; 1773 s->strm->total_out += len; 1774 } 1775 } while (last == 0); 1776 1777 /* Update the sliding window with the last s->w_size bytes of the copied 1778 * data, or append all of the copied data to the existing window if less 1779 * than s->w_size bytes were copied. Also update the number of bytes to 1780 * insert in the hash tables, in the event that deflateParams() switches to 1781 * a non-zero compression level. 1782 */ 1783 used -= s->strm->avail_in; /* number of input bytes directly copied */ 1784 if (used) { 1785 /* If any input was used, then no unused input remains in the window, 1786 * therefore s->block_start == s->strstart. 1787 */ 1788 if (used >= s->w_size) { /* supplant the previous history */ 1789 s->matches = 2; /* clear hash */ 1790 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size); 1791 s->strstart = s->w_size; 1792 s->insert = s->strstart; 1793 } 1794 else { 1795 if (s->window_size - s->strstart <= used) { 1796 /* Slide the window down. */ 1797 s->strstart -= s->w_size; 1798 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1799 if (s->matches < 2) 1800 s->matches++; /* add a pending slide_hash() */ 1801 if (s->insert > s->strstart) 1802 s->insert = s->strstart; 1803 } 1804 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used); 1805 s->strstart += used; 1806 s->insert += MIN(used, s->w_size - s->insert); 1807 } 1808 s->block_start = s->strstart; 1809 } 1810 if (s->high_water < s->strstart) 1811 s->high_water = s->strstart; 1812 1813 /* If the last block was written to next_out, then done. */ 1814 if (last) 1815 return finish_done; 1816 1817 /* If flushing and all input has been consumed, then done. */ 1818 if (flush != Z_NO_FLUSH && flush != Z_FINISH && 1819 s->strm->avail_in == 0 && (long)s->strstart == s->block_start) 1820 return block_done; 1821 1822 /* Fill the window with any remaining input. */ 1823 have = s->window_size - s->strstart; 1824 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) { 1825 /* Slide the window down. */ 1826 s->block_start -= s->w_size; 1827 s->strstart -= s->w_size; 1828 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1829 if (s->matches < 2) 1830 s->matches++; /* add a pending slide_hash() */ 1831 have += s->w_size; /* more space now */ 1832 if (s->insert > s->strstart) 1833 s->insert = s->strstart; 1834 } 1835 if (have > s->strm->avail_in) 1836 have = s->strm->avail_in; 1837 if (have) { 1838 read_buf(s->strm, s->window + s->strstart, have); 1839 s->strstart += have; 1840 s->insert += MIN(have, s->w_size - s->insert); 1841 } 1842 if (s->high_water < s->strstart) 1843 s->high_water = s->strstart; 1844 1845 /* There was not enough avail_out to write a complete worthy or flushed 1846 * stored block to next_out. Write a stored block to pending instead, if we 1847 * have enough input for a worthy block, or if flushing and there is enough 1848 * room for the remaining input as a stored block in the pending buffer. 1849 */ 1850 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1851 /* maximum stored block length that will fit in pending: */ 1852 have = MIN(s->pending_buf_size - have, MAX_STORED); 1853 min_block = MIN(have, s->w_size); 1854 left = s->strstart - s->block_start; 1855 if (left >= min_block || 1856 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH && 1857 s->strm->avail_in == 0 && left <= have)) { 1858 len = MIN(left, have); 1859 last = flush == Z_FINISH && s->strm->avail_in == 0 && 1860 len == left ? 1 : 0; 1861 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last); 1862 s->block_start += len; 1863 flush_pending(s->strm); 1864 } 1865 1866 /* We've done all we can with the available input and output. */ 1867 return last ? finish_started : need_more; 1868 } 1869 1870 /* =========================================================================== 1871 * Compress as much as possible from the input stream, return the current 1872 * block state. 1873 * This function does not perform lazy evaluation of matches and inserts 1874 * new strings in the dictionary only for unmatched strings or for short 1875 * matches. It is used only for the fast compression options. 1876 */ 1877 local block_state deflate_fast(s, flush) 1878 deflate_state *s; 1879 int flush; 1880 { 1881 IPos hash_head; /* head of the hash chain */ 1882 int bflush; /* set if current block must be flushed */ 1883 1884 for (;;) { 1885 /* Make sure that we always have enough lookahead, except 1886 * at the end of the input file. We need MAX_MATCH bytes 1887 * for the next match, plus MIN_MATCH bytes to insert the 1888 * string following the next match. 1889 */ 1890 if (s->lookahead < MIN_LOOKAHEAD) { 1891 fill_window(s); 1892 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1893 return need_more; 1894 } 1895 if (s->lookahead == 0) break; /* flush the current block */ 1896 } 1897 1898 /* Insert the string window[strstart .. strstart + 2] in the 1899 * dictionary, and set hash_head to the head of the hash chain: 1900 */ 1901 hash_head = NIL; 1902 if (s->lookahead >= MIN_MATCH) { 1903 INSERT_STRING(s, s->strstart, hash_head); 1904 } 1905 1906 /* Find the longest match, discarding those <= prev_length. 1907 * At this point we have always match_length < MIN_MATCH 1908 */ 1909 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { 1910 /* To simplify the code, we prevent matches with the string 1911 * of window index 0 (in particular we have to avoid a match 1912 * of the string with itself at the start of the input file). 1913 */ 1914 s->match_length = longest_match (s, hash_head); 1915 /* longest_match() sets match_start */ 1916 } 1917 if (s->match_length >= MIN_MATCH) { 1918 check_match(s, s->strstart, s->match_start, s->match_length); 1919 1920 _tr_tally_dist(s, s->strstart - s->match_start, 1921 s->match_length - MIN_MATCH, bflush); 1922 1923 s->lookahead -= s->match_length; 1924 1925 /* Insert new strings in the hash table only if the match length 1926 * is not too large. This saves time but degrades compression. 1927 */ 1928 #ifndef FASTEST 1929 if (s->match_length <= s->max_insert_length && 1930 s->lookahead >= MIN_MATCH) { 1931 s->match_length--; /* string at strstart already in table */ 1932 do { 1933 s->strstart++; 1934 INSERT_STRING(s, s->strstart, hash_head); 1935 /* strstart never exceeds WSIZE-MAX_MATCH, so there are 1936 * always MIN_MATCH bytes ahead. 1937 */ 1938 } while (--s->match_length != 0); 1939 s->strstart++; 1940 } else 1941 #endif 1942 { 1943 s->strstart += s->match_length; 1944 s->match_length = 0; 1945 s->ins_h = s->window[s->strstart]; 1946 UPDATE_HASH(s, s->ins_h, s->window[s->strstart + 1]); 1947 #if MIN_MATCH != 3 1948 Call UPDATE_HASH() MIN_MATCH-3 more times 1949 #endif 1950 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not 1951 * matter since it will be recomputed at next deflate call. 1952 */ 1953 } 1954 } else { 1955 /* No match, output a literal byte */ 1956 Tracevv((stderr,"%c", s->window[s->strstart])); 1957 _tr_tally_lit(s, s->window[s->strstart], bflush); 1958 s->lookahead--; 1959 s->strstart++; 1960 } 1961 if (bflush) FLUSH_BLOCK(s, 0); 1962 } 1963 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 1964 if (flush == Z_FINISH) { 1965 FLUSH_BLOCK(s, 1); 1966 return finish_done; 1967 } 1968 if (s->sym_next) 1969 FLUSH_BLOCK(s, 0); 1970 return block_done; 1971 } 1972 1973 #ifndef FASTEST 1974 /* =========================================================================== 1975 * Same as above, but achieves better compression. We use a lazy 1976 * evaluation for matches: a match is finally adopted only if there is 1977 * no better match at the next window position. 1978 */ 1979 local block_state deflate_slow(s, flush) 1980 deflate_state *s; 1981 int flush; 1982 { 1983 IPos hash_head; /* head of hash chain */ 1984 int bflush; /* set if current block must be flushed */ 1985 1986 /* Process the input block. */ 1987 for (;;) { 1988 /* Make sure that we always have enough lookahead, except 1989 * at the end of the input file. We need MAX_MATCH bytes 1990 * for the next match, plus MIN_MATCH bytes to insert the 1991 * string following the next match. 1992 */ 1993 if (s->lookahead < MIN_LOOKAHEAD) { 1994 fill_window(s); 1995 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1996 return need_more; 1997 } 1998 if (s->lookahead == 0) break; /* flush the current block */ 1999 } 2000 2001 /* Insert the string window[strstart .. strstart + 2] in the 2002 * dictionary, and set hash_head to the head of the hash chain: 2003 */ 2004 hash_head = NIL; 2005 if (s->lookahead >= MIN_MATCH) { 2006 INSERT_STRING(s, s->strstart, hash_head); 2007 } 2008 2009 /* Find the longest match, discarding those <= prev_length. 2010 */ 2011 s->prev_length = s->match_length, s->prev_match = s->match_start; 2012 s->match_length = MIN_MATCH-1; 2013 2014 if (hash_head != NIL && s->prev_length < s->max_lazy_match && 2015 s->strstart - hash_head <= MAX_DIST(s)) { 2016 /* To simplify the code, we prevent matches with the string 2017 * of window index 0 (in particular we have to avoid a match 2018 * of the string with itself at the start of the input file). 2019 */ 2020 s->match_length = longest_match (s, hash_head); 2021 /* longest_match() sets match_start */ 2022 2023 if (s->match_length <= 5 && (s->strategy == Z_FILTERED 2024 #if TOO_FAR <= 32767 2025 || (s->match_length == MIN_MATCH && 2026 s->strstart - s->match_start > TOO_FAR) 2027 #endif 2028 )) { 2029 2030 /* If prev_match is also MIN_MATCH, match_start is garbage 2031 * but we will ignore the current match anyway. 2032 */ 2033 s->match_length = MIN_MATCH-1; 2034 } 2035 } 2036 /* If there was a match at the previous step and the current 2037 * match is not better, output the previous match: 2038 */ 2039 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { 2040 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; 2041 /* Do not insert strings in hash table beyond this. */ 2042 2043 check_match(s, s->strstart - 1, s->prev_match, s->prev_length); 2044 2045 _tr_tally_dist(s, s->strstart - 1 - s->prev_match, 2046 s->prev_length - MIN_MATCH, bflush); 2047 2048 /* Insert in hash table all strings up to the end of the match. 2049 * strstart - 1 and strstart are already inserted. If there is not 2050 * enough lookahead, the last two strings are not inserted in 2051 * the hash table. 2052 */ 2053 s->lookahead -= s->prev_length - 1; 2054 s->prev_length -= 2; 2055 do { 2056 if (++s->strstart <= max_insert) { 2057 INSERT_STRING(s, s->strstart, hash_head); 2058 } 2059 } while (--s->prev_length != 0); 2060 s->match_available = 0; 2061 s->match_length = MIN_MATCH-1; 2062 s->strstart++; 2063 2064 if (bflush) FLUSH_BLOCK(s, 0); 2065 2066 } else if (s->match_available) { 2067 /* If there was no match at the previous position, output a 2068 * single literal. If there was a match but the current match 2069 * is longer, truncate the previous match to a single literal. 2070 */ 2071 Tracevv((stderr,"%c", s->window[s->strstart - 1])); 2072 _tr_tally_lit(s, s->window[s->strstart - 1], bflush); 2073 if (bflush) { 2074 FLUSH_BLOCK_ONLY(s, 0); 2075 } 2076 s->strstart++; 2077 s->lookahead--; 2078 if (s->strm->avail_out == 0) return need_more; 2079 } else { 2080 /* There is no previous match to compare with, wait for 2081 * the next step to decide. 2082 */ 2083 s->match_available = 1; 2084 s->strstart++; 2085 s->lookahead--; 2086 } 2087 } 2088 Assert (flush != Z_NO_FLUSH, "no flush?"); 2089 if (s->match_available) { 2090 Tracevv((stderr,"%c", s->window[s->strstart - 1])); 2091 _tr_tally_lit(s, s->window[s->strstart - 1], bflush); 2092 s->match_available = 0; 2093 } 2094 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 2095 if (flush == Z_FINISH) { 2096 FLUSH_BLOCK(s, 1); 2097 return finish_done; 2098 } 2099 if (s->sym_next) 2100 FLUSH_BLOCK(s, 0); 2101 return block_done; 2102 } 2103 #endif /* FASTEST */ 2104 2105 /* =========================================================================== 2106 * For Z_RLE, simply look for runs of bytes, generate matches only of distance 2107 * one. Do not maintain a hash table. (It will be regenerated if this run of 2108 * deflate switches away from Z_RLE.) 2109 */ 2110 local block_state deflate_rle(s, flush) 2111 deflate_state *s; 2112 int flush; 2113 { 2114 int bflush; /* set if current block must be flushed */ 2115 uInt prev; /* byte at distance one to match */ 2116 Bytef *scan, *strend; /* scan goes up to strend for length of run */ 2117 2118 for (;;) { 2119 /* Make sure that we always have enough lookahead, except 2120 * at the end of the input file. We need MAX_MATCH bytes 2121 * for the longest run, plus one for the unrolled loop. 2122 */ 2123 if (s->lookahead <= MAX_MATCH) { 2124 fill_window(s); 2125 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) { 2126 return need_more; 2127 } 2128 if (s->lookahead == 0) break; /* flush the current block */ 2129 } 2130 2131 /* See how many times the previous byte repeats */ 2132 s->match_length = 0; 2133 if (s->lookahead >= MIN_MATCH && s->strstart > 0) { 2134 scan = s->window + s->strstart - 1; 2135 prev = *scan; 2136 if (prev == *++scan && prev == *++scan && prev == *++scan) { 2137 strend = s->window + s->strstart + MAX_MATCH; 2138 do { 2139 } while (prev == *++scan && prev == *++scan && 2140 prev == *++scan && prev == *++scan && 2141 prev == *++scan && prev == *++scan && 2142 prev == *++scan && prev == *++scan && 2143 scan < strend); 2144 s->match_length = MAX_MATCH - (uInt)(strend - scan); 2145 if (s->match_length > s->lookahead) 2146 s->match_length = s->lookahead; 2147 } 2148 Assert(scan <= s->window + (uInt)(s->window_size - 1), 2149 "wild scan"); 2150 } 2151 2152 /* Emit match if have run of MIN_MATCH or longer, else emit literal */ 2153 if (s->match_length >= MIN_MATCH) { 2154 check_match(s, s->strstart, s->strstart - 1, s->match_length); 2155 2156 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush); 2157 2158 s->lookahead -= s->match_length; 2159 s->strstart += s->match_length; 2160 s->match_length = 0; 2161 } else { 2162 /* No match, output a literal byte */ 2163 Tracevv((stderr,"%c", s->window[s->strstart])); 2164 _tr_tally_lit(s, s->window[s->strstart], bflush); 2165 s->lookahead--; 2166 s->strstart++; 2167 } 2168 if (bflush) FLUSH_BLOCK(s, 0); 2169 } 2170 s->insert = 0; 2171 if (flush == Z_FINISH) { 2172 FLUSH_BLOCK(s, 1); 2173 return finish_done; 2174 } 2175 if (s->sym_next) 2176 FLUSH_BLOCK(s, 0); 2177 return block_done; 2178 } 2179 2180 /* =========================================================================== 2181 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table. 2182 * (It will be regenerated if this run of deflate switches away from Huffman.) 2183 */ 2184 local block_state deflate_huff(s, flush) 2185 deflate_state *s; 2186 int flush; 2187 { 2188 int bflush; /* set if current block must be flushed */ 2189 2190 for (;;) { 2191 /* Make sure that we have a literal to write. */ 2192 if (s->lookahead == 0) { 2193 fill_window(s); 2194 if (s->lookahead == 0) { 2195 if (flush == Z_NO_FLUSH) 2196 return need_more; 2197 break; /* flush the current block */ 2198 } 2199 } 2200 2201 /* Output a literal byte */ 2202 s->match_length = 0; 2203 Tracevv((stderr,"%c", s->window[s->strstart])); 2204 _tr_tally_lit(s, s->window[s->strstart], bflush); 2205 s->lookahead--; 2206 s->strstart++; 2207 if (bflush) FLUSH_BLOCK(s, 0); 2208 } 2209 s->insert = 0; 2210 if (flush == Z_FINISH) { 2211 FLUSH_BLOCK(s, 1); 2212 return finish_done; 2213 } 2214 if (s->sym_next) 2215 FLUSH_BLOCK(s, 0); 2216 return block_done; 2217 } 2218