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 void fill_window OF((deflate_state *s));
77 local block_state deflate_stored OF((deflate_state *s, int flush));
78 local block_state deflate_fast OF((deflate_state *s, int flush));
80 local block_state deflate_slow OF((deflate_state *s, int flush));
82 local block_state deflate_rle OF((deflate_state *s, int flush));
83 local block_state deflate_huff OF((deflate_state *s, int flush));
84 local void lm_init OF((deflate_state *s));
85 local void putShortMSB OF((deflate_state *s, uInt b));
86 local void flush_pending OF((z_streamp strm));
87 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
89 void match_init OF((void)); /* asm code initialization */
90 uInt longest_match OF((deflate_state *s, IPos cur_match));
92 local uInt longest_match OF((deflate_state *s, IPos cur_match));
96 local void check_match OF((deflate_state *s, IPos start, IPos match,
100 /* ===========================================================================
105 /* Tail of hash chains */
108 # define TOO_FAR 4096
110 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
112 /* Values for max_lazy_match, good_match and max_chain_length, depending on
113 * the desired pack level (0..9). The values given below have been tuned to
114 * exclude worst case performance for pathological files. Better values may be
115 * found for specific files.
117 typedef struct config_s {
118 ush good_length; /* reduce lazy search above this match length */
119 ush max_lazy; /* do not perform lazy search above this match length */
120 ush nice_length; /* quit search above this match length */
126 local const config configuration_table[2] = {
127 /* good lazy nice chain */
128 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
129 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
131 local const config configuration_table[10] = {
132 /* good lazy nice chain */
133 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
134 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
135 /* 2 */ {4, 5, 16, 8, deflate_fast},
136 /* 3 */ {4, 6, 32, 32, deflate_fast},
138 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
139 /* 5 */ {8, 16, 32, 32, deflate_slow},
140 /* 6 */ {8, 16, 128, 128, deflate_slow},
141 /* 7 */ {8, 32, 128, 256, deflate_slow},
142 /* 8 */ {32, 128, 258, 1024, deflate_slow},
143 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
146 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
147 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
152 /* result of memcmp for equal strings */
154 #ifndef NO_DUMMY_DECL
155 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
158 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
159 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
161 /* ===========================================================================
162 * Update a hash value with the given input byte
163 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
164 * input characters, so that a running hash key can be computed from the
165 * previous key instead of complete recalculation each time.
167 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
170 /* ===========================================================================
171 * Insert string str in the dictionary and set match_head to the previous head
172 * of the hash chain (the most recent string with same hash key). Return
173 * the previous length of the hash chain.
174 * If this file is compiled with -DFASTEST, the compression level is forced
175 * to 1, and no hash chains are maintained.
176 * IN assertion: all calls to to INSERT_STRING are made with consecutive
177 * input characters and the first MIN_MATCH bytes of str are valid
178 * (except for the last MIN_MATCH-1 bytes of the input file).
181 #define INSERT_STRING(s, str, match_head) \
182 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
183 match_head = s->head[s->ins_h], \
184 s->head[s->ins_h] = (Pos)(str))
186 #define INSERT_STRING(s, str, match_head) \
187 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
188 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
189 s->head[s->ins_h] = (Pos)(str))
192 /* ===========================================================================
193 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
194 * prev[] will be initialized on the fly.
196 #define CLEAR_HASH(s) \
197 s->head[s->hash_size-1] = NIL; \
198 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
200 /* ========================================================================= */
201 int ZEXPORT deflateInit_(strm, level, version, stream_size)
207 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
208 Z_DEFAULT_STRATEGY, version, stream_size);
209 /* To do: ignore strm->next_in if we use it as window */
212 /* ========================================================================= */
213 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
214 version, stream_size)
226 static const char my_version[] = ZLIB_VERSION;
229 /* We overlay pending_buf and d_buf+l_buf. This works since the average
230 * output size for (length,distance) codes is <= 24 bits.
233 if (version == Z_NULL || version[0] != my_version[0] ||
234 stream_size != sizeof(z_stream)) {
235 return Z_VERSION_ERROR;
237 if (strm == Z_NULL) return Z_STREAM_ERROR;
240 if (strm->zalloc == (alloc_func)0) {
242 return Z_STREAM_ERROR;
244 strm->zalloc = zcalloc;
245 strm->opaque = (voidpf)0;
248 if (strm->zfree == (free_func)0)
250 return Z_STREAM_ERROR;
252 strm->zfree = zcfree;
256 if (level != 0) level = 1;
258 if (level == Z_DEFAULT_COMPRESSION) level = 6;
261 if (windowBits < 0) { /* suppress zlib wrapper */
263 windowBits = -windowBits;
266 else if (windowBits > 15) {
267 wrap = 2; /* write gzip wrapper instead */
271 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
272 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
273 strategy < 0 || strategy > Z_FIXED) {
274 return Z_STREAM_ERROR;
276 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
277 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
278 if (s == Z_NULL) return Z_MEM_ERROR;
279 strm->state = (struct internal_state FAR *)s;
284 s->w_bits = windowBits;
285 s->w_size = 1 << s->w_bits;
286 s->w_mask = s->w_size - 1;
288 s->hash_bits = memLevel + 7;
289 s->hash_size = 1 << s->hash_bits;
290 s->hash_mask = s->hash_size - 1;
291 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
293 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
294 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
295 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
297 s->high_water = 0; /* nothing written to s->window yet */
299 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
301 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
302 s->pending_buf = (uchf *) overlay;
303 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
305 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
306 s->pending_buf == Z_NULL) {
307 s->status = FINISH_STATE;
308 strm->msg = ERR_MSG(Z_MEM_ERROR);
312 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
313 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
316 s->strategy = strategy;
317 s->method = (Byte)method;
319 return deflateReset(strm);
322 /* ========================================================================= */
323 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
325 const Bytef *dictionary;
332 z_const unsigned char *next;
334 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL)
335 return Z_STREAM_ERROR;
338 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
339 return Z_STREAM_ERROR;
341 /* when using zlib wrappers, compute Adler-32 for provided dictionary */
343 strm->adler = adler32(strm->adler, dictionary, dictLength);
344 s->wrap = 0; /* avoid computing Adler-32 in read_buf */
346 /* if dictionary would fill window, just replace the history */
347 if (dictLength >= s->w_size) {
348 if (wrap == 0) { /* already empty otherwise */
354 dictionary += dictLength - s->w_size; /* use the tail */
355 dictLength = s->w_size;
358 /* insert dictionary into window and hash */
359 avail = strm->avail_in;
360 next = strm->next_in;
361 strm->avail_in = dictLength;
362 strm->next_in = (z_const Bytef *)dictionary;
364 while (s->lookahead >= MIN_MATCH) {
366 n = s->lookahead - (MIN_MATCH-1);
368 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
370 s->prev[str & s->w_mask] = s->head[s->ins_h];
372 s->head[s->ins_h] = (Pos)str;
376 s->lookahead = MIN_MATCH-1;
379 s->strstart += s->lookahead;
380 s->block_start = (long)s->strstart;
381 s->insert = s->lookahead;
383 s->match_length = s->prev_length = MIN_MATCH-1;
384 s->match_available = 0;
385 strm->next_in = next;
386 strm->avail_in = avail;
391 /* ========================================================================= */
392 int ZEXPORT deflateResetKeep (strm)
397 if (strm == Z_NULL || strm->state == Z_NULL ||
398 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
399 return Z_STREAM_ERROR;
402 strm->total_in = strm->total_out = 0;
403 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
404 strm->data_type = Z_UNKNOWN;
406 s = (deflate_state *)strm->state;
408 s->pending_out = s->pending_buf;
411 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
413 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
416 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
418 adler32(0L, Z_NULL, 0);
419 s->last_flush = Z_NO_FLUSH;
426 /* ========================================================================= */
427 int ZEXPORT deflateReset (strm)
432 ret = deflateResetKeep(strm);
434 lm_init(strm->state);
438 /* ========================================================================= */
439 int ZEXPORT deflateSetHeader (strm, head)
443 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
444 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
445 strm->state->gzhead = head;
449 /* ========================================================================= */
450 int ZEXPORT deflatePending (strm, pending, bits)
455 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
456 if (pending != Z_NULL)
457 *pending = strm->state->pending;
459 *bits = strm->state->bi_valid;
463 /* ========================================================================= */
464 int ZEXPORT deflatePrime (strm, bits, value)
472 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
474 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
477 put = Buf_size - s->bi_valid;
480 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
489 /* ========================================================================= */
490 int ZEXPORT deflateParams(strm, level, strategy)
499 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
503 if (level != 0) level = 1;
505 if (level == Z_DEFAULT_COMPRESSION) level = 6;
507 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
508 return Z_STREAM_ERROR;
510 func = configuration_table[s->level].func;
512 if ((strategy != s->strategy || func != configuration_table[level].func)) {
513 /* Flush the last buffer: */
514 err = deflate(strm, Z_BLOCK);
515 if (err == Z_BUF_ERROR && s->pending == 0)
518 if (s->level != level) {
520 s->max_lazy_match = configuration_table[level].max_lazy;
521 s->good_match = configuration_table[level].good_length;
522 s->nice_match = configuration_table[level].nice_length;
523 s->max_chain_length = configuration_table[level].max_chain;
525 s->strategy = strategy;
529 /* ========================================================================= */
530 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
539 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
541 s->good_match = good_length;
542 s->max_lazy_match = max_lazy;
543 s->nice_match = nice_length;
544 s->max_chain_length = max_chain;
548 /* =========================================================================
549 * For the default windowBits of 15 and memLevel of 8, this function returns
550 * a close to exact, as well as small, upper bound on the compressed size.
551 * They are coded as constants here for a reason--if the #define's are
552 * changed, then this function needs to be changed as well. The return
553 * value for 15 and 8 only works for those exact settings.
555 * For any setting other than those defaults for windowBits and memLevel,
556 * the value returned is a conservative worst case for the maximum expansion
557 * resulting from using fixed blocks instead of stored blocks, which deflate
558 * can emit on compressed data for some combinations of the parameters.
560 * This function could be more sophisticated to provide closer upper bounds for
561 * every combination of windowBits and memLevel. But even the conservative
562 * upper bound of about 14% expansion does not seem onerous for output buffer
565 uLong ZEXPORT deflateBound(strm, sourceLen)
570 uLong complen, wraplen;
573 /* conservative upper bound for compressed data */
574 complen = sourceLen +
575 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
577 /* if can't get parameters, return conservative bound plus zlib wrapper */
578 if (strm == Z_NULL || strm->state == Z_NULL)
581 /* compute wrapper length */
584 case 0: /* raw deflate */
587 case 1: /* zlib wrapper */
588 wraplen = 6 + (s->strstart ? 4 : 0);
590 case 2: /* gzip wrapper */
592 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
593 if (s->gzhead->extra != Z_NULL)
594 wraplen += 2 + s->gzhead->extra_len;
595 str = s->gzhead->name;
600 str = s->gzhead->comment;
609 default: /* for compiler happiness */
613 /* if not default parameters, return conservative bound */
614 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
615 return complen + wraplen;
617 /* default settings: return tight bound for that case */
618 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
619 (sourceLen >> 25) + 13 - 6 + wraplen;
622 /* =========================================================================
623 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
624 * IN assertion: the stream state is correct and there is enough room in
627 local void putShortMSB (s, b)
631 put_byte(s, (Byte)(b >> 8));
632 put_byte(s, (Byte)(b & 0xff));
635 /* =========================================================================
636 * Flush as much pending output as possible. All deflate() output goes
637 * through this function so some applications may wish to modify it
638 * to avoid allocating a large strm->next_out buffer and copying into it.
639 * (See also read_buf()).
641 local void flush_pending(strm)
645 deflate_state *s = strm->state;
649 if (len > strm->avail_out) len = strm->avail_out;
650 if (len == 0) return;
652 zmemcpy(strm->next_out, s->pending_out, len);
653 strm->next_out += len;
654 s->pending_out += len;
655 strm->total_out += len;
656 strm->avail_out -= len;
658 if (s->pending == 0) {
659 s->pending_out = s->pending_buf;
663 /* ========================================================================= */
664 int ZEXPORT deflate (strm, flush)
668 int old_flush; /* value of flush param for previous deflate call */
671 if (strm == Z_NULL || strm->state == Z_NULL ||
672 flush > Z_BLOCK || flush < 0) {
673 return Z_STREAM_ERROR;
677 if (strm->next_out == Z_NULL ||
678 (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
679 (s->status == FINISH_STATE && flush != Z_FINISH)) {
680 ERR_RETURN(strm, Z_STREAM_ERROR);
682 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
684 s->strm = strm; /* just in case */
685 old_flush = s->last_flush;
686 s->last_flush = flush;
688 /* Write the header */
689 if (s->status == INIT_STATE) {
692 strm->adler = crc32(0L, Z_NULL, 0);
696 if (s->gzhead == Z_NULL) {
702 put_byte(s, s->level == 9 ? 2 :
703 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
705 put_byte(s, OS_CODE);
706 s->status = BUSY_STATE;
709 put_byte(s, (s->gzhead->text ? 1 : 0) +
710 (s->gzhead->hcrc ? 2 : 0) +
711 (s->gzhead->extra == Z_NULL ? 0 : 4) +
712 (s->gzhead->name == Z_NULL ? 0 : 8) +
713 (s->gzhead->comment == Z_NULL ? 0 : 16)
715 put_byte(s, (Byte)(s->gzhead->time & 0xff));
716 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
717 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
718 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
719 put_byte(s, s->level == 9 ? 2 :
720 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
722 put_byte(s, s->gzhead->os & 0xff);
723 if (s->gzhead->extra != Z_NULL) {
724 put_byte(s, s->gzhead->extra_len & 0xff);
725 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
728 strm->adler = crc32(strm->adler, s->pending_buf,
731 s->status = EXTRA_STATE;
737 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
740 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
742 else if (s->level < 6)
744 else if (s->level == 6)
748 header |= (level_flags << 6);
749 if (s->strstart != 0) header |= PRESET_DICT;
750 header += 31 - (header % 31);
752 s->status = BUSY_STATE;
753 putShortMSB(s, header);
755 /* Save the adler32 of the preset dictionary: */
756 if (s->strstart != 0) {
757 putShortMSB(s, (uInt)(strm->adler >> 16));
758 putShortMSB(s, (uInt)(strm->adler & 0xffff));
760 strm->adler = adler32(0L, Z_NULL, 0);
764 if (s->status == EXTRA_STATE) {
765 if (s->gzhead->extra != Z_NULL) {
766 uInt beg = s->pending; /* start of bytes to update crc */
768 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
769 if (s->pending == s->pending_buf_size) {
770 if (s->gzhead->hcrc && s->pending > beg)
771 strm->adler = crc32(strm->adler, s->pending_buf + beg,
775 if (s->pending == s->pending_buf_size)
778 put_byte(s, s->gzhead->extra[s->gzindex]);
781 if (s->gzhead->hcrc && s->pending > beg)
782 strm->adler = crc32(strm->adler, s->pending_buf + beg,
784 if (s->gzindex == s->gzhead->extra_len) {
786 s->status = NAME_STATE;
790 s->status = NAME_STATE;
792 if (s->status == NAME_STATE) {
793 if (s->gzhead->name != Z_NULL) {
794 uInt beg = s->pending; /* start of bytes to update crc */
798 if (s->pending == s->pending_buf_size) {
799 if (s->gzhead->hcrc && s->pending > beg)
800 strm->adler = crc32(strm->adler, s->pending_buf + beg,
804 if (s->pending == s->pending_buf_size) {
809 val = s->gzhead->name[s->gzindex++];
812 if (s->gzhead->hcrc && s->pending > beg)
813 strm->adler = crc32(strm->adler, s->pending_buf + beg,
817 s->status = COMMENT_STATE;
821 s->status = COMMENT_STATE;
823 if (s->status == COMMENT_STATE) {
824 if (s->gzhead->comment != Z_NULL) {
825 uInt beg = s->pending; /* start of bytes to update crc */
829 if (s->pending == s->pending_buf_size) {
830 if (s->gzhead->hcrc && s->pending > beg)
831 strm->adler = crc32(strm->adler, s->pending_buf + beg,
835 if (s->pending == s->pending_buf_size) {
840 val = s->gzhead->comment[s->gzindex++];
843 if (s->gzhead->hcrc && s->pending > beg)
844 strm->adler = crc32(strm->adler, s->pending_buf + beg,
847 s->status = HCRC_STATE;
850 s->status = HCRC_STATE;
852 if (s->status == HCRC_STATE) {
853 if (s->gzhead->hcrc) {
854 if (s->pending + 2 > s->pending_buf_size)
856 if (s->pending + 2 <= s->pending_buf_size) {
857 put_byte(s, (Byte)(strm->adler & 0xff));
858 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
859 strm->adler = crc32(0L, Z_NULL, 0);
860 s->status = BUSY_STATE;
864 s->status = BUSY_STATE;
868 /* Flush as much pending output as possible */
869 if (s->pending != 0) {
871 if (strm->avail_out == 0) {
872 /* Since avail_out is 0, deflate will be called again with
873 * more output space, but possibly with both pending and
874 * avail_in equal to zero. There won't be anything to do,
875 * but this is not an error situation so make sure we
876 * return OK instead of BUF_ERROR at next call of deflate:
882 /* Make sure there is something to do and avoid duplicate consecutive
883 * flushes. For repeated and useless calls with Z_FINISH, we keep
884 * returning Z_STREAM_END instead of Z_BUF_ERROR.
886 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
888 ERR_RETURN(strm, Z_BUF_ERROR);
891 /* User must not provide more input after the first FINISH: */
892 if (s->status == FINISH_STATE && strm->avail_in != 0) {
893 ERR_RETURN(strm, Z_BUF_ERROR);
896 /* Start a new block or continue the current one.
898 if (strm->avail_in != 0 || s->lookahead != 0 ||
899 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
902 bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
903 (s->strategy == Z_RLE ? deflate_rle(s, flush) :
904 (*(configuration_table[s->level].func))(s, flush));
906 if (bstate == finish_started || bstate == finish_done) {
907 s->status = FINISH_STATE;
909 if (bstate == need_more || bstate == finish_started) {
910 if (strm->avail_out == 0) {
911 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
914 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
915 * of deflate should use the same flush parameter to make sure
916 * that the flush is complete. So we don't have to output an
917 * empty block here, this will be done at next call. This also
918 * ensures that for a very small output buffer, we emit at most
922 if (bstate == block_done) {
923 if (flush == Z_PARTIAL_FLUSH) {
925 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
926 _tr_stored_block(s, (char*)0, 0L, 0);
927 /* For a full flush, this empty block will be recognized
928 * as a special marker by inflate_sync().
930 if (flush == Z_FULL_FLUSH) {
931 CLEAR_HASH(s); /* forget history */
932 if (s->lookahead == 0) {
940 if (strm->avail_out == 0) {
941 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
946 Assert(strm->avail_out > 0, "bug2");
948 if (flush != Z_FINISH) return Z_OK;
949 if (s->wrap <= 0) return Z_STREAM_END;
951 /* Write the trailer */
954 put_byte(s, (Byte)(strm->adler & 0xff));
955 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
956 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
957 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
958 put_byte(s, (Byte)(strm->total_in & 0xff));
959 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
960 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
961 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
966 putShortMSB(s, (uInt)(strm->adler >> 16));
967 putShortMSB(s, (uInt)(strm->adler & 0xffff));
970 /* If avail_out is zero, the application will call deflate again
973 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
974 return s->pending != 0 ? Z_OK : Z_STREAM_END;
977 /* ========================================================================= */
978 int ZEXPORT deflateEnd (strm)
983 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
985 status = strm->state->status;
986 if (status != INIT_STATE &&
987 status != EXTRA_STATE &&
988 status != NAME_STATE &&
989 status != COMMENT_STATE &&
990 status != HCRC_STATE &&
991 status != BUSY_STATE &&
992 status != FINISH_STATE) {
993 return Z_STREAM_ERROR;
996 /* Deallocate in reverse order of allocations: */
997 TRY_FREE(strm, strm->state->pending_buf);
998 TRY_FREE(strm, strm->state->head);
999 TRY_FREE(strm, strm->state->prev);
1000 TRY_FREE(strm, strm->state->window);
1002 ZFREE(strm, strm->state);
1003 strm->state = Z_NULL;
1005 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1008 /* =========================================================================
1009 * Copy the source state to the destination state.
1010 * To simplify the source, this is not supported for 16-bit MSDOS (which
1011 * doesn't have enough memory anyway to duplicate compression states).
1013 int ZEXPORT deflateCopy (dest, source)
1018 return Z_STREAM_ERROR;
1025 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
1026 return Z_STREAM_ERROR;
1031 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1033 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1034 if (ds == Z_NULL) return Z_MEM_ERROR;
1035 dest->state = (struct internal_state FAR *) ds;
1036 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1039 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1040 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
1041 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
1042 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1043 ds->pending_buf = (uchf *) overlay;
1045 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1046 ds->pending_buf == Z_NULL) {
1050 /* following zmemcpy do not work for 16-bit MSDOS */
1051 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1052 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1053 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1054 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1056 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1057 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1058 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1060 ds->l_desc.dyn_tree = ds->dyn_ltree;
1061 ds->d_desc.dyn_tree = ds->dyn_dtree;
1062 ds->bl_desc.dyn_tree = ds->bl_tree;
1065 #endif /* MAXSEG_64K */
1068 /* ===========================================================================
1069 * Read a new buffer from the current input stream, update the adler32
1070 * and total number of bytes read. All deflate() input goes through
1071 * this function so some applications may wish to modify it to avoid
1072 * allocating a large strm->next_in buffer and copying from it.
1073 * (See also flush_pending()).
1075 local int read_buf(strm, buf, size)
1080 unsigned len = strm->avail_in;
1082 if (len > size) len = size;
1083 if (len == 0) return 0;
1085 strm->avail_in -= len;
1087 zmemcpy(buf, strm->next_in, len);
1088 if (strm->state->wrap == 1) {
1089 strm->adler = adler32(strm->adler, buf, len);
1092 else if (strm->state->wrap == 2) {
1093 strm->adler = crc32(strm->adler, buf, len);
1096 strm->next_in += len;
1097 strm->total_in += len;
1102 /* ===========================================================================
1103 * Initialize the "longest match" routines for a new zlib stream
1105 local void lm_init (s)
1108 s->window_size = (ulg)2L*s->w_size;
1112 /* Set the default configuration parameters:
1114 s->max_lazy_match = configuration_table[s->level].max_lazy;
1115 s->good_match = configuration_table[s->level].good_length;
1116 s->nice_match = configuration_table[s->level].nice_length;
1117 s->max_chain_length = configuration_table[s->level].max_chain;
1120 s->block_start = 0L;
1123 s->match_length = s->prev_length = MIN_MATCH-1;
1124 s->match_available = 0;
1128 match_init(); /* initialize the asm code */
1134 /* ===========================================================================
1135 * Set match_start to the longest match starting at the given string and
1136 * return its length. Matches shorter or equal to prev_length are discarded,
1137 * in which case the result is equal to prev_length and match_start is
1139 * IN assertions: cur_match is the head of the hash chain for the current
1140 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1141 * OUT assertion: the match length is not greater than s->lookahead.
1144 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1145 * match.S. The code will be functionally equivalent.
1147 local uInt longest_match(s, cur_match)
1149 IPos cur_match; /* current match */
1151 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1152 register Bytef *scan = s->window + s->strstart; /* current string */
1153 register Bytef *match; /* matched string */
1154 register int len; /* length of current match */
1155 int best_len = s->prev_length; /* best match length so far */
1156 int nice_match = s->nice_match; /* stop if match long enough */
1157 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1158 s->strstart - (IPos)MAX_DIST(s) : NIL;
1159 /* Stop when cur_match becomes <= limit. To simplify the code,
1160 * we prevent matches with the string of window index 0.
1162 Posf *prev = s->prev;
1163 uInt wmask = s->w_mask;
1166 /* Compare two bytes at a time. Note: this is not always beneficial.
1167 * Try with and without -DUNALIGNED_OK to check.
1169 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1170 register ush scan_start = *(ushf*)scan;
1171 register ush scan_end = *(ushf*)(scan+best_len-1);
1173 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1174 register Byte scan_end1 = scan[best_len-1];
1175 register Byte scan_end = scan[best_len];
1178 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1179 * It is easy to get rid of this optimization if necessary.
1181 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1183 /* Do not waste too much time if we already have a good match: */
1184 if (s->prev_length >= s->good_match) {
1187 /* Do not look for matches beyond the end of the input. This is necessary
1188 * to make deflate deterministic.
1190 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1192 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1195 Assert(cur_match < s->strstart, "no future");
1196 match = s->window + cur_match;
1198 /* Skip to next match if the match length cannot increase
1199 * or if the match length is less than 2. Note that the checks below
1200 * for insufficient lookahead only occur occasionally for performance
1201 * reasons. Therefore uninitialized memory will be accessed, and
1202 * conditional jumps will be made that depend on those values.
1203 * However the length of the match is limited to the lookahead, so
1204 * the output of deflate is not affected by the uninitialized values.
1206 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1207 /* This code assumes sizeof(unsigned short) == 2. Do not use
1208 * UNALIGNED_OK if your compiler uses a different size.
1210 if (*(ushf*)(match+best_len-1) != scan_end ||
1211 *(ushf*)match != scan_start) continue;
1213 /* It is not necessary to compare scan[2] and match[2] since they are
1214 * always equal when the other bytes match, given that the hash keys
1215 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1216 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1217 * lookahead only every 4th comparison; the 128th check will be made
1218 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1219 * necessary to put more guard bytes at the end of the window, or
1220 * to check more often for insufficient lookahead.
1222 Assert(scan[2] == match[2], "scan[2]?");
1225 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1226 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1227 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1228 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1230 /* The funny "do {}" generates better code on most compilers */
1232 /* Here, scan <= window+strstart+257 */
1233 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1234 if (*scan == *match) scan++;
1236 len = (MAX_MATCH - 1) - (int)(strend-scan);
1237 scan = strend - (MAX_MATCH-1);
1239 #else /* UNALIGNED_OK */
1241 if (match[best_len] != scan_end ||
1242 match[best_len-1] != scan_end1 ||
1244 *++match != scan[1]) continue;
1246 /* The check at best_len-1 can be removed because it will be made
1247 * again later. (This heuristic is not always a win.)
1248 * It is not necessary to compare scan[2] and match[2] since they
1249 * are always equal when the other bytes match, given that
1250 * the hash keys are equal and that HASH_BITS >= 8.
1253 Assert(*scan == *match, "match[2]?");
1255 /* We check for insufficient lookahead only every 8th comparison;
1256 * the 256th check will be made at strstart+258.
1259 } while (*++scan == *++match && *++scan == *++match &&
1260 *++scan == *++match && *++scan == *++match &&
1261 *++scan == *++match && *++scan == *++match &&
1262 *++scan == *++match && *++scan == *++match &&
1265 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1267 len = MAX_MATCH - (int)(strend - scan);
1268 scan = strend - MAX_MATCH;
1270 #endif /* UNALIGNED_OK */
1272 if (len > best_len) {
1273 s->match_start = cur_match;
1275 if (len >= nice_match) break;
1277 scan_end = *(ushf*)(scan+best_len-1);
1279 scan_end1 = scan[best_len-1];
1280 scan_end = scan[best_len];
1283 } while ((cur_match = prev[cur_match & wmask]) > limit
1284 && --chain_length != 0);
1286 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1287 return s->lookahead;
1293 /* ---------------------------------------------------------------------------
1294 * Optimized version for FASTEST only
1296 local uInt longest_match(s, cur_match)
1298 IPos cur_match; /* current match */
1300 register Bytef *scan = s->window + s->strstart; /* current string */
1301 register Bytef *match; /* matched string */
1302 register int len; /* length of current match */
1303 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1305 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1306 * It is easy to get rid of this optimization if necessary.
1308 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1310 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1312 Assert(cur_match < s->strstart, "no future");
1314 match = s->window + cur_match;
1316 /* Return failure if the match length is less than 2:
1318 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1320 /* The check at best_len-1 can be removed because it will be made
1321 * again later. (This heuristic is not always a win.)
1322 * It is not necessary to compare scan[2] and match[2] since they
1323 * are always equal when the other bytes match, given that
1324 * the hash keys are equal and that HASH_BITS >= 8.
1326 scan += 2, match += 2;
1327 Assert(*scan == *match, "match[2]?");
1329 /* We check for insufficient lookahead only every 8th comparison;
1330 * the 256th check will be made at strstart+258.
1333 } while (*++scan == *++match && *++scan == *++match &&
1334 *++scan == *++match && *++scan == *++match &&
1335 *++scan == *++match && *++scan == *++match &&
1336 *++scan == *++match && *++scan == *++match &&
1339 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1341 len = MAX_MATCH - (int)(strend - scan);
1343 if (len < MIN_MATCH) return MIN_MATCH - 1;
1345 s->match_start = cur_match;
1346 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1349 #endif /* FASTEST */
1352 /* ===========================================================================
1353 * Check that the match at match_start is indeed a match.
1355 local void check_match(s, start, match, length)
1360 /* check that the match is indeed a match */
1361 if (zmemcmp(s->window + match,
1362 s->window + start, length) != EQUAL) {
1363 fprintf(stderr, " start %u, match %u, length %d\n",
1364 start, match, length);
1366 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1367 } while (--length != 0);
1368 z_error("invalid match");
1370 if (z_verbose > 1) {
1371 fprintf(stderr,"\\[%d,%d]", start-match, length);
1372 do { putc(s->window[start++], stderr); } while (--length != 0);
1376 # define check_match(s, start, match, length)
1377 #endif /* ZLIB_DEBUG */
1379 /* ===========================================================================
1380 * Fill the window when the lookahead becomes insufficient.
1381 * Updates strstart and lookahead.
1383 * IN assertion: lookahead < MIN_LOOKAHEAD
1384 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1385 * At least one byte has been read, or avail_in == 0; reads are
1386 * performed for at least two bytes (required for the zip translate_eol
1387 * option -- not supported here).
1389 local void fill_window(s)
1392 register unsigned n, m;
1394 unsigned more; /* Amount of free space at the end of the window. */
1395 uInt wsize = s->w_size;
1397 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1400 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1402 /* Deal with !@#$% 64K limit: */
1403 if (sizeof(int) <= 2) {
1404 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1407 } else if (more == (unsigned)(-1)) {
1408 /* Very unlikely, but possible on 16 bit machine if
1409 * strstart == 0 && lookahead == 1 (input done a byte at time)
1415 /* If the window is almost full and there is insufficient lookahead,
1416 * move the upper half to the lower one to make room in the upper half.
1418 if (s->strstart >= wsize+MAX_DIST(s)) {
1420 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1421 s->match_start -= wsize;
1422 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1423 s->block_start -= (long) wsize;
1425 /* Slide the hash table (could be avoided with 32 bit values
1426 at the expense of memory usage). We slide even when level == 0
1427 to keep the hash table consistent if we switch back to level > 0
1428 later. (Using level 0 permanently is not an optimal usage of
1429 zlib, so we don't care about this pathological case.)
1435 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1443 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1444 /* If n is not on any hash chain, prev[n] is garbage but
1445 * its value will never be used.
1451 if (s->strm->avail_in == 0) break;
1453 /* If there was no sliding:
1454 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1455 * more == window_size - lookahead - strstart
1456 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1457 * => more >= window_size - 2*WSIZE + 2
1458 * In the BIG_MEM or MMAP case (not yet supported),
1459 * window_size == input_size + MIN_LOOKAHEAD &&
1460 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1461 * Otherwise, window_size == 2*WSIZE so more >= 2.
1462 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1464 Assert(more >= 2, "more < 2");
1466 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1469 /* Initialize the hash value now that we have some input: */
1470 if (s->lookahead + s->insert >= MIN_MATCH) {
1471 uInt str = s->strstart - s->insert;
1472 s->ins_h = s->window[str];
1473 UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1475 Call UPDATE_HASH() MIN_MATCH-3 more times
1478 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1480 s->prev[str & s->w_mask] = s->head[s->ins_h];
1482 s->head[s->ins_h] = (Pos)str;
1485 if (s->lookahead + s->insert < MIN_MATCH)
1489 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1490 * but this is not important since only literal bytes will be emitted.
1493 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1495 /* If the WIN_INIT bytes after the end of the current data have never been
1496 * written, then zero those bytes in order to avoid memory check reports of
1497 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1498 * the longest match routines. Update the high water mark for the next
1499 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1500 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1502 if (s->high_water < s->window_size) {
1503 ulg curr = s->strstart + (ulg)(s->lookahead);
1506 if (s->high_water < curr) {
1507 /* Previous high water mark below current data -- zero WIN_INIT
1508 * bytes or up to end of window, whichever is less.
1510 init = s->window_size - curr;
1511 if (init > WIN_INIT)
1513 zmemzero(s->window + curr, (unsigned)init);
1514 s->high_water = curr + init;
1516 else if (s->high_water < (ulg)curr + WIN_INIT) {
1517 /* High water mark at or above current data, but below current data
1518 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1519 * to end of window, whichever is less.
1521 init = (ulg)curr + WIN_INIT - s->high_water;
1522 if (init > s->window_size - s->high_water)
1523 init = s->window_size - s->high_water;
1524 zmemzero(s->window + s->high_water, (unsigned)init);
1525 s->high_water += init;
1529 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1530 "not enough room for search");
1533 /* ===========================================================================
1534 * Flush the current block, with given end-of-file flag.
1535 * IN assertion: strstart is set to the end of the current match.
1537 #define FLUSH_BLOCK_ONLY(s, last) { \
1538 _tr_flush_block(s, (s->block_start >= 0L ? \
1539 (charf *)&s->window[(unsigned)s->block_start] : \
1541 (ulg)((long)s->strstart - s->block_start), \
1543 s->block_start = s->strstart; \
1544 flush_pending(s->strm); \
1545 Tracev((stderr,"[FLUSH]")); \
1548 /* Same but force premature exit if necessary. */
1549 #define FLUSH_BLOCK(s, last) { \
1550 FLUSH_BLOCK_ONLY(s, last); \
1551 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1554 /* ===========================================================================
1555 * Copy without compression as much as possible from the input stream, return
1556 * the current block state.
1557 * This function does not insert new strings in the dictionary since
1558 * uncompressible data is probably not useful. This function is used
1559 * only for the level=0 compression option.
1560 * NOTE: this function should be optimized to avoid extra copying from
1561 * window to pending_buf.
1563 local block_state deflate_stored(s, flush)
1567 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1568 * to pending_buf_size, and each stored block has a 5 byte header:
1570 ulg max_block_size = 0xffff;
1573 if (max_block_size > s->pending_buf_size - 5) {
1574 max_block_size = s->pending_buf_size - 5;
1577 /* Copy as much as possible from input to output: */
1579 /* Fill the window as much as possible: */
1580 if (s->lookahead <= 1) {
1582 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1583 s->block_start >= (long)s->w_size, "slide too late");
1586 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1588 if (s->lookahead == 0) break; /* flush the current block */
1590 Assert(s->block_start >= 0L, "block gone");
1592 s->strstart += s->lookahead;
1595 /* Emit a stored block if pending_buf will be full: */
1596 max_start = s->block_start + max_block_size;
1597 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1598 /* strstart == 0 is possible when wraparound on 16-bit machine */
1599 s->lookahead = (uInt)(s->strstart - max_start);
1600 s->strstart = (uInt)max_start;
1603 /* Flush if we may have to slide, otherwise block_start may become
1604 * negative and the data will be gone:
1606 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1611 if (flush == Z_FINISH) {
1615 if ((long)s->strstart > s->block_start)
1620 /* ===========================================================================
1621 * Compress as much as possible from the input stream, return the current
1623 * This function does not perform lazy evaluation of matches and inserts
1624 * new strings in the dictionary only for unmatched strings or for short
1625 * matches. It is used only for the fast compression options.
1627 local block_state deflate_fast(s, flush)
1631 IPos hash_head; /* head of the hash chain */
1632 int bflush; /* set if current block must be flushed */
1635 /* Make sure that we always have enough lookahead, except
1636 * at the end of the input file. We need MAX_MATCH bytes
1637 * for the next match, plus MIN_MATCH bytes to insert the
1638 * string following the next match.
1640 if (s->lookahead < MIN_LOOKAHEAD) {
1642 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1645 if (s->lookahead == 0) break; /* flush the current block */
1648 /* Insert the string window[strstart .. strstart+2] in the
1649 * dictionary, and set hash_head to the head of the hash chain:
1652 if (s->lookahead >= MIN_MATCH) {
1653 INSERT_STRING(s, s->strstart, hash_head);
1656 /* Find the longest match, discarding those <= prev_length.
1657 * At this point we have always match_length < MIN_MATCH
1659 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1660 /* To simplify the code, we prevent matches with the string
1661 * of window index 0 (in particular we have to avoid a match
1662 * of the string with itself at the start of the input file).
1664 s->match_length = longest_match (s, hash_head);
1665 /* longest_match() sets match_start */
1667 if (s->match_length >= MIN_MATCH) {
1668 check_match(s, s->strstart, s->match_start, s->match_length);
1670 _tr_tally_dist(s, s->strstart - s->match_start,
1671 s->match_length - MIN_MATCH, bflush);
1673 s->lookahead -= s->match_length;
1675 /* Insert new strings in the hash table only if the match length
1676 * is not too large. This saves time but degrades compression.
1679 if (s->match_length <= s->max_insert_length &&
1680 s->lookahead >= MIN_MATCH) {
1681 s->match_length--; /* string at strstart already in table */
1684 INSERT_STRING(s, s->strstart, hash_head);
1685 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1686 * always MIN_MATCH bytes ahead.
1688 } while (--s->match_length != 0);
1693 s->strstart += s->match_length;
1694 s->match_length = 0;
1695 s->ins_h = s->window[s->strstart];
1696 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1698 Call UPDATE_HASH() MIN_MATCH-3 more times
1700 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1701 * matter since it will be recomputed at next deflate call.
1705 /* No match, output a literal byte */
1706 Tracevv((stderr,"%c", s->window[s->strstart]));
1707 _tr_tally_lit (s, s->window[s->strstart], bflush);
1711 if (bflush) FLUSH_BLOCK(s, 0);
1713 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1714 if (flush == Z_FINISH) {
1724 /* ===========================================================================
1725 * Same as above, but achieves better compression. We use a lazy
1726 * evaluation for matches: a match is finally adopted only if there is
1727 * no better match at the next window position.
1729 local block_state deflate_slow(s, flush)
1733 IPos hash_head; /* head of hash chain */
1734 int bflush; /* set if current block must be flushed */
1736 /* Process the input block. */
1738 /* Make sure that we always have enough lookahead, except
1739 * at the end of the input file. We need MAX_MATCH bytes
1740 * for the next match, plus MIN_MATCH bytes to insert the
1741 * string following the next match.
1743 if (s->lookahead < MIN_LOOKAHEAD) {
1745 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1748 if (s->lookahead == 0) break; /* flush the current block */
1751 /* Insert the string window[strstart .. strstart+2] in the
1752 * dictionary, and set hash_head to the head of the hash chain:
1755 if (s->lookahead >= MIN_MATCH) {
1756 INSERT_STRING(s, s->strstart, hash_head);
1759 /* Find the longest match, discarding those <= prev_length.
1761 s->prev_length = s->match_length, s->prev_match = s->match_start;
1762 s->match_length = MIN_MATCH-1;
1764 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1765 s->strstart - hash_head <= MAX_DIST(s)) {
1766 /* To simplify the code, we prevent matches with the string
1767 * of window index 0 (in particular we have to avoid a match
1768 * of the string with itself at the start of the input file).
1770 s->match_length = longest_match (s, hash_head);
1771 /* longest_match() sets match_start */
1773 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1774 #if TOO_FAR <= 32767
1775 || (s->match_length == MIN_MATCH &&
1776 s->strstart - s->match_start > TOO_FAR)
1780 /* If prev_match is also MIN_MATCH, match_start is garbage
1781 * but we will ignore the current match anyway.
1783 s->match_length = MIN_MATCH-1;
1786 /* If there was a match at the previous step and the current
1787 * match is not better, output the previous match:
1789 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1790 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1791 /* Do not insert strings in hash table beyond this. */
1793 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1795 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1796 s->prev_length - MIN_MATCH, bflush);
1798 /* Insert in hash table all strings up to the end of the match.
1799 * strstart-1 and strstart are already inserted. If there is not
1800 * enough lookahead, the last two strings are not inserted in
1803 s->lookahead -= s->prev_length-1;
1804 s->prev_length -= 2;
1806 if (++s->strstart <= max_insert) {
1807 INSERT_STRING(s, s->strstart, hash_head);
1809 } while (--s->prev_length != 0);
1810 s->match_available = 0;
1811 s->match_length = MIN_MATCH-1;
1814 if (bflush) FLUSH_BLOCK(s, 0);
1816 } else if (s->match_available) {
1817 /* If there was no match at the previous position, output a
1818 * single literal. If there was a match but the current match
1819 * is longer, truncate the previous match to a single literal.
1821 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1822 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1824 FLUSH_BLOCK_ONLY(s, 0);
1828 if (s->strm->avail_out == 0) return need_more;
1830 /* There is no previous match to compare with, wait for
1831 * the next step to decide.
1833 s->match_available = 1;
1838 Assert (flush != Z_NO_FLUSH, "no flush?");
1839 if (s->match_available) {
1840 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1841 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1842 s->match_available = 0;
1844 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1845 if (flush == Z_FINISH) {
1853 #endif /* FASTEST */
1855 /* ===========================================================================
1856 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1857 * one. Do not maintain a hash table. (It will be regenerated if this run of
1858 * deflate switches away from Z_RLE.)
1860 local block_state deflate_rle(s, flush)
1864 int bflush; /* set if current block must be flushed */
1865 uInt prev; /* byte at distance one to match */
1866 Bytef *scan, *strend; /* scan goes up to strend for length of run */
1869 /* Make sure that we always have enough lookahead, except
1870 * at the end of the input file. We need MAX_MATCH bytes
1871 * for the longest run, plus one for the unrolled loop.
1873 if (s->lookahead <= MAX_MATCH) {
1875 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
1878 if (s->lookahead == 0) break; /* flush the current block */
1881 /* See how many times the previous byte repeats */
1882 s->match_length = 0;
1883 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
1884 scan = s->window + s->strstart - 1;
1886 if (prev == *++scan && prev == *++scan && prev == *++scan) {
1887 strend = s->window + s->strstart + MAX_MATCH;
1889 } while (prev == *++scan && prev == *++scan &&
1890 prev == *++scan && prev == *++scan &&
1891 prev == *++scan && prev == *++scan &&
1892 prev == *++scan && prev == *++scan &&
1894 s->match_length = MAX_MATCH - (int)(strend - scan);
1895 if (s->match_length > s->lookahead)
1896 s->match_length = s->lookahead;
1898 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
1901 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1902 if (s->match_length >= MIN_MATCH) {
1903 check_match(s, s->strstart, s->strstart - 1, s->match_length);
1905 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
1907 s->lookahead -= s->match_length;
1908 s->strstart += s->match_length;
1909 s->match_length = 0;
1911 /* No match, output a literal byte */
1912 Tracevv((stderr,"%c", s->window[s->strstart]));
1913 _tr_tally_lit (s, s->window[s->strstart], bflush);
1917 if (bflush) FLUSH_BLOCK(s, 0);
1920 if (flush == Z_FINISH) {
1929 /* ===========================================================================
1930 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
1931 * (It will be regenerated if this run of deflate switches away from Huffman.)
1933 local block_state deflate_huff(s, flush)
1937 int bflush; /* set if current block must be flushed */
1940 /* Make sure that we have a literal to write. */
1941 if (s->lookahead == 0) {
1943 if (s->lookahead == 0) {
1944 if (flush == Z_NO_FLUSH)
1946 break; /* flush the current block */
1950 /* Output a literal byte */
1951 s->match_length = 0;
1952 Tracevv((stderr,"%c", s->window[s->strstart]));
1953 _tr_tally_lit (s, s->window[s->strstart], bflush);
1956 if (bflush) FLUSH_BLOCK(s, 0);
1959 if (flush == Z_FINISH) {