1 /* deflate.c -- compress data using the deflation algorithm 2 * Copyright (C) 1995-2005 Jean-loup Gailly. 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://www.ietf.org/rfc/rfc1951.txt 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 1.2.3 Copyright 1995-2005 Jean-loup Gailly "; 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 void fill_window OF((deflate_state *s)); 77 local block_state deflate_stored OF((deflate_state *s, int flush)); 78 local block_state deflate_fast OF((deflate_state *s, int flush)); 79 #ifndef FASTEST 80 local block_state deflate_slow OF((deflate_state *s, int flush)); 81 #endif 82 local void lm_init OF((deflate_state *s)); 83 local void putShortMSB OF((deflate_state *s, uInt b)); 84 local void flush_pending OF((z_streamp strm)); 85 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); 86 #ifndef FASTEST 87 #ifdef ASMV 88 void match_init OF((void)); /* asm code initialization */ 89 uInt longest_match OF((deflate_state *s, IPos cur_match)); 90 #else 91 local uInt longest_match OF((deflate_state *s, IPos cur_match)); 92 #endif 93 #endif 94 local uInt longest_match_fast OF((deflate_state *s, IPos cur_match)); 95 96 #ifdef DEBUG 97 local void check_match OF((deflate_state *s, IPos start, IPos match, 98 int length)); 99 #endif 100 101 /* =========================================================================== 102 * Local data 103 */ 104 105 #define NIL 0 106 /* Tail of hash chains */ 107 108 #ifndef TOO_FAR 109 # define TOO_FAR 4096 110 #endif 111 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ 112 113 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) 114 /* Minimum amount of lookahead, except at the end of the input file. 115 * See deflate.c for comments about the MIN_MATCH+1. 116 */ 117 118 /* Values for max_lazy_match, good_match and max_chain_length, depending on 119 * the desired pack level (0..9). The values given below have been tuned to 120 * exclude worst case performance for pathological files. Better values may be 121 * found for specific files. 122 */ 123 typedef struct config_s { 124 ush good_length; /* reduce lazy search above this match length */ 125 ush max_lazy; /* do not perform lazy search above this match length */ 126 ush nice_length; /* quit search above this match length */ 127 ush max_chain; 128 compress_func func; 129 } config; 130 131 #ifdef FASTEST 132 local const config configuration_table[2] = { 133 /* good lazy nice chain */ 134 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 135 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */ 136 #else 137 local const config configuration_table[10] = { 138 /* good lazy nice chain */ 139 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 140 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */ 141 /* 2 */ {4, 5, 16, 8, deflate_fast}, 142 /* 3 */ {4, 6, 32, 32, deflate_fast}, 143 144 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ 145 /* 5 */ {8, 16, 32, 32, deflate_slow}, 146 /* 6 */ {8, 16, 128, 128, deflate_slow}, 147 /* 7 */ {8, 32, 128, 256, deflate_slow}, 148 /* 8 */ {32, 128, 258, 1024, deflate_slow}, 149 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */ 150 #endif 151 152 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 153 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different 154 * meaning. 155 */ 156 157 #define EQUAL 0 158 /* result of memcmp for equal strings */ 159 160 #ifndef NO_DUMMY_DECL 161 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */ 162 #endif 163 164 /* =========================================================================== 165 * Update a hash value with the given input byte 166 * IN assertion: all calls to to UPDATE_HASH are made with consecutive 167 * input characters, so that a running hash key can be computed from the 168 * previous key instead of complete recalculation each time. 169 */ 170 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask) 171 172 173 /* =========================================================================== 174 * Insert string str in the dictionary and set match_head to the previous head 175 * of the hash chain (the most recent string with same hash key). Return 176 * the previous length of the hash chain. 177 * If this file is compiled with -DFASTEST, the compression level is forced 178 * to 1, and no hash chains are maintained. 179 * IN assertion: all calls to to INSERT_STRING are made with consecutive 180 * input characters and the first MIN_MATCH bytes of str are valid 181 * (except for the last MIN_MATCH-1 bytes of the input file). 182 */ 183 #ifdef FASTEST 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->head[s->ins_h], \ 187 s->head[s->ins_h] = (Pos)(str)) 188 #else 189 #define INSERT_STRING(s, str, match_head) \ 190 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 191 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \ 192 s->head[s->ins_h] = (Pos)(str)) 193 #endif 194 195 /* =========================================================================== 196 * Initialize the hash table (avoiding 64K overflow for 16 bit systems). 197 * prev[] will be initialized on the fly. 198 */ 199 #define CLEAR_HASH(s) \ 200 s->head[s->hash_size-1] = NIL; \ 201 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head)); 202 203 /* ========================================================================= */ 204 int ZEXPORT deflateInit_( 205 z_streamp strm, 206 int level, 207 const char *version, 208 int stream_size) 209 { 210 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, 211 Z_DEFAULT_STRATEGY, version, stream_size); 212 /* To do: ignore strm->next_in if we use it as window */ 213 } 214 215 /* ========================================================================= */ 216 int ZEXPORT deflateInit2_( 217 z_streamp strm, 218 int level, 219 int method, 220 int windowBits, 221 int memLevel, 222 int strategy, 223 const char *version, 224 int stream_size) 225 { 226 deflate_state *s; 227 int wrap = 1; 228 static const char my_version[] = ZLIB_VERSION; 229 230 ushf *overlay; 231 /* We overlay pending_buf and d_buf+l_buf. This works since the average 232 * output size for (length,distance) codes is <= 24 bits. 233 */ 234 235 if (version == Z_NULL || version[0] != my_version[0] || 236 stream_size != sizeof(z_stream)) { 237 return Z_VERSION_ERROR; 238 } 239 if (strm == Z_NULL) return Z_STREAM_ERROR; 240 241 strm->msg = Z_NULL; 242 if (strm->zalloc == (alloc_func)0) { 243 strm->zalloc = zcalloc; 244 strm->opaque = (voidpf)0; 245 } 246 if (strm->zfree == (free_func)0) strm->zfree = zcfree; 247 248 #ifdef FASTEST 249 if (level != 0) level = 1; 250 #else 251 if (level == Z_DEFAULT_COMPRESSION) level = 6; 252 #endif 253 254 if (windowBits < 0) { /* suppress zlib wrapper */ 255 wrap = 0; 256 windowBits = -windowBits; 257 } 258 #ifdef GZIP 259 else if (windowBits > 15) { 260 wrap = 2; /* write gzip wrapper instead */ 261 windowBits -= 16; 262 } 263 #endif 264 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || 265 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || 266 strategy < 0 || strategy > Z_FIXED) { 267 return Z_STREAM_ERROR; 268 } 269 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ 270 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); 271 if (s == Z_NULL) return Z_MEM_ERROR; 272 strm->state = (struct internal_state FAR *)s; 273 s->strm = strm; 274 275 s->wrap = wrap; 276 s->gzhead = Z_NULL; 277 s->w_bits = windowBits; 278 s->w_size = 1 << s->w_bits; 279 s->w_mask = s->w_size - 1; 280 281 s->hash_bits = memLevel + 7; 282 s->hash_size = 1 << s->hash_bits; 283 s->hash_mask = s->hash_size - 1; 284 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); 285 286 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); 287 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); 288 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); 289 290 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ 291 292 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2); 293 s->pending_buf = (uchf *) overlay; 294 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); 295 296 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || 297 s->pending_buf == Z_NULL) { 298 s->status = FINISH_STATE; 299 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR); 300 deflateEnd (strm); 301 return Z_MEM_ERROR; 302 } 303 s->d_buf = overlay + s->lit_bufsize/sizeof(ush); 304 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; 305 306 s->level = level; 307 s->strategy = strategy; 308 s->method = (Byte)method; 309 310 return deflateReset(strm); 311 } 312 313 /* ========================================================================= */ 314 int ZEXPORT deflateSetDictionary ( 315 z_streamp strm, 316 const Bytef *dictionary, 317 uInt dictLength) 318 { 319 deflate_state *s; 320 uInt length = dictLength; 321 uInt n; 322 IPos hash_head = 0; 323 324 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL || 325 strm->state->wrap == 2 || 326 (strm->state->wrap == 1 && strm->state->status != INIT_STATE)) 327 return Z_STREAM_ERROR; 328 329 s = strm->state; 330 if (s->wrap) 331 strm->adler = adler32(strm->adler, dictionary, dictLength); 332 333 if (length < MIN_MATCH) return Z_OK; 334 if (length > MAX_DIST(s)) { 335 length = MAX_DIST(s); 336 dictionary += dictLength - length; /* use the tail of the dictionary */ 337 } 338 zmemcpy(s->window, dictionary, length); 339 s->strstart = length; 340 s->block_start = (long)length; 341 342 /* Insert all strings in the hash table (except for the last two bytes). 343 * s->lookahead stays null, so s->ins_h will be recomputed at the next 344 * call of fill_window. 345 */ 346 s->ins_h = s->window[0]; 347 UPDATE_HASH(s, s->ins_h, s->window[1]); 348 for (n = 0; n <= length - MIN_MATCH; n++) { 349 INSERT_STRING(s, n, hash_head); 350 } 351 if (hash_head) hash_head = 0; /* to make compiler happy */ 352 return Z_OK; 353 } 354 355 /* ========================================================================= */ 356 int ZEXPORT deflateReset ( 357 z_streamp strm) 358 { 359 deflate_state *s; 360 361 if (strm == Z_NULL || strm->state == Z_NULL || 362 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) { 363 return Z_STREAM_ERROR; 364 } 365 366 strm->total_in = strm->total_out = 0; 367 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ 368 strm->data_type = Z_UNKNOWN; 369 370 s = (deflate_state *)strm->state; 371 s->pending = 0; 372 s->pending_out = s->pending_buf; 373 374 if (s->wrap < 0) { 375 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ 376 } 377 s->status = s->wrap ? INIT_STATE : BUSY_STATE; 378 strm->adler = 379 #ifdef GZIP 380 s->wrap == 2 ? crc32(0L, Z_NULL, 0) : 381 #endif 382 adler32(0L, Z_NULL, 0); 383 s->last_flush = Z_NO_FLUSH; 384 385 _tr_init(s); 386 lm_init(s); 387 388 return Z_OK; 389 } 390 391 /* ========================================================================= */ 392 int ZEXPORT deflateSetHeader ( 393 z_streamp strm, 394 gz_headerp head) 395 { 396 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; 397 if (strm->state->wrap != 2) return Z_STREAM_ERROR; 398 strm->state->gzhead = head; 399 return Z_OK; 400 } 401 402 /* ========================================================================= */ 403 int ZEXPORT deflatePrime ( 404 z_streamp strm, 405 int bits, 406 int value) 407 { 408 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; 409 strm->state->bi_valid = bits; 410 strm->state->bi_buf = (ush)(value & ((1 << bits) - 1)); 411 return Z_OK; 412 } 413 414 /* ========================================================================= */ 415 int ZEXPORT deflateParams( 416 z_streamp strm, 417 int level, 418 int strategy) 419 { 420 deflate_state *s; 421 compress_func func; 422 int err = Z_OK; 423 424 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; 425 s = strm->state; 426 427 #ifdef FASTEST 428 if (level != 0) level = 1; 429 #else 430 if (level == Z_DEFAULT_COMPRESSION) level = 6; 431 #endif 432 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { 433 return Z_STREAM_ERROR; 434 } 435 func = configuration_table[s->level].func; 436 437 if (func != configuration_table[level].func && strm->total_in != 0) { 438 /* Flush the last buffer: */ 439 err = deflate(strm, Z_PARTIAL_FLUSH); 440 } 441 if (s->level != level) { 442 s->level = level; 443 s->max_lazy_match = configuration_table[level].max_lazy; 444 s->good_match = configuration_table[level].good_length; 445 s->nice_match = configuration_table[level].nice_length; 446 s->max_chain_length = configuration_table[level].max_chain; 447 } 448 s->strategy = strategy; 449 return err; 450 } 451 452 /* ========================================================================= */ 453 int ZEXPORT deflateTune( 454 z_streamp strm, 455 int good_length, 456 int max_lazy, 457 int nice_length, 458 int max_chain) 459 { 460 deflate_state *s; 461 462 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; 463 s = strm->state; 464 s->good_match = good_length; 465 s->max_lazy_match = max_lazy; 466 s->nice_match = nice_length; 467 s->max_chain_length = max_chain; 468 return Z_OK; 469 } 470 471 /* ========================================================================= 472 * For the default windowBits of 15 and memLevel of 8, this function returns 473 * a close to exact, as well as small, upper bound on the compressed size. 474 * They are coded as constants here for a reason--if the #define's are 475 * changed, then this function needs to be changed as well. The return 476 * value for 15 and 8 only works for those exact settings. 477 * 478 * For any setting other than those defaults for windowBits and memLevel, 479 * the value returned is a conservative worst case for the maximum expansion 480 * resulting from using fixed blocks instead of stored blocks, which deflate 481 * can emit on compressed data for some combinations of the parameters. 482 * 483 * This function could be more sophisticated to provide closer upper bounds 484 * for every combination of windowBits and memLevel, as well as wrap. 485 * But even the conservative upper bound of about 14% expansion does not 486 * seem onerous for output buffer allocation. 487 */ 488 uLong ZEXPORT deflateBound( 489 z_streamp strm, 490 uLong sourceLen) 491 { 492 deflate_state *s; 493 uLong destLen; 494 495 /* conservative upper bound */ 496 destLen = sourceLen + 497 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 11; 498 499 /* if can't get parameters, return conservative bound */ 500 if (strm == Z_NULL || strm->state == Z_NULL) 501 return destLen; 502 503 /* if not default parameters, return conservative bound */ 504 s = strm->state; 505 if (s->w_bits != 15 || s->hash_bits != 8 + 7) 506 return destLen; 507 508 /* default settings: return tight bound for that case */ 509 return compressBound(sourceLen); 510 } 511 512 /* ========================================================================= 513 * Put a short in the pending buffer. The 16-bit value is put in MSB order. 514 * IN assertion: the stream state is correct and there is enough room in 515 * pending_buf. 516 */ 517 local void putShortMSB ( 518 deflate_state *s, 519 uInt b) 520 { 521 put_byte(s, (Byte)(b >> 8)); 522 put_byte(s, (Byte)(b & 0xff)); 523 } 524 525 /* ========================================================================= 526 * Flush as much pending output as possible. All deflate() output goes 527 * through this function so some applications may wish to modify it 528 * to avoid allocating a large strm->next_out buffer and copying into it. 529 * (See also read_buf()). 530 */ 531 local void flush_pending( 532 z_streamp strm) 533 { 534 unsigned len = strm->state->pending; 535 536 if (len > strm->avail_out) len = strm->avail_out; 537 if (len == 0) return; 538 539 zmemcpy(strm->next_out, strm->state->pending_out, len); 540 strm->next_out += len; 541 strm->state->pending_out += len; 542 strm->total_out += len; 543 strm->avail_out -= len; 544 strm->state->pending -= len; 545 if (strm->state->pending == 0) { 546 strm->state->pending_out = strm->state->pending_buf; 547 } 548 } 549 550 /* ========================================================================= */ 551 int ZEXPORT deflate ( 552 z_streamp strm, 553 int flush) 554 { 555 int old_flush; /* value of flush param for previous deflate call */ 556 deflate_state *s; 557 558 if (strm == Z_NULL || strm->state == Z_NULL || 559 flush > Z_FINISH || flush < 0) { 560 return Z_STREAM_ERROR; 561 } 562 s = strm->state; 563 564 if (strm->next_out == Z_NULL || 565 (strm->next_in == Z_NULL && strm->avail_in != 0) || 566 (s->status == FINISH_STATE && flush != Z_FINISH)) { 567 ERR_RETURN(strm, Z_STREAM_ERROR); 568 } 569 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); 570 571 s->strm = strm; /* just in case */ 572 old_flush = s->last_flush; 573 s->last_flush = flush; 574 575 /* Write the header */ 576 if (s->status == INIT_STATE) { 577 #ifdef GZIP 578 if (s->wrap == 2) { 579 strm->adler = crc32(0L, Z_NULL, 0); 580 put_byte(s, 31); 581 put_byte(s, 139); 582 put_byte(s, 8); 583 if (s->gzhead == NULL) { 584 put_byte(s, 0); 585 put_byte(s, 0); 586 put_byte(s, 0); 587 put_byte(s, 0); 588 put_byte(s, 0); 589 put_byte(s, s->level == 9 ? 2 : 590 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 591 4 : 0)); 592 put_byte(s, OS_CODE); 593 s->status = BUSY_STATE; 594 } 595 else { 596 put_byte(s, (s->gzhead->text ? 1 : 0) + 597 (s->gzhead->hcrc ? 2 : 0) + 598 (s->gzhead->extra == Z_NULL ? 0 : 4) + 599 (s->gzhead->name == Z_NULL ? 0 : 8) + 600 (s->gzhead->comment == Z_NULL ? 0 : 16) 601 ); 602 put_byte(s, (Byte)(s->gzhead->time & 0xff)); 603 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); 604 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); 605 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); 606 put_byte(s, s->level == 9 ? 2 : 607 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 608 4 : 0)); 609 put_byte(s, s->gzhead->os & 0xff); 610 if (s->gzhead->extra != NULL) { 611 put_byte(s, s->gzhead->extra_len & 0xff); 612 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); 613 } 614 if (s->gzhead->hcrc) 615 strm->adler = crc32(strm->adler, s->pending_buf, 616 s->pending); 617 s->gzindex = 0; 618 s->status = EXTRA_STATE; 619 } 620 } 621 else 622 #endif 623 { 624 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; 625 uInt level_flags; 626 627 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) 628 level_flags = 0; 629 else if (s->level < 6) 630 level_flags = 1; 631 else if (s->level == 6) 632 level_flags = 2; 633 else 634 level_flags = 3; 635 header |= (level_flags << 6); 636 if (s->strstart != 0) header |= PRESET_DICT; 637 header += 31 - (header % 31); 638 639 s->status = BUSY_STATE; 640 putShortMSB(s, header); 641 642 /* Save the adler32 of the preset dictionary: */ 643 if (s->strstart != 0) { 644 putShortMSB(s, (uInt)(strm->adler >> 16)); 645 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 646 } 647 strm->adler = adler32(0L, Z_NULL, 0); 648 } 649 } 650 #ifdef GZIP 651 if (s->status == EXTRA_STATE) { 652 if (s->gzhead->extra != NULL) { 653 uInt beg = s->pending; /* start of bytes to update crc */ 654 655 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) { 656 if (s->pending == s->pending_buf_size) { 657 if (s->gzhead->hcrc && s->pending > beg) 658 strm->adler = crc32(strm->adler, s->pending_buf + beg, 659 s->pending - beg); 660 flush_pending(strm); 661 beg = s->pending; 662 if (s->pending == s->pending_buf_size) 663 break; 664 } 665 put_byte(s, s->gzhead->extra[s->gzindex]); 666 s->gzindex++; 667 } 668 if (s->gzhead->hcrc && s->pending > beg) 669 strm->adler = crc32(strm->adler, s->pending_buf + beg, 670 s->pending - beg); 671 if (s->gzindex == s->gzhead->extra_len) { 672 s->gzindex = 0; 673 s->status = NAME_STATE; 674 } 675 } 676 else 677 s->status = NAME_STATE; 678 } 679 if (s->status == NAME_STATE) { 680 if (s->gzhead->name != NULL) { 681 uInt beg = s->pending; /* start of bytes to update crc */ 682 int val; 683 684 do { 685 if (s->pending == s->pending_buf_size) { 686 if (s->gzhead->hcrc && s->pending > beg) 687 strm->adler = crc32(strm->adler, s->pending_buf + beg, 688 s->pending - beg); 689 flush_pending(strm); 690 beg = s->pending; 691 if (s->pending == s->pending_buf_size) { 692 val = 1; 693 break; 694 } 695 } 696 val = s->gzhead->name[s->gzindex++]; 697 put_byte(s, val); 698 } while (val != 0); 699 if (s->gzhead->hcrc && s->pending > beg) 700 strm->adler = crc32(strm->adler, s->pending_buf + beg, 701 s->pending - beg); 702 if (val == 0) { 703 s->gzindex = 0; 704 s->status = COMMENT_STATE; 705 } 706 } 707 else 708 s->status = COMMENT_STATE; 709 } 710 if (s->status == COMMENT_STATE) { 711 if (s->gzhead->comment != NULL) { 712 uInt beg = s->pending; /* start of bytes to update crc */ 713 int val; 714 715 do { 716 if (s->pending == s->pending_buf_size) { 717 if (s->gzhead->hcrc && s->pending > beg) 718 strm->adler = crc32(strm->adler, s->pending_buf + beg, 719 s->pending - beg); 720 flush_pending(strm); 721 beg = s->pending; 722 if (s->pending == s->pending_buf_size) { 723 val = 1; 724 break; 725 } 726 } 727 val = s->gzhead->comment[s->gzindex++]; 728 put_byte(s, val); 729 } while (val != 0); 730 if (s->gzhead->hcrc && s->pending > beg) 731 strm->adler = crc32(strm->adler, s->pending_buf + beg, 732 s->pending - beg); 733 if (val == 0) 734 s->status = HCRC_STATE; 735 } 736 else 737 s->status = HCRC_STATE; 738 } 739 if (s->status == HCRC_STATE) { 740 if (s->gzhead->hcrc) { 741 if (s->pending + 2 > s->pending_buf_size) 742 flush_pending(strm); 743 if (s->pending + 2 <= s->pending_buf_size) { 744 put_byte(s, (Byte)(strm->adler & 0xff)); 745 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 746 strm->adler = crc32(0L, Z_NULL, 0); 747 s->status = BUSY_STATE; 748 } 749 } 750 else 751 s->status = BUSY_STATE; 752 } 753 #endif 754 755 /* Flush as much pending output as possible */ 756 if (s->pending != 0) { 757 flush_pending(strm); 758 if (strm->avail_out == 0) { 759 /* Since avail_out is 0, deflate will be called again with 760 * more output space, but possibly with both pending and 761 * avail_in equal to zero. There won't be anything to do, 762 * but this is not an error situation so make sure we 763 * return OK instead of BUF_ERROR at next call of deflate: 764 */ 765 s->last_flush = -1; 766 return Z_OK; 767 } 768 769 /* Make sure there is something to do and avoid duplicate consecutive 770 * flushes. For repeated and useless calls with Z_FINISH, we keep 771 * returning Z_STREAM_END instead of Z_BUF_ERROR. 772 */ 773 } else if (strm->avail_in == 0 && flush <= old_flush && 774 flush != Z_FINISH) { 775 ERR_RETURN(strm, Z_BUF_ERROR); 776 } 777 778 /* User must not provide more input after the first FINISH: */ 779 if (s->status == FINISH_STATE && strm->avail_in != 0) { 780 ERR_RETURN(strm, Z_BUF_ERROR); 781 } 782 783 /* Start a new block or continue the current one. 784 */ 785 if (strm->avail_in != 0 || s->lookahead != 0 || 786 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { 787 block_state bstate; 788 789 bstate = (*(configuration_table[s->level].func))(s, flush); 790 791 if (bstate == finish_started || bstate == finish_done) { 792 s->status = FINISH_STATE; 793 } 794 if (bstate == need_more || bstate == finish_started) { 795 if (strm->avail_out == 0) { 796 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ 797 } 798 return Z_OK; 799 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call 800 * of deflate should use the same flush parameter to make sure 801 * that the flush is complete. So we don't have to output an 802 * empty block here, this will be done at next call. This also 803 * ensures that for a very small output buffer, we emit at most 804 * one empty block. 805 */ 806 } 807 if (bstate == block_done) { 808 if (flush == Z_PARTIAL_FLUSH) { 809 _tr_align(s); 810 } else { /* FULL_FLUSH or SYNC_FLUSH */ 811 _tr_stored_block(s, (char*)0, 0L, 0); 812 /* For a full flush, this empty block will be recognized 813 * as a special marker by inflate_sync(). 814 */ 815 if (flush == Z_FULL_FLUSH) { 816 CLEAR_HASH(s); /* forget history */ 817 } 818 } 819 flush_pending(strm); 820 if (strm->avail_out == 0) { 821 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ 822 return Z_OK; 823 } 824 } 825 } 826 Assert(strm->avail_out > 0, "bug2"); 827 828 if (flush != Z_FINISH) return Z_OK; 829 if (s->wrap <= 0) return Z_STREAM_END; 830 831 /* Write the trailer */ 832 #ifdef GZIP 833 if (s->wrap == 2) { 834 put_byte(s, (Byte)(strm->adler & 0xff)); 835 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 836 put_byte(s, (Byte)((strm->adler >> 16) & 0xff)); 837 put_byte(s, (Byte)((strm->adler >> 24) & 0xff)); 838 put_byte(s, (Byte)(strm->total_in & 0xff)); 839 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff)); 840 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff)); 841 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff)); 842 } 843 else 844 #endif 845 { 846 putShortMSB(s, (uInt)(strm->adler >> 16)); 847 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 848 } 849 flush_pending(strm); 850 /* If avail_out is zero, the application will call deflate again 851 * to flush the rest. 852 */ 853 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ 854 return s->pending != 0 ? Z_OK : Z_STREAM_END; 855 } 856 857 /* ========================================================================= */ 858 int ZEXPORT deflateEnd ( 859 z_streamp strm) 860 { 861 int status; 862 863 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; 864 865 status = strm->state->status; 866 if (status != INIT_STATE && 867 status != EXTRA_STATE && 868 status != NAME_STATE && 869 status != COMMENT_STATE && 870 status != HCRC_STATE && 871 status != BUSY_STATE && 872 status != FINISH_STATE) { 873 return Z_STREAM_ERROR; 874 } 875 876 /* Deallocate in reverse order of allocations: */ 877 TRY_FREE(strm, strm->state->pending_buf); 878 TRY_FREE(strm, strm->state->head); 879 TRY_FREE(strm, strm->state->prev); 880 TRY_FREE(strm, strm->state->window); 881 882 ZFREE(strm, strm->state); 883 strm->state = Z_NULL; 884 885 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; 886 } 887 888 /* ========================================================================= 889 * Copy the source state to the destination state. 890 * To simplify the source, this is not supported for 16-bit MSDOS (which 891 * doesn't have enough memory anyway to duplicate compression states). 892 */ 893 int ZEXPORT deflateCopy ( 894 z_streamp dest, 895 z_streamp source) 896 { 897 #ifdef MAXSEG_64K 898 return Z_STREAM_ERROR; 899 #else 900 deflate_state *ds; 901 deflate_state *ss; 902 ushf *overlay; 903 904 905 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) { 906 return Z_STREAM_ERROR; 907 } 908 909 ss = source->state; 910 911 zmemcpy(dest, source, sizeof(z_stream)); 912 913 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); 914 if (ds == Z_NULL) return Z_MEM_ERROR; 915 dest->state = (struct internal_state FAR *) ds; 916 zmemcpy(ds, ss, sizeof(deflate_state)); 917 ds->strm = dest; 918 919 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); 920 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); 921 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); 922 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2); 923 ds->pending_buf = (uchf *) overlay; 924 925 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || 926 ds->pending_buf == Z_NULL) { 927 deflateEnd (dest); 928 return Z_MEM_ERROR; 929 } 930 /* following zmemcpy do not work for 16-bit MSDOS */ 931 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); 932 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos)); 933 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos)); 934 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); 935 936 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); 937 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); 938 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; 939 940 ds->l_desc.dyn_tree = ds->dyn_ltree; 941 ds->d_desc.dyn_tree = ds->dyn_dtree; 942 ds->bl_desc.dyn_tree = ds->bl_tree; 943 944 return Z_OK; 945 #endif /* MAXSEG_64K */ 946 } 947 948 /* =========================================================================== 949 * Read a new buffer from the current input stream, update the adler32 950 * and total number of bytes read. All deflate() input goes through 951 * this function so some applications may wish to modify it to avoid 952 * allocating a large strm->next_in buffer and copying from it. 953 * (See also flush_pending()). 954 */ 955 local int read_buf( 956 z_streamp strm, 957 Bytef *buf, 958 unsigned size) 959 { 960 unsigned len = strm->avail_in; 961 962 if (len > size) len = size; 963 if (len == 0) return 0; 964 965 strm->avail_in -= len; 966 967 if (strm->state->wrap == 1) { 968 strm->adler = adler32(strm->adler, strm->next_in, len); 969 } 970 #ifdef GZIP 971 else if (strm->state->wrap == 2) { 972 strm->adler = crc32(strm->adler, strm->next_in, len); 973 } 974 #endif 975 zmemcpy(buf, strm->next_in, len); 976 strm->next_in += len; 977 strm->total_in += len; 978 979 return (int)len; 980 } 981 982 /* =========================================================================== 983 * Initialize the "longest match" routines for a new zlib stream 984 */ 985 local void lm_init ( 986 deflate_state *s) 987 { 988 s->window_size = (ulg)2L*s->w_size; 989 990 CLEAR_HASH(s); 991 992 /* Set the default configuration parameters: 993 */ 994 s->max_lazy_match = configuration_table[s->level].max_lazy; 995 s->good_match = configuration_table[s->level].good_length; 996 s->nice_match = configuration_table[s->level].nice_length; 997 s->max_chain_length = configuration_table[s->level].max_chain; 998 999 s->strstart = 0; 1000 s->block_start = 0L; 1001 s->lookahead = 0; 1002 s->match_length = s->prev_length = MIN_MATCH-1; 1003 s->match_available = 0; 1004 s->ins_h = 0; 1005 #ifndef FASTEST 1006 #ifdef ASMV 1007 match_init(); /* initialize the asm code */ 1008 #endif 1009 #endif 1010 } 1011 1012 #ifndef FASTEST 1013 /* =========================================================================== 1014 * Set match_start to the longest match starting at the given string and 1015 * return its length. Matches shorter or equal to prev_length are discarded, 1016 * in which case the result is equal to prev_length and match_start is 1017 * garbage. 1018 * IN assertions: cur_match is the head of the hash chain for the current 1019 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 1020 * OUT assertion: the match length is not greater than s->lookahead. 1021 */ 1022 #ifndef ASMV 1023 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or 1024 * match.S. The code will be functionally equivalent. 1025 */ 1026 local uInt longest_match( 1027 deflate_state *s, 1028 IPos cur_match) 1029 { 1030 unsigned chain_length = s->max_chain_length;/* max hash chain length */ 1031 register Bytef *scan = s->window + s->strstart; /* current string */ 1032 register Bytef *match; /* matched string */ 1033 register int len; /* length of current match */ 1034 int best_len = s->prev_length; /* best match length so far */ 1035 int nice_match = s->nice_match; /* stop if match long enough */ 1036 IPos limit = s->strstart > (IPos)MAX_DIST(s) ? 1037 s->strstart - (IPos)MAX_DIST(s) : NIL; 1038 /* Stop when cur_match becomes <= limit. To simplify the code, 1039 * we prevent matches with the string of window index 0. 1040 */ 1041 Posf *prev = s->prev; 1042 uInt wmask = s->w_mask; 1043 1044 #ifdef UNALIGNED_OK 1045 /* Compare two bytes at a time. Note: this is not always beneficial. 1046 * Try with and without -DUNALIGNED_OK to check. 1047 */ 1048 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; 1049 register ush scan_start = *(ushf*)scan; 1050 register ush scan_end = *(ushf*)(scan+best_len-1); 1051 #else 1052 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1053 register Byte scan_end1 = scan[best_len-1]; 1054 register Byte scan_end = scan[best_len]; 1055 #endif 1056 1057 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1058 * It is easy to get rid of this optimization if necessary. 1059 */ 1060 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1061 1062 /* Do not waste too much time if we already have a good match: */ 1063 if (s->prev_length >= s->good_match) { 1064 chain_length >>= 2; 1065 } 1066 /* Do not look for matches beyond the end of the input. This is necessary 1067 * to make deflate deterministic. 1068 */ 1069 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; 1070 1071 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); 1072 1073 do { 1074 Assert(cur_match < s->strstart, "no future"); 1075 match = s->window + cur_match; 1076 1077 /* Skip to next match if the match length cannot increase 1078 * or if the match length is less than 2. Note that the checks below 1079 * for insufficient lookahead only occur occasionally for performance 1080 * reasons. Therefore uninitialized memory will be accessed, and 1081 * conditional jumps will be made that depend on those values. 1082 * However the length of the match is limited to the lookahead, so 1083 * the output of deflate is not affected by the uninitialized values. 1084 */ 1085 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) 1086 /* This code assumes sizeof(unsigned short) == 2. Do not use 1087 * UNALIGNED_OK if your compiler uses a different size. 1088 */ 1089 if (*(ushf*)(match+best_len-1) != scan_end || 1090 *(ushf*)match != scan_start) continue; 1091 1092 /* It is not necessary to compare scan[2] and match[2] since they are 1093 * always equal when the other bytes match, given that the hash keys 1094 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at 1095 * strstart+3, +5, ... up to strstart+257. We check for insufficient 1096 * lookahead only every 4th comparison; the 128th check will be made 1097 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is 1098 * necessary to put more guard bytes at the end of the window, or 1099 * to check more often for insufficient lookahead. 1100 */ 1101 Assert(scan[2] == match[2], "scan[2]?"); 1102 scan++, match++; 1103 do { 1104 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1105 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1106 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1107 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && 1108 scan < strend); 1109 /* The funny "do {}" generates better code on most compilers */ 1110 1111 /* Here, scan <= window+strstart+257 */ 1112 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1113 if (*scan == *match) scan++; 1114 1115 len = (MAX_MATCH - 1) - (int)(strend-scan); 1116 scan = strend - (MAX_MATCH-1); 1117 1118 #else /* UNALIGNED_OK */ 1119 1120 if (match[best_len] != scan_end || 1121 match[best_len-1] != scan_end1 || 1122 *match != *scan || 1123 *++match != scan[1]) continue; 1124 1125 /* The check at best_len-1 can be removed because it will be made 1126 * again later. (This heuristic is not always a win.) 1127 * It is not necessary to compare scan[2] and match[2] since they 1128 * are always equal when the other bytes match, given that 1129 * the hash keys are equal and that HASH_BITS >= 8. 1130 */ 1131 scan += 2, match++; 1132 Assert(*scan == *match, "match[2]?"); 1133 1134 /* We check for insufficient lookahead only every 8th comparison; 1135 * the 256th check will be made at strstart+258. 1136 */ 1137 do { 1138 } while (*++scan == *++match && *++scan == *++match && 1139 *++scan == *++match && *++scan == *++match && 1140 *++scan == *++match && *++scan == *++match && 1141 *++scan == *++match && *++scan == *++match && 1142 scan < strend); 1143 1144 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1145 1146 len = MAX_MATCH - (int)(strend - scan); 1147 scan = strend - MAX_MATCH; 1148 1149 #endif /* UNALIGNED_OK */ 1150 1151 if (len > best_len) { 1152 s->match_start = cur_match; 1153 best_len = len; 1154 if (len >= nice_match) break; 1155 #ifdef UNALIGNED_OK 1156 scan_end = *(ushf*)(scan+best_len-1); 1157 #else 1158 scan_end1 = scan[best_len-1]; 1159 scan_end = scan[best_len]; 1160 #endif 1161 } 1162 } while ((cur_match = prev[cur_match & wmask]) > limit 1163 && --chain_length != 0); 1164 1165 if ((uInt)best_len <= s->lookahead) return (uInt)best_len; 1166 return s->lookahead; 1167 } 1168 #endif /* ASMV */ 1169 #endif /* FASTEST */ 1170 1171 /* --------------------------------------------------------------------------- 1172 * Optimized version for level == 1 or strategy == Z_RLE only 1173 */ 1174 local uInt longest_match_fast( 1175 deflate_state *s, 1176 IPos cur_match) 1177 { 1178 register Bytef *scan = s->window + s->strstart; /* current string */ 1179 register Bytef *match; /* matched string */ 1180 register int len; /* length of current match */ 1181 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1182 1183 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1184 * It is easy to get rid of this optimization if necessary. 1185 */ 1186 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1187 1188 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); 1189 1190 Assert(cur_match < s->strstart, "no future"); 1191 1192 match = s->window + cur_match; 1193 1194 /* Return failure if the match length is less than 2: 1195 */ 1196 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; 1197 1198 /* The check at best_len-1 can be removed because it will be made 1199 * again later. (This heuristic is not always a win.) 1200 * It is not necessary to compare scan[2] and match[2] since they 1201 * are always equal when the other bytes match, given that 1202 * the hash keys are equal and that HASH_BITS >= 8. 1203 */ 1204 scan += 2, match += 2; 1205 Assert(*scan == *match, "match[2]?"); 1206 1207 /* We check for insufficient lookahead only every 8th comparison; 1208 * the 256th check will be made at strstart+258. 1209 */ 1210 do { 1211 } while (*++scan == *++match && *++scan == *++match && 1212 *++scan == *++match && *++scan == *++match && 1213 *++scan == *++match && *++scan == *++match && 1214 *++scan == *++match && *++scan == *++match && 1215 scan < strend); 1216 1217 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 1218 1219 len = MAX_MATCH - (int)(strend - scan); 1220 1221 if (len < MIN_MATCH) return MIN_MATCH - 1; 1222 1223 s->match_start = cur_match; 1224 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead; 1225 } 1226 1227 #ifdef DEBUG 1228 /* =========================================================================== 1229 * Check that the match at match_start is indeed a match. 1230 */ 1231 local void check_match( 1232 deflate_state *s, 1233 IPos start, 1234 IPos match, 1235 int length) 1236 { 1237 /* check that the match is indeed a match */ 1238 if (zmemcmp(s->window + match, 1239 s->window + start, length) != EQUAL) { 1240 fprintf(stderr, " start %u, match %u, length %d\n", 1241 start, match, length); 1242 do { 1243 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); 1244 } while (--length != 0); 1245 z_error("invalid match"); 1246 } 1247 if (z_verbose > 1) { 1248 fprintf(stderr,"\\[%d,%d]", start-match, length); 1249 do { putc(s->window[start++], stderr); } while (--length != 0); 1250 } 1251 } 1252 #else 1253 # define check_match(s, start, match, length) 1254 #endif /* DEBUG */ 1255 1256 /* =========================================================================== 1257 * Fill the window when the lookahead becomes insufficient. 1258 * Updates strstart and lookahead. 1259 * 1260 * IN assertion: lookahead < MIN_LOOKAHEAD 1261 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD 1262 * At least one byte has been read, or avail_in == 0; reads are 1263 * performed for at least two bytes (required for the zip translate_eol 1264 * option -- not supported here). 1265 */ 1266 local void fill_window( 1267 deflate_state *s) 1268 { 1269 register unsigned n, m; 1270 register Posf *p; 1271 unsigned more; /* Amount of free space at the end of the window. */ 1272 uInt wsize = s->w_size; 1273 1274 do { 1275 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); 1276 1277 /* Deal with !@#$% 64K limit: */ 1278 if (sizeof(int) <= 2) { 1279 if (more == 0 && s->strstart == 0 && s->lookahead == 0) { 1280 more = wsize; 1281 1282 } else if (more == (unsigned)(-1)) { 1283 /* Very unlikely, but possible on 16 bit machine if 1284 * strstart == 0 && lookahead == 1 (input done a byte at time) 1285 */ 1286 more--; 1287 } 1288 } 1289 1290 /* If the window is almost full and there is insufficient lookahead, 1291 * move the upper half to the lower one to make room in the upper half. 1292 */ 1293 if (s->strstart >= wsize+MAX_DIST(s)) { 1294 1295 zmemcpy(s->window, s->window+wsize, (unsigned)wsize); 1296 s->match_start -= wsize; 1297 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ 1298 s->block_start -= (long) wsize; 1299 1300 /* Slide the hash table (could be avoided with 32 bit values 1301 at the expense of memory usage). We slide even when level == 0 1302 to keep the hash table consistent if we switch back to level > 0 1303 later. (Using level 0 permanently is not an optimal usage of 1304 zlib, so we don't care about this pathological case.) 1305 */ 1306 /* %%% avoid this when Z_RLE */ 1307 n = s->hash_size; 1308 p = &s->head[n]; 1309 do { 1310 m = *--p; 1311 *p = (Pos)(m >= wsize ? m-wsize : NIL); 1312 } while (--n); 1313 1314 n = wsize; 1315 #ifndef FASTEST 1316 p = &s->prev[n]; 1317 do { 1318 m = *--p; 1319 *p = (Pos)(m >= wsize ? m-wsize : NIL); 1320 /* If n is not on any hash chain, prev[n] is garbage but 1321 * its value will never be used. 1322 */ 1323 } while (--n); 1324 #endif 1325 more += wsize; 1326 } 1327 if (s->strm->avail_in == 0) return; 1328 1329 /* If there was no sliding: 1330 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && 1331 * more == window_size - lookahead - strstart 1332 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) 1333 * => more >= window_size - 2*WSIZE + 2 1334 * In the BIG_MEM or MMAP case (not yet supported), 1335 * window_size == input_size + MIN_LOOKAHEAD && 1336 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. 1337 * Otherwise, window_size == 2*WSIZE so more >= 2. 1338 * If there was sliding, more >= WSIZE. So in all cases, more >= 2. 1339 */ 1340 Assert(more >= 2, "more < 2"); 1341 1342 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); 1343 s->lookahead += n; 1344 1345 /* Initialize the hash value now that we have some input: */ 1346 if (s->lookahead >= MIN_MATCH) { 1347 s->ins_h = s->window[s->strstart]; 1348 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); 1349 #if MIN_MATCH != 3 1350 Call UPDATE_HASH() MIN_MATCH-3 more times 1351 #endif 1352 } 1353 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, 1354 * but this is not important since only literal bytes will be emitted. 1355 */ 1356 1357 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); 1358 } 1359 1360 /* =========================================================================== 1361 * Flush the current block, with given end-of-file flag. 1362 * IN assertion: strstart is set to the end of the current match. 1363 */ 1364 #define FLUSH_BLOCK_ONLY(s, eof) { \ 1365 _tr_flush_block(s, (s->block_start >= 0L ? \ 1366 (charf *)&s->window[(unsigned)s->block_start] : \ 1367 (charf *)Z_NULL), \ 1368 (ulg)((long)s->strstart - s->block_start), \ 1369 (eof)); \ 1370 s->block_start = s->strstart; \ 1371 flush_pending(s->strm); \ 1372 Tracev((stderr,"[FLUSH]")); \ 1373 } 1374 1375 /* Same but force premature exit if necessary. */ 1376 #define FLUSH_BLOCK(s, eof) { \ 1377 FLUSH_BLOCK_ONLY(s, eof); \ 1378 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \ 1379 } 1380 1381 /* =========================================================================== 1382 * Copy without compression as much as possible from the input stream, return 1383 * the current block state. 1384 * This function does not insert new strings in the dictionary since 1385 * uncompressible data is probably not useful. This function is used 1386 * only for the level=0 compression option. 1387 * NOTE: this function should be optimized to avoid extra copying from 1388 * window to pending_buf. 1389 */ 1390 local block_state deflate_stored( 1391 deflate_state *s, 1392 int flush) 1393 { 1394 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited 1395 * to pending_buf_size, and each stored block has a 5 byte header: 1396 */ 1397 ulg max_block_size = 0xffff; 1398 ulg max_start; 1399 1400 if (max_block_size > s->pending_buf_size - 5) { 1401 max_block_size = s->pending_buf_size - 5; 1402 } 1403 1404 /* Copy as much as possible from input to output: */ 1405 for (;;) { 1406 /* Fill the window as much as possible: */ 1407 if (s->lookahead <= 1) { 1408 1409 Assert(s->strstart < s->w_size+MAX_DIST(s) || 1410 s->block_start >= (long)s->w_size, "slide too late"); 1411 1412 fill_window(s); 1413 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more; 1414 1415 if (s->lookahead == 0) break; /* flush the current block */ 1416 } 1417 Assert(s->block_start >= 0L, "block gone"); 1418 1419 s->strstart += s->lookahead; 1420 s->lookahead = 0; 1421 1422 /* Emit a stored block if pending_buf will be full: */ 1423 max_start = s->block_start + max_block_size; 1424 if (s->strstart == 0 || (ulg)s->strstart >= max_start) { 1425 /* strstart == 0 is possible when wraparound on 16-bit machine */ 1426 s->lookahead = (uInt)(s->strstart - max_start); 1427 s->strstart = (uInt)max_start; 1428 FLUSH_BLOCK(s, 0); 1429 } 1430 /* Flush if we may have to slide, otherwise block_start may become 1431 * negative and the data will be gone: 1432 */ 1433 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) { 1434 FLUSH_BLOCK(s, 0); 1435 } 1436 } 1437 FLUSH_BLOCK(s, flush == Z_FINISH); 1438 return flush == Z_FINISH ? finish_done : block_done; 1439 } 1440 1441 /* =========================================================================== 1442 * Compress as much as possible from the input stream, return the current 1443 * block state. 1444 * This function does not perform lazy evaluation of matches and inserts 1445 * new strings in the dictionary only for unmatched strings or for short 1446 * matches. It is used only for the fast compression options. 1447 */ 1448 local block_state deflate_fast( 1449 deflate_state *s, 1450 int flush) 1451 { 1452 IPos hash_head = NIL; /* head of the hash chain */ 1453 int bflush; /* set if current block must be flushed */ 1454 1455 for (;;) { 1456 /* Make sure that we always have enough lookahead, except 1457 * at the end of the input file. We need MAX_MATCH bytes 1458 * for the next match, plus MIN_MATCH bytes to insert the 1459 * string following the next match. 1460 */ 1461 if (s->lookahead < MIN_LOOKAHEAD) { 1462 fill_window(s); 1463 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1464 return need_more; 1465 } 1466 if (s->lookahead == 0) break; /* flush the current block */ 1467 } 1468 1469 /* Insert the string window[strstart .. strstart+2] in the 1470 * dictionary, and set hash_head to the head of the hash chain: 1471 */ 1472 if (s->lookahead >= MIN_MATCH) { 1473 INSERT_STRING(s, s->strstart, hash_head); 1474 } 1475 1476 /* Find the longest match, discarding those <= prev_length. 1477 * At this point we have always match_length < MIN_MATCH 1478 */ 1479 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { 1480 /* To simplify the code, we prevent matches with the string 1481 * of window index 0 (in particular we have to avoid a match 1482 * of the string with itself at the start of the input file). 1483 */ 1484 #ifdef FASTEST 1485 if ((s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) || 1486 (s->strategy == Z_RLE && s->strstart - hash_head == 1)) { 1487 s->match_length = longest_match_fast (s, hash_head); 1488 } 1489 #else 1490 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) { 1491 s->match_length = longest_match (s, hash_head); 1492 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) { 1493 s->match_length = longest_match_fast (s, hash_head); 1494 } 1495 #endif 1496 /* longest_match() or longest_match_fast() sets match_start */ 1497 } 1498 if (s->match_length >= MIN_MATCH) { 1499 check_match(s, s->strstart, s->match_start, s->match_length); 1500 1501 _tr_tally_dist(s, s->strstart - s->match_start, 1502 s->match_length - MIN_MATCH, bflush); 1503 1504 s->lookahead -= s->match_length; 1505 1506 /* Insert new strings in the hash table only if the match length 1507 * is not too large. This saves time but degrades compression. 1508 */ 1509 #ifndef FASTEST 1510 if (s->match_length <= s->max_insert_length && 1511 s->lookahead >= MIN_MATCH) { 1512 s->match_length--; /* string at strstart already in table */ 1513 do { 1514 s->strstart++; 1515 INSERT_STRING(s, s->strstart, hash_head); 1516 /* strstart never exceeds WSIZE-MAX_MATCH, so there are 1517 * always MIN_MATCH bytes ahead. 1518 */ 1519 } while (--s->match_length != 0); 1520 s->strstart++; 1521 } else 1522 #endif 1523 { 1524 s->strstart += s->match_length; 1525 s->match_length = 0; 1526 s->ins_h = s->window[s->strstart]; 1527 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); 1528 #if MIN_MATCH != 3 1529 Call UPDATE_HASH() MIN_MATCH-3 more times 1530 #endif 1531 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not 1532 * matter since it will be recomputed at next deflate call. 1533 */ 1534 } 1535 } else { 1536 /* No match, output a literal byte */ 1537 Tracevv((stderr,"%c", s->window[s->strstart])); 1538 _tr_tally_lit (s, s->window[s->strstart], bflush); 1539 s->lookahead--; 1540 s->strstart++; 1541 } 1542 if (bflush) FLUSH_BLOCK(s, 0); 1543 } 1544 FLUSH_BLOCK(s, flush == Z_FINISH); 1545 return flush == Z_FINISH ? finish_done : block_done; 1546 } 1547 1548 #ifndef FASTEST 1549 /* =========================================================================== 1550 * Same as above, but achieves better compression. We use a lazy 1551 * evaluation for matches: a match is finally adopted only if there is 1552 * no better match at the next window position. 1553 */ 1554 local block_state deflate_slow( 1555 deflate_state *s, 1556 int flush) 1557 { 1558 IPos hash_head = NIL; /* head of hash chain */ 1559 int bflush; /* set if current block must be flushed */ 1560 1561 /* Process the input block. */ 1562 for (;;) { 1563 /* Make sure that we always have enough lookahead, except 1564 * at the end of the input file. We need MAX_MATCH bytes 1565 * for the next match, plus MIN_MATCH bytes to insert the 1566 * string following the next match. 1567 */ 1568 if (s->lookahead < MIN_LOOKAHEAD) { 1569 fill_window(s); 1570 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1571 return need_more; 1572 } 1573 if (s->lookahead == 0) break; /* flush the current block */ 1574 } 1575 1576 /* Insert the string window[strstart .. strstart+2] in the 1577 * dictionary, and set hash_head to the head of the hash chain: 1578 */ 1579 if (s->lookahead >= MIN_MATCH) { 1580 INSERT_STRING(s, s->strstart, hash_head); 1581 } 1582 1583 /* Find the longest match, discarding those <= prev_length. 1584 */ 1585 s->prev_length = s->match_length, s->prev_match = s->match_start; 1586 s->match_length = MIN_MATCH-1; 1587 1588 if (hash_head != NIL && s->prev_length < s->max_lazy_match && 1589 s->strstart - hash_head <= MAX_DIST(s)) { 1590 /* To simplify the code, we prevent matches with the string 1591 * of window index 0 (in particular we have to avoid a match 1592 * of the string with itself at the start of the input file). 1593 */ 1594 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) { 1595 s->match_length = longest_match (s, hash_head); 1596 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) { 1597 s->match_length = longest_match_fast (s, hash_head); 1598 } 1599 /* longest_match() or longest_match_fast() sets match_start */ 1600 1601 if (s->match_length <= 5 && (s->strategy == Z_FILTERED 1602 #if TOO_FAR <= 32767 1603 || (s->match_length == MIN_MATCH && 1604 s->strstart - s->match_start > TOO_FAR) 1605 #endif 1606 )) { 1607 1608 /* If prev_match is also MIN_MATCH, match_start is garbage 1609 * but we will ignore the current match anyway. 1610 */ 1611 s->match_length = MIN_MATCH-1; 1612 } 1613 } 1614 /* If there was a match at the previous step and the current 1615 * match is not better, output the previous match: 1616 */ 1617 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { 1618 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; 1619 /* Do not insert strings in hash table beyond this. */ 1620 1621 check_match(s, s->strstart-1, s->prev_match, s->prev_length); 1622 1623 _tr_tally_dist(s, s->strstart -1 - s->prev_match, 1624 s->prev_length - MIN_MATCH, bflush); 1625 1626 /* Insert in hash table all strings up to the end of the match. 1627 * strstart-1 and strstart are already inserted. If there is not 1628 * enough lookahead, the last two strings are not inserted in 1629 * the hash table. 1630 */ 1631 s->lookahead -= s->prev_length-1; 1632 s->prev_length -= 2; 1633 do { 1634 if (++s->strstart <= max_insert) { 1635 INSERT_STRING(s, s->strstart, hash_head); 1636 } 1637 } while (--s->prev_length != 0); 1638 s->match_available = 0; 1639 s->match_length = MIN_MATCH-1; 1640 s->strstart++; 1641 1642 if (bflush) FLUSH_BLOCK(s, 0); 1643 1644 } else if (s->match_available) { 1645 /* If there was no match at the previous position, output a 1646 * single literal. If there was a match but the current match 1647 * is longer, truncate the previous match to a single literal. 1648 */ 1649 Tracevv((stderr,"%c", s->window[s->strstart-1])); 1650 _tr_tally_lit(s, s->window[s->strstart-1], bflush); 1651 if (bflush) { 1652 FLUSH_BLOCK_ONLY(s, 0); 1653 } 1654 s->strstart++; 1655 s->lookahead--; 1656 if (s->strm->avail_out == 0) return need_more; 1657 } else { 1658 /* There is no previous match to compare with, wait for 1659 * the next step to decide. 1660 */ 1661 s->match_available = 1; 1662 s->strstart++; 1663 s->lookahead--; 1664 } 1665 } 1666 Assert (flush != Z_NO_FLUSH, "no flush?"); 1667 if (s->match_available) { 1668 Tracevv((stderr,"%c", s->window[s->strstart-1])); 1669 _tr_tally_lit(s, s->window[s->strstart-1], bflush); 1670 s->match_available = 0; 1671 } 1672 FLUSH_BLOCK(s, flush == Z_FINISH); 1673 return flush == Z_FINISH ? finish_done : block_done; 1674 } 1675 #endif /* FASTEST */ 1676 1677 #if 0 1678 /* =========================================================================== 1679 * For Z_RLE, simply look for runs of bytes, generate matches only of distance 1680 * one. Do not maintain a hash table. (It will be regenerated if this run of 1681 * deflate switches away from Z_RLE.) 1682 */ 1683 local block_state deflate_rle( 1684 deflate_state *s, 1685 int flush) 1686 { 1687 int bflush; /* set if current block must be flushed */ 1688 uInt run; /* length of run */ 1689 uInt max; /* maximum length of run */ 1690 uInt prev; /* byte at distance one to match */ 1691 Bytef *scan; /* scan for end of run */ 1692 1693 for (;;) { 1694 /* Make sure that we always have enough lookahead, except 1695 * at the end of the input file. We need MAX_MATCH bytes 1696 * for the longest encodable run. 1697 */ 1698 if (s->lookahead < MAX_MATCH) { 1699 fill_window(s); 1700 if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) { 1701 return need_more; 1702 } 1703 if (s->lookahead == 0) break; /* flush the current block */ 1704 } 1705 1706 /* See how many times the previous byte repeats */ 1707 run = 0; 1708 if (s->strstart > 0) { /* if there is a previous byte, that is */ 1709 max = s->lookahead < MAX_MATCH ? s->lookahead : MAX_MATCH; 1710 scan = s->window + s->strstart - 1; 1711 prev = *scan++; 1712 do { 1713 if (*scan++ != prev) 1714 break; 1715 } while (++run < max); 1716 } 1717 1718 /* Emit match if have run of MIN_MATCH or longer, else emit literal */ 1719 if (run >= MIN_MATCH) { 1720 check_match(s, s->strstart, s->strstart - 1, run); 1721 _tr_tally_dist(s, 1, run - MIN_MATCH, bflush); 1722 s->lookahead -= run; 1723 s->strstart += run; 1724 } else { 1725 /* No match, output a literal byte */ 1726 Tracevv((stderr,"%c", s->window[s->strstart])); 1727 _tr_tally_lit (s, s->window[s->strstart], bflush); 1728 s->lookahead--; 1729 s->strstart++; 1730 } 1731 if (bflush) FLUSH_BLOCK(s, 0); 1732 } 1733 FLUSH_BLOCK(s, flush == Z_FINISH); 1734 return flush == Z_FINISH ? finish_done : block_done; 1735 } 1736 #endif 1737