1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2013 Jean-loup Gailly and Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
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).
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.
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.
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40 * Available in http://tools.ietf.org/html/rfc1951
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
54 const char deflate_copyright[] =
55 " deflate 1.2.8.1 Copyright 1995-2013 Jean-loup Gailly and Mark Adler ";
57 If you use the zlib library in a product, an acknowledgment is welcome
58 in the documentation of your product. If for some reason you cannot
59 include such an acknowledgment, I would appreciate that you keep this
60 copyright string in the executable of your product.
63 /* ===========================================================================
64 * Function prototypes.
67 need_more, /* block not completed, need more input or more output */
68 block_done, /* block flush performed */
69 finish_started, /* finish started, need only more output at next deflate */
70 finish_done /* finish done, accept no more input or output */
73 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
74 /* Compression function. Returns the block state after the call. */
76 local int deflateStateCheck OF((z_streamp strm));
77 local void fill_window OF((deflate_state *s));
78 local block_state deflate_stored OF((deflate_state *s, int flush));
79 local block_state deflate_fast OF((deflate_state *s, int flush));
81 local block_state deflate_slow OF((deflate_state *s, int flush));
83 local block_state deflate_rle OF((deflate_state *s, int flush));
84 local block_state deflate_huff OF((deflate_state *s, int flush));
85 local void lm_init OF((deflate_state *s));
86 local void putShortMSB OF((deflate_state *s, uInt b));
87 local void flush_pending OF((z_streamp strm));
88 local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
90 void match_init OF((void)); /* asm code initialization */
91 uInt longest_match OF((deflate_state *s, IPos cur_match));
93 local uInt longest_match OF((deflate_state *s, IPos cur_match));
97 local void check_match OF((deflate_state *s, IPos start, IPos match,
101 /* ===========================================================================
106 /* Tail of hash chains */
109 # define TOO_FAR 4096
111 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
113 /* Values for max_lazy_match, good_match and max_chain_length, depending on
114 * the desired pack level (0..9). The values given below have been tuned to
115 * exclude worst case performance for pathological files. Better values may be
116 * found for specific files.
118 typedef struct config_s {
119 ush good_length; /* reduce lazy search above this match length */
120 ush max_lazy; /* do not perform lazy search above this match length */
121 ush nice_length; /* quit search above this match length */
127 local const config configuration_table[2] = {
128 /* good lazy nice chain */
129 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
130 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
132 local const config configuration_table[10] = {
133 /* good lazy nice chain */
134 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
135 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
136 /* 2 */ {4, 5, 16, 8, deflate_fast},
137 /* 3 */ {4, 6, 32, 32, deflate_fast},
139 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
140 /* 5 */ {8, 16, 32, 32, deflate_slow},
141 /* 6 */ {8, 16, 128, 128, deflate_slow},
142 /* 7 */ {8, 32, 128, 256, deflate_slow},
143 /* 8 */ {32, 128, 258, 1024, deflate_slow},
144 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
147 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
148 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
152 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
153 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
155 /* ===========================================================================
156 * Update a hash value with the given input byte
157 * IN assertion: all calls to UPDATE_HASH are made with consecutive input
158 * characters, so that a running hash key can be computed from the previous
159 * key instead of complete recalculation each time.
161 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
164 /* ===========================================================================
165 * Insert string str in the dictionary and set match_head to the previous head
166 * of the hash chain (the most recent string with same hash key). Return
167 * the previous length of the hash chain.
168 * If this file is compiled with -DFASTEST, the compression level is forced
169 * to 1, and no hash chains are maintained.
170 * IN assertion: all calls to INSERT_STRING are made with consecutive input
171 * characters and the first MIN_MATCH bytes of str are valid (except for
172 * the last MIN_MATCH-1 bytes of the input file).
175 #define INSERT_STRING(s, str, match_head) \
176 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
177 match_head = s->head[s->ins_h], \
178 s->head[s->ins_h] = (Pos)(str))
180 #define INSERT_STRING(s, str, match_head) \
181 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
182 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
183 s->head[s->ins_h] = (Pos)(str))
186 /* ===========================================================================
187 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
188 * prev[] will be initialized on the fly.
190 #define CLEAR_HASH(s) \
191 s->head[s->hash_size-1] = NIL; \
192 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
194 /* ========================================================================= */
195 int ZEXPORT deflateInit_(strm, level, version, stream_size)
201 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
202 Z_DEFAULT_STRATEGY, version, stream_size);
203 /* To do: ignore strm->next_in if we use it as window */
206 /* ========================================================================= */
207 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
208 version, stream_size)
220 static const char my_version[] = ZLIB_VERSION;
223 /* We overlay pending_buf and d_buf+l_buf. This works since the average
224 * output size for (length,distance) codes is <= 24 bits.
227 if (version == Z_NULL || version[0] != my_version[0] ||
228 stream_size != sizeof(z_stream)) {
229 return Z_VERSION_ERROR;
231 if (strm == Z_NULL) return Z_STREAM_ERROR;
234 if (strm->zalloc == (alloc_func)0) {
236 return Z_STREAM_ERROR;
238 strm->zalloc = zcalloc;
239 strm->opaque = (voidpf)0;
242 if (strm->zfree == (free_func)0)
244 return Z_STREAM_ERROR;
246 strm->zfree = zcfree;
250 if (level != 0) level = 1;
252 if (level == Z_DEFAULT_COMPRESSION) level = 6;
255 if (windowBits < 0) { /* suppress zlib wrapper */
257 windowBits = -windowBits;
260 else if (windowBits > 15) {
261 wrap = 2; /* write gzip wrapper instead */
265 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
266 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
267 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
268 return Z_STREAM_ERROR;
270 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
271 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
272 if (s == Z_NULL) return Z_MEM_ERROR;
273 strm->state = (struct internal_state FAR *)s;
275 s->status = INIT_STATE; /* to pass state test in deflateReset() */
279 s->w_bits = (uInt)windowBits;
280 s->w_size = 1 << s->w_bits;
281 s->w_mask = s->w_size - 1;
283 s->hash_bits = (uInt)memLevel + 7;
284 s->hash_size = 1 << s->hash_bits;
285 s->hash_mask = s->hash_size - 1;
286 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
288 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
289 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
290 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
292 s->high_water = 0; /* nothing written to s->window yet */
294 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
296 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
297 s->pending_buf = (uchf *) overlay;
298 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
300 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
301 s->pending_buf == Z_NULL) {
302 s->status = FINISH_STATE;
303 strm->msg = ERR_MSG(Z_MEM_ERROR);
307 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
308 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
311 s->strategy = strategy;
312 s->method = (Byte)method;
314 return deflateReset(strm);
317 /* =========================================================================
318 * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
320 local int deflateStateCheck (strm)
324 if (strm == Z_NULL ||
325 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
328 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
330 s->status != GZIP_STATE &&
332 s->status != EXTRA_STATE &&
333 s->status != NAME_STATE &&
334 s->status != COMMENT_STATE &&
335 s->status != HCRC_STATE &&
336 s->status != BUSY_STATE &&
337 s->status != FINISH_STATE))
342 /* ========================================================================= */
343 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
345 const Bytef *dictionary;
352 z_const unsigned char *next;
354 if (deflateStateCheck(strm) || dictionary == Z_NULL)
355 return Z_STREAM_ERROR;
358 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
359 return Z_STREAM_ERROR;
361 /* when using zlib wrappers, compute Adler-32 for provided dictionary */
363 strm->adler = adler32(strm->adler, dictionary, dictLength);
364 s->wrap = 0; /* avoid computing Adler-32 in read_buf */
366 /* if dictionary would fill window, just replace the history */
367 if (dictLength >= s->w_size) {
368 if (wrap == 0) { /* already empty otherwise */
374 dictionary += dictLength - s->w_size; /* use the tail */
375 dictLength = s->w_size;
378 /* insert dictionary into window and hash */
379 avail = strm->avail_in;
380 next = strm->next_in;
381 strm->avail_in = dictLength;
382 strm->next_in = (z_const Bytef *)dictionary;
384 while (s->lookahead >= MIN_MATCH) {
386 n = s->lookahead - (MIN_MATCH-1);
388 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
390 s->prev[str & s->w_mask] = s->head[s->ins_h];
392 s->head[s->ins_h] = (Pos)str;
396 s->lookahead = MIN_MATCH-1;
399 s->strstart += s->lookahead;
400 s->block_start = (long)s->strstart;
401 s->insert = s->lookahead;
403 s->match_length = s->prev_length = MIN_MATCH-1;
404 s->match_available = 0;
405 strm->next_in = next;
406 strm->avail_in = avail;
411 /* ========================================================================= */
412 int ZEXPORT deflateResetKeep (strm)
417 if (deflateStateCheck(strm)) {
418 return Z_STREAM_ERROR;
421 strm->total_in = strm->total_out = 0;
422 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
423 strm->data_type = Z_UNKNOWN;
425 s = (deflate_state *)strm->state;
427 s->pending_out = s->pending_buf;
430 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
434 s->wrap == 2 ? GZIP_STATE :
436 s->wrap ? INIT_STATE : BUSY_STATE;
439 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
441 adler32(0L, Z_NULL, 0);
442 s->last_flush = Z_NO_FLUSH;
449 /* ========================================================================= */
450 int ZEXPORT deflateReset (strm)
455 ret = deflateResetKeep(strm);
457 lm_init(strm->state);
461 /* ========================================================================= */
462 int ZEXPORT deflateSetHeader (strm, head)
466 if (deflateStateCheck(strm) || strm->state->wrap != 2)
467 return Z_STREAM_ERROR;
468 strm->state->gzhead = head;
472 /* ========================================================================= */
473 int ZEXPORT deflatePending (strm, pending, bits)
478 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
479 if (pending != Z_NULL)
480 *pending = strm->state->pending;
482 *bits = strm->state->bi_valid;
486 /* ========================================================================= */
487 int ZEXPORT deflatePrime (strm, bits, value)
495 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
497 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
500 put = Buf_size - s->bi_valid;
503 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
512 /* ========================================================================= */
513 int ZEXPORT deflateParams(strm, level, strategy)
522 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
526 if (level != 0) level = 1;
528 if (level == Z_DEFAULT_COMPRESSION) level = 6;
530 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
531 return Z_STREAM_ERROR;
533 func = configuration_table[s->level].func;
535 if ((strategy != s->strategy || func != configuration_table[level].func)) {
536 /* Flush the last buffer: */
537 err = deflate(strm, Z_BLOCK);
538 if (err == Z_BUF_ERROR && s->pending == 0)
541 if (s->level != level) {
543 s->max_lazy_match = configuration_table[level].max_lazy;
544 s->good_match = configuration_table[level].good_length;
545 s->nice_match = configuration_table[level].nice_length;
546 s->max_chain_length = configuration_table[level].max_chain;
548 s->strategy = strategy;
552 /* ========================================================================= */
553 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
562 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
564 s->good_match = (uInt)good_length;
565 s->max_lazy_match = (uInt)max_lazy;
566 s->nice_match = nice_length;
567 s->max_chain_length = (uInt)max_chain;
571 /* =========================================================================
572 * For the default windowBits of 15 and memLevel of 8, this function returns
573 * a close to exact, as well as small, upper bound on the compressed size.
574 * They are coded as constants here for a reason--if the #define's are
575 * changed, then this function needs to be changed as well. The return
576 * value for 15 and 8 only works for those exact settings.
578 * For any setting other than those defaults for windowBits and memLevel,
579 * the value returned is a conservative worst case for the maximum expansion
580 * resulting from using fixed blocks instead of stored blocks, which deflate
581 * can emit on compressed data for some combinations of the parameters.
583 * This function could be more sophisticated to provide closer upper bounds for
584 * every combination of windowBits and memLevel. But even the conservative
585 * upper bound of about 14% expansion does not seem onerous for output buffer
588 uLong ZEXPORT deflateBound(strm, sourceLen)
593 uLong complen, wraplen;
595 /* conservative upper bound for compressed data */
596 complen = sourceLen +
597 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
599 /* if can't get parameters, return conservative bound plus zlib wrapper */
600 if (deflateStateCheck(strm))
603 /* compute wrapper length */
606 case 0: /* raw deflate */
609 case 1: /* zlib wrapper */
610 wraplen = 6 + (s->strstart ? 4 : 0);
613 case 2: /* gzip wrapper */
615 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
617 if (s->gzhead->extra != Z_NULL)
618 wraplen += 2 + s->gzhead->extra_len;
619 str = s->gzhead->name;
624 str = s->gzhead->comment;
634 default: /* for compiler happiness */
638 /* if not default parameters, return conservative bound */
639 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
640 return complen + wraplen;
642 /* default settings: return tight bound for that case */
643 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
644 (sourceLen >> 25) + 13 - 6 + wraplen;
647 /* =========================================================================
648 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
649 * IN assertion: the stream state is correct and there is enough room in
652 local void putShortMSB (s, b)
656 put_byte(s, (Byte)(b >> 8));
657 put_byte(s, (Byte)(b & 0xff));
660 /* =========================================================================
661 * Flush as much pending output as possible. All deflate() output goes
662 * through this function so some applications may wish to modify it
663 * to avoid allocating a large strm->next_out buffer and copying into it.
664 * (See also read_buf()).
666 local void flush_pending(strm)
670 deflate_state *s = strm->state;
674 if (len > strm->avail_out) len = strm->avail_out;
675 if (len == 0) return;
677 zmemcpy(strm->next_out, s->pending_out, len);
678 strm->next_out += len;
679 s->pending_out += len;
680 strm->total_out += len;
681 strm->avail_out -= len;
683 if (s->pending == 0) {
684 s->pending_out = s->pending_buf;
688 /* ===========================================================================
689 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
691 #define HCRC_UPDATE(beg) \
693 if (s->gzhead->hcrc && s->pending > (beg)) \
694 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
695 s->pending - (beg)); \
698 /* ========================================================================= */
699 int ZEXPORT deflate (strm, flush)
703 int old_flush; /* value of flush param for previous deflate call */
706 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
707 return Z_STREAM_ERROR;
711 if (strm->next_out == Z_NULL ||
712 (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
713 (s->status == FINISH_STATE && flush != Z_FINISH)) {
714 ERR_RETURN(strm, Z_STREAM_ERROR);
716 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
718 old_flush = s->last_flush;
719 s->last_flush = flush;
721 /* Flush as much pending output as possible */
722 if (s->pending != 0) {
724 if (strm->avail_out == 0) {
725 /* Since avail_out is 0, deflate will be called again with
726 * more output space, but possibly with both pending and
727 * avail_in equal to zero. There won't be anything to do,
728 * but this is not an error situation so make sure we
729 * return OK instead of BUF_ERROR at next call of deflate:
735 /* Make sure there is something to do and avoid duplicate consecutive
736 * flushes. For repeated and useless calls with Z_FINISH, we keep
737 * returning Z_STREAM_END instead of Z_BUF_ERROR.
739 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
741 ERR_RETURN(strm, Z_BUF_ERROR);
744 /* User must not provide more input after the first FINISH: */
745 if (s->status == FINISH_STATE && strm->avail_in != 0) {
746 ERR_RETURN(strm, Z_BUF_ERROR);
749 /* Write the header */
750 if (s->status == INIT_STATE) {
752 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
755 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
757 else if (s->level < 6)
759 else if (s->level == 6)
763 header |= (level_flags << 6);
764 if (s->strstart != 0) header |= PRESET_DICT;
765 header += 31 - (header % 31);
767 putShortMSB(s, header);
769 /* Save the adler32 of the preset dictionary: */
770 if (s->strstart != 0) {
771 putShortMSB(s, (uInt)(strm->adler >> 16));
772 putShortMSB(s, (uInt)(strm->adler & 0xffff));
774 strm->adler = adler32(0L, Z_NULL, 0);
775 s->status = BUSY_STATE;
777 /* Compression must start with an empty pending buffer */
779 if (s->pending != 0) {
785 if (s->status == GZIP_STATE) {
787 strm->adler = crc32(0L, Z_NULL, 0);
791 if (s->gzhead == Z_NULL) {
797 put_byte(s, s->level == 9 ? 2 :
798 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
800 put_byte(s, OS_CODE);
801 s->status = BUSY_STATE;
803 /* Compression must start with an empty pending buffer */
805 if (s->pending != 0) {
811 put_byte(s, (s->gzhead->text ? 1 : 0) +
812 (s->gzhead->hcrc ? 2 : 0) +
813 (s->gzhead->extra == Z_NULL ? 0 : 4) +
814 (s->gzhead->name == Z_NULL ? 0 : 8) +
815 (s->gzhead->comment == Z_NULL ? 0 : 16)
817 put_byte(s, (Byte)(s->gzhead->time & 0xff));
818 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
819 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
820 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
821 put_byte(s, s->level == 9 ? 2 :
822 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
824 put_byte(s, s->gzhead->os & 0xff);
825 if (s->gzhead->extra != Z_NULL) {
826 put_byte(s, s->gzhead->extra_len & 0xff);
827 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
830 strm->adler = crc32(strm->adler, s->pending_buf,
833 s->status = EXTRA_STATE;
836 if (s->status == EXTRA_STATE) {
837 if (s->gzhead->extra != Z_NULL) {
838 ulg beg = s->pending; /* start of bytes to update crc */
839 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
840 while (s->pending + left > s->pending_buf_size) {
841 uInt copy = s->pending_buf_size - s->pending;
842 zmemcpy(s->pending_buf + s->pending,
843 s->gzhead->extra + s->gzindex, copy);
844 s->pending = s->pending_buf_size;
848 if (s->pending != 0) {
855 zmemcpy(s->pending_buf + s->pending,
856 s->gzhead->extra + s->gzindex, left);
861 s->status = NAME_STATE;
863 if (s->status == NAME_STATE) {
864 if (s->gzhead->name != Z_NULL) {
865 ulg beg = s->pending; /* start of bytes to update crc */
868 if (s->pending == s->pending_buf_size) {
871 if (s->pending != 0) {
877 val = s->gzhead->name[s->gzindex++];
883 s->status = COMMENT_STATE;
885 if (s->status == COMMENT_STATE) {
886 if (s->gzhead->comment != Z_NULL) {
887 ulg beg = s->pending; /* start of bytes to update crc */
890 if (s->pending == s->pending_buf_size) {
893 if (s->pending != 0) {
899 val = s->gzhead->comment[s->gzindex++];
904 s->status = HCRC_STATE;
906 if (s->status == HCRC_STATE) {
907 if (s->gzhead->hcrc) {
908 if (s->pending + 2 > s->pending_buf_size) {
910 if (s->pending != 0) {
915 put_byte(s, (Byte)(strm->adler & 0xff));
916 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
917 strm->adler = crc32(0L, Z_NULL, 0);
919 s->status = BUSY_STATE;
921 /* Compression must start with an empty pending buffer */
923 if (s->pending != 0) {
930 /* Start a new block or continue the current one.
932 if (strm->avail_in != 0 || s->lookahead != 0 ||
933 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
936 bstate = s->level == 0 ? deflate_stored(s, flush) :
937 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
938 s->strategy == Z_RLE ? deflate_rle(s, flush) :
939 (*(configuration_table[s->level].func))(s, flush);
941 if (bstate == finish_started || bstate == finish_done) {
942 s->status = FINISH_STATE;
944 if (bstate == need_more || bstate == finish_started) {
945 if (strm->avail_out == 0) {
946 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
949 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
950 * of deflate should use the same flush parameter to make sure
951 * that the flush is complete. So we don't have to output an
952 * empty block here, this will be done at next call. This also
953 * ensures that for a very small output buffer, we emit at most
957 if (bstate == block_done) {
958 if (flush == Z_PARTIAL_FLUSH) {
960 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
961 _tr_stored_block(s, (char*)0, 0L, 0);
962 /* For a full flush, this empty block will be recognized
963 * as a special marker by inflate_sync().
965 if (flush == Z_FULL_FLUSH) {
966 CLEAR_HASH(s); /* forget history */
967 if (s->lookahead == 0) {
975 if (strm->avail_out == 0) {
976 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
981 Assert(strm->avail_out > 0, "bug2");
983 if (flush != Z_FINISH) return Z_OK;
984 if (s->wrap <= 0) return Z_STREAM_END;
986 /* Write the trailer */
989 put_byte(s, (Byte)(strm->adler & 0xff));
990 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
991 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
992 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
993 put_byte(s, (Byte)(strm->total_in & 0xff));
994 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
995 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
996 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1001 putShortMSB(s, (uInt)(strm->adler >> 16));
1002 putShortMSB(s, (uInt)(strm->adler & 0xffff));
1004 flush_pending(strm);
1005 /* If avail_out is zero, the application will call deflate again
1006 * to flush the rest.
1008 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1009 return s->pending != 0 ? Z_OK : Z_STREAM_END;
1012 /* ========================================================================= */
1013 int ZEXPORT deflateEnd (strm)
1018 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1020 status = strm->state->status;
1022 /* Deallocate in reverse order of allocations: */
1023 TRY_FREE(strm, strm->state->pending_buf);
1024 TRY_FREE(strm, strm->state->head);
1025 TRY_FREE(strm, strm->state->prev);
1026 TRY_FREE(strm, strm->state->window);
1028 ZFREE(strm, strm->state);
1029 strm->state = Z_NULL;
1031 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1034 /* =========================================================================
1035 * Copy the source state to the destination state.
1036 * To simplify the source, this is not supported for 16-bit MSDOS (which
1037 * doesn't have enough memory anyway to duplicate compression states).
1039 int ZEXPORT deflateCopy (dest, source)
1044 return Z_STREAM_ERROR;
1051 if (deflateStateCheck(source) || dest == Z_NULL) {
1052 return Z_STREAM_ERROR;
1057 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1059 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1060 if (ds == Z_NULL) return Z_MEM_ERROR;
1061 dest->state = (struct internal_state FAR *) ds;
1062 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1065 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1066 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
1067 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
1068 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1069 ds->pending_buf = (uchf *) overlay;
1071 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1072 ds->pending_buf == Z_NULL) {
1076 /* following zmemcpy do not work for 16-bit MSDOS */
1077 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1078 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1079 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1080 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1082 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1083 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1084 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1086 ds->l_desc.dyn_tree = ds->dyn_ltree;
1087 ds->d_desc.dyn_tree = ds->dyn_dtree;
1088 ds->bl_desc.dyn_tree = ds->bl_tree;
1091 #endif /* MAXSEG_64K */
1094 /* ===========================================================================
1095 * Read a new buffer from the current input stream, update the adler32
1096 * and total number of bytes read. All deflate() input goes through
1097 * this function so some applications may wish to modify it to avoid
1098 * allocating a large strm->next_in buffer and copying from it.
1099 * (See also flush_pending()).
1101 local unsigned read_buf(strm, buf, size)
1106 unsigned len = strm->avail_in;
1108 if (len > size) len = size;
1109 if (len == 0) return 0;
1111 strm->avail_in -= len;
1113 zmemcpy(buf, strm->next_in, len);
1114 if (strm->state->wrap == 1) {
1115 strm->adler = adler32(strm->adler, buf, len);
1118 else if (strm->state->wrap == 2) {
1119 strm->adler = crc32(strm->adler, buf, len);
1122 strm->next_in += len;
1123 strm->total_in += len;
1128 /* ===========================================================================
1129 * Initialize the "longest match" routines for a new zlib stream
1131 local void lm_init (s)
1134 s->window_size = (ulg)2L*s->w_size;
1138 /* Set the default configuration parameters:
1140 s->max_lazy_match = configuration_table[s->level].max_lazy;
1141 s->good_match = configuration_table[s->level].good_length;
1142 s->nice_match = configuration_table[s->level].nice_length;
1143 s->max_chain_length = configuration_table[s->level].max_chain;
1146 s->block_start = 0L;
1149 s->match_length = s->prev_length = MIN_MATCH-1;
1150 s->match_available = 0;
1154 match_init(); /* initialize the asm code */
1160 /* ===========================================================================
1161 * Set match_start to the longest match starting at the given string and
1162 * return its length. Matches shorter or equal to prev_length are discarded,
1163 * in which case the result is equal to prev_length and match_start is
1165 * IN assertions: cur_match is the head of the hash chain for the current
1166 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1167 * OUT assertion: the match length is not greater than s->lookahead.
1170 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1171 * match.S. The code will be functionally equivalent.
1173 local uInt longest_match(s, cur_match)
1175 IPos cur_match; /* current match */
1177 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1178 register Bytef *scan = s->window + s->strstart; /* current string */
1179 register Bytef *match; /* matched string */
1180 register int len; /* length of current match */
1181 int best_len = (int)s->prev_length; /* best match length so far */
1182 int nice_match = s->nice_match; /* stop if match long enough */
1183 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1184 s->strstart - (IPos)MAX_DIST(s) : NIL;
1185 /* Stop when cur_match becomes <= limit. To simplify the code,
1186 * we prevent matches with the string of window index 0.
1188 Posf *prev = s->prev;
1189 uInt wmask = s->w_mask;
1192 /* Compare two bytes at a time. Note: this is not always beneficial.
1193 * Try with and without -DUNALIGNED_OK to check.
1195 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1196 register ush scan_start = *(ushf*)scan;
1197 register ush scan_end = *(ushf*)(scan+best_len-1);
1199 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1200 register Byte scan_end1 = scan[best_len-1];
1201 register Byte scan_end = scan[best_len];
1204 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1205 * It is easy to get rid of this optimization if necessary.
1207 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1209 /* Do not waste too much time if we already have a good match: */
1210 if (s->prev_length >= s->good_match) {
1213 /* Do not look for matches beyond the end of the input. This is necessary
1214 * to make deflate deterministic.
1216 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1218 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1221 Assert(cur_match < s->strstart, "no future");
1222 match = s->window + cur_match;
1224 /* Skip to next match if the match length cannot increase
1225 * or if the match length is less than 2. Note that the checks below
1226 * for insufficient lookahead only occur occasionally for performance
1227 * reasons. Therefore uninitialized memory will be accessed, and
1228 * conditional jumps will be made that depend on those values.
1229 * However the length of the match is limited to the lookahead, so
1230 * the output of deflate is not affected by the uninitialized values.
1232 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1233 /* This code assumes sizeof(unsigned short) == 2. Do not use
1234 * UNALIGNED_OK if your compiler uses a different size.
1236 if (*(ushf*)(match+best_len-1) != scan_end ||
1237 *(ushf*)match != scan_start) continue;
1239 /* It is not necessary to compare scan[2] and match[2] since they are
1240 * always equal when the other bytes match, given that the hash keys
1241 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1242 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1243 * lookahead only every 4th comparison; the 128th check will be made
1244 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1245 * necessary to put more guard bytes at the end of the window, or
1246 * to check more often for insufficient lookahead.
1248 Assert(scan[2] == match[2], "scan[2]?");
1251 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1252 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1253 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1254 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1256 /* The funny "do {}" generates better code on most compilers */
1258 /* Here, scan <= window+strstart+257 */
1259 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1260 if (*scan == *match) scan++;
1262 len = (MAX_MATCH - 1) - (int)(strend-scan);
1263 scan = strend - (MAX_MATCH-1);
1265 #else /* UNALIGNED_OK */
1267 if (match[best_len] != scan_end ||
1268 match[best_len-1] != scan_end1 ||
1270 *++match != scan[1]) continue;
1272 /* The check at best_len-1 can be removed because it will be made
1273 * again later. (This heuristic is not always a win.)
1274 * It is not necessary to compare scan[2] and match[2] since they
1275 * are always equal when the other bytes match, given that
1276 * the hash keys are equal and that HASH_BITS >= 8.
1279 Assert(*scan == *match, "match[2]?");
1281 /* We check for insufficient lookahead only every 8th comparison;
1282 * the 256th check will be made at strstart+258.
1285 } while (*++scan == *++match && *++scan == *++match &&
1286 *++scan == *++match && *++scan == *++match &&
1287 *++scan == *++match && *++scan == *++match &&
1288 *++scan == *++match && *++scan == *++match &&
1291 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1293 len = MAX_MATCH - (int)(strend - scan);
1294 scan = strend - MAX_MATCH;
1296 #endif /* UNALIGNED_OK */
1298 if (len > best_len) {
1299 s->match_start = cur_match;
1301 if (len >= nice_match) break;
1303 scan_end = *(ushf*)(scan+best_len-1);
1305 scan_end1 = scan[best_len-1];
1306 scan_end = scan[best_len];
1309 } while ((cur_match = prev[cur_match & wmask]) > limit
1310 && --chain_length != 0);
1312 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1313 return s->lookahead;
1319 /* ---------------------------------------------------------------------------
1320 * Optimized version for FASTEST only
1322 local uInt longest_match(s, cur_match)
1324 IPos cur_match; /* current match */
1326 register Bytef *scan = s->window + s->strstart; /* current string */
1327 register Bytef *match; /* matched string */
1328 register int len; /* length of current match */
1329 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1331 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1332 * It is easy to get rid of this optimization if necessary.
1334 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1336 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1338 Assert(cur_match < s->strstart, "no future");
1340 match = s->window + cur_match;
1342 /* Return failure if the match length is less than 2:
1344 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1346 /* The check at best_len-1 can be removed because it will be made
1347 * again later. (This heuristic is not always a win.)
1348 * It is not necessary to compare scan[2] and match[2] since they
1349 * are always equal when the other bytes match, given that
1350 * the hash keys are equal and that HASH_BITS >= 8.
1352 scan += 2, match += 2;
1353 Assert(*scan == *match, "match[2]?");
1355 /* We check for insufficient lookahead only every 8th comparison;
1356 * the 256th check will be made at strstart+258.
1359 } while (*++scan == *++match && *++scan == *++match &&
1360 *++scan == *++match && *++scan == *++match &&
1361 *++scan == *++match && *++scan == *++match &&
1362 *++scan == *++match && *++scan == *++match &&
1365 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1367 len = MAX_MATCH - (int)(strend - scan);
1369 if (len < MIN_MATCH) return MIN_MATCH - 1;
1371 s->match_start = cur_match;
1372 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1375 #endif /* FASTEST */
1380 /* result of memcmp for equal strings */
1382 /* ===========================================================================
1383 * Check that the match at match_start is indeed a match.
1385 local void check_match(s, start, match, length)
1390 /* check that the match is indeed a match */
1391 if (zmemcmp(s->window + match,
1392 s->window + start, length) != EQUAL) {
1393 fprintf(stderr, " start %u, match %u, length %d\n",
1394 start, match, length);
1396 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1397 } while (--length != 0);
1398 z_error("invalid match");
1400 if (z_verbose > 1) {
1401 fprintf(stderr,"\\[%d,%d]", start-match, length);
1402 do { putc(s->window[start++], stderr); } while (--length != 0);
1406 # define check_match(s, start, match, length)
1407 #endif /* ZLIB_DEBUG */
1409 /* ===========================================================================
1410 * Fill the window when the lookahead becomes insufficient.
1411 * Updates strstart and lookahead.
1413 * IN assertion: lookahead < MIN_LOOKAHEAD
1414 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1415 * At least one byte has been read, or avail_in == 0; reads are
1416 * performed for at least two bytes (required for the zip translate_eol
1417 * option -- not supported here).
1419 local void fill_window(s)
1422 register unsigned n, m;
1424 unsigned more; /* Amount of free space at the end of the window. */
1425 uInt wsize = s->w_size;
1427 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1430 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1432 /* Deal with !@#$% 64K limit: */
1433 if (sizeof(int) <= 2) {
1434 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1437 } else if (more == (unsigned)(-1)) {
1438 /* Very unlikely, but possible on 16 bit machine if
1439 * strstart == 0 && lookahead == 1 (input done a byte at time)
1445 /* If the window is almost full and there is insufficient lookahead,
1446 * move the upper half to the lower one to make room in the upper half.
1448 if (s->strstart >= wsize+MAX_DIST(s)) {
1450 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1451 s->match_start -= wsize;
1452 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1453 s->block_start -= (long) wsize;
1455 /* Slide the hash table (could be avoided with 32 bit values
1456 at the expense of memory usage). We slide even when level == 0
1457 to keep the hash table consistent if we switch back to level > 0
1458 later. (Using level 0 permanently is not an optimal usage of
1459 zlib, so we don't care about this pathological case.)
1465 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1473 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1474 /* If n is not on any hash chain, prev[n] is garbage but
1475 * its value will never be used.
1481 if (s->strm->avail_in == 0) break;
1483 /* If there was no sliding:
1484 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1485 * more == window_size - lookahead - strstart
1486 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1487 * => more >= window_size - 2*WSIZE + 2
1488 * In the BIG_MEM or MMAP case (not yet supported),
1489 * window_size == input_size + MIN_LOOKAHEAD &&
1490 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1491 * Otherwise, window_size == 2*WSIZE so more >= 2.
1492 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1494 Assert(more >= 2, "more < 2");
1496 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1499 /* Initialize the hash value now that we have some input: */
1500 if (s->lookahead + s->insert >= MIN_MATCH) {
1501 uInt str = s->strstart - s->insert;
1502 s->ins_h = s->window[str];
1503 UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1505 Call UPDATE_HASH() MIN_MATCH-3 more times
1508 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1510 s->prev[str & s->w_mask] = s->head[s->ins_h];
1512 s->head[s->ins_h] = (Pos)str;
1515 if (s->lookahead + s->insert < MIN_MATCH)
1519 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1520 * but this is not important since only literal bytes will be emitted.
1523 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1525 /* If the WIN_INIT bytes after the end of the current data have never been
1526 * written, then zero those bytes in order to avoid memory check reports of
1527 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1528 * the longest match routines. Update the high water mark for the next
1529 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1530 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1532 if (s->high_water < s->window_size) {
1533 ulg curr = s->strstart + (ulg)(s->lookahead);
1536 if (s->high_water < curr) {
1537 /* Previous high water mark below current data -- zero WIN_INIT
1538 * bytes or up to end of window, whichever is less.
1540 init = s->window_size - curr;
1541 if (init > WIN_INIT)
1543 zmemzero(s->window + curr, (unsigned)init);
1544 s->high_water = curr + init;
1546 else if (s->high_water < (ulg)curr + WIN_INIT) {
1547 /* High water mark at or above current data, but below current data
1548 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1549 * to end of window, whichever is less.
1551 init = (ulg)curr + WIN_INIT - s->high_water;
1552 if (init > s->window_size - s->high_water)
1553 init = s->window_size - s->high_water;
1554 zmemzero(s->window + s->high_water, (unsigned)init);
1555 s->high_water += init;
1559 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1560 "not enough room for search");
1563 /* ===========================================================================
1564 * Flush the current block, with given end-of-file flag.
1565 * IN assertion: strstart is set to the end of the current match.
1567 #define FLUSH_BLOCK_ONLY(s, last) { \
1568 _tr_flush_block(s, (s->block_start >= 0L ? \
1569 (charf *)&s->window[(unsigned)s->block_start] : \
1571 (ulg)((long)s->strstart - s->block_start), \
1573 s->block_start = s->strstart; \
1574 flush_pending(s->strm); \
1575 Tracev((stderr,"[FLUSH]")); \
1578 /* Same but force premature exit if necessary. */
1579 #define FLUSH_BLOCK(s, last) { \
1580 FLUSH_BLOCK_ONLY(s, last); \
1581 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1584 /* ===========================================================================
1585 * Copy without compression as much as possible from the input stream, return
1586 * the current block state.
1587 * This function does not insert new strings in the dictionary since
1588 * uncompressible data is probably not useful. This function is used
1589 * only for the level=0 compression option.
1590 * NOTE: this function should be optimized to avoid extra copying from
1591 * window to pending_buf.
1593 local block_state deflate_stored(s, flush)
1597 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1598 * to pending_buf_size, and each stored block has a 5 byte header:
1600 ulg max_block_size = 0xffff;
1603 if (max_block_size > s->pending_buf_size - 5) {
1604 max_block_size = s->pending_buf_size - 5;
1607 /* Copy as much as possible from input to output: */
1609 /* Fill the window as much as possible: */
1610 if (s->lookahead <= 1) {
1612 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1613 s->block_start >= (long)s->w_size, "slide too late");
1616 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1618 if (s->lookahead == 0) break; /* flush the current block */
1620 Assert(s->block_start >= 0L, "block gone");
1622 s->strstart += s->lookahead;
1625 /* Emit a stored block if pending_buf will be full: */
1626 max_start = max_block_size + (ulg)s->block_start;
1627 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1628 /* strstart == 0 is possible when wraparound on 16-bit machine */
1629 s->lookahead = (uInt)(s->strstart - max_start);
1630 s->strstart = (uInt)max_start;
1633 /* Flush if we may have to slide, otherwise block_start may become
1634 * negative and the data will be gone:
1636 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1641 if (flush == Z_FINISH) {
1645 if ((long)s->strstart > s->block_start)
1650 /* ===========================================================================
1651 * Compress as much as possible from the input stream, return the current
1653 * This function does not perform lazy evaluation of matches and inserts
1654 * new strings in the dictionary only for unmatched strings or for short
1655 * matches. It is used only for the fast compression options.
1657 local block_state deflate_fast(s, flush)
1661 IPos hash_head; /* head of the hash chain */
1662 int bflush; /* set if current block must be flushed */
1665 /* Make sure that we always have enough lookahead, except
1666 * at the end of the input file. We need MAX_MATCH bytes
1667 * for the next match, plus MIN_MATCH bytes to insert the
1668 * string following the next match.
1670 if (s->lookahead < MIN_LOOKAHEAD) {
1672 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1675 if (s->lookahead == 0) break; /* flush the current block */
1678 /* Insert the string window[strstart .. strstart+2] in the
1679 * dictionary, and set hash_head to the head of the hash chain:
1682 if (s->lookahead >= MIN_MATCH) {
1683 INSERT_STRING(s, s->strstart, hash_head);
1686 /* Find the longest match, discarding those <= prev_length.
1687 * At this point we have always match_length < MIN_MATCH
1689 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1690 /* To simplify the code, we prevent matches with the string
1691 * of window index 0 (in particular we have to avoid a match
1692 * of the string with itself at the start of the input file).
1694 s->match_length = longest_match (s, hash_head);
1695 /* longest_match() sets match_start */
1697 if (s->match_length >= MIN_MATCH) {
1698 check_match(s, s->strstart, s->match_start, s->match_length);
1700 _tr_tally_dist(s, s->strstart - s->match_start,
1701 s->match_length - MIN_MATCH, bflush);
1703 s->lookahead -= s->match_length;
1705 /* Insert new strings in the hash table only if the match length
1706 * is not too large. This saves time but degrades compression.
1709 if (s->match_length <= s->max_insert_length &&
1710 s->lookahead >= MIN_MATCH) {
1711 s->match_length--; /* string at strstart already in table */
1714 INSERT_STRING(s, s->strstart, hash_head);
1715 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1716 * always MIN_MATCH bytes ahead.
1718 } while (--s->match_length != 0);
1723 s->strstart += s->match_length;
1724 s->match_length = 0;
1725 s->ins_h = s->window[s->strstart];
1726 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1728 Call UPDATE_HASH() MIN_MATCH-3 more times
1730 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1731 * matter since it will be recomputed at next deflate call.
1735 /* No match, output a literal byte */
1736 Tracevv((stderr,"%c", s->window[s->strstart]));
1737 _tr_tally_lit (s, s->window[s->strstart], bflush);
1741 if (bflush) FLUSH_BLOCK(s, 0);
1743 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1744 if (flush == Z_FINISH) {
1754 /* ===========================================================================
1755 * Same as above, but achieves better compression. We use a lazy
1756 * evaluation for matches: a match is finally adopted only if there is
1757 * no better match at the next window position.
1759 local block_state deflate_slow(s, flush)
1763 IPos hash_head; /* head of hash chain */
1764 int bflush; /* set if current block must be flushed */
1766 /* Process the input block. */
1768 /* Make sure that we always have enough lookahead, except
1769 * at the end of the input file. We need MAX_MATCH bytes
1770 * for the next match, plus MIN_MATCH bytes to insert the
1771 * string following the next match.
1773 if (s->lookahead < MIN_LOOKAHEAD) {
1775 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1778 if (s->lookahead == 0) break; /* flush the current block */
1781 /* Insert the string window[strstart .. strstart+2] in the
1782 * dictionary, and set hash_head to the head of the hash chain:
1785 if (s->lookahead >= MIN_MATCH) {
1786 INSERT_STRING(s, s->strstart, hash_head);
1789 /* Find the longest match, discarding those <= prev_length.
1791 s->prev_length = s->match_length, s->prev_match = s->match_start;
1792 s->match_length = MIN_MATCH-1;
1794 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1795 s->strstart - hash_head <= MAX_DIST(s)) {
1796 /* To simplify the code, we prevent matches with the string
1797 * of window index 0 (in particular we have to avoid a match
1798 * of the string with itself at the start of the input file).
1800 s->match_length = longest_match (s, hash_head);
1801 /* longest_match() sets match_start */
1803 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1804 #if TOO_FAR <= 32767
1805 || (s->match_length == MIN_MATCH &&
1806 s->strstart - s->match_start > TOO_FAR)
1810 /* If prev_match is also MIN_MATCH, match_start is garbage
1811 * but we will ignore the current match anyway.
1813 s->match_length = MIN_MATCH-1;
1816 /* If there was a match at the previous step and the current
1817 * match is not better, output the previous match:
1819 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1820 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1821 /* Do not insert strings in hash table beyond this. */
1823 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1825 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1826 s->prev_length - MIN_MATCH, bflush);
1828 /* Insert in hash table all strings up to the end of the match.
1829 * strstart-1 and strstart are already inserted. If there is not
1830 * enough lookahead, the last two strings are not inserted in
1833 s->lookahead -= s->prev_length-1;
1834 s->prev_length -= 2;
1836 if (++s->strstart <= max_insert) {
1837 INSERT_STRING(s, s->strstart, hash_head);
1839 } while (--s->prev_length != 0);
1840 s->match_available = 0;
1841 s->match_length = MIN_MATCH-1;
1844 if (bflush) FLUSH_BLOCK(s, 0);
1846 } else if (s->match_available) {
1847 /* If there was no match at the previous position, output a
1848 * single literal. If there was a match but the current match
1849 * is longer, truncate the previous match to a single literal.
1851 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1852 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1854 FLUSH_BLOCK_ONLY(s, 0);
1858 if (s->strm->avail_out == 0) return need_more;
1860 /* There is no previous match to compare with, wait for
1861 * the next step to decide.
1863 s->match_available = 1;
1868 Assert (flush != Z_NO_FLUSH, "no flush?");
1869 if (s->match_available) {
1870 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1871 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1872 s->match_available = 0;
1874 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1875 if (flush == Z_FINISH) {
1883 #endif /* FASTEST */
1885 /* ===========================================================================
1886 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1887 * one. Do not maintain a hash table. (It will be regenerated if this run of
1888 * deflate switches away from Z_RLE.)
1890 local block_state deflate_rle(s, flush)
1894 int bflush; /* set if current block must be flushed */
1895 uInt prev; /* byte at distance one to match */
1896 Bytef *scan, *strend; /* scan goes up to strend for length of run */
1899 /* Make sure that we always have enough lookahead, except
1900 * at the end of the input file. We need MAX_MATCH bytes
1901 * for the longest run, plus one for the unrolled loop.
1903 if (s->lookahead <= MAX_MATCH) {
1905 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
1908 if (s->lookahead == 0) break; /* flush the current block */
1911 /* See how many times the previous byte repeats */
1912 s->match_length = 0;
1913 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
1914 scan = s->window + s->strstart - 1;
1916 if (prev == *++scan && prev == *++scan && prev == *++scan) {
1917 strend = s->window + s->strstart + MAX_MATCH;
1919 } while (prev == *++scan && prev == *++scan &&
1920 prev == *++scan && prev == *++scan &&
1921 prev == *++scan && prev == *++scan &&
1922 prev == *++scan && prev == *++scan &&
1924 s->match_length = MAX_MATCH - (uInt)(strend - scan);
1925 if (s->match_length > s->lookahead)
1926 s->match_length = s->lookahead;
1928 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
1931 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1932 if (s->match_length >= MIN_MATCH) {
1933 check_match(s, s->strstart, s->strstart - 1, s->match_length);
1935 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
1937 s->lookahead -= s->match_length;
1938 s->strstart += s->match_length;
1939 s->match_length = 0;
1941 /* No match, output a literal byte */
1942 Tracevv((stderr,"%c", s->window[s->strstart]));
1943 _tr_tally_lit (s, s->window[s->strstart], bflush);
1947 if (bflush) FLUSH_BLOCK(s, 0);
1950 if (flush == Z_FINISH) {
1959 /* ===========================================================================
1960 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
1961 * (It will be regenerated if this run of deflate switches away from Huffman.)
1963 local block_state deflate_huff(s, flush)
1967 int bflush; /* set if current block must be flushed */
1970 /* Make sure that we have a literal to write. */
1971 if (s->lookahead == 0) {
1973 if (s->lookahead == 0) {
1974 if (flush == Z_NO_FLUSH)
1976 break; /* flush the current block */
1980 /* Output a literal byte */
1981 s->match_length = 0;
1982 Tracevv((stderr,"%c", s->window[s->strstart]));
1983 _tr_tally_lit (s, s->window[s->strstart], bflush);
1986 if (bflush) FLUSH_BLOCK(s, 0);
1989 if (flush == Z_FINISH) {