1 /* deflate.c -- compress data using the deflation algorithm
\r
2 * Copyright (C) 1995-2005 Jean-loup Gailly.
\r
3 * For conditions of distribution and use, see copyright notice in zlib.h
\r
9 * The "deflation" process depends on being able to identify portions
\r
10 * of the input text which are identical to earlier input (within a
\r
11 * sliding window trailing behind the input currently being processed).
\r
13 * The most straightforward technique turns out to be the fastest for
\r
14 * most input files: try all possible matches and select the longest.
\r
15 * The key feature of this algorithm is that insertions into the string
\r
16 * dictionary are very simple and thus fast, and deletions are avoided
\r
17 * completely. Insertions are performed at each input character, whereas
\r
18 * string matches are performed only when the previous match ends. So it
\r
19 * is preferable to spend more time in matches to allow very fast string
\r
20 * insertions and avoid deletions. The matching algorithm for small
\r
21 * strings is inspired from that of Rabin & Karp. A brute force approach
\r
22 * is used to find longer strings when a small match has been found.
\r
23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
\r
24 * (by Leonid Broukhis).
\r
25 * A previous version of this file used a more sophisticated algorithm
\r
26 * (by Fiala and Greene) which is guaranteed to run in linear amortized
\r
27 * time, but has a larger average cost, uses more memory and is patented.
\r
28 * However the F&G algorithm may be faster for some highly redundant
\r
29 * files if the parameter max_chain_length (described below) is too large.
\r
33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
\r
34 * I found it in 'freeze' written by Leonid Broukhis.
\r
35 * Thanks to many people for bug reports and testing.
\r
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
\r
40 * Available in http://www.ietf.org/rfc/rfc1951.txt
\r
42 * A description of the Rabin and Karp algorithm is given in the book
\r
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
\r
45 * Fiala,E.R., and Greene,D.H.
\r
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
\r
52 #include "deflate.h"
\r
54 const char deflate_copyright[] =
\r
55 " deflate 1.2.3 Copyright 1995-2005 Jean-loup Gailly ";
\r
57 If you use the zlib library in a product, an acknowledgment is welcome
\r
58 in the documentation of your product. If for some reason you cannot
\r
59 include such an acknowledgment, I would appreciate that you keep this
\r
60 copyright string in the executable of your product.
\r
63 /* ===========================================================================
\r
64 * Function prototypes.
\r
67 need_more, /* block not completed, need more input or more output */
\r
68 block_done, /* block flush performed */
\r
69 finish_started, /* finish started, need only more output at next deflate */
\r
70 finish_done /* finish done, accept no more input or output */
\r
73 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
\r
74 /* Compression function. Returns the block state after the call. */
\r
76 local void fill_window OF((deflate_state *s));
\r
77 local block_state deflate_stored OF((deflate_state *s, int flush));
\r
78 local block_state deflate_fast OF((deflate_state *s, int flush));
\r
80 local block_state deflate_slow OF((deflate_state *s, int flush));
\r
82 local void lm_init OF((deflate_state *s));
\r
83 local void putShortMSB OF((deflate_state *s, uInt b));
\r
84 local void flush_pending OF((z_streamp strm));
\r
85 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
\r
88 void match_init OF((void)); /* asm code initialization */
\r
89 uInt longest_match OF((deflate_state *s, IPos cur_match));
\r
91 local uInt longest_match OF((deflate_state *s, IPos cur_match));
\r
94 local uInt longest_match_fast OF((deflate_state *s, IPos cur_match));
\r
97 local void check_match OF((deflate_state *s, IPos start, IPos match,
\r
101 /* ===========================================================================
\r
106 /* Tail of hash chains */
\r
109 # define TOO_FAR 4096
\r
111 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
\r
113 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
\r
114 /* Minimum amount of lookahead, except at the end of the input file.
\r
115 * See deflate.c for comments about the MIN_MATCH+1.
\r
118 /* Values for max_lazy_match, good_match and max_chain_length, depending on
\r
119 * the desired pack level (0..9). The values given below have been tuned to
\r
120 * exclude worst case performance for pathological files. Better values may be
\r
121 * found for specific files.
\r
123 typedef struct config_s {
\r
124 ush good_length; /* reduce lazy search above this match length */
\r
125 ush max_lazy; /* do not perform lazy search above this match length */
\r
126 ush nice_length; /* quit search above this match length */
\r
128 compress_func func;
\r
132 local const config configuration_table[2] = {
\r
133 /* good lazy nice chain */
\r
134 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
\r
135 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
\r
137 local const config configuration_table[10] = {
\r
138 /* good lazy nice chain */
\r
139 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
\r
140 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
\r
141 /* 2 */ {4, 5, 16, 8, deflate_fast},
\r
142 /* 3 */ {4, 6, 32, 32, deflate_fast},
\r
144 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
\r
145 /* 5 */ {8, 16, 32, 32, deflate_slow},
\r
146 /* 6 */ {8, 16, 128, 128, deflate_slow},
\r
147 /* 7 */ {8, 32, 128, 256, deflate_slow},
\r
148 /* 8 */ {32, 128, 258, 1024, deflate_slow},
\r
149 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
\r
152 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
\r
153 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
\r
158 /* result of memcmp for equal strings */
\r
160 #ifndef NO_DUMMY_DECL
\r
161 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
\r
164 /* ===========================================================================
\r
165 * Update a hash value with the given input byte
\r
166 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
\r
167 * input characters, so that a running hash key can be computed from the
\r
168 * previous key instead of complete recalculation each time.
\r
170 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
\r
173 /* ===========================================================================
\r
174 * Insert string str in the dictionary and set match_head to the previous head
\r
175 * of the hash chain (the most recent string with same hash key). Return
\r
176 * the previous length of the hash chain.
\r
177 * If this file is compiled with -DFASTEST, the compression level is forced
\r
178 * to 1, and no hash chains are maintained.
\r
179 * IN assertion: all calls to to INSERT_STRING are made with consecutive
\r
180 * input characters and the first MIN_MATCH bytes of str are valid
\r
181 * (except for the last MIN_MATCH-1 bytes of the input file).
\r
184 #define INSERT_STRING(s, str, match_head) \
\r
185 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
\r
186 match_head = s->head[s->ins_h], \
\r
187 s->head[s->ins_h] = (Pos)(str))
\r
189 #define INSERT_STRING(s, str, match_head) \
\r
190 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
\r
191 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
\r
192 s->head[s->ins_h] = (Pos)(str))
\r
195 /* ===========================================================================
\r
196 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
\r
197 * prev[] will be initialized on the fly.
\r
199 #define CLEAR_HASH(s) \
\r
200 s->head[s->hash_size-1] = NIL; \
\r
201 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
\r
203 /* ========================================================================= */
\r
204 int ZEXPORT deflateInit_(strm, level, version, stream_size)
\r
207 const char *version;
\r
210 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
\r
211 Z_DEFAULT_STRATEGY, version, stream_size);
\r
212 /* To do: ignore strm->next_in if we use it as window */
\r
215 /* ========================================================================= */
\r
216 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
\r
217 version, stream_size)
\r
224 const char *version;
\r
229 static const char my_version[] = ZLIB_VERSION;
\r
232 /* We overlay pending_buf and d_buf+l_buf. This works since the average
\r
233 * output size for (length,distance) codes is <= 24 bits.
\r
236 if (version == Z_NULL || version[0] != my_version[0] ||
\r
237 stream_size != sizeof(z_stream)) {
\r
238 return Z_VERSION_ERROR;
\r
240 if (strm == Z_NULL) return Z_STREAM_ERROR;
\r
242 strm->msg = Z_NULL;
\r
243 if (strm->zalloc == (alloc_func)0) {
\r
244 strm->zalloc = zcalloc;
\r
245 strm->opaque = (voidpf)0;
\r
247 if (strm->zfree == (free_func)0) strm->zfree = zcfree;
\r
250 if (level != 0) level = 1;
\r
252 if (level == Z_DEFAULT_COMPRESSION) level = 6;
\r
255 if (windowBits < 0) { /* suppress zlib wrapper */
\r
257 windowBits = -windowBits;
\r
260 else if (windowBits > 15) {
\r
261 wrap = 2; /* write gzip wrapper instead */
\r
265 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
\r
266 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
\r
267 strategy < 0 || strategy > Z_FIXED) {
\r
268 return Z_STREAM_ERROR;
\r
270 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
\r
271 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
\r
272 if (s == Z_NULL) return Z_MEM_ERROR;
\r
273 strm->state = (struct internal_state FAR *)s;
\r
277 s->gzhead = Z_NULL;
\r
278 s->w_bits = windowBits;
\r
279 s->w_size = 1 << s->w_bits;
\r
280 s->w_mask = s->w_size - 1;
\r
282 s->hash_bits = memLevel + 7;
\r
283 s->hash_size = 1 << s->hash_bits;
\r
284 s->hash_mask = s->hash_size - 1;
\r
285 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
\r
287 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
\r
288 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
\r
289 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
\r
291 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
\r
293 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
\r
294 s->pending_buf = (uchf *) overlay;
\r
295 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
\r
297 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
\r
298 s->pending_buf == Z_NULL) {
\r
299 s->status = FINISH_STATE;
\r
300 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
\r
302 return Z_MEM_ERROR;
\r
304 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
\r
305 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
\r
308 s->strategy = strategy;
\r
309 s->method = (Byte)method;
\r
311 return deflateReset(strm);
\r
314 /* ========================================================================= */
\r
315 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
\r
317 const Bytef *dictionary;
\r
321 uInt length = dictLength;
\r
323 IPos hash_head = 0;
\r
325 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
\r
326 strm->state->wrap == 2 ||
\r
327 (strm->state->wrap == 1 && strm->state->status != INIT_STATE))
\r
328 return Z_STREAM_ERROR;
\r
332 strm->adler = adler32(strm->adler, dictionary, dictLength);
\r
334 if (length < MIN_MATCH) return Z_OK;
\r
335 if (length > MAX_DIST(s)) {
\r
336 length = MAX_DIST(s);
\r
337 dictionary += dictLength - length; /* use the tail of the dictionary */
\r
339 zmemcpy(s->window, dictionary, length);
\r
340 s->strstart = length;
\r
341 s->block_start = (long)length;
\r
343 /* Insert all strings in the hash table (except for the last two bytes).
\r
344 * s->lookahead stays null, so s->ins_h will be recomputed at the next
\r
345 * call of fill_window.
\r
347 s->ins_h = s->window[0];
\r
348 UPDATE_HASH(s, s->ins_h, s->window[1]);
\r
349 for (n = 0; n <= length - MIN_MATCH; n++) {
\r
350 INSERT_STRING(s, n, hash_head);
\r
352 if (hash_head) hash_head = 0; /* to make compiler happy */
\r
356 /* ========================================================================= */
\r
357 int ZEXPORT deflateReset (strm)
\r
362 if (strm == Z_NULL || strm->state == Z_NULL ||
\r
363 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
\r
364 return Z_STREAM_ERROR;
\r
367 strm->total_in = strm->total_out = 0;
\r
368 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
\r
369 strm->data_type = Z_UNKNOWN;
\r
371 s = (deflate_state *)strm->state;
\r
373 s->pending_out = s->pending_buf;
\r
376 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
\r
378 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
\r
381 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
\r
383 adler32(0L, Z_NULL, 0);
\r
384 s->last_flush = Z_NO_FLUSH;
\r
392 /* ========================================================================= */
\r
393 int ZEXPORT deflateSetHeader (strm, head)
\r
397 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
\r
398 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
\r
399 strm->state->gzhead = head;
\r
403 /* ========================================================================= */
\r
404 int ZEXPORT deflatePrime (strm, bits, value)
\r
409 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
\r
410 strm->state->bi_valid = bits;
\r
411 strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));
\r
415 /* ========================================================================= */
\r
416 int ZEXPORT deflateParams(strm, level, strategy)
\r
422 compress_func func;
\r
425 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
\r
429 if (level != 0) level = 1;
\r
431 if (level == Z_DEFAULT_COMPRESSION) level = 6;
\r
433 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
\r
434 return Z_STREAM_ERROR;
\r
436 func = configuration_table[s->level].func;
\r
438 if (func != configuration_table[level].func && strm->total_in != 0) {
\r
439 /* Flush the last buffer: */
\r
440 err = deflate(strm, Z_PARTIAL_FLUSH);
\r
442 if (s->level != level) {
\r
444 s->max_lazy_match = configuration_table[level].max_lazy;
\r
445 s->good_match = configuration_table[level].good_length;
\r
446 s->nice_match = configuration_table[level].nice_length;
\r
447 s->max_chain_length = configuration_table[level].max_chain;
\r
449 s->strategy = strategy;
\r
453 /* ========================================================================= */
\r
454 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
\r
463 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
\r
465 s->good_match = good_length;
\r
466 s->max_lazy_match = max_lazy;
\r
467 s->nice_match = nice_length;
\r
468 s->max_chain_length = max_chain;
\r
472 /* =========================================================================
\r
473 * For the default windowBits of 15 and memLevel of 8, this function returns
\r
474 * a close to exact, as well as small, upper bound on the compressed size.
\r
475 * They are coded as constants here for a reason--if the #define's are
\r
476 * changed, then this function needs to be changed as well. The return
\r
477 * value for 15 and 8 only works for those exact settings.
\r
479 * For any setting other than those defaults for windowBits and memLevel,
\r
480 * the value returned is a conservative worst case for the maximum expansion
\r
481 * resulting from using fixed blocks instead of stored blocks, which deflate
\r
482 * can emit on compressed data for some combinations of the parameters.
\r
484 * This function could be more sophisticated to provide closer upper bounds
\r
485 * for every combination of windowBits and memLevel, as well as wrap.
\r
486 * But even the conservative upper bound of about 14% expansion does not
\r
487 * seem onerous for output buffer allocation.
\r
489 uLong ZEXPORT deflateBound(strm, sourceLen)
\r
496 /* conservative upper bound */
\r
497 destLen = sourceLen +
\r
498 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 11;
\r
500 /* if can't get parameters, return conservative bound */
\r
501 if (strm == Z_NULL || strm->state == Z_NULL)
\r
504 /* if not default parameters, return conservative bound */
\r
506 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
\r
509 /* default settings: return tight bound for that case */
\r
510 return compressBound(sourceLen);
\r
513 /* =========================================================================
\r
514 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
\r
515 * IN assertion: the stream state is correct and there is enough room in
\r
518 local void putShortMSB (s, b)
\r
522 put_byte(s, (Byte)(b >> 8));
\r
523 put_byte(s, (Byte)(b & 0xff));
\r
526 /* =========================================================================
\r
527 * Flush as much pending output as possible. All deflate() output goes
\r
528 * through this function so some applications may wish to modify it
\r
529 * to avoid allocating a large strm->next_out buffer and copying into it.
\r
530 * (See also read_buf()).
\r
532 local void flush_pending(strm)
\r
535 unsigned len = strm->state->pending;
\r
537 if (len > strm->avail_out) len = strm->avail_out;
\r
538 if (len == 0) return;
\r
540 zmemcpy(strm->next_out, strm->state->pending_out, len);
\r
541 strm->next_out += len;
\r
542 strm->state->pending_out += len;
\r
543 strm->total_out += len;
\r
544 strm->avail_out -= len;
\r
545 strm->state->pending -= len;
\r
546 if (strm->state->pending == 0) {
\r
547 strm->state->pending_out = strm->state->pending_buf;
\r
551 /* ========================================================================= */
\r
552 int ZEXPORT deflate (strm, flush)
\r
556 int old_flush; /* value of flush param for previous deflate call */
\r
559 if (strm == Z_NULL || strm->state == Z_NULL ||
\r
560 flush > Z_FINISH || flush < 0) {
\r
561 return Z_STREAM_ERROR;
\r
565 if (strm->next_out == Z_NULL ||
\r
566 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
\r
567 (s->status == FINISH_STATE && flush != Z_FINISH)) {
\r
568 ERR_RETURN(strm, Z_STREAM_ERROR);
\r
570 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
\r
572 s->strm = strm; /* just in case */
\r
573 old_flush = s->last_flush;
\r
574 s->last_flush = flush;
\r
576 /* Write the header */
\r
577 if (s->status == INIT_STATE) {
\r
579 if (s->wrap == 2) {
\r
580 strm->adler = crc32(0L, Z_NULL, 0);
\r
584 if (s->gzhead == NULL) {
\r
590 put_byte(s, s->level == 9 ? 2 :
\r
591 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
\r
593 put_byte(s, OS_CODE);
\r
594 s->status = BUSY_STATE;
\r
597 put_byte(s, (s->gzhead->text ? 1 : 0) +
\r
598 (s->gzhead->hcrc ? 2 : 0) +
\r
599 (s->gzhead->extra == Z_NULL ? 0 : 4) +
\r
600 (s->gzhead->name == Z_NULL ? 0 : 8) +
\r
601 (s->gzhead->comment == Z_NULL ? 0 : 16)
\r
603 put_byte(s, (Byte)(s->gzhead->time & 0xff));
\r
604 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
\r
605 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
\r
606 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
\r
607 put_byte(s, s->level == 9 ? 2 :
\r
608 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
\r
610 put_byte(s, s->gzhead->os & 0xff);
\r
611 if (s->gzhead->extra != NULL) {
\r
612 put_byte(s, s->gzhead->extra_len & 0xff);
\r
613 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
\r
615 if (s->gzhead->hcrc)
\r
616 strm->adler = crc32(strm->adler, s->pending_buf,
\r
619 s->status = EXTRA_STATE;
\r
625 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
\r
628 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
\r
630 else if (s->level < 6)
\r
632 else if (s->level == 6)
\r
636 header |= (level_flags << 6);
\r
637 if (s->strstart != 0) header |= PRESET_DICT;
\r
638 header += 31 - (header % 31);
\r
640 s->status = BUSY_STATE;
\r
641 putShortMSB(s, header);
\r
643 /* Save the adler32 of the preset dictionary: */
\r
644 if (s->strstart != 0) {
\r
645 putShortMSB(s, (uInt)(strm->adler >> 16));
\r
646 putShortMSB(s, (uInt)(strm->adler & 0xffff));
\r
648 strm->adler = adler32(0L, Z_NULL, 0);
\r
652 if (s->status == EXTRA_STATE) {
\r
653 if (s->gzhead->extra != NULL) {
\r
654 uInt beg = s->pending; /* start of bytes to update crc */
\r
656 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
\r
657 if (s->pending == s->pending_buf_size) {
\r
658 if (s->gzhead->hcrc && s->pending > beg)
\r
659 strm->adler = crc32(strm->adler, s->pending_buf + beg,
\r
661 flush_pending(strm);
\r
663 if (s->pending == s->pending_buf_size)
\r
666 put_byte(s, s->gzhead->extra[s->gzindex]);
\r
669 if (s->gzhead->hcrc && s->pending > beg)
\r
670 strm->adler = crc32(strm->adler, s->pending_buf + beg,
\r
672 if (s->gzindex == s->gzhead->extra_len) {
\r
674 s->status = NAME_STATE;
\r
678 s->status = NAME_STATE;
\r
680 if (s->status == NAME_STATE) {
\r
681 if (s->gzhead->name != NULL) {
\r
682 uInt beg = s->pending; /* start of bytes to update crc */
\r
686 if (s->pending == s->pending_buf_size) {
\r
687 if (s->gzhead->hcrc && s->pending > beg)
\r
688 strm->adler = crc32(strm->adler, s->pending_buf + beg,
\r
690 flush_pending(strm);
\r
692 if (s->pending == s->pending_buf_size) {
\r
697 val = s->gzhead->name[s->gzindex++];
\r
699 } while (val != 0);
\r
700 if (s->gzhead->hcrc && s->pending > beg)
\r
701 strm->adler = crc32(strm->adler, s->pending_buf + beg,
\r
705 s->status = COMMENT_STATE;
\r
709 s->status = COMMENT_STATE;
\r
711 if (s->status == COMMENT_STATE) {
\r
712 if (s->gzhead->comment != NULL) {
\r
713 uInt beg = s->pending; /* start of bytes to update crc */
\r
717 if (s->pending == s->pending_buf_size) {
\r
718 if (s->gzhead->hcrc && s->pending > beg)
\r
719 strm->adler = crc32(strm->adler, s->pending_buf + beg,
\r
721 flush_pending(strm);
\r
723 if (s->pending == s->pending_buf_size) {
\r
728 val = s->gzhead->comment[s->gzindex++];
\r
730 } while (val != 0);
\r
731 if (s->gzhead->hcrc && s->pending > beg)
\r
732 strm->adler = crc32(strm->adler, s->pending_buf + beg,
\r
735 s->status = HCRC_STATE;
\r
738 s->status = HCRC_STATE;
\r
740 if (s->status == HCRC_STATE) {
\r
741 if (s->gzhead->hcrc) {
\r
742 if (s->pending + 2 > s->pending_buf_size)
\r
743 flush_pending(strm);
\r
744 if (s->pending + 2 <= s->pending_buf_size) {
\r
745 put_byte(s, (Byte)(strm->adler & 0xff));
\r
746 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
\r
747 strm->adler = crc32(0L, Z_NULL, 0);
\r
748 s->status = BUSY_STATE;
\r
752 s->status = BUSY_STATE;
\r
756 /* Flush as much pending output as possible */
\r
757 if (s->pending != 0) {
\r
758 flush_pending(strm);
\r
759 if (strm->avail_out == 0) {
\r
760 /* Since avail_out is 0, deflate will be called again with
\r
761 * more output space, but possibly with both pending and
\r
762 * avail_in equal to zero. There won't be anything to do,
\r
763 * but this is not an error situation so make sure we
\r
764 * return OK instead of BUF_ERROR at next call of deflate:
\r
766 s->last_flush = -1;
\r
770 /* Make sure there is something to do and avoid duplicate consecutive
\r
771 * flushes. For repeated and useless calls with Z_FINISH, we keep
\r
772 * returning Z_STREAM_END instead of Z_BUF_ERROR.
\r
774 } else if (strm->avail_in == 0 && flush <= old_flush &&
\r
775 flush != Z_FINISH) {
\r
776 ERR_RETURN(strm, Z_BUF_ERROR);
\r
779 /* User must not provide more input after the first FINISH: */
\r
780 if (s->status == FINISH_STATE && strm->avail_in != 0) {
\r
781 ERR_RETURN(strm, Z_BUF_ERROR);
\r
784 /* Start a new block or continue the current one.
\r
786 if (strm->avail_in != 0 || s->lookahead != 0 ||
\r
787 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
\r
788 block_state bstate;
\r
790 bstate = (*(configuration_table[s->level].func))(s, flush);
\r
792 if (bstate == finish_started || bstate == finish_done) {
\r
793 s->status = FINISH_STATE;
\r
795 if (bstate == need_more || bstate == finish_started) {
\r
796 if (strm->avail_out == 0) {
\r
797 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
\r
800 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
\r
801 * of deflate should use the same flush parameter to make sure
\r
802 * that the flush is complete. So we don't have to output an
\r
803 * empty block here, this will be done at next call. This also
\r
804 * ensures that for a very small output buffer, we emit at most
\r
808 if (bstate == block_done) {
\r
809 if (flush == Z_PARTIAL_FLUSH) {
\r
811 } else { /* FULL_FLUSH or SYNC_FLUSH */
\r
812 _tr_stored_block(s, (char*)0, 0L, 0);
\r
813 /* For a full flush, this empty block will be recognized
\r
814 * as a special marker by inflate_sync().
\r
816 if (flush == Z_FULL_FLUSH) {
\r
817 CLEAR_HASH(s); /* forget history */
\r
820 flush_pending(strm);
\r
821 if (strm->avail_out == 0) {
\r
822 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
\r
827 Assert(strm->avail_out > 0, "bug2");
\r
829 if (flush != Z_FINISH) return Z_OK;
\r
830 if (s->wrap <= 0) return Z_STREAM_END;
\r
832 /* Write the trailer */
\r
834 if (s->wrap == 2) {
\r
835 put_byte(s, (Byte)(strm->adler & 0xff));
\r
836 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
\r
837 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
\r
838 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
\r
839 put_byte(s, (Byte)(strm->total_in & 0xff));
\r
840 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
\r
841 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
\r
842 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
\r
847 putShortMSB(s, (uInt)(strm->adler >> 16));
\r
848 putShortMSB(s, (uInt)(strm->adler & 0xffff));
\r
850 flush_pending(strm);
\r
851 /* If avail_out is zero, the application will call deflate again
\r
852 * to flush the rest.
\r
854 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
\r
855 return s->pending != 0 ? Z_OK : Z_STREAM_END;
\r
858 /* ========================================================================= */
\r
859 int ZEXPORT deflateEnd (strm)
\r
864 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
\r
866 status = strm->state->status;
\r
867 if (status != INIT_STATE &&
\r
868 status != EXTRA_STATE &&
\r
869 status != NAME_STATE &&
\r
870 status != COMMENT_STATE &&
\r
871 status != HCRC_STATE &&
\r
872 status != BUSY_STATE &&
\r
873 status != FINISH_STATE) {
\r
874 return Z_STREAM_ERROR;
\r
877 /* Deallocate in reverse order of allocations: */
\r
878 TRY_FREE(strm, strm->state->pending_buf);
\r
879 TRY_FREE(strm, strm->state->head);
\r
880 TRY_FREE(strm, strm->state->prev);
\r
881 TRY_FREE(strm, strm->state->window);
\r
883 ZFREE(strm, strm->state);
\r
884 strm->state = Z_NULL;
\r
886 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
\r
889 /* =========================================================================
\r
890 * Copy the source state to the destination state.
\r
891 * To simplify the source, this is not supported for 16-bit MSDOS (which
\r
892 * doesn't have enough memory anyway to duplicate compression states).
\r
894 int ZEXPORT deflateCopy (dest, source)
\r
899 return Z_STREAM_ERROR;
\r
906 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
\r
907 return Z_STREAM_ERROR;
\r
910 ss = source->state;
\r
912 zmemcpy(dest, source, sizeof(z_stream));
\r
914 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
\r
915 if (ds == Z_NULL) return Z_MEM_ERROR;
\r
916 dest->state = (struct internal_state FAR *) ds;
\r
917 zmemcpy(ds, ss, sizeof(deflate_state));
\r
920 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
\r
921 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
\r
922 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
\r
923 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
\r
924 ds->pending_buf = (uchf *) overlay;
\r
926 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
\r
927 ds->pending_buf == Z_NULL) {
\r
929 return Z_MEM_ERROR;
\r
931 /* following zmemcpy do not work for 16-bit MSDOS */
\r
932 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
\r
933 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
\r
934 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
\r
935 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
\r
937 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
\r
938 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
\r
939 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
\r
941 ds->l_desc.dyn_tree = ds->dyn_ltree;
\r
942 ds->d_desc.dyn_tree = ds->dyn_dtree;
\r
943 ds->bl_desc.dyn_tree = ds->bl_tree;
\r
946 #endif /* MAXSEG_64K */
\r
949 /* ===========================================================================
\r
950 * Read a new buffer from the current input stream, update the adler32
\r
951 * and total number of bytes read. All deflate() input goes through
\r
952 * this function so some applications may wish to modify it to avoid
\r
953 * allocating a large strm->next_in buffer and copying from it.
\r
954 * (See also flush_pending()).
\r
956 local int read_buf(strm, buf, size)
\r
961 unsigned len = strm->avail_in;
\r
963 if (len > size) len = size;
\r
964 if (len == 0) return 0;
\r
966 strm->avail_in -= len;
\r
968 if (strm->state->wrap == 1) {
\r
969 strm->adler = adler32(strm->adler, strm->next_in, len);
\r
972 else if (strm->state->wrap == 2) {
\r
973 strm->adler = crc32(strm->adler, strm->next_in, len);
\r
976 zmemcpy(buf, strm->next_in, len);
\r
977 strm->next_in += len;
\r
978 strm->total_in += len;
\r
983 /* ===========================================================================
\r
984 * Initialize the "longest match" routines for a new zlib stream
\r
986 local void lm_init (s)
\r
989 s->window_size = (ulg)2L*s->w_size;
\r
993 /* Set the default configuration parameters:
\r
995 s->max_lazy_match = configuration_table[s->level].max_lazy;
\r
996 s->good_match = configuration_table[s->level].good_length;
\r
997 s->nice_match = configuration_table[s->level].nice_length;
\r
998 s->max_chain_length = configuration_table[s->level].max_chain;
\r
1001 s->block_start = 0L;
\r
1003 s->match_length = s->prev_length = MIN_MATCH-1;
\r
1004 s->match_available = 0;
\r
1008 match_init(); /* initialize the asm code */
\r
1014 /* ===========================================================================
\r
1015 * Set match_start to the longest match starting at the given string and
\r
1016 * return its length. Matches shorter or equal to prev_length are discarded,
\r
1017 * in which case the result is equal to prev_length and match_start is
\r
1019 * IN assertions: cur_match is the head of the hash chain for the current
\r
1020 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
\r
1021 * OUT assertion: the match length is not greater than s->lookahead.
\r
1024 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
\r
1025 * match.S. The code will be functionally equivalent.
\r
1027 local uInt longest_match(s, cur_match)
\r
1029 IPos cur_match; /* current match */
\r
1031 unsigned chain_length = s->max_chain_length;/* max hash chain length */
\r
1032 register Bytef *scan = s->window + s->strstart; /* current string */
\r
1033 register Bytef *match; /* matched string */
\r
1034 register int len; /* length of current match */
\r
1035 int best_len = s->prev_length; /* best match length so far */
\r
1036 int nice_match = s->nice_match; /* stop if match long enough */
\r
1037 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
\r
1038 s->strstart - (IPos)MAX_DIST(s) : NIL;
\r
1039 /* Stop when cur_match becomes <= limit. To simplify the code,
\r
1040 * we prevent matches with the string of window index 0.
\r
1042 Posf *prev = s->prev;
\r
1043 uInt wmask = s->w_mask;
\r
1045 #ifdef UNALIGNED_OK
\r
1046 /* Compare two bytes at a time. Note: this is not always beneficial.
\r
1047 * Try with and without -DUNALIGNED_OK to check.
\r
1049 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
\r
1050 register ush scan_start = *(ushf*)scan;
\r
1051 register ush scan_end = *(ushf*)(scan+best_len-1);
\r
1053 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
\r
1054 register Byte scan_end1 = scan[best_len-1];
\r
1055 register Byte scan_end = scan[best_len];
\r
1058 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
\r
1059 * It is easy to get rid of this optimization if necessary.
\r
1061 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
\r
1063 /* Do not waste too much time if we already have a good match: */
\r
1064 if (s->prev_length >= s->good_match) {
\r
1065 chain_length >>= 2;
\r
1067 /* Do not look for matches beyond the end of the input. This is necessary
\r
1068 * to make deflate deterministic.
\r
1070 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
\r
1072 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
\r
1075 Assert(cur_match < s->strstart, "no future");
\r
1076 match = s->window + cur_match;
\r
1078 /* Skip to next match if the match length cannot increase
\r
1079 * or if the match length is less than 2. Note that the checks below
\r
1080 * for insufficient lookahead only occur occasionally for performance
\r
1081 * reasons. Therefore uninitialized memory will be accessed, and
\r
1082 * conditional jumps will be made that depend on those values.
\r
1083 * However the length of the match is limited to the lookahead, so
\r
1084 * the output of deflate is not affected by the uninitialized values.
\r
1086 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
\r
1087 /* This code assumes sizeof(unsigned short) == 2. Do not use
\r
1088 * UNALIGNED_OK if your compiler uses a different size.
\r
1090 if (*(ushf*)(match+best_len-1) != scan_end ||
\r
1091 *(ushf*)match != scan_start) continue;
\r
1093 /* It is not necessary to compare scan[2] and match[2] since they are
\r
1094 * always equal when the other bytes match, given that the hash keys
\r
1095 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
\r
1096 * strstart+3, +5, ... up to strstart+257. We check for insufficient
\r
1097 * lookahead only every 4th comparison; the 128th check will be made
\r
1098 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
\r
1099 * necessary to put more guard bytes at the end of the window, or
\r
1100 * to check more often for insufficient lookahead.
\r
1102 Assert(scan[2] == match[2], "scan[2]?");
\r
1105 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
\r
1106 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
\r
1107 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
\r
1108 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
\r
1110 /* The funny "do {}" generates better code on most compilers */
\r
1112 /* Here, scan <= window+strstart+257 */
\r
1113 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
\r
1114 if (*scan == *match) scan++;
\r
1116 len = (MAX_MATCH - 1) - (int)(strend-scan);
\r
1117 scan = strend - (MAX_MATCH-1);
\r
1119 #else /* UNALIGNED_OK */
\r
1121 if (match[best_len] != scan_end ||
\r
1122 match[best_len-1] != scan_end1 ||
\r
1123 *match != *scan ||
\r
1124 *++match != scan[1]) continue;
\r
1126 /* The check at best_len-1 can be removed because it will be made
\r
1127 * again later. (This heuristic is not always a win.)
\r
1128 * It is not necessary to compare scan[2] and match[2] since they
\r
1129 * are always equal when the other bytes match, given that
\r
1130 * the hash keys are equal and that HASH_BITS >= 8.
\r
1132 scan += 2, match++;
\r
1133 Assert(*scan == *match, "match[2]?");
\r
1135 /* We check for insufficient lookahead only every 8th comparison;
\r
1136 * the 256th check will be made at strstart+258.
\r
1139 } while (*++scan == *++match && *++scan == *++match &&
\r
1140 *++scan == *++match && *++scan == *++match &&
\r
1141 *++scan == *++match && *++scan == *++match &&
\r
1142 *++scan == *++match && *++scan == *++match &&
\r
1145 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
\r
1147 len = MAX_MATCH - (int)(strend - scan);
\r
1148 scan = strend - MAX_MATCH;
\r
1150 #endif /* UNALIGNED_OK */
\r
1152 if (len > best_len) {
\r
1153 s->match_start = cur_match;
\r
1155 if (len >= nice_match) break;
\r
1156 #ifdef UNALIGNED_OK
\r
1157 scan_end = *(ushf*)(scan+best_len-1);
\r
1159 scan_end1 = scan[best_len-1];
\r
1160 scan_end = scan[best_len];
\r
1163 } while ((cur_match = prev[cur_match & wmask]) > limit
\r
1164 && --chain_length != 0);
\r
1166 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
\r
1167 return s->lookahead;
\r
1170 #endif /* FASTEST */
\r
1172 /* ---------------------------------------------------------------------------
\r
1173 * Optimized version for level == 1 or strategy == Z_RLE only
\r
1175 local uInt longest_match_fast(s, cur_match)
\r
1177 IPos cur_match; /* current match */
\r
1179 register Bytef *scan = s->window + s->strstart; /* current string */
\r
1180 register Bytef *match; /* matched string */
\r
1181 register int len; /* length of current match */
\r
1182 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
\r
1184 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
\r
1185 * It is easy to get rid of this optimization if necessary.
\r
1187 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
\r
1189 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
\r
1191 Assert(cur_match < s->strstart, "no future");
\r
1193 match = s->window + cur_match;
\r
1195 /* Return failure if the match length is less than 2:
\r
1197 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
\r
1199 /* The check at best_len-1 can be removed because it will be made
\r
1200 * again later. (This heuristic is not always a win.)
\r
1201 * It is not necessary to compare scan[2] and match[2] since they
\r
1202 * are always equal when the other bytes match, given that
\r
1203 * the hash keys are equal and that HASH_BITS >= 8.
\r
1205 scan += 2, match += 2;
\r
1206 Assert(*scan == *match, "match[2]?");
\r
1208 /* We check for insufficient lookahead only every 8th comparison;
\r
1209 * the 256th check will be made at strstart+258.
\r
1212 } while (*++scan == *++match && *++scan == *++match &&
\r
1213 *++scan == *++match && *++scan == *++match &&
\r
1214 *++scan == *++match && *++scan == *++match &&
\r
1215 *++scan == *++match && *++scan == *++match &&
\r
1218 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
\r
1220 len = MAX_MATCH - (int)(strend - scan);
\r
1222 if (len < MIN_MATCH) return MIN_MATCH - 1;
\r
1224 s->match_start = cur_match;
\r
1225 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
\r
1229 /* ===========================================================================
\r
1230 * Check that the match at match_start is indeed a match.
\r
1232 local void check_match(s, start, match, length)
\r
1234 IPos start, match;
\r
1237 /* check that the match is indeed a match */
\r
1238 if (zmemcmp(s->window + match,
\r
1239 s->window + start, length) != EQUAL) {
\r
1240 fprintf(stderr, " start %u, match %u, length %d\n",
\r
1241 start, match, length);
\r
1243 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
\r
1244 } while (--length != 0);
\r
1245 z_error("invalid match");
\r
1247 if (z_verbose > 1) {
\r
1248 fprintf(stderr,"\\[%d,%d]", start-match, length);
\r
1249 do { putc(s->window[start++], stderr); } while (--length != 0);
\r
1253 # define check_match(s, start, match, length)
\r
1254 #endif /* DEBUG */
\r
1256 /* ===========================================================================
\r
1257 * Fill the window when the lookahead becomes insufficient.
\r
1258 * Updates strstart and lookahead.
\r
1260 * IN assertion: lookahead < MIN_LOOKAHEAD
\r
1261 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
\r
1262 * At least one byte has been read, or avail_in == 0; reads are
\r
1263 * performed for at least two bytes (required for the zip translate_eol
\r
1264 * option -- not supported here).
\r
1266 local void fill_window(s)
\r
1269 register unsigned n, m;
\r
1271 unsigned more; /* Amount of free space at the end of the window. */
\r
1272 uInt wsize = s->w_size;
\r
1275 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
\r
1277 /* Deal with !@#$% 64K limit: */
\r
1278 if (sizeof(int) <= 2) {
\r
1279 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
\r
1282 } else if (more == (unsigned)(-1)) {
\r
1283 /* Very unlikely, but possible on 16 bit machine if
\r
1284 * strstart == 0 && lookahead == 1 (input done a byte at time)
\r
1290 /* If the window is almost full and there is insufficient lookahead,
\r
1291 * move the upper half to the lower one to make room in the upper half.
\r
1293 if (s->strstart >= wsize+MAX_DIST(s)) {
\r
1295 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
\r
1296 s->match_start -= wsize;
\r
1297 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
\r
1298 s->block_start -= (long) wsize;
\r
1300 /* Slide the hash table (could be avoided with 32 bit values
\r
1301 at the expense of memory usage). We slide even when level == 0
\r
1302 to keep the hash table consistent if we switch back to level > 0
\r
1303 later. (Using level 0 permanently is not an optimal usage of
\r
1304 zlib, so we don't care about this pathological case.)
\r
1306 /* %%% avoid this when Z_RLE */
\r
1311 *p = (Pos)(m >= wsize ? m-wsize : NIL);
\r
1319 *p = (Pos)(m >= wsize ? m-wsize : NIL);
\r
1320 /* If n is not on any hash chain, prev[n] is garbage but
\r
1321 * its value will never be used.
\r
1327 if (s->strm->avail_in == 0) return;
\r
1329 /* If there was no sliding:
\r
1330 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
\r
1331 * more == window_size - lookahead - strstart
\r
1332 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
\r
1333 * => more >= window_size - 2*WSIZE + 2
\r
1334 * In the BIG_MEM or MMAP case (not yet supported),
\r
1335 * window_size == input_size + MIN_LOOKAHEAD &&
\r
1336 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
\r
1337 * Otherwise, window_size == 2*WSIZE so more >= 2.
\r
1338 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
\r
1340 Assert(more >= 2, "more < 2");
\r
1342 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
\r
1343 s->lookahead += n;
\r
1345 /* Initialize the hash value now that we have some input: */
\r
1346 if (s->lookahead >= MIN_MATCH) {
\r
1347 s->ins_h = s->window[s->strstart];
\r
1348 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
\r
1349 #if MIN_MATCH != 3
\r
1350 Call UPDATE_HASH() MIN_MATCH-3 more times
\r
1353 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
\r
1354 * but this is not important since only literal bytes will be emitted.
\r
1357 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
\r
1360 /* ===========================================================================
\r
1361 * Flush the current block, with given end-of-file flag.
\r
1362 * IN assertion: strstart is set to the end of the current match.
\r
1364 #define FLUSH_BLOCK_ONLY(s, eof) { \
\r
1365 _tr_flush_block(s, (s->block_start >= 0L ? \
\r
1366 (charf *)&s->window[(unsigned)s->block_start] : \
\r
1367 (charf *)Z_NULL), \
\r
1368 (ulg)((long)s->strstart - s->block_start), \
\r
1370 s->block_start = s->strstart; \
\r
1371 flush_pending(s->strm); \
\r
1372 Tracev((stderr,"[FLUSH]")); \
\r
1375 /* Same but force premature exit if necessary. */
\r
1376 #define FLUSH_BLOCK(s, eof) { \
\r
1377 FLUSH_BLOCK_ONLY(s, eof); \
\r
1378 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
\r
1381 /* ===========================================================================
\r
1382 * Copy without compression as much as possible from the input stream, return
\r
1383 * the current block state.
\r
1384 * This function does not insert new strings in the dictionary since
\r
1385 * uncompressible data is probably not useful. This function is used
\r
1386 * only for the level=0 compression option.
\r
1387 * NOTE: this function should be optimized to avoid extra copying from
\r
1388 * window to pending_buf.
\r
1390 local block_state deflate_stored(s, flush)
\r
1394 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
\r
1395 * to pending_buf_size, and each stored block has a 5 byte header:
\r
1397 ulg max_block_size = 0xffff;
\r
1400 if (max_block_size > s->pending_buf_size - 5) {
\r
1401 max_block_size = s->pending_buf_size - 5;
\r
1404 /* Copy as much as possible from input to output: */
\r
1406 /* Fill the window as much as possible: */
\r
1407 if (s->lookahead <= 1) {
\r
1409 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
\r
1410 s->block_start >= (long)s->w_size, "slide too late");
\r
1413 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
\r
1415 if (s->lookahead == 0) break; /* flush the current block */
\r
1417 Assert(s->block_start >= 0L, "block gone");
\r
1419 s->strstart += s->lookahead;
\r
1422 /* Emit a stored block if pending_buf will be full: */
\r
1423 max_start = s->block_start + max_block_size;
\r
1424 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
\r
1425 /* strstart == 0 is possible when wraparound on 16-bit machine */
\r
1426 s->lookahead = (uInt)(s->strstart - max_start);
\r
1427 s->strstart = (uInt)max_start;
\r
1428 FLUSH_BLOCK(s, 0);
\r
1430 /* Flush if we may have to slide, otherwise block_start may become
\r
1431 * negative and the data will be gone:
\r
1433 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
\r
1434 FLUSH_BLOCK(s, 0);
\r
1437 FLUSH_BLOCK(s, flush == Z_FINISH);
\r
1438 return flush == Z_FINISH ? finish_done : block_done;
\r
1441 /* ===========================================================================
\r
1442 * Compress as much as possible from the input stream, return the current
\r
1444 * This function does not perform lazy evaluation of matches and inserts
\r
1445 * new strings in the dictionary only for unmatched strings or for short
\r
1446 * matches. It is used only for the fast compression options.
\r
1448 local block_state deflate_fast(s, flush)
\r
1452 IPos hash_head = NIL; /* head of the hash chain */
\r
1453 int bflush; /* set if current block must be flushed */
\r
1456 /* Make sure that we always have enough lookahead, except
\r
1457 * at the end of the input file. We need MAX_MATCH bytes
\r
1458 * for the next match, plus MIN_MATCH bytes to insert the
\r
1459 * string following the next match.
\r
1461 if (s->lookahead < MIN_LOOKAHEAD) {
\r
1463 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
\r
1466 if (s->lookahead == 0) break; /* flush the current block */
\r
1469 /* Insert the string window[strstart .. strstart+2] in the
\r
1470 * dictionary, and set hash_head to the head of the hash chain:
\r
1472 if (s->lookahead >= MIN_MATCH) {
\r
1473 INSERT_STRING(s, s->strstart, hash_head);
\r
1476 /* Find the longest match, discarding those <= prev_length.
\r
1477 * At this point we have always match_length < MIN_MATCH
\r
1479 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
\r
1480 /* To simplify the code, we prevent matches with the string
\r
1481 * of window index 0 (in particular we have to avoid a match
\r
1482 * of the string with itself at the start of the input file).
\r
1485 if ((s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) ||
\r
1486 (s->strategy == Z_RLE && s->strstart - hash_head == 1)) {
\r
1487 s->match_length = longest_match_fast (s, hash_head);
\r
1490 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) {
\r
1491 s->match_length = longest_match (s, hash_head);
\r
1492 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) {
\r
1493 s->match_length = longest_match_fast (s, hash_head);
\r
1496 /* longest_match() or longest_match_fast() sets match_start */
\r
1498 if (s->match_length >= MIN_MATCH) {
\r
1499 check_match(s, s->strstart, s->match_start, s->match_length);
\r
1501 _tr_tally_dist(s, s->strstart - s->match_start,
\r
1502 s->match_length - MIN_MATCH, bflush);
\r
1504 s->lookahead -= s->match_length;
\r
1506 /* Insert new strings in the hash table only if the match length
\r
1507 * is not too large. This saves time but degrades compression.
\r
1510 if (s->match_length <= s->max_insert_length &&
\r
1511 s->lookahead >= MIN_MATCH) {
\r
1512 s->match_length--; /* string at strstart already in table */
\r
1515 INSERT_STRING(s, s->strstart, hash_head);
\r
1516 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
\r
1517 * always MIN_MATCH bytes ahead.
\r
1519 } while (--s->match_length != 0);
\r
1524 s->strstart += s->match_length;
\r
1525 s->match_length = 0;
\r
1526 s->ins_h = s->window[s->strstart];
\r
1527 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
\r
1528 #if MIN_MATCH != 3
\r
1529 Call UPDATE_HASH() MIN_MATCH-3 more times
\r
1531 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
\r
1532 * matter since it will be recomputed at next deflate call.
\r
1536 /* No match, output a literal byte */
\r
1537 Tracevv((stderr,"%c", s->window[s->strstart]));
\r
1538 _tr_tally_lit (s, s->window[s->strstart], bflush);
\r
1542 if (bflush) FLUSH_BLOCK(s, 0);
\r
1544 FLUSH_BLOCK(s, flush == Z_FINISH);
\r
1545 return flush == Z_FINISH ? finish_done : block_done;
\r
1549 /* ===========================================================================
\r
1550 * Same as above, but achieves better compression. We use a lazy
\r
1551 * evaluation for matches: a match is finally adopted only if there is
\r
1552 * no better match at the next window position.
\r
1554 local block_state deflate_slow(s, flush)
\r
1558 IPos hash_head = NIL; /* head of hash chain */
\r
1559 int bflush; /* set if current block must be flushed */
\r
1561 /* Process the input block. */
\r
1563 /* Make sure that we always have enough lookahead, except
\r
1564 * at the end of the input file. We need MAX_MATCH bytes
\r
1565 * for the next match, plus MIN_MATCH bytes to insert the
\r
1566 * string following the next match.
\r
1568 if (s->lookahead < MIN_LOOKAHEAD) {
\r
1570 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
\r
1573 if (s->lookahead == 0) break; /* flush the current block */
\r
1576 /* Insert the string window[strstart .. strstart+2] in the
\r
1577 * dictionary, and set hash_head to the head of the hash chain:
\r
1579 if (s->lookahead >= MIN_MATCH) {
\r
1580 INSERT_STRING(s, s->strstart, hash_head);
\r
1583 /* Find the longest match, discarding those <= prev_length.
\r
1585 s->prev_length = s->match_length, s->prev_match = s->match_start;
\r
1586 s->match_length = MIN_MATCH-1;
\r
1588 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
\r
1589 s->strstart - hash_head <= MAX_DIST(s)) {
\r
1590 /* To simplify the code, we prevent matches with the string
\r
1591 * of window index 0 (in particular we have to avoid a match
\r
1592 * of the string with itself at the start of the input file).
\r
1594 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) {
\r
1595 s->match_length = longest_match (s, hash_head);
\r
1596 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) {
\r
1597 s->match_length = longest_match_fast (s, hash_head);
\r
1599 /* longest_match() or longest_match_fast() sets match_start */
\r
1601 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
\r
1602 #if TOO_FAR <= 32767
\r
1603 || (s->match_length == MIN_MATCH &&
\r
1604 s->strstart - s->match_start > TOO_FAR)
\r
1608 /* If prev_match is also MIN_MATCH, match_start is garbage
\r
1609 * but we will ignore the current match anyway.
\r
1611 s->match_length = MIN_MATCH-1;
\r
1614 /* If there was a match at the previous step and the current
\r
1615 * match is not better, output the previous match:
\r
1617 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
\r
1618 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
\r
1619 /* Do not insert strings in hash table beyond this. */
\r
1621 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
\r
1623 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
\r
1624 s->prev_length - MIN_MATCH, bflush);
\r
1626 /* Insert in hash table all strings up to the end of the match.
\r
1627 * strstart-1 and strstart are already inserted. If there is not
\r
1628 * enough lookahead, the last two strings are not inserted in
\r
1631 s->lookahead -= s->prev_length-1;
\r
1632 s->prev_length -= 2;
\r
1634 if (++s->strstart <= max_insert) {
\r
1635 INSERT_STRING(s, s->strstart, hash_head);
\r
1637 } while (--s->prev_length != 0);
\r
1638 s->match_available = 0;
\r
1639 s->match_length = MIN_MATCH-1;
\r
1642 if (bflush) FLUSH_BLOCK(s, 0);
\r
1644 } else if (s->match_available) {
\r
1645 /* If there was no match at the previous position, output a
\r
1646 * single literal. If there was a match but the current match
\r
1647 * is longer, truncate the previous match to a single literal.
\r
1649 Tracevv((stderr,"%c", s->window[s->strstart-1]));
\r
1650 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
\r
1652 FLUSH_BLOCK_ONLY(s, 0);
\r
1656 if (s->strm->avail_out == 0) return need_more;
\r
1658 /* There is no previous match to compare with, wait for
\r
1659 * the next step to decide.
\r
1661 s->match_available = 1;
\r
1666 Assert (flush != Z_NO_FLUSH, "no flush?");
\r
1667 if (s->match_available) {
\r
1668 Tracevv((stderr,"%c", s->window[s->strstart-1]));
\r
1669 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
\r
1670 s->match_available = 0;
\r
1672 FLUSH_BLOCK(s, flush == Z_FINISH);
\r
1673 return flush == Z_FINISH ? finish_done : block_done;
\r
1675 #endif /* FASTEST */
\r
1678 /* ===========================================================================
\r
1679 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
\r
1680 * one. Do not maintain a hash table. (It will be regenerated if this run of
\r
1681 * deflate switches away from Z_RLE.)
\r
1683 local block_state deflate_rle(s, flush)
\r
1687 int bflush; /* set if current block must be flushed */
\r
1688 uInt run; /* length of run */
\r
1689 uInt max; /* maximum length of run */
\r
1690 uInt prev; /* byte at distance one to match */
\r
1691 Bytef *scan; /* scan for end of run */
\r
1694 /* Make sure that we always have enough lookahead, except
\r
1695 * at the end of the input file. We need MAX_MATCH bytes
\r
1696 * for the longest encodable run.
\r
1698 if (s->lookahead < MAX_MATCH) {
\r
1700 if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {
\r
1703 if (s->lookahead == 0) break; /* flush the current block */
\r
1706 /* See how many times the previous byte repeats */
\r
1708 if (s->strstart > 0) { /* if there is a previous byte, that is */
\r
1709 max = s->lookahead < MAX_MATCH ? s->lookahead : MAX_MATCH;
\r
1710 scan = s->window + s->strstart - 1;
\r
1713 if (*scan++ != prev)
\r
1715 } while (++run < max);
\r
1718 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
\r
1719 if (run >= MIN_MATCH) {
\r
1720 check_match(s, s->strstart, s->strstart - 1, run);
\r
1721 _tr_tally_dist(s, 1, run - MIN_MATCH, bflush);
\r
1722 s->lookahead -= run;
\r
1723 s->strstart += run;
\r
1725 /* No match, output a literal byte */
\r
1726 Tracevv((stderr,"%c", s->window[s->strstart]));
\r
1727 _tr_tally_lit (s, s->window[s->strstart], bflush);
\r
1731 if (bflush) FLUSH_BLOCK(s, 0);
\r
1733 FLUSH_BLOCK(s, flush == Z_FINISH);
\r
1734 return flush == Z_FINISH ? finish_done : block_done;
\r