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 #include "deflate.h" 51 52 /* 53 If you use the zlib library in a product, an acknowledgment is welcome 54 in the documentation of your product. If for some reason you cannot 55 include such an acknowledgment, I would appreciate that you keep this 56 copyright string in the executable of your product. 57 */ 58 59 typedef enum { 60 need_more, /* block not completed, need more input or more output */ 61 block_done, /* block flush performed */ 62 finish_started, /* finish started, need only more output at next deflate */ 63 finish_done /* finish done, accept no more input or output */ 64 } block_state; 65 66 typedef block_state (*compress_func)(deflate_state *s, int flush); 67 /* Compression function. Returns the block state after the call. */ 68 69 local block_state deflate_stored(deflate_state *s, int flush); 70 local block_state deflate_fast(deflate_state *s, int flush); 71 #ifndef FASTEST 72 local block_state deflate_slow(deflate_state *s, int flush); 73 #endif 74 local block_state deflate_rle(deflate_state *s, int flush); 75 local block_state deflate_huff(deflate_state *s, int flush); 76 77 /* =========================================================================== 78 * Local data 79 */ 80 81 #define NIL 0 82 /* Tail of hash chains */ 83 84 #ifndef TOO_FAR 85 # define TOO_FAR 4096 86 #endif 87 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ 88 89 /* Values for max_lazy_match, good_match and max_chain_length, depending on 90 * the desired pack level (0..9). The values given below have been tuned to 91 * exclude worst case performance for pathological files. Better values may be 92 * found for specific files. 93 */ 94 typedef struct config_s { 95 ush good_length; /* reduce lazy search above this match length */ 96 ush max_lazy; /* do not perform lazy search above this match length */ 97 ush nice_length; /* quit search above this match length */ 98 ush max_chain; 99 compress_func func; 100 } config; 101 102 #ifdef FASTEST 103 local const config configuration_table[2] = { 104 /* good lazy nice chain */ 105 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 106 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */ 107 #else 108 local const config configuration_table[10] = { 109 /* good lazy nice chain */ 110 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 111 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */ 112 /* 2 */ {4, 5, 16, 8, deflate_fast}, 113 /* 3 */ {4, 6, 32, 32, deflate_fast}, 114 115 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ 116 /* 5 */ {8, 16, 32, 32, deflate_slow}, 117 /* 6 */ {8, 16, 128, 128, deflate_slow}, 118 /* 7 */ {8, 32, 128, 256, deflate_slow}, 119 /* 8 */ {32, 128, 258, 1024, deflate_slow}, 120 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */ 121 #endif 122 123 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 124 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different 125 * meaning. 126 */ 127 128 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */ 129 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0)) 130 131 /* =========================================================================== 132 * Update a hash value with the given input byte 133 * IN assertion: all calls to UPDATE_HASH are made with consecutive input 134 * characters, so that a running hash key can be computed from the previous 135 * key instead of complete recalculation each time. 136 */ 137 #define UPDATE_HASH(s,h,c) (h = (((h) << s->hash_shift) ^ (c)) & s->hash_mask) 138 139 140 /* =========================================================================== 141 * Insert string str in the dictionary and set match_head to the previous head 142 * of the hash chain (the most recent string with same hash key). Return 143 * the previous length of the hash chain. 144 * If this file is compiled with -DFASTEST, the compression level is forced 145 * to 1, and no hash chains are maintained. 146 * IN assertion: all calls to INSERT_STRING are made with consecutive input 147 * characters and the first MIN_MATCH bytes of str are valid (except for 148 * the last MIN_MATCH-1 bytes of the input file). 149 */ 150 #ifdef FASTEST 151 #define INSERT_STRING(s, str, match_head) \ 152 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 153 match_head = s->head[s->ins_h], \ 154 s->head[s->ins_h] = (Pos)(str)) 155 #else 156 #define INSERT_STRING(s, str, match_head) \ 157 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 158 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \ 159 s->head[s->ins_h] = (Pos)(str)) 160 #endif 161 162 /* =========================================================================== 163 * Initialize the hash table (avoiding 64K overflow for 16 bit systems). 164 * prev[] will be initialized on the fly. 165 */ 166 #define CLEAR_HASH(s) \ 167 do { \ 168 s->head[s->hash_size - 1] = NIL; \ 169 zmemzero((Bytef *)s->head, \ 170 (unsigned)(s->hash_size - 1)*sizeof(*s->head)); \ 171 } while (0) 172 173 /* =========================================================================== 174 * Slide the hash table when sliding the window down (could be avoided with 32 175 * bit values at the expense of memory usage). We slide even when level == 0 to 176 * keep the hash table consistent if we switch back to level > 0 later. 177 */ 178 local void slide_hash(deflate_state *s) { 179 unsigned n, m; 180 Posf *p; 181 uInt wsize = s->w_size; 182 183 n = s->hash_size; 184 p = &s->head[n]; 185 do { 186 m = *--p; 187 *p = (Pos)(m >= wsize ? m - wsize : NIL); 188 } while (--n); 189 n = wsize; 190 #ifndef FASTEST 191 p = &s->prev[n]; 192 do { 193 m = *--p; 194 *p = (Pos)(m >= wsize ? m - wsize : NIL); 195 /* If n is not on any hash chain, prev[n] is garbage but 196 * its value will never be used. 197 */ 198 } while (--n); 199 #endif 200 } 201 202 /* =========================================================================== 203 * Read a new buffer from the current input stream, update the adler32 204 * and total number of bytes read. All deflate() input goes through 205 * this function so some applications may wish to modify it to avoid 206 * allocating a large strm->next_in buffer and copying from it. 207 * (See also flush_pending()). 208 */ 209 local unsigned read_buf(z_streamp strm, Bytef *buf, unsigned size) { 210 unsigned len = strm->avail_in; 211 212 if (len > size) len = size; 213 if (len == 0) return 0; 214 215 strm->avail_in -= len; 216 217 zmemcpy(buf, strm->next_in, len); 218 if (strm->state->wrap == 1) { 219 strm->adler = adler32(strm->adler, buf, len); 220 } 221 #ifdef GZIP 222 else if (strm->state->wrap == 2) { 223 strm->adler = crc32(strm->adler, buf, len); 224 } 225 #endif 226 strm->next_in += len; 227 strm->total_in += len; 228 229 return len; 230 } 231 232 /* =========================================================================== 233 * Fill the window when the lookahead becomes insufficient. 234 * Updates strstart and lookahead. 235 * 236 * IN assertion: lookahead < MIN_LOOKAHEAD 237 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD 238 * At least one byte has been read, or avail_in == 0; reads are 239 * performed for at least two bytes (required for the zip translate_eol 240 * option -- not supported here). 241 */ 242 local void fill_window(deflate_state *s) { 243 unsigned n; 244 unsigned more; /* Amount of free space at the end of the window. */ 245 uInt wsize = s->w_size; 246 247 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead"); 248 249 do { 250 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); 251 252 /* Deal with !@#$% 64K limit: */ 253 if (sizeof(int) <= 2) { 254 if (more == 0 && s->strstart == 0 && s->lookahead == 0) { 255 more = wsize; 256 257 } else if (more == (unsigned)(-1)) { 258 /* Very unlikely, but possible on 16 bit machine if 259 * strstart == 0 && lookahead == 1 (input done a byte at time) 260 */ 261 more--; 262 } 263 } 264 265 /* If the window is almost full and there is insufficient lookahead, 266 * move the upper half to the lower one to make room in the upper half. 267 */ 268 if (s->strstart >= wsize + MAX_DIST(s)) { 269 270 zmemcpy(s->window, s->window + wsize, (unsigned)wsize - more); 271 s->match_start -= wsize; 272 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ 273 s->block_start -= (long) wsize; 274 if (s->insert > s->strstart) 275 s->insert = s->strstart; 276 slide_hash(s); 277 more += wsize; 278 } 279 if (s->strm->avail_in == 0) break; 280 281 /* If there was no sliding: 282 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && 283 * more == window_size - lookahead - strstart 284 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) 285 * => more >= window_size - 2*WSIZE + 2 286 * In the BIG_MEM or MMAP case (not yet supported), 287 * window_size == input_size + MIN_LOOKAHEAD && 288 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. 289 * Otherwise, window_size == 2*WSIZE so more >= 2. 290 * If there was sliding, more >= WSIZE. So in all cases, more >= 2. 291 */ 292 Assert(more >= 2, "more < 2"); 293 294 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); 295 s->lookahead += n; 296 297 /* Initialize the hash value now that we have some input: */ 298 if (s->lookahead + s->insert >= MIN_MATCH) { 299 uInt str = s->strstart - s->insert; 300 s->ins_h = s->window[str]; 301 UPDATE_HASH(s, s->ins_h, s->window[str + 1]); 302 #if MIN_MATCH != 3 303 Call UPDATE_HASH() MIN_MATCH-3 more times 304 #endif 305 while (s->insert) { 306 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 307 #ifndef FASTEST 308 s->prev[str & s->w_mask] = s->head[s->ins_h]; 309 #endif 310 s->head[s->ins_h] = (Pos)str; 311 str++; 312 s->insert--; 313 if (s->lookahead + s->insert < MIN_MATCH) 314 break; 315 } 316 } 317 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, 318 * but this is not important since only literal bytes will be emitted. 319 */ 320 321 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); 322 323 /* If the WIN_INIT bytes after the end of the current data have never been 324 * written, then zero those bytes in order to avoid memory check reports of 325 * the use of uninitialized (or uninitialised as Julian writes) bytes by 326 * the longest match routines. Update the high water mark for the next 327 * time through here. WIN_INIT is set to MAX_MATCH since the longest match 328 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead. 329 */ 330 if (s->high_water < s->window_size) { 331 ulg curr = s->strstart + (ulg)(s->lookahead); 332 ulg init; 333 334 if (s->high_water < curr) { 335 /* Previous high water mark below current data -- zero WIN_INIT 336 * bytes or up to end of window, whichever is less. 337 */ 338 init = s->window_size - curr; 339 if (init > WIN_INIT) 340 init = WIN_INIT; 341 zmemzero(s->window + curr, (unsigned)init); 342 s->high_water = curr + init; 343 } 344 else if (s->high_water < (ulg)curr + WIN_INIT) { 345 /* High water mark at or above current data, but below current data 346 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up 347 * to end of window, whichever is less. 348 */ 349 init = (ulg)curr + WIN_INIT - s->high_water; 350 if (init > s->window_size - s->high_water) 351 init = s->window_size - s->high_water; 352 zmemzero(s->window + s->high_water, (unsigned)init); 353 s->high_water += init; 354 } 355 } 356 357 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 358 "not enough room for search"); 359 } 360 361 /* ========================================================================= */ 362 int ZEXPORT deflateInit_(z_streamp strm, int level, const char *version, 363 int stream_size) { 364 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, 365 Z_DEFAULT_STRATEGY, version, stream_size); 366 /* To do: ignore strm->next_in if we use it as window */ 367 } 368 369 /* ========================================================================= */ 370 int ZEXPORT deflateInit2_(z_streamp strm, int level, int method, 371 int windowBits, int memLevel, int strategy, 372 const char *version, int stream_size) { 373 deflate_state *s; 374 int wrap = 1; 375 static const char my_version[] = ZLIB_VERSION; 376 377 if (version == Z_NULL || version[0] != my_version[0] || 378 stream_size != sizeof(z_stream)) { 379 return Z_VERSION_ERROR; 380 } 381 if (strm == Z_NULL) return Z_STREAM_ERROR; 382 383 strm->msg = Z_NULL; 384 if (strm->zalloc == (alloc_func)0) { 385 #ifdef Z_SOLO 386 return Z_STREAM_ERROR; 387 #else 388 strm->zalloc = zcalloc; 389 strm->opaque = (voidpf)0; 390 #endif 391 } 392 if (strm->zfree == (free_func)0) 393 #ifdef Z_SOLO 394 return Z_STREAM_ERROR; 395 #else 396 strm->zfree = zcfree; 397 #endif 398 399 #ifdef FASTEST 400 if (level != 0) level = 1; 401 #else 402 if (level == Z_DEFAULT_COMPRESSION) level = 6; 403 #endif 404 405 if (windowBits < 0) { /* suppress zlib wrapper */ 406 wrap = 0; 407 if (windowBits < -15) 408 return Z_STREAM_ERROR; 409 windowBits = -windowBits; 410 } 411 #ifdef GZIP 412 else if (windowBits > 15) { 413 wrap = 2; /* write gzip wrapper instead */ 414 windowBits -= 16; 415 } 416 #endif 417 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || 418 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || 419 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) { 420 return Z_STREAM_ERROR; 421 } 422 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ 423 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); 424 if (s == Z_NULL) return Z_MEM_ERROR; 425 strm->state = (struct internal_state FAR *)s; 426 s->strm = strm; 427 s->status = INIT_STATE; /* to pass state test in deflateReset() */ 428 429 s->wrap = wrap; 430 s->gzhead = Z_NULL; 431 s->w_bits = (uInt)windowBits; 432 s->w_size = 1 << s->w_bits; 433 s->w_mask = s->w_size - 1; 434 435 s->hash_bits = (uInt)memLevel + 7; 436 s->hash_size = 1 << s->hash_bits; 437 s->hash_mask = s->hash_size - 1; 438 s->hash_shift = ((s->hash_bits + MIN_MATCH-1) / MIN_MATCH); 439 440 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); 441 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); 442 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); 443 444 s->high_water = 0; /* nothing written to s->window yet */ 445 446 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ 447 448 /* We overlay pending_buf and sym_buf. This works since the average size 449 * for length/distance pairs over any compressed block is assured to be 31 450 * bits or less. 451 * 452 * Analysis: The longest fixed codes are a length code of 8 bits plus 5 453 * extra bits, for lengths 131 to 257. The longest fixed distance codes are 454 * 5 bits plus 13 extra bits, for distances 16385 to 32768. The longest 455 * possible fixed-codes length/distance pair is then 31 bits total. 456 * 457 * sym_buf starts one-fourth of the way into pending_buf. So there are 458 * three bytes in sym_buf for every four bytes in pending_buf. Each symbol 459 * in sym_buf is three bytes -- two for the distance and one for the 460 * literal/length. As each symbol is consumed, the pointer to the next 461 * sym_buf value to read moves forward three bytes. From that symbol, up to 462 * 31 bits are written to pending_buf. The closest the written pending_buf 463 * bits gets to the next sym_buf symbol to read is just before the last 464 * code is written. At that time, 31*(n - 2) bits have been written, just 465 * after 24*(n - 2) bits have been consumed from sym_buf. sym_buf starts at 466 * 8*n bits into pending_buf. (Note that the symbol buffer fills when n - 1 467 * symbols are written.) The closest the writing gets to what is unread is 468 * then n + 14 bits. Here n is lit_bufsize, which is 16384 by default, and 469 * can range from 128 to 32768. 470 * 471 * Therefore, at a minimum, there are 142 bits of space between what is 472 * written and what is read in the overlain buffers, so the symbols cannot 473 * be overwritten by the compressed data. That space is actually 139 bits, 474 * due to the three-bit fixed-code block header. 475 * 476 * That covers the case where either Z_FIXED is specified, forcing fixed 477 * codes, or when the use of fixed codes is chosen, because that choice 478 * results in a smaller compressed block than dynamic codes. That latter 479 * condition then assures that the above analysis also covers all dynamic 480 * blocks. A dynamic-code block will only be chosen to be emitted if it has 481 * fewer bits than a fixed-code block would for the same set of symbols. 482 * Therefore its average symbol length is assured to be less than 31. So 483 * the compressed data for a dynamic block also cannot overwrite the 484 * symbols from which it is being constructed. 485 */ 486 487 s->pending_buf = (uchf *) ZALLOC(strm, s->lit_bufsize, 4); 488 s->pending_buf_size = (ulg)s->lit_bufsize * 4; 489 490 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || 491 s->pending_buf == Z_NULL) { 492 s->status = FINISH_STATE; 493 strm->msg = ERR_MSG(Z_MEM_ERROR); 494 deflateEnd (strm); 495 return Z_MEM_ERROR; 496 } 497 s->sym_buf = s->pending_buf + s->lit_bufsize; 498 s->sym_end = (s->lit_bufsize - 1) * 3; 499 /* We avoid equality with lit_bufsize*3 because of wraparound at 64K 500 * on 16 bit machines and because stored blocks are restricted to 501 * 64K-1 bytes. 502 */ 503 504 s->level = level; 505 s->strategy = strategy; 506 s->method = (Byte)method; 507 508 return deflateReset(strm); 509 } 510 511 /* ========================================================================= 512 * Check for a valid deflate stream state. Return 0 if ok, 1 if not. 513 */ 514 local int deflateStateCheck(z_streamp strm) { 515 deflate_state *s; 516 if (strm == Z_NULL || 517 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) 518 return 1; 519 s = strm->state; 520 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE && 521 #ifdef GZIP 522 s->status != GZIP_STATE && 523 #endif 524 s->status != EXTRA_STATE && 525 s->status != NAME_STATE && 526 s->status != COMMENT_STATE && 527 s->status != HCRC_STATE && 528 s->status != BUSY_STATE && 529 s->status != FINISH_STATE)) 530 return 1; 531 return 0; 532 } 533 534 /* ========================================================================= */ 535 int ZEXPORT deflateSetDictionary(z_streamp strm, const Bytef *dictionary, 536 uInt dictLength) { 537 deflate_state *s; 538 uInt str, n; 539 int wrap; 540 unsigned avail; 541 z_const unsigned char *next; 542 543 if (deflateStateCheck(strm) || dictionary == Z_NULL) 544 return Z_STREAM_ERROR; 545 s = strm->state; 546 wrap = s->wrap; 547 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead) 548 return Z_STREAM_ERROR; 549 550 /* when using zlib wrappers, compute Adler-32 for provided dictionary */ 551 if (wrap == 1) 552 strm->adler = adler32(strm->adler, dictionary, dictLength); 553 s->wrap = 0; /* avoid computing Adler-32 in read_buf */ 554 555 /* if dictionary would fill window, just replace the history */ 556 if (dictLength >= s->w_size) { 557 if (wrap == 0) { /* already empty otherwise */ 558 CLEAR_HASH(s); 559 s->strstart = 0; 560 s->block_start = 0L; 561 s->insert = 0; 562 } 563 dictionary += dictLength - s->w_size; /* use the tail */ 564 dictLength = s->w_size; 565 } 566 567 /* insert dictionary into window and hash */ 568 avail = strm->avail_in; 569 next = strm->next_in; 570 strm->avail_in = dictLength; 571 strm->next_in = (z_const Bytef *)dictionary; 572 fill_window(s); 573 while (s->lookahead >= MIN_MATCH) { 574 str = s->strstart; 575 n = s->lookahead - (MIN_MATCH-1); 576 do { 577 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); 578 #ifndef FASTEST 579 s->prev[str & s->w_mask] = s->head[s->ins_h]; 580 #endif 581 s->head[s->ins_h] = (Pos)str; 582 str++; 583 } while (--n); 584 s->strstart = str; 585 s->lookahead = MIN_MATCH-1; 586 fill_window(s); 587 } 588 s->strstart += s->lookahead; 589 s->block_start = (long)s->strstart; 590 s->insert = s->lookahead; 591 s->lookahead = 0; 592 s->match_length = s->prev_length = MIN_MATCH-1; 593 s->match_available = 0; 594 strm->next_in = next; 595 strm->avail_in = avail; 596 s->wrap = wrap; 597 return Z_OK; 598 } 599 600 /* ========================================================================= */ 601 int ZEXPORT deflateGetDictionary(z_streamp strm, Bytef *dictionary, 602 uInt *dictLength) { 603 deflate_state *s; 604 uInt len; 605 606 if (deflateStateCheck(strm)) 607 return Z_STREAM_ERROR; 608 s = strm->state; 609 len = s->strstart + s->lookahead; 610 if (len > s->w_size) 611 len = s->w_size; 612 if (dictionary != Z_NULL && len) 613 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len); 614 if (dictLength != Z_NULL) 615 *dictLength = len; 616 return Z_OK; 617 } 618 619 /* ========================================================================= */ 620 int ZEXPORT deflateResetKeep(z_streamp strm) { 621 deflate_state *s; 622 623 if (deflateStateCheck(strm)) { 624 return Z_STREAM_ERROR; 625 } 626 627 strm->total_in = strm->total_out = 0; 628 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ 629 strm->data_type = Z_UNKNOWN; 630 631 s = (deflate_state *)strm->state; 632 s->pending = 0; 633 s->pending_out = s->pending_buf; 634 635 if (s->wrap < 0) { 636 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ 637 } 638 s->status = 639 #ifdef GZIP 640 s->wrap == 2 ? GZIP_STATE : 641 #endif 642 INIT_STATE; 643 strm->adler = 644 #ifdef GZIP 645 s->wrap == 2 ? crc32(0L, Z_NULL, 0) : 646 #endif 647 adler32(0L, Z_NULL, 0); 648 s->last_flush = -2; 649 650 _tr_init(s); 651 652 return Z_OK; 653 } 654 655 /* =========================================================================== 656 * Initialize the "longest match" routines for a new zlib stream 657 */ 658 local void lm_init(deflate_state *s) { 659 s->window_size = (ulg)2L*s->w_size; 660 661 CLEAR_HASH(s); 662 663 /* Set the default configuration parameters: 664 */ 665 s->max_lazy_match = configuration_table[s->level].max_lazy; 666 s->good_match = configuration_table[s->level].good_length; 667 s->nice_match = configuration_table[s->level].nice_length; 668 s->max_chain_length = configuration_table[s->level].max_chain; 669 670 s->strstart = 0; 671 s->block_start = 0L; 672 s->lookahead = 0; 673 s->insert = 0; 674 s->match_length = s->prev_length = MIN_MATCH-1; 675 s->match_available = 0; 676 s->ins_h = 0; 677 } 678 679 /* ========================================================================= */ 680 int ZEXPORT deflateReset(z_streamp strm) { 681 int ret; 682 683 ret = deflateResetKeep(strm); 684 if (ret == Z_OK) 685 lm_init(strm->state); 686 return ret; 687 } 688 689 /* ========================================================================= */ 690 int ZEXPORT deflateSetHeader(z_streamp strm, gz_headerp head) { 691 if (deflateStateCheck(strm) || strm->state->wrap != 2) 692 return Z_STREAM_ERROR; 693 strm->state->gzhead = head; 694 return Z_OK; 695 } 696 697 /* ========================================================================= */ 698 int ZEXPORT deflatePending(z_streamp strm, unsigned *pending, int *bits) { 699 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 700 if (pending != Z_NULL) 701 *pending = strm->state->pending; 702 if (bits != Z_NULL) 703 *bits = strm->state->bi_valid; 704 return Z_OK; 705 } 706 707 /* ========================================================================= */ 708 int ZEXPORT deflatePrime(z_streamp strm, int bits, int value) { 709 deflate_state *s; 710 int put; 711 712 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 713 s = strm->state; 714 if (bits < 0 || bits > 16 || 715 s->sym_buf < s->pending_out + ((Buf_size + 7) >> 3)) 716 return Z_BUF_ERROR; 717 do { 718 put = Buf_size - s->bi_valid; 719 if (put > bits) 720 put = bits; 721 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid); 722 s->bi_valid += put; 723 _tr_flush_bits(s); 724 value >>= put; 725 bits -= put; 726 } while (bits); 727 return Z_OK; 728 } 729 730 /* ========================================================================= */ 731 int ZEXPORT deflateParams(z_streamp strm, int level, int strategy) { 732 deflate_state *s; 733 compress_func func; 734 735 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 736 s = strm->state; 737 738 #ifdef FASTEST 739 if (level != 0) level = 1; 740 #else 741 if (level == Z_DEFAULT_COMPRESSION) level = 6; 742 #endif 743 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { 744 return Z_STREAM_ERROR; 745 } 746 func = configuration_table[s->level].func; 747 748 if ((strategy != s->strategy || func != configuration_table[level].func) && 749 s->last_flush != -2) { 750 /* Flush the last buffer: */ 751 int err = deflate(strm, Z_BLOCK); 752 if (err == Z_STREAM_ERROR) 753 return err; 754 if (strm->avail_in || (s->strstart - s->block_start) + s->lookahead) 755 return Z_BUF_ERROR; 756 } 757 if (s->level != level) { 758 if (s->level == 0 && s->matches != 0) { 759 if (s->matches == 1) 760 slide_hash(s); 761 else 762 CLEAR_HASH(s); 763 s->matches = 0; 764 } 765 s->level = level; 766 s->max_lazy_match = configuration_table[level].max_lazy; 767 s->good_match = configuration_table[level].good_length; 768 s->nice_match = configuration_table[level].nice_length; 769 s->max_chain_length = configuration_table[level].max_chain; 770 } 771 s->strategy = strategy; 772 return Z_OK; 773 } 774 775 /* ========================================================================= */ 776 int ZEXPORT deflateTune(z_streamp strm, int good_length, int max_lazy, 777 int nice_length, int max_chain) { 778 deflate_state *s; 779 780 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 781 s = strm->state; 782 s->good_match = (uInt)good_length; 783 s->max_lazy_match = (uInt)max_lazy; 784 s->nice_match = nice_length; 785 s->max_chain_length = (uInt)max_chain; 786 return Z_OK; 787 } 788 789 /* ========================================================================= 790 * For the default windowBits of 15 and memLevel of 8, this function returns a 791 * close to exact, as well as small, upper bound on the compressed size. This 792 * is an expansion of ~0.03%, plus a small constant. 793 * 794 * For any setting other than those defaults for windowBits and memLevel, one 795 * of two worst case bounds is returned. This is at most an expansion of ~4% or 796 * ~13%, plus a small constant. 797 * 798 * Both the 0.03% and 4% derive from the overhead of stored blocks. The first 799 * one is for stored blocks of 16383 bytes (memLevel == 8), whereas the second 800 * is for stored blocks of 127 bytes (the worst case memLevel == 1). The 801 * expansion results from five bytes of header for each stored block. 802 * 803 * The larger expansion of 13% results from a window size less than or equal to 804 * the symbols buffer size (windowBits <= memLevel + 7). In that case some of 805 * the data being compressed may have slid out of the sliding window, impeding 806 * a stored block from being emitted. Then the only choice is a fixed or 807 * dynamic block, where a fixed block limits the maximum expansion to 9 bits 808 * per 8-bit byte, plus 10 bits for every block. The smallest block size for 809 * which this can occur is 255 (memLevel == 2). 810 * 811 * Shifts are used to approximate divisions, for speed. 812 */ 813 uLong ZEXPORT deflateBound(z_streamp strm, uLong sourceLen) { 814 deflate_state *s; 815 uLong fixedlen, storelen, wraplen; 816 817 /* upper bound for fixed blocks with 9-bit literals and length 255 818 (memLevel == 2, which is the lowest that may not use stored blocks) -- 819 ~13% overhead plus a small constant */ 820 fixedlen = sourceLen + (sourceLen >> 3) + (sourceLen >> 8) + 821 (sourceLen >> 9) + 4; 822 823 /* upper bound for stored blocks with length 127 (memLevel == 1) -- 824 ~4% overhead plus a small constant */ 825 storelen = sourceLen + (sourceLen >> 5) + (sourceLen >> 7) + 826 (sourceLen >> 11) + 7; 827 828 /* if can't get parameters, return larger bound plus a zlib wrapper */ 829 if (deflateStateCheck(strm)) 830 return (fixedlen > storelen ? fixedlen : storelen) + 6; 831 832 /* compute wrapper length */ 833 s = strm->state; 834 switch (s->wrap) { 835 case 0: /* raw deflate */ 836 wraplen = 0; 837 break; 838 case 1: /* zlib wrapper */ 839 wraplen = 6 + (s->strstart ? 4 : 0); 840 break; 841 #ifdef GZIP 842 case 2: /* gzip wrapper */ 843 wraplen = 18; 844 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */ 845 Bytef *str; 846 if (s->gzhead->extra != Z_NULL) 847 wraplen += 2 + s->gzhead->extra_len; 848 str = s->gzhead->name; 849 if (str != Z_NULL) 850 do { 851 wraplen++; 852 } while (*str++); 853 str = s->gzhead->comment; 854 if (str != Z_NULL) 855 do { 856 wraplen++; 857 } while (*str++); 858 if (s->gzhead->hcrc) 859 wraplen += 2; 860 } 861 break; 862 #endif 863 default: /* for compiler happiness */ 864 wraplen = 6; 865 } 866 867 /* if not default parameters, return one of the conservative bounds */ 868 if (s->w_bits != 15 || s->hash_bits != 8 + 7) 869 return (s->w_bits <= s->hash_bits && s->level ? fixedlen : storelen) + 870 wraplen; 871 872 /* default settings: return tight bound for that case -- ~0.03% overhead 873 plus a small constant */ 874 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) + 875 (sourceLen >> 25) + 13 - 6 + wraplen; 876 } 877 878 /* ========================================================================= 879 * Put a short in the pending buffer. The 16-bit value is put in MSB order. 880 * IN assertion: the stream state is correct and there is enough room in 881 * pending_buf. 882 */ 883 local void putShortMSB(deflate_state *s, uInt b) { 884 put_byte(s, (Byte)(b >> 8)); 885 put_byte(s, (Byte)(b & 0xff)); 886 } 887 888 /* ========================================================================= 889 * Flush as much pending output as possible. All deflate() output, except for 890 * some deflate_stored() output, goes through this function so some 891 * applications may wish to modify it to avoid allocating a large 892 * strm->next_out buffer and copying into it. (See also read_buf()). 893 */ 894 local void flush_pending(z_streamp strm) { 895 unsigned len; 896 deflate_state *s = strm->state; 897 898 _tr_flush_bits(s); 899 len = s->pending; 900 if (len > strm->avail_out) len = strm->avail_out; 901 if (len == 0) return; 902 903 zmemcpy(strm->next_out, s->pending_out, len); 904 strm->next_out += len; 905 s->pending_out += len; 906 strm->total_out += len; 907 strm->avail_out -= len; 908 s->pending -= len; 909 if (s->pending == 0) { 910 s->pending_out = s->pending_buf; 911 } 912 } 913 914 /* =========================================================================== 915 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1]. 916 */ 917 #define HCRC_UPDATE(beg) \ 918 do { \ 919 if (s->gzhead->hcrc && s->pending > (beg)) \ 920 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \ 921 s->pending - (beg)); \ 922 } while (0) 923 924 /* ========================================================================= */ 925 int ZEXPORT deflate(z_streamp strm, int flush) { 926 int old_flush; /* value of flush param for previous deflate call */ 927 deflate_state *s; 928 929 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) { 930 return Z_STREAM_ERROR; 931 } 932 s = strm->state; 933 934 if (strm->next_out == Z_NULL || 935 (strm->avail_in != 0 && strm->next_in == Z_NULL) || 936 (s->status == FINISH_STATE && flush != Z_FINISH)) { 937 ERR_RETURN(strm, Z_STREAM_ERROR); 938 } 939 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); 940 941 old_flush = s->last_flush; 942 s->last_flush = flush; 943 944 /* Flush as much pending output as possible */ 945 if (s->pending != 0) { 946 flush_pending(strm); 947 if (strm->avail_out == 0) { 948 /* Since avail_out is 0, deflate will be called again with 949 * more output space, but possibly with both pending and 950 * avail_in equal to zero. There won't be anything to do, 951 * but this is not an error situation so make sure we 952 * return OK instead of BUF_ERROR at next call of deflate: 953 */ 954 s->last_flush = -1; 955 return Z_OK; 956 } 957 958 /* Make sure there is something to do and avoid duplicate consecutive 959 * flushes. For repeated and useless calls with Z_FINISH, we keep 960 * returning Z_STREAM_END instead of Z_BUF_ERROR. 961 */ 962 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) && 963 flush != Z_FINISH) { 964 ERR_RETURN(strm, Z_BUF_ERROR); 965 } 966 967 /* User must not provide more input after the first FINISH: */ 968 if (s->status == FINISH_STATE && strm->avail_in != 0) { 969 ERR_RETURN(strm, Z_BUF_ERROR); 970 } 971 972 /* Write the header */ 973 if (s->status == INIT_STATE && s->wrap == 0) 974 s->status = BUSY_STATE; 975 if (s->status == INIT_STATE) { 976 /* zlib header */ 977 uInt header = (Z_DEFLATED + ((s->w_bits - 8) << 4)) << 8; 978 uInt level_flags; 979 980 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) 981 level_flags = 0; 982 else if (s->level < 6) 983 level_flags = 1; 984 else if (s->level == 6) 985 level_flags = 2; 986 else 987 level_flags = 3; 988 header |= (level_flags << 6); 989 if (s->strstart != 0) header |= PRESET_DICT; 990 header += 31 - (header % 31); 991 992 putShortMSB(s, header); 993 994 /* Save the adler32 of the preset dictionary: */ 995 if (s->strstart != 0) { 996 putShortMSB(s, (uInt)(strm->adler >> 16)); 997 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 998 } 999 strm->adler = adler32(0L, Z_NULL, 0); 1000 s->status = BUSY_STATE; 1001 1002 /* Compression must start with an empty pending buffer */ 1003 flush_pending(strm); 1004 if (s->pending != 0) { 1005 s->last_flush = -1; 1006 return Z_OK; 1007 } 1008 } 1009 #ifdef GZIP 1010 if (s->status == GZIP_STATE) { 1011 /* gzip header */ 1012 strm->adler = crc32(0L, Z_NULL, 0); 1013 put_byte(s, 31); 1014 put_byte(s, 139); 1015 put_byte(s, 8); 1016 if (s->gzhead == Z_NULL) { 1017 put_byte(s, 0); 1018 put_byte(s, 0); 1019 put_byte(s, 0); 1020 put_byte(s, 0); 1021 put_byte(s, 0); 1022 put_byte(s, s->level == 9 ? 2 : 1023 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 1024 4 : 0)); 1025 put_byte(s, OS_CODE); 1026 s->status = BUSY_STATE; 1027 1028 /* Compression must start with an empty pending buffer */ 1029 flush_pending(strm); 1030 if (s->pending != 0) { 1031 s->last_flush = -1; 1032 return Z_OK; 1033 } 1034 } 1035 else { 1036 put_byte(s, (s->gzhead->text ? 1 : 0) + 1037 (s->gzhead->hcrc ? 2 : 0) + 1038 (s->gzhead->extra == Z_NULL ? 0 : 4) + 1039 (s->gzhead->name == Z_NULL ? 0 : 8) + 1040 (s->gzhead->comment == Z_NULL ? 0 : 16) 1041 ); 1042 put_byte(s, (Byte)(s->gzhead->time & 0xff)); 1043 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); 1044 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); 1045 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); 1046 put_byte(s, s->level == 9 ? 2 : 1047 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? 1048 4 : 0)); 1049 put_byte(s, s->gzhead->os & 0xff); 1050 if (s->gzhead->extra != Z_NULL) { 1051 put_byte(s, s->gzhead->extra_len & 0xff); 1052 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); 1053 } 1054 if (s->gzhead->hcrc) 1055 strm->adler = crc32(strm->adler, s->pending_buf, 1056 s->pending); 1057 s->gzindex = 0; 1058 s->status = EXTRA_STATE; 1059 } 1060 } 1061 if (s->status == EXTRA_STATE) { 1062 if (s->gzhead->extra != Z_NULL) { 1063 ulg beg = s->pending; /* start of bytes to update crc */ 1064 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex; 1065 while (s->pending + left > s->pending_buf_size) { 1066 uInt copy = s->pending_buf_size - s->pending; 1067 zmemcpy(s->pending_buf + s->pending, 1068 s->gzhead->extra + s->gzindex, copy); 1069 s->pending = s->pending_buf_size; 1070 HCRC_UPDATE(beg); 1071 s->gzindex += copy; 1072 flush_pending(strm); 1073 if (s->pending != 0) { 1074 s->last_flush = -1; 1075 return Z_OK; 1076 } 1077 beg = 0; 1078 left -= copy; 1079 } 1080 zmemcpy(s->pending_buf + s->pending, 1081 s->gzhead->extra + s->gzindex, left); 1082 s->pending += left; 1083 HCRC_UPDATE(beg); 1084 s->gzindex = 0; 1085 } 1086 s->status = NAME_STATE; 1087 } 1088 if (s->status == NAME_STATE) { 1089 if (s->gzhead->name != Z_NULL) { 1090 ulg beg = s->pending; /* start of bytes to update crc */ 1091 int val; 1092 do { 1093 if (s->pending == s->pending_buf_size) { 1094 HCRC_UPDATE(beg); 1095 flush_pending(strm); 1096 if (s->pending != 0) { 1097 s->last_flush = -1; 1098 return Z_OK; 1099 } 1100 beg = 0; 1101 } 1102 val = s->gzhead->name[s->gzindex++]; 1103 put_byte(s, val); 1104 } while (val != 0); 1105 HCRC_UPDATE(beg); 1106 s->gzindex = 0; 1107 } 1108 s->status = COMMENT_STATE; 1109 } 1110 if (s->status == COMMENT_STATE) { 1111 if (s->gzhead->comment != Z_NULL) { 1112 ulg beg = s->pending; /* start of bytes to update crc */ 1113 int val; 1114 do { 1115 if (s->pending == s->pending_buf_size) { 1116 HCRC_UPDATE(beg); 1117 flush_pending(strm); 1118 if (s->pending != 0) { 1119 s->last_flush = -1; 1120 return Z_OK; 1121 } 1122 beg = 0; 1123 } 1124 val = s->gzhead->comment[s->gzindex++]; 1125 put_byte(s, val); 1126 } while (val != 0); 1127 HCRC_UPDATE(beg); 1128 } 1129 s->status = HCRC_STATE; 1130 } 1131 if (s->status == HCRC_STATE) { 1132 if (s->gzhead->hcrc) { 1133 if (s->pending + 2 > s->pending_buf_size) { 1134 flush_pending(strm); 1135 if (s->pending != 0) { 1136 s->last_flush = -1; 1137 return Z_OK; 1138 } 1139 } 1140 put_byte(s, (Byte)(strm->adler & 0xff)); 1141 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 1142 strm->adler = crc32(0L, Z_NULL, 0); 1143 } 1144 s->status = BUSY_STATE; 1145 1146 /* Compression must start with an empty pending buffer */ 1147 flush_pending(strm); 1148 if (s->pending != 0) { 1149 s->last_flush = -1; 1150 return Z_OK; 1151 } 1152 } 1153 #endif 1154 1155 /* Start a new block or continue the current one. 1156 */ 1157 if (strm->avail_in != 0 || s->lookahead != 0 || 1158 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { 1159 block_state bstate; 1160 1161 bstate = s->level == 0 ? deflate_stored(s, flush) : 1162 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : 1163 s->strategy == Z_RLE ? deflate_rle(s, flush) : 1164 (*(configuration_table[s->level].func))(s, flush); 1165 1166 if (bstate == finish_started || bstate == finish_done) { 1167 s->status = FINISH_STATE; 1168 } 1169 if (bstate == need_more || bstate == finish_started) { 1170 if (strm->avail_out == 0) { 1171 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ 1172 } 1173 return Z_OK; 1174 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call 1175 * of deflate should use the same flush parameter to make sure 1176 * that the flush is complete. So we don't have to output an 1177 * empty block here, this will be done at next call. This also 1178 * ensures that for a very small output buffer, we emit at most 1179 * one empty block. 1180 */ 1181 } 1182 if (bstate == block_done) { 1183 if (flush == Z_PARTIAL_FLUSH) { 1184 _tr_align(s); 1185 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */ 1186 _tr_stored_block(s, (char*)0, 0L, 0); 1187 /* For a full flush, this empty block will be recognized 1188 * as a special marker by inflate_sync(). 1189 */ 1190 if (flush == Z_FULL_FLUSH) { 1191 CLEAR_HASH(s); /* forget history */ 1192 if (s->lookahead == 0) { 1193 s->strstart = 0; 1194 s->block_start = 0L; 1195 s->insert = 0; 1196 } 1197 } 1198 } 1199 flush_pending(strm); 1200 if (strm->avail_out == 0) { 1201 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ 1202 return Z_OK; 1203 } 1204 } 1205 } 1206 1207 if (flush != Z_FINISH) return Z_OK; 1208 if (s->wrap <= 0) return Z_STREAM_END; 1209 1210 /* Write the trailer */ 1211 #ifdef GZIP 1212 if (s->wrap == 2) { 1213 put_byte(s, (Byte)(strm->adler & 0xff)); 1214 put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); 1215 put_byte(s, (Byte)((strm->adler >> 16) & 0xff)); 1216 put_byte(s, (Byte)((strm->adler >> 24) & 0xff)); 1217 put_byte(s, (Byte)(strm->total_in & 0xff)); 1218 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff)); 1219 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff)); 1220 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff)); 1221 } 1222 else 1223 #endif 1224 { 1225 putShortMSB(s, (uInt)(strm->adler >> 16)); 1226 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 1227 } 1228 flush_pending(strm); 1229 /* If avail_out is zero, the application will call deflate again 1230 * to flush the rest. 1231 */ 1232 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ 1233 return s->pending != 0 ? Z_OK : Z_STREAM_END; 1234 } 1235 1236 /* ========================================================================= */ 1237 int ZEXPORT deflateEnd(z_streamp strm) { 1238 int status; 1239 1240 if (deflateStateCheck(strm)) return Z_STREAM_ERROR; 1241 1242 status = strm->state->status; 1243 1244 /* Deallocate in reverse order of allocations: */ 1245 TRY_FREE(strm, strm->state->pending_buf); 1246 TRY_FREE(strm, strm->state->head); 1247 TRY_FREE(strm, strm->state->prev); 1248 TRY_FREE(strm, strm->state->window); 1249 1250 ZFREE(strm, strm->state); 1251 strm->state = Z_NULL; 1252 1253 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; 1254 } 1255 1256 /* ========================================================================= 1257 * Copy the source state to the destination state. 1258 * To simplify the source, this is not supported for 16-bit MSDOS (which 1259 * doesn't have enough memory anyway to duplicate compression states). 1260 */ 1261 int ZEXPORT deflateCopy(z_streamp dest, z_streamp source) { 1262 #ifdef MAXSEG_64K 1263 (void)dest; 1264 (void)source; 1265 return Z_STREAM_ERROR; 1266 #else 1267 deflate_state *ds; 1268 deflate_state *ss; 1269 1270 1271 if (deflateStateCheck(source) || dest == Z_NULL) { 1272 return Z_STREAM_ERROR; 1273 } 1274 1275 ss = source->state; 1276 1277 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream)); 1278 1279 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); 1280 if (ds == Z_NULL) return Z_MEM_ERROR; 1281 dest->state = (struct internal_state FAR *) ds; 1282 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state)); 1283 ds->strm = dest; 1284 1285 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); 1286 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); 1287 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); 1288 ds->pending_buf = (uchf *) ZALLOC(dest, ds->lit_bufsize, 4); 1289 1290 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || 1291 ds->pending_buf == Z_NULL) { 1292 deflateEnd (dest); 1293 return Z_MEM_ERROR; 1294 } 1295 /* following zmemcpy do not work for 16-bit MSDOS */ 1296 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); 1297 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos)); 1298 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos)); 1299 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); 1300 1301 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); 1302 ds->sym_buf = ds->pending_buf + ds->lit_bufsize; 1303 1304 ds->l_desc.dyn_tree = ds->dyn_ltree; 1305 ds->d_desc.dyn_tree = ds->dyn_dtree; 1306 ds->bl_desc.dyn_tree = ds->bl_tree; 1307 1308 return Z_OK; 1309 #endif /* MAXSEG_64K */ 1310 } 1311 1312 #ifndef FASTEST 1313 /* =========================================================================== 1314 * Set match_start to the longest match starting at the given string and 1315 * return its length. Matches shorter or equal to prev_length are discarded, 1316 * in which case the result is equal to prev_length and match_start is 1317 * garbage. 1318 * IN assertions: cur_match is the head of the hash chain for the current 1319 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 1320 * OUT assertion: the match length is not greater than s->lookahead. 1321 */ 1322 local uInt longest_match(deflate_state *s, IPos cur_match) { 1323 unsigned chain_length = s->max_chain_length;/* max hash chain length */ 1324 register Bytef *scan = s->window + s->strstart; /* current string */ 1325 register Bytef *match; /* matched string */ 1326 register int len; /* length of current match */ 1327 int best_len = (int)s->prev_length; /* best match length so far */ 1328 int nice_match = s->nice_match; /* stop if match long enough */ 1329 IPos limit = s->strstart > (IPos)MAX_DIST(s) ? 1330 s->strstart - (IPos)MAX_DIST(s) : NIL; 1331 /* Stop when cur_match becomes <= limit. To simplify the code, 1332 * we prevent matches with the string of window index 0. 1333 */ 1334 Posf *prev = s->prev; 1335 uInt wmask = s->w_mask; 1336 1337 #ifdef UNALIGNED_OK 1338 /* Compare two bytes at a time. Note: this is not always beneficial. 1339 * Try with and without -DUNALIGNED_OK to check. 1340 */ 1341 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; 1342 register ush scan_start = *(ushf*)scan; 1343 register ush scan_end = *(ushf*)(scan + best_len - 1); 1344 #else 1345 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1346 register Byte scan_end1 = scan[best_len - 1]; 1347 register Byte scan_end = scan[best_len]; 1348 #endif 1349 1350 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1351 * It is easy to get rid of this optimization if necessary. 1352 */ 1353 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1354 1355 /* Do not waste too much time if we already have a good match: */ 1356 if (s->prev_length >= s->good_match) { 1357 chain_length >>= 2; 1358 } 1359 /* Do not look for matches beyond the end of the input. This is necessary 1360 * to make deflate deterministic. 1361 */ 1362 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead; 1363 1364 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 1365 "need lookahead"); 1366 1367 do { 1368 Assert(cur_match < s->strstart, "no future"); 1369 match = s->window + cur_match; 1370 1371 /* Skip to next match if the match length cannot increase 1372 * or if the match length is less than 2. Note that the checks below 1373 * for insufficient lookahead only occur occasionally for performance 1374 * reasons. Therefore uninitialized memory will be accessed, and 1375 * conditional jumps will be made that depend on those values. 1376 * However the length of the match is limited to the lookahead, so 1377 * the output of deflate is not affected by the uninitialized values. 1378 */ 1379 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) 1380 /* This code assumes sizeof(unsigned short) == 2. Do not use 1381 * UNALIGNED_OK if your compiler uses a different size. 1382 */ 1383 if (*(ushf*)(match + best_len - 1) != scan_end || 1384 *(ushf*)match != scan_start) continue; 1385 1386 /* It is not necessary to compare scan[2] and match[2] since they are 1387 * always equal when the other bytes match, given that the hash keys 1388 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at 1389 * strstart + 3, + 5, up to strstart + 257. We check for insufficient 1390 * lookahead only every 4th comparison; the 128th check will be made 1391 * at strstart + 257. If MAX_MATCH-2 is not a multiple of 8, it is 1392 * necessary to put more guard bytes at the end of the window, or 1393 * to check more often for insufficient lookahead. 1394 */ 1395 Assert(scan[2] == match[2], "scan[2]?"); 1396 scan++, match++; 1397 do { 1398 } while (*(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1399 *(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1400 *(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1401 *(ushf*)(scan += 2) == *(ushf*)(match += 2) && 1402 scan < strend); 1403 /* The funny "do {}" generates better code on most compilers */ 1404 1405 /* Here, scan <= window + strstart + 257 */ 1406 Assert(scan <= s->window + (unsigned)(s->window_size - 1), 1407 "wild scan"); 1408 if (*scan == *match) scan++; 1409 1410 len = (MAX_MATCH - 1) - (int)(strend - scan); 1411 scan = strend - (MAX_MATCH-1); 1412 1413 #else /* UNALIGNED_OK */ 1414 1415 if (match[best_len] != scan_end || 1416 match[best_len - 1] != scan_end1 || 1417 *match != *scan || 1418 *++match != scan[1]) continue; 1419 1420 /* The check at best_len - 1 can be removed because it will be made 1421 * again later. (This heuristic is not always a win.) 1422 * It is not necessary to compare scan[2] and match[2] since they 1423 * are always equal when the other bytes match, given that 1424 * the hash keys are equal and that HASH_BITS >= 8. 1425 */ 1426 scan += 2, match++; 1427 Assert(*scan == *match, "match[2]?"); 1428 1429 /* We check for insufficient lookahead only every 8th comparison; 1430 * the 256th check will be made at strstart + 258. 1431 */ 1432 do { 1433 } while (*++scan == *++match && *++scan == *++match && 1434 *++scan == *++match && *++scan == *++match && 1435 *++scan == *++match && *++scan == *++match && 1436 *++scan == *++match && *++scan == *++match && 1437 scan < strend); 1438 1439 Assert(scan <= s->window + (unsigned)(s->window_size - 1), 1440 "wild scan"); 1441 1442 len = MAX_MATCH - (int)(strend - scan); 1443 scan = strend - MAX_MATCH; 1444 1445 #endif /* UNALIGNED_OK */ 1446 1447 if (len > best_len) { 1448 s->match_start = cur_match; 1449 best_len = len; 1450 if (len >= nice_match) break; 1451 #ifdef UNALIGNED_OK 1452 scan_end = *(ushf*)(scan + best_len - 1); 1453 #else 1454 scan_end1 = scan[best_len - 1]; 1455 scan_end = scan[best_len]; 1456 #endif 1457 } 1458 } while ((cur_match = prev[cur_match & wmask]) > limit 1459 && --chain_length != 0); 1460 1461 if ((uInt)best_len <= s->lookahead) return (uInt)best_len; 1462 return s->lookahead; 1463 } 1464 1465 #else /* FASTEST */ 1466 1467 /* --------------------------------------------------------------------------- 1468 * Optimized version for FASTEST only 1469 */ 1470 local uInt longest_match(deflate_state *s, IPos cur_match) { 1471 register Bytef *scan = s->window + s->strstart; /* current string */ 1472 register Bytef *match; /* matched string */ 1473 register int len; /* length of current match */ 1474 register Bytef *strend = s->window + s->strstart + MAX_MATCH; 1475 1476 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 1477 * It is easy to get rid of this optimization if necessary. 1478 */ 1479 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 1480 1481 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, 1482 "need lookahead"); 1483 1484 Assert(cur_match < s->strstart, "no future"); 1485 1486 match = s->window + cur_match; 1487 1488 /* Return failure if the match length is less than 2: 1489 */ 1490 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; 1491 1492 /* The check at best_len - 1 can be removed because it will be made 1493 * again later. (This heuristic is not always a win.) 1494 * It is not necessary to compare scan[2] and match[2] since they 1495 * are always equal when the other bytes match, given that 1496 * the hash keys are equal and that HASH_BITS >= 8. 1497 */ 1498 scan += 2, match += 2; 1499 Assert(*scan == *match, "match[2]?"); 1500 1501 /* We check for insufficient lookahead only every 8th comparison; 1502 * the 256th check will be made at strstart + 258. 1503 */ 1504 do { 1505 } while (*++scan == *++match && *++scan == *++match && 1506 *++scan == *++match && *++scan == *++match && 1507 *++scan == *++match && *++scan == *++match && 1508 *++scan == *++match && *++scan == *++match && 1509 scan < strend); 1510 1511 Assert(scan <= s->window + (unsigned)(s->window_size - 1), "wild scan"); 1512 1513 len = MAX_MATCH - (int)(strend - scan); 1514 1515 if (len < MIN_MATCH) return MIN_MATCH - 1; 1516 1517 s->match_start = cur_match; 1518 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead; 1519 } 1520 1521 #endif /* FASTEST */ 1522 1523 #ifdef ZLIB_DEBUG 1524 1525 #define EQUAL 0 1526 /* result of memcmp for equal strings */ 1527 1528 /* =========================================================================== 1529 * Check that the match at match_start is indeed a match. 1530 */ 1531 local void check_match(deflate_state *s, IPos start, IPos match, int length) { 1532 /* check that the match is indeed a match */ 1533 if (zmemcmp(s->window + match, 1534 s->window + start, length) != EQUAL) { 1535 fprintf(stderr, " start %u, match %u, length %d\n", 1536 start, match, length); 1537 do { 1538 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); 1539 } while (--length != 0); 1540 z_error("invalid match"); 1541 } 1542 if (z_verbose > 1) { 1543 fprintf(stderr,"\\[%d,%d]", start - match, length); 1544 do { putc(s->window[start++], stderr); } while (--length != 0); 1545 } 1546 } 1547 #else 1548 # define check_match(s, start, match, length) 1549 #endif /* ZLIB_DEBUG */ 1550 1551 /* =========================================================================== 1552 * Flush the current block, with given end-of-file flag. 1553 * IN assertion: strstart is set to the end of the current match. 1554 */ 1555 #define FLUSH_BLOCK_ONLY(s, last) { \ 1556 _tr_flush_block(s, (s->block_start >= 0L ? \ 1557 (charf *)&s->window[(unsigned)s->block_start] : \ 1558 (charf *)Z_NULL), \ 1559 (ulg)((long)s->strstart - s->block_start), \ 1560 (last)); \ 1561 s->block_start = s->strstart; \ 1562 flush_pending(s->strm); \ 1563 Tracev((stderr,"[FLUSH]")); \ 1564 } 1565 1566 /* Same but force premature exit if necessary. */ 1567 #define FLUSH_BLOCK(s, last) { \ 1568 FLUSH_BLOCK_ONLY(s, last); \ 1569 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \ 1570 } 1571 1572 /* Maximum stored block length in deflate format (not including header). */ 1573 #define MAX_STORED 65535 1574 1575 /* Minimum of a and b. */ 1576 #define MIN(a, b) ((a) > (b) ? (b) : (a)) 1577 1578 /* =========================================================================== 1579 * Copy without compression as much as possible from the input stream, return 1580 * the current block state. 1581 * 1582 * In case deflateParams() is used to later switch to a non-zero compression 1583 * level, s->matches (otherwise unused when storing) keeps track of the number 1584 * of hash table slides to perform. If s->matches is 1, then one hash table 1585 * slide will be done when switching. If s->matches is 2, the maximum value 1586 * allowed here, then the hash table will be cleared, since two or more slides 1587 * is the same as a clear. 1588 * 1589 * deflate_stored() is written to minimize the number of times an input byte is 1590 * copied. It is most efficient with large input and output buffers, which 1591 * maximizes the opportunities to have a single copy from next_in to next_out. 1592 */ 1593 local block_state deflate_stored(deflate_state *s, int flush) { 1594 /* Smallest worthy block size when not flushing or finishing. By default 1595 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For 1596 * large input and output buffers, the stored block size will be larger. 1597 */ 1598 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size); 1599 1600 /* Copy as many min_block or larger stored blocks directly to next_out as 1601 * possible. If flushing, copy the remaining available input to next_out as 1602 * stored blocks, if there is enough space. 1603 */ 1604 unsigned len, left, have, last = 0; 1605 unsigned used = s->strm->avail_in; 1606 do { 1607 /* Set len to the maximum size block that we can copy directly with the 1608 * available input data and output space. Set left to how much of that 1609 * would be copied from what's left in the window. 1610 */ 1611 len = MAX_STORED; /* maximum deflate stored block length */ 1612 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1613 if (s->strm->avail_out < have) /* need room for header */ 1614 break; 1615 /* maximum stored block length that will fit in avail_out: */ 1616 have = s->strm->avail_out - have; 1617 left = s->strstart - s->block_start; /* bytes left in window */ 1618 if (len > (ulg)left + s->strm->avail_in) 1619 len = left + s->strm->avail_in; /* limit len to the input */ 1620 if (len > have) 1621 len = have; /* limit len to the output */ 1622 1623 /* If the stored block would be less than min_block in length, or if 1624 * unable to copy all of the available input when flushing, then try 1625 * copying to the window and the pending buffer instead. Also don't 1626 * write an empty block when flushing -- deflate() does that. 1627 */ 1628 if (len < min_block && ((len == 0 && flush != Z_FINISH) || 1629 flush == Z_NO_FLUSH || 1630 len != left + s->strm->avail_in)) 1631 break; 1632 1633 /* Make a dummy stored block in pending to get the header bytes, 1634 * including any pending bits. This also updates the debugging counts. 1635 */ 1636 last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0; 1637 _tr_stored_block(s, (char *)0, 0L, last); 1638 1639 /* Replace the lengths in the dummy stored block with len. */ 1640 s->pending_buf[s->pending - 4] = len; 1641 s->pending_buf[s->pending - 3] = len >> 8; 1642 s->pending_buf[s->pending - 2] = ~len; 1643 s->pending_buf[s->pending - 1] = ~len >> 8; 1644 1645 /* Write the stored block header bytes. */ 1646 flush_pending(s->strm); 1647 1648 #ifdef ZLIB_DEBUG 1649 /* Update debugging counts for the data about to be copied. */ 1650 s->compressed_len += len << 3; 1651 s->bits_sent += len << 3; 1652 #endif 1653 1654 /* Copy uncompressed bytes from the window to next_out. */ 1655 if (left) { 1656 if (left > len) 1657 left = len; 1658 zmemcpy(s->strm->next_out, s->window + s->block_start, left); 1659 s->strm->next_out += left; 1660 s->strm->avail_out -= left; 1661 s->strm->total_out += left; 1662 s->block_start += left; 1663 len -= left; 1664 } 1665 1666 /* Copy uncompressed bytes directly from next_in to next_out, updating 1667 * the check value. 1668 */ 1669 if (len) { 1670 read_buf(s->strm, s->strm->next_out, len); 1671 s->strm->next_out += len; 1672 s->strm->avail_out -= len; 1673 s->strm->total_out += len; 1674 } 1675 } while (last == 0); 1676 1677 /* Update the sliding window with the last s->w_size bytes of the copied 1678 * data, or append all of the copied data to the existing window if less 1679 * than s->w_size bytes were copied. Also update the number of bytes to 1680 * insert in the hash tables, in the event that deflateParams() switches to 1681 * a non-zero compression level. 1682 */ 1683 used -= s->strm->avail_in; /* number of input bytes directly copied */ 1684 if (used) { 1685 /* If any input was used, then no unused input remains in the window, 1686 * therefore s->block_start == s->strstart. 1687 */ 1688 if (used >= s->w_size) { /* supplant the previous history */ 1689 s->matches = 2; /* clear hash */ 1690 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size); 1691 s->strstart = s->w_size; 1692 s->insert = s->strstart; 1693 } 1694 else { 1695 if (s->window_size - s->strstart <= used) { 1696 /* Slide the window down. */ 1697 s->strstart -= s->w_size; 1698 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1699 if (s->matches < 2) 1700 s->matches++; /* add a pending slide_hash() */ 1701 if (s->insert > s->strstart) 1702 s->insert = s->strstart; 1703 } 1704 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used); 1705 s->strstart += used; 1706 s->insert += MIN(used, s->w_size - s->insert); 1707 } 1708 s->block_start = s->strstart; 1709 } 1710 if (s->high_water < s->strstart) 1711 s->high_water = s->strstart; 1712 1713 /* If the last block was written to next_out, then done. */ 1714 if (last) 1715 return finish_done; 1716 1717 /* If flushing and all input has been consumed, then done. */ 1718 if (flush != Z_NO_FLUSH && flush != Z_FINISH && 1719 s->strm->avail_in == 0 && (long)s->strstart == s->block_start) 1720 return block_done; 1721 1722 /* Fill the window with any remaining input. */ 1723 have = s->window_size - s->strstart; 1724 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) { 1725 /* Slide the window down. */ 1726 s->block_start -= s->w_size; 1727 s->strstart -= s->w_size; 1728 zmemcpy(s->window, s->window + s->w_size, s->strstart); 1729 if (s->matches < 2) 1730 s->matches++; /* add a pending slide_hash() */ 1731 have += s->w_size; /* more space now */ 1732 if (s->insert > s->strstart) 1733 s->insert = s->strstart; 1734 } 1735 if (have > s->strm->avail_in) 1736 have = s->strm->avail_in; 1737 if (have) { 1738 read_buf(s->strm, s->window + s->strstart, have); 1739 s->strstart += have; 1740 s->insert += MIN(have, s->w_size - s->insert); 1741 } 1742 if (s->high_water < s->strstart) 1743 s->high_water = s->strstart; 1744 1745 /* There was not enough avail_out to write a complete worthy or flushed 1746 * stored block to next_out. Write a stored block to pending instead, if we 1747 * have enough input for a worthy block, or if flushing and there is enough 1748 * room for the remaining input as a stored block in the pending buffer. 1749 */ 1750 have = (s->bi_valid + 42) >> 3; /* number of header bytes */ 1751 /* maximum stored block length that will fit in pending: */ 1752 have = MIN(s->pending_buf_size - have, MAX_STORED); 1753 min_block = MIN(have, s->w_size); 1754 left = s->strstart - s->block_start; 1755 if (left >= min_block || 1756 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH && 1757 s->strm->avail_in == 0 && left <= have)) { 1758 len = MIN(left, have); 1759 last = flush == Z_FINISH && s->strm->avail_in == 0 && 1760 len == left ? 1 : 0; 1761 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last); 1762 s->block_start += len; 1763 flush_pending(s->strm); 1764 } 1765 1766 /* We've done all we can with the available input and output. */ 1767 return last ? finish_started : need_more; 1768 } 1769 1770 /* =========================================================================== 1771 * Compress as much as possible from the input stream, return the current 1772 * block state. 1773 * This function does not perform lazy evaluation of matches and inserts 1774 * new strings in the dictionary only for unmatched strings or for short 1775 * matches. It is used only for the fast compression options. 1776 */ 1777 local block_state deflate_fast(deflate_state *s, int flush) { 1778 IPos hash_head; /* head of the hash chain */ 1779 int bflush; /* set if current block must be flushed */ 1780 1781 for (;;) { 1782 /* Make sure that we always have enough lookahead, except 1783 * at the end of the input file. We need MAX_MATCH bytes 1784 * for the next match, plus MIN_MATCH bytes to insert the 1785 * string following the next match. 1786 */ 1787 if (s->lookahead < MIN_LOOKAHEAD) { 1788 fill_window(s); 1789 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1790 return need_more; 1791 } 1792 if (s->lookahead == 0) break; /* flush the current block */ 1793 } 1794 1795 /* Insert the string window[strstart .. strstart + 2] in the 1796 * dictionary, and set hash_head to the head of the hash chain: 1797 */ 1798 hash_head = NIL; 1799 if (s->lookahead >= MIN_MATCH) { 1800 INSERT_STRING(s, s->strstart, hash_head); 1801 } 1802 1803 /* Find the longest match, discarding those <= prev_length. 1804 * At this point we have always match_length < MIN_MATCH 1805 */ 1806 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { 1807 /* To simplify the code, we prevent matches with the string 1808 * of window index 0 (in particular we have to avoid a match 1809 * of the string with itself at the start of the input file). 1810 */ 1811 s->match_length = longest_match (s, hash_head); 1812 /* longest_match() sets match_start */ 1813 } 1814 if (s->match_length >= MIN_MATCH) { 1815 check_match(s, s->strstart, s->match_start, s->match_length); 1816 1817 _tr_tally_dist(s, s->strstart - s->match_start, 1818 s->match_length - MIN_MATCH, bflush); 1819 1820 s->lookahead -= s->match_length; 1821 1822 /* Insert new strings in the hash table only if the match length 1823 * is not too large. This saves time but degrades compression. 1824 */ 1825 #ifndef FASTEST 1826 if (s->match_length <= s->max_insert_length && 1827 s->lookahead >= MIN_MATCH) { 1828 s->match_length--; /* string at strstart already in table */ 1829 do { 1830 s->strstart++; 1831 INSERT_STRING(s, s->strstart, hash_head); 1832 /* strstart never exceeds WSIZE-MAX_MATCH, so there are 1833 * always MIN_MATCH bytes ahead. 1834 */ 1835 } while (--s->match_length != 0); 1836 s->strstart++; 1837 } else 1838 #endif 1839 { 1840 s->strstart += s->match_length; 1841 s->match_length = 0; 1842 s->ins_h = s->window[s->strstart]; 1843 UPDATE_HASH(s, s->ins_h, s->window[s->strstart + 1]); 1844 #if MIN_MATCH != 3 1845 Call UPDATE_HASH() MIN_MATCH-3 more times 1846 #endif 1847 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not 1848 * matter since it will be recomputed at next deflate call. 1849 */ 1850 } 1851 } else { 1852 /* No match, output a literal byte */ 1853 Tracevv((stderr,"%c", s->window[s->strstart])); 1854 _tr_tally_lit(s, s->window[s->strstart], bflush); 1855 s->lookahead--; 1856 s->strstart++; 1857 } 1858 if (bflush) FLUSH_BLOCK(s, 0); 1859 } 1860 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 1861 if (flush == Z_FINISH) { 1862 FLUSH_BLOCK(s, 1); 1863 return finish_done; 1864 } 1865 if (s->sym_next) 1866 FLUSH_BLOCK(s, 0); 1867 return block_done; 1868 } 1869 1870 #ifndef FASTEST 1871 /* =========================================================================== 1872 * Same as above, but achieves better compression. We use a lazy 1873 * evaluation for matches: a match is finally adopted only if there is 1874 * no better match at the next window position. 1875 */ 1876 local block_state deflate_slow(deflate_state *s, int flush) { 1877 IPos hash_head; /* head of hash chain */ 1878 int bflush; /* set if current block must be flushed */ 1879 1880 /* Process the input block. */ 1881 for (;;) { 1882 /* Make sure that we always have enough lookahead, except 1883 * at the end of the input file. We need MAX_MATCH bytes 1884 * for the next match, plus MIN_MATCH bytes to insert the 1885 * string following the next match. 1886 */ 1887 if (s->lookahead < MIN_LOOKAHEAD) { 1888 fill_window(s); 1889 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1890 return need_more; 1891 } 1892 if (s->lookahead == 0) break; /* flush the current block */ 1893 } 1894 1895 /* Insert the string window[strstart .. strstart + 2] in the 1896 * dictionary, and set hash_head to the head of the hash chain: 1897 */ 1898 hash_head = NIL; 1899 if (s->lookahead >= MIN_MATCH) { 1900 INSERT_STRING(s, s->strstart, hash_head); 1901 } 1902 1903 /* Find the longest match, discarding those <= prev_length. 1904 */ 1905 s->prev_length = s->match_length, s->prev_match = s->match_start; 1906 s->match_length = MIN_MATCH-1; 1907 1908 if (hash_head != NIL && s->prev_length < s->max_lazy_match && 1909 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 <= 5 && (s->strategy == Z_FILTERED 1918 #if TOO_FAR <= 32767 1919 || (s->match_length == MIN_MATCH && 1920 s->strstart - s->match_start > TOO_FAR) 1921 #endif 1922 )) { 1923 1924 /* If prev_match is also MIN_MATCH, match_start is garbage 1925 * but we will ignore the current match anyway. 1926 */ 1927 s->match_length = MIN_MATCH-1; 1928 } 1929 } 1930 /* If there was a match at the previous step and the current 1931 * match is not better, output the previous match: 1932 */ 1933 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { 1934 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; 1935 /* Do not insert strings in hash table beyond this. */ 1936 1937 check_match(s, s->strstart - 1, s->prev_match, s->prev_length); 1938 1939 _tr_tally_dist(s, s->strstart - 1 - s->prev_match, 1940 s->prev_length - MIN_MATCH, bflush); 1941 1942 /* Insert in hash table all strings up to the end of the match. 1943 * strstart - 1 and strstart are already inserted. If there is not 1944 * enough lookahead, the last two strings are not inserted in 1945 * the hash table. 1946 */ 1947 s->lookahead -= s->prev_length - 1; 1948 s->prev_length -= 2; 1949 do { 1950 if (++s->strstart <= max_insert) { 1951 INSERT_STRING(s, s->strstart, hash_head); 1952 } 1953 } while (--s->prev_length != 0); 1954 s->match_available = 0; 1955 s->match_length = MIN_MATCH-1; 1956 s->strstart++; 1957 1958 if (bflush) FLUSH_BLOCK(s, 0); 1959 1960 } else if (s->match_available) { 1961 /* If there was no match at the previous position, output a 1962 * single literal. If there was a match but the current match 1963 * is longer, truncate the previous match to a single literal. 1964 */ 1965 Tracevv((stderr,"%c", s->window[s->strstart - 1])); 1966 _tr_tally_lit(s, s->window[s->strstart - 1], bflush); 1967 if (bflush) { 1968 FLUSH_BLOCK_ONLY(s, 0); 1969 } 1970 s->strstart++; 1971 s->lookahead--; 1972 if (s->strm->avail_out == 0) return need_more; 1973 } else { 1974 /* There is no previous match to compare with, wait for 1975 * the next step to decide. 1976 */ 1977 s->match_available = 1; 1978 s->strstart++; 1979 s->lookahead--; 1980 } 1981 } 1982 Assert (flush != Z_NO_FLUSH, "no flush?"); 1983 if (s->match_available) { 1984 Tracevv((stderr,"%c", s->window[s->strstart - 1])); 1985 _tr_tally_lit(s, s->window[s->strstart - 1], bflush); 1986 s->match_available = 0; 1987 } 1988 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; 1989 if (flush == Z_FINISH) { 1990 FLUSH_BLOCK(s, 1); 1991 return finish_done; 1992 } 1993 if (s->sym_next) 1994 FLUSH_BLOCK(s, 0); 1995 return block_done; 1996 } 1997 #endif /* FASTEST */ 1998 1999 /* =========================================================================== 2000 * For Z_RLE, simply look for runs of bytes, generate matches only of distance 2001 * one. Do not maintain a hash table. (It will be regenerated if this run of 2002 * deflate switches away from Z_RLE.) 2003 */ 2004 local block_state deflate_rle(deflate_state *s, int flush) { 2005 int bflush; /* set if current block must be flushed */ 2006 uInt prev; /* byte at distance one to match */ 2007 Bytef *scan, *strend; /* scan goes up to strend for length of run */ 2008 2009 for (;;) { 2010 /* Make sure that we always have enough lookahead, except 2011 * at the end of the input file. We need MAX_MATCH bytes 2012 * for the longest run, plus one for the unrolled loop. 2013 */ 2014 if (s->lookahead <= MAX_MATCH) { 2015 fill_window(s); 2016 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) { 2017 return need_more; 2018 } 2019 if (s->lookahead == 0) break; /* flush the current block */ 2020 } 2021 2022 /* See how many times the previous byte repeats */ 2023 s->match_length = 0; 2024 if (s->lookahead >= MIN_MATCH && s->strstart > 0) { 2025 scan = s->window + s->strstart - 1; 2026 prev = *scan; 2027 if (prev == *++scan && prev == *++scan && prev == *++scan) { 2028 strend = s->window + s->strstart + MAX_MATCH; 2029 do { 2030 } while (prev == *++scan && prev == *++scan && 2031 prev == *++scan && prev == *++scan && 2032 prev == *++scan && prev == *++scan && 2033 prev == *++scan && prev == *++scan && 2034 scan < strend); 2035 s->match_length = MAX_MATCH - (uInt)(strend - scan); 2036 if (s->match_length > s->lookahead) 2037 s->match_length = s->lookahead; 2038 } 2039 Assert(scan <= s->window + (uInt)(s->window_size - 1), 2040 "wild scan"); 2041 } 2042 2043 /* Emit match if have run of MIN_MATCH or longer, else emit literal */ 2044 if (s->match_length >= MIN_MATCH) { 2045 check_match(s, s->strstart, s->strstart - 1, s->match_length); 2046 2047 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush); 2048 2049 s->lookahead -= s->match_length; 2050 s->strstart += s->match_length; 2051 s->match_length = 0; 2052 } else { 2053 /* No match, output a literal byte */ 2054 Tracevv((stderr,"%c", s->window[s->strstart])); 2055 _tr_tally_lit(s, s->window[s->strstart], bflush); 2056 s->lookahead--; 2057 s->strstart++; 2058 } 2059 if (bflush) FLUSH_BLOCK(s, 0); 2060 } 2061 s->insert = 0; 2062 if (flush == Z_FINISH) { 2063 FLUSH_BLOCK(s, 1); 2064 return finish_done; 2065 } 2066 if (s->sym_next) 2067 FLUSH_BLOCK(s, 0); 2068 return block_done; 2069 } 2070 2071 /* =========================================================================== 2072 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table. 2073 * (It will be regenerated if this run of deflate switches away from Huffman.) 2074 */ 2075 local block_state deflate_huff(deflate_state *s, int flush) { 2076 int bflush; /* set if current block must be flushed */ 2077 2078 for (;;) { 2079 /* Make sure that we have a literal to write. */ 2080 if (s->lookahead == 0) { 2081 fill_window(s); 2082 if (s->lookahead == 0) { 2083 if (flush == Z_NO_FLUSH) 2084 return need_more; 2085 break; /* flush the current block */ 2086 } 2087 } 2088 2089 /* Output a literal byte */ 2090 s->match_length = 0; 2091 Tracevv((stderr,"%c", s->window[s->strstart])); 2092 _tr_tally_lit(s, s->window[s->strstart], bflush); 2093 s->lookahead--; 2094 s->strstart++; 2095 if (bflush) FLUSH_BLOCK(s, 0); 2096 } 2097 s->insert = 0; 2098 if (flush == Z_FINISH) { 2099 FLUSH_BLOCK(s, 1); 2100 return finish_done; 2101 } 2102 if (s->sym_next) 2103 FLUSH_BLOCK(s, 0); 2104 return block_done; 2105 } 2106