1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2017 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 1.2.11.1 Copyright 1995-2017 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 slide_hash OF((deflate_state *s));
78 local void fill_window OF((deflate_state *s));
79 local block_state deflate_stored OF((deflate_state *s, int flush));
80 local block_state deflate_fast OF((deflate_state *s, int flush));
82 local block_state deflate_slow OF((deflate_state *s, int flush));
84 local block_state deflate_rle OF((deflate_state *s, int flush));
85 local block_state deflate_huff OF((deflate_state *s, int flush));
86 local void lm_init OF((deflate_state *s));
87 local void putShortMSB OF((deflate_state *s, uInt b));
88 local void flush_pending OF((z_streamp strm));
89 local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
91 # pragma message("Assembler code may have bugs -- use at your own risk")
92 void match_init OF((void)); /* asm code initialization */
93 uInt longest_match OF((deflate_state *s, IPos cur_match));
95 local uInt longest_match OF((deflate_state *s, IPos cur_match));
99 local void check_match OF((deflate_state *s, IPos start, IPos match,
103 /* ===========================================================================
108 /* Tail of hash chains */
111 # define TOO_FAR 4096
113 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
115 /* Values for max_lazy_match, good_match and max_chain_length, depending on
116 * the desired pack level (0..9). The values given below have been tuned to
117 * exclude worst case performance for pathological files. Better values may be
118 * found for specific files.
120 typedef struct config_s {
121 ush good_length; /* reduce lazy search above this match length */
122 ush max_lazy; /* do not perform lazy search above this match length */
123 ush nice_length; /* quit search above this match length */
129 local const config configuration_table[2] = {
130 /* good lazy nice chain */
131 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
132 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
134 local const config configuration_table[10] = {
135 /* good lazy nice chain */
136 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
137 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
138 /* 2 */ {4, 5, 16, 8, deflate_fast},
139 /* 3 */ {4, 6, 32, 32, deflate_fast},
141 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
142 /* 5 */ {8, 16, 32, 32, deflate_slow},
143 /* 6 */ {8, 16, 128, 128, deflate_slow},
144 /* 7 */ {8, 32, 128, 256, deflate_slow},
145 /* 8 */ {32, 128, 258, 1024, deflate_slow},
146 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
149 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
150 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
154 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
155 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
157 /* ===========================================================================
158 * Update a hash value with the given input byte
159 * IN assertion: all calls to UPDATE_HASH are made with consecutive input
160 * characters, so that a running hash key can be computed from the previous
161 * key instead of complete recalculation each time.
163 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
166 /* ===========================================================================
167 * Insert string str in the dictionary and set match_head to the previous head
168 * of the hash chain (the most recent string with same hash key). Return
169 * the previous length of the hash chain.
170 * If this file is compiled with -DFASTEST, the compression level is forced
171 * to 1, and no hash chains are maintained.
172 * IN assertion: all calls to INSERT_STRING are made with consecutive input
173 * characters and the first MIN_MATCH bytes of str are valid (except for
174 * the last MIN_MATCH-1 bytes of the input file).
177 #define INSERT_STRING(s, str, match_head) \
178 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
179 match_head = s->head[s->ins_h], \
180 s->head[s->ins_h] = (Pos)(str))
182 #define INSERT_STRING(s, str, match_head) \
183 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
184 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
185 s->head[s->ins_h] = (Pos)(str))
188 /* ===========================================================================
189 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
190 * prev[] will be initialized on the fly.
192 #define CLEAR_HASH(s) \
194 s->head[s->hash_size-1] = NIL; \
195 zmemzero((Bytef *)s->head, \
196 (unsigned)(s->hash_size-1)*sizeof(*s->head)); \
199 /* ===========================================================================
200 * Slide the hash table when sliding the window down (could be avoided with 32
201 * bit values at the expense of memory usage). We slide even when level == 0 to
202 * keep the hash table consistent if we switch back to level > 0 later.
204 local void slide_hash(s)
209 uInt wsize = s->w_size;
215 *p = (Pos)(m >= wsize ? m - wsize : NIL);
222 *p = (Pos)(m >= wsize ? m - wsize : NIL);
223 /* If n is not on any hash chain, prev[n] is garbage but
224 * its value will never be used.
230 /* ========================================================================= */
231 int ZEXPORT deflateInit_(strm, level, version, stream_size)
237 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
238 Z_DEFAULT_STRATEGY, version, stream_size);
239 /* To do: ignore strm->next_in if we use it as window */
242 /* ========================================================================= */
243 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
244 version, stream_size)
256 static const char my_version[] = ZLIB_VERSION;
259 /* We overlay pending_buf and d_buf+l_buf. This works since the average
260 * output size for (length,distance) codes is <= 24 bits.
263 if (version == Z_NULL || version[0] != my_version[0] ||
264 stream_size != sizeof(z_stream)) {
265 return Z_VERSION_ERROR;
267 if (strm == Z_NULL) return Z_STREAM_ERROR;
270 if (strm->zalloc == (alloc_func)0) {
272 return Z_STREAM_ERROR;
274 strm->zalloc = zcalloc;
275 strm->opaque = (voidpf)0;
278 if (strm->zfree == (free_func)0)
280 return Z_STREAM_ERROR;
282 strm->zfree = zcfree;
286 if (level != 0) level = 1;
288 if (level == Z_DEFAULT_COMPRESSION) level = 6;
291 if (windowBits < 0) { /* suppress zlib wrapper */
293 windowBits = -windowBits;
296 else if (windowBits > 15) {
297 wrap = 2; /* write gzip wrapper instead */
301 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
302 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
303 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
304 return Z_STREAM_ERROR;
306 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
307 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
308 if (s == Z_NULL) return Z_MEM_ERROR;
309 strm->state = (struct internal_state FAR *)s;
311 s->status = INIT_STATE; /* to pass state test in deflateReset() */
315 s->w_bits = (uInt)windowBits;
316 s->w_size = 1 << s->w_bits;
317 s->w_mask = s->w_size - 1;
319 s->hash_bits = (uInt)memLevel + 7;
320 s->hash_size = 1 << s->hash_bits;
321 s->hash_mask = s->hash_size - 1;
322 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
324 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
325 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
326 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
328 s->high_water = 0; /* nothing written to s->window yet */
330 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
332 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
333 s->pending_buf = (uchf *) overlay;
334 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
336 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
337 s->pending_buf == Z_NULL) {
338 s->status = FINISH_STATE;
339 strm->msg = ERR_MSG(Z_MEM_ERROR);
343 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
344 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
347 s->strategy = strategy;
348 s->method = (Byte)method;
350 return deflateReset(strm);
353 /* =========================================================================
354 * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
356 local int deflateStateCheck (strm)
360 if (strm == Z_NULL ||
361 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
364 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
366 s->status != GZIP_STATE &&
368 s->status != EXTRA_STATE &&
369 s->status != NAME_STATE &&
370 s->status != COMMENT_STATE &&
371 s->status != HCRC_STATE &&
372 s->status != BUSY_STATE &&
373 s->status != FINISH_STATE))
378 /* ========================================================================= */
379 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
381 const Bytef *dictionary;
388 z_const unsigned char *next;
390 if (deflateStateCheck(strm) || dictionary == Z_NULL)
391 return Z_STREAM_ERROR;
394 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
395 return Z_STREAM_ERROR;
397 /* when using zlib wrappers, compute Adler-32 for provided dictionary */
399 strm->adler = adler32(strm->adler, dictionary, dictLength);
400 s->wrap = 0; /* avoid computing Adler-32 in read_buf */
402 /* if dictionary would fill window, just replace the history */
403 if (dictLength >= s->w_size) {
404 if (wrap == 0) { /* already empty otherwise */
410 dictionary += dictLength - s->w_size; /* use the tail */
411 dictLength = s->w_size;
414 /* insert dictionary into window and hash */
415 avail = strm->avail_in;
416 next = strm->next_in;
417 strm->avail_in = dictLength;
418 strm->next_in = (z_const Bytef *)dictionary;
420 while (s->lookahead >= MIN_MATCH) {
422 n = s->lookahead - (MIN_MATCH-1);
424 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
426 s->prev[str & s->w_mask] = s->head[s->ins_h];
428 s->head[s->ins_h] = (Pos)str;
432 s->lookahead = MIN_MATCH-1;
435 s->strstart += s->lookahead;
436 s->block_start = (long)s->strstart;
437 s->insert = s->lookahead;
439 s->match_length = s->prev_length = MIN_MATCH-1;
440 s->match_available = 0;
441 strm->next_in = next;
442 strm->avail_in = avail;
447 /* ========================================================================= */
448 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
456 if (deflateStateCheck(strm))
457 return Z_STREAM_ERROR;
459 len = s->strstart + s->lookahead;
462 if (dictionary != Z_NULL && len)
463 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
464 if (dictLength != Z_NULL)
469 /* ========================================================================= */
470 int ZEXPORT deflateResetKeep (strm)
475 if (deflateStateCheck(strm)) {
476 return Z_STREAM_ERROR;
479 strm->total_in = strm->total_out = 0;
480 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
481 strm->data_type = Z_UNKNOWN;
483 s = (deflate_state *)strm->state;
485 s->pending_out = s->pending_buf;
488 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
492 s->wrap == 2 ? GZIP_STATE :
497 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
499 adler32(0L, Z_NULL, 0);
507 /* ========================================================================= */
508 int ZEXPORT deflateReset (strm)
513 ret = deflateResetKeep(strm);
515 lm_init(strm->state);
519 /* ========================================================================= */
520 int ZEXPORT deflateSetHeader (strm, head)
524 if (deflateStateCheck(strm) || strm->state->wrap != 2)
525 return Z_STREAM_ERROR;
526 strm->state->gzhead = head;
530 /* ========================================================================= */
531 int ZEXPORT deflatePending (strm, pending, bits)
536 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
537 if (pending != Z_NULL)
538 *pending = strm->state->pending;
540 *bits = strm->state->bi_valid;
544 /* ========================================================================= */
545 int ZEXPORT deflatePrime (strm, bits, value)
553 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
555 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
558 put = Buf_size - s->bi_valid;
561 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
570 /* ========================================================================= */
571 int ZEXPORT deflateParams(strm, level, strategy)
579 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
583 if (level != 0) level = 1;
585 if (level == Z_DEFAULT_COMPRESSION) level = 6;
587 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
588 return Z_STREAM_ERROR;
590 func = configuration_table[s->level].func;
592 if ((strategy != s->strategy || func != configuration_table[level].func) &&
593 s->last_flush != -2) {
594 /* Flush the last buffer: */
595 int err = deflate(strm, Z_BLOCK);
596 if (err == Z_STREAM_ERROR)
598 if (strm->avail_in || (s->strstart - s->block_start) + s->lookahead)
601 if (s->level != level) {
602 if (s->level == 0 && s->matches != 0) {
610 s->max_lazy_match = configuration_table[level].max_lazy;
611 s->good_match = configuration_table[level].good_length;
612 s->nice_match = configuration_table[level].nice_length;
613 s->max_chain_length = configuration_table[level].max_chain;
615 s->strategy = strategy;
619 /* ========================================================================= */
620 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
629 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
631 s->good_match = (uInt)good_length;
632 s->max_lazy_match = (uInt)max_lazy;
633 s->nice_match = nice_length;
634 s->max_chain_length = (uInt)max_chain;
638 /* =========================================================================
639 * For the default windowBits of 15 and memLevel of 8, this function returns
640 * a close to exact, as well as small, upper bound on the compressed size.
641 * They are coded as constants here for a reason--if the #define's are
642 * changed, then this function needs to be changed as well. The return
643 * value for 15 and 8 only works for those exact settings.
645 * For any setting other than those defaults for windowBits and memLevel,
646 * the value returned is a conservative worst case for the maximum expansion
647 * resulting from using fixed blocks instead of stored blocks, which deflate
648 * can emit on compressed data for some combinations of the parameters.
650 * This function could be more sophisticated to provide closer upper bounds for
651 * every combination of windowBits and memLevel. But even the conservative
652 * upper bound of about 14% expansion does not seem onerous for output buffer
655 uLong ZEXPORT deflateBound(strm, sourceLen)
660 uLong complen, wraplen;
662 /* conservative upper bound for compressed data */
663 complen = sourceLen +
664 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
666 /* if can't get parameters, return conservative bound plus zlib wrapper */
667 if (deflateStateCheck(strm))
670 /* compute wrapper length */
673 case 0: /* raw deflate */
676 case 1: /* zlib wrapper */
677 wraplen = 6 + (s->strstart ? 4 : 0);
680 case 2: /* gzip wrapper */
682 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
684 if (s->gzhead->extra != Z_NULL)
685 wraplen += 2 + s->gzhead->extra_len;
686 str = s->gzhead->name;
691 str = s->gzhead->comment;
701 default: /* for compiler happiness */
705 /* if not default parameters, return conservative bound */
706 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
707 return complen + wraplen;
709 /* default settings: return tight bound for that case */
710 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
711 (sourceLen >> 25) + 13 - 6 + wraplen;
714 /* =========================================================================
715 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
716 * IN assertion: the stream state is correct and there is enough room in
719 local void putShortMSB (s, b)
723 put_byte(s, (Byte)(b >> 8));
724 put_byte(s, (Byte)(b & 0xff));
727 /* =========================================================================
728 * Flush as much pending output as possible. All deflate() output, except for
729 * some deflate_stored() output, goes through this function so some
730 * applications may wish to modify it to avoid allocating a large
731 * strm->next_out buffer and copying into it. (See also read_buf()).
733 local void flush_pending(strm)
737 deflate_state *s = strm->state;
741 if (len > strm->avail_out) len = strm->avail_out;
742 if (len == 0) return;
744 zmemcpy(strm->next_out, s->pending_out, len);
745 strm->next_out += len;
746 s->pending_out += len;
747 strm->total_out += len;
748 strm->avail_out -= len;
750 if (s->pending == 0) {
751 s->pending_out = s->pending_buf;
755 /* ===========================================================================
756 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
758 #define HCRC_UPDATE(beg) \
760 if (s->gzhead->hcrc && s->pending > (beg)) \
761 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
762 s->pending - (beg)); \
765 /* ========================================================================= */
766 int ZEXPORT deflate (strm, flush)
770 int old_flush; /* value of flush param for previous deflate call */
773 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
774 return Z_STREAM_ERROR;
778 if (strm->next_out == Z_NULL ||
779 (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
780 (s->status == FINISH_STATE && flush != Z_FINISH)) {
781 ERR_RETURN(strm, Z_STREAM_ERROR);
783 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
785 old_flush = s->last_flush;
786 s->last_flush = flush;
788 /* Flush as much pending output as possible */
789 if (s->pending != 0) {
791 if (strm->avail_out == 0) {
792 /* Since avail_out is 0, deflate will be called again with
793 * more output space, but possibly with both pending and
794 * avail_in equal to zero. There won't be anything to do,
795 * but this is not an error situation so make sure we
796 * return OK instead of BUF_ERROR at next call of deflate:
802 /* Make sure there is something to do and avoid duplicate consecutive
803 * flushes. For repeated and useless calls with Z_FINISH, we keep
804 * returning Z_STREAM_END instead of Z_BUF_ERROR.
806 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
808 ERR_RETURN(strm, Z_BUF_ERROR);
811 /* User must not provide more input after the first FINISH: */
812 if (s->status == FINISH_STATE && strm->avail_in != 0) {
813 ERR_RETURN(strm, Z_BUF_ERROR);
816 /* Write the header */
817 if (s->status == INIT_STATE && s->wrap == 0)
818 s->status = BUSY_STATE;
819 if (s->status == INIT_STATE) {
821 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
824 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
826 else if (s->level < 6)
828 else if (s->level == 6)
832 header |= (level_flags << 6);
833 if (s->strstart != 0) header |= PRESET_DICT;
834 header += 31 - (header % 31);
836 putShortMSB(s, header);
838 /* Save the adler32 of the preset dictionary: */
839 if (s->strstart != 0) {
840 putShortMSB(s, (uInt)(strm->adler >> 16));
841 putShortMSB(s, (uInt)(strm->adler & 0xffff));
843 strm->adler = adler32(0L, Z_NULL, 0);
844 s->status = BUSY_STATE;
846 /* Compression must start with an empty pending buffer */
848 if (s->pending != 0) {
854 if (s->status == GZIP_STATE) {
856 strm->adler = crc32(0L, Z_NULL, 0);
860 if (s->gzhead == Z_NULL) {
866 put_byte(s, s->level == 9 ? 2 :
867 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
869 put_byte(s, OS_CODE);
870 s->status = BUSY_STATE;
872 /* Compression must start with an empty pending buffer */
874 if (s->pending != 0) {
880 put_byte(s, (s->gzhead->text ? 1 : 0) +
881 (s->gzhead->hcrc ? 2 : 0) +
882 (s->gzhead->extra == Z_NULL ? 0 : 4) +
883 (s->gzhead->name == Z_NULL ? 0 : 8) +
884 (s->gzhead->comment == Z_NULL ? 0 : 16)
886 put_byte(s, (Byte)(s->gzhead->time & 0xff));
887 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
888 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
889 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
890 put_byte(s, s->level == 9 ? 2 :
891 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
893 put_byte(s, s->gzhead->os & 0xff);
894 if (s->gzhead->extra != Z_NULL) {
895 put_byte(s, s->gzhead->extra_len & 0xff);
896 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
899 strm->adler = crc32(strm->adler, s->pending_buf,
902 s->status = EXTRA_STATE;
905 if (s->status == EXTRA_STATE) {
906 if (s->gzhead->extra != Z_NULL) {
907 ulg beg = s->pending; /* start of bytes to update crc */
908 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
909 while (s->pending + left > s->pending_buf_size) {
910 uInt copy = s->pending_buf_size - s->pending;
911 zmemcpy(s->pending_buf + s->pending,
912 s->gzhead->extra + s->gzindex, copy);
913 s->pending = s->pending_buf_size;
917 if (s->pending != 0) {
924 zmemcpy(s->pending_buf + s->pending,
925 s->gzhead->extra + s->gzindex, left);
930 s->status = NAME_STATE;
932 if (s->status == NAME_STATE) {
933 if (s->gzhead->name != Z_NULL) {
934 ulg beg = s->pending; /* start of bytes to update crc */
937 if (s->pending == s->pending_buf_size) {
940 if (s->pending != 0) {
946 val = s->gzhead->name[s->gzindex++];
952 s->status = COMMENT_STATE;
954 if (s->status == COMMENT_STATE) {
955 if (s->gzhead->comment != Z_NULL) {
956 ulg beg = s->pending; /* start of bytes to update crc */
959 if (s->pending == s->pending_buf_size) {
962 if (s->pending != 0) {
968 val = s->gzhead->comment[s->gzindex++];
973 s->status = HCRC_STATE;
975 if (s->status == HCRC_STATE) {
976 if (s->gzhead->hcrc) {
977 if (s->pending + 2 > s->pending_buf_size) {
979 if (s->pending != 0) {
984 put_byte(s, (Byte)(strm->adler & 0xff));
985 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
986 strm->adler = crc32(0L, Z_NULL, 0);
988 s->status = BUSY_STATE;
990 /* Compression must start with an empty pending buffer */
992 if (s->pending != 0) {
999 /* Start a new block or continue the current one.
1001 if (strm->avail_in != 0 || s->lookahead != 0 ||
1002 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
1005 bstate = s->level == 0 ? deflate_stored(s, flush) :
1006 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
1007 s->strategy == Z_RLE ? deflate_rle(s, flush) :
1008 (*(configuration_table[s->level].func))(s, flush);
1010 if (bstate == finish_started || bstate == finish_done) {
1011 s->status = FINISH_STATE;
1013 if (bstate == need_more || bstate == finish_started) {
1014 if (strm->avail_out == 0) {
1015 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1018 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1019 * of deflate should use the same flush parameter to make sure
1020 * that the flush is complete. So we don't have to output an
1021 * empty block here, this will be done at next call. This also
1022 * ensures that for a very small output buffer, we emit at most
1026 if (bstate == block_done) {
1027 if (flush == Z_PARTIAL_FLUSH) {
1029 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1030 _tr_stored_block(s, (char*)0, 0L, 0);
1031 /* For a full flush, this empty block will be recognized
1032 * as a special marker by inflate_sync().
1034 if (flush == Z_FULL_FLUSH) {
1035 CLEAR_HASH(s); /* forget history */
1036 if (s->lookahead == 0) {
1038 s->block_start = 0L;
1043 flush_pending(strm);
1044 if (strm->avail_out == 0) {
1045 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1051 if (flush != Z_FINISH) return Z_OK;
1052 if (s->wrap <= 0) return Z_STREAM_END;
1054 /* Write the trailer */
1057 put_byte(s, (Byte)(strm->adler & 0xff));
1058 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1059 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1060 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1061 put_byte(s, (Byte)(strm->total_in & 0xff));
1062 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1063 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1064 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1069 putShortMSB(s, (uInt)(strm->adler >> 16));
1070 putShortMSB(s, (uInt)(strm->adler & 0xffff));
1072 flush_pending(strm);
1073 /* If avail_out is zero, the application will call deflate again
1074 * to flush the rest.
1076 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1077 return s->pending != 0 ? Z_OK : Z_STREAM_END;
1080 /* ========================================================================= */
1081 int ZEXPORT deflateEnd (strm)
1086 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1088 status = strm->state->status;
1090 /* Deallocate in reverse order of allocations: */
1091 TRY_FREE(strm, strm->state->pending_buf);
1092 TRY_FREE(strm, strm->state->head);
1093 TRY_FREE(strm, strm->state->prev);
1094 TRY_FREE(strm, strm->state->window);
1096 ZFREE(strm, strm->state);
1097 strm->state = Z_NULL;
1099 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1102 /* =========================================================================
1103 * Copy the source state to the destination state.
1104 * To simplify the source, this is not supported for 16-bit MSDOS (which
1105 * doesn't have enough memory anyway to duplicate compression states).
1107 int ZEXPORT deflateCopy (dest, source)
1112 return Z_STREAM_ERROR;
1119 if (deflateStateCheck(source) || dest == Z_NULL) {
1120 return Z_STREAM_ERROR;
1125 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1127 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1128 if (ds == Z_NULL) return Z_MEM_ERROR;
1129 dest->state = (struct internal_state FAR *) ds;
1130 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1133 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1134 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
1135 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
1136 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1137 ds->pending_buf = (uchf *) overlay;
1139 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1140 ds->pending_buf == Z_NULL) {
1144 /* following zmemcpy do not work for 16-bit MSDOS */
1145 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1146 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1147 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1148 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1150 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1151 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1152 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1154 ds->l_desc.dyn_tree = ds->dyn_ltree;
1155 ds->d_desc.dyn_tree = ds->dyn_dtree;
1156 ds->bl_desc.dyn_tree = ds->bl_tree;
1159 #endif /* MAXSEG_64K */
1162 /* ===========================================================================
1163 * Read a new buffer from the current input stream, update the adler32
1164 * and total number of bytes read. All deflate() input goes through
1165 * this function so some applications may wish to modify it to avoid
1166 * allocating a large strm->next_in buffer and copying from it.
1167 * (See also flush_pending()).
1169 local unsigned read_buf(strm, buf, size)
1174 unsigned len = strm->avail_in;
1176 if (len > size) len = size;
1177 if (len == 0) return 0;
1179 strm->avail_in -= len;
1181 zmemcpy(buf, strm->next_in, len);
1182 if (strm->state->wrap == 1) {
1183 strm->adler = adler32(strm->adler, buf, len);
1186 else if (strm->state->wrap == 2) {
1187 strm->adler = crc32(strm->adler, buf, len);
1190 strm->next_in += len;
1191 strm->total_in += len;
1196 /* ===========================================================================
1197 * Initialize the "longest match" routines for a new zlib stream
1199 local void lm_init (s)
1202 s->window_size = (ulg)2L*s->w_size;
1206 /* Set the default configuration parameters:
1208 s->max_lazy_match = configuration_table[s->level].max_lazy;
1209 s->good_match = configuration_table[s->level].good_length;
1210 s->nice_match = configuration_table[s->level].nice_length;
1211 s->max_chain_length = configuration_table[s->level].max_chain;
1214 s->block_start = 0L;
1217 s->match_length = s->prev_length = MIN_MATCH-1;
1218 s->match_available = 0;
1222 match_init(); /* initialize the asm code */
1228 /* ===========================================================================
1229 * Set match_start to the longest match starting at the given string and
1230 * return its length. Matches shorter or equal to prev_length are discarded,
1231 * in which case the result is equal to prev_length and match_start is
1233 * IN assertions: cur_match is the head of the hash chain for the current
1234 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1235 * OUT assertion: the match length is not greater than s->lookahead.
1238 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1239 * match.S. The code will be functionally equivalent.
1241 local uInt longest_match(s, cur_match)
1243 IPos cur_match; /* current match */
1245 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1246 register Bytef *scan = s->window + s->strstart; /* current string */
1247 register Bytef *match; /* matched string */
1248 register int len; /* length of current match */
1249 int best_len = (int)s->prev_length; /* best match length so far */
1250 int nice_match = s->nice_match; /* stop if match long enough */
1251 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1252 s->strstart - (IPos)MAX_DIST(s) : NIL;
1253 /* Stop when cur_match becomes <= limit. To simplify the code,
1254 * we prevent matches with the string of window index 0.
1256 Posf *prev = s->prev;
1257 uInt wmask = s->w_mask;
1260 /* Compare two bytes at a time. Note: this is not always beneficial.
1261 * Try with and without -DUNALIGNED_OK to check.
1263 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1264 register ush scan_start = *(ushf*)scan;
1265 register ush scan_end = *(ushf*)(scan+best_len-1);
1267 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1268 register Byte scan_end1 = scan[best_len-1];
1269 register Byte scan_end = scan[best_len];
1272 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1273 * It is easy to get rid of this optimization if necessary.
1275 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1277 /* Do not waste too much time if we already have a good match: */
1278 if (s->prev_length >= s->good_match) {
1281 /* Do not look for matches beyond the end of the input. This is necessary
1282 * to make deflate deterministic.
1284 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1286 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1289 Assert(cur_match < s->strstart, "no future");
1290 match = s->window + cur_match;
1292 /* Skip to next match if the match length cannot increase
1293 * or if the match length is less than 2. Note that the checks below
1294 * for insufficient lookahead only occur occasionally for performance
1295 * reasons. Therefore uninitialized memory will be accessed, and
1296 * conditional jumps will be made that depend on those values.
1297 * However the length of the match is limited to the lookahead, so
1298 * the output of deflate is not affected by the uninitialized values.
1300 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1301 /* This code assumes sizeof(unsigned short) == 2. Do not use
1302 * UNALIGNED_OK if your compiler uses a different size.
1304 if (*(ushf*)(match+best_len-1) != scan_end ||
1305 *(ushf*)match != scan_start) continue;
1307 /* It is not necessary to compare scan[2] and match[2] since they are
1308 * always equal when the other bytes match, given that the hash keys
1309 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1310 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1311 * lookahead only every 4th comparison; the 128th check will be made
1312 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1313 * necessary to put more guard bytes at the end of the window, or
1314 * to check more often for insufficient lookahead.
1316 Assert(scan[2] == match[2], "scan[2]?");
1319 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1320 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1321 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1322 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1324 /* The funny "do {}" generates better code on most compilers */
1326 /* Here, scan <= window+strstart+257 */
1327 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1328 if (*scan == *match) scan++;
1330 len = (MAX_MATCH - 1) - (int)(strend-scan);
1331 scan = strend - (MAX_MATCH-1);
1333 #else /* UNALIGNED_OK */
1335 if (match[best_len] != scan_end ||
1336 match[best_len-1] != scan_end1 ||
1338 *++match != scan[1]) continue;
1340 /* The check at best_len-1 can be removed because it will be made
1341 * again later. (This heuristic is not always a win.)
1342 * It is not necessary to compare scan[2] and match[2] since they
1343 * are always equal when the other bytes match, given that
1344 * the hash keys are equal and that HASH_BITS >= 8.
1347 Assert(*scan == *match, "match[2]?");
1349 /* We check for insufficient lookahead only every 8th comparison;
1350 * the 256th check will be made at strstart+258.
1353 } while (*++scan == *++match && *++scan == *++match &&
1354 *++scan == *++match && *++scan == *++match &&
1355 *++scan == *++match && *++scan == *++match &&
1356 *++scan == *++match && *++scan == *++match &&
1359 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1361 len = MAX_MATCH - (int)(strend - scan);
1362 scan = strend - MAX_MATCH;
1364 #endif /* UNALIGNED_OK */
1366 if (len > best_len) {
1367 s->match_start = cur_match;
1369 if (len >= nice_match) break;
1371 scan_end = *(ushf*)(scan+best_len-1);
1373 scan_end1 = scan[best_len-1];
1374 scan_end = scan[best_len];
1377 } while ((cur_match = prev[cur_match & wmask]) > limit
1378 && --chain_length != 0);
1380 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1381 return s->lookahead;
1387 /* ---------------------------------------------------------------------------
1388 * Optimized version for FASTEST only
1390 local uInt longest_match(s, cur_match)
1392 IPos cur_match; /* current match */
1394 register Bytef *scan = s->window + s->strstart; /* current string */
1395 register Bytef *match; /* matched string */
1396 register int len; /* length of current match */
1397 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1399 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1400 * It is easy to get rid of this optimization if necessary.
1402 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1404 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1406 Assert(cur_match < s->strstart, "no future");
1408 match = s->window + cur_match;
1410 /* Return failure if the match length is less than 2:
1412 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1414 /* The check at best_len-1 can be removed because it will be made
1415 * again later. (This heuristic is not always a win.)
1416 * It is not necessary to compare scan[2] and match[2] since they
1417 * are always equal when the other bytes match, given that
1418 * the hash keys are equal and that HASH_BITS >= 8.
1420 scan += 2, match += 2;
1421 Assert(*scan == *match, "match[2]?");
1423 /* We check for insufficient lookahead only every 8th comparison;
1424 * the 256th check will be made at strstart+258.
1427 } while (*++scan == *++match && *++scan == *++match &&
1428 *++scan == *++match && *++scan == *++match &&
1429 *++scan == *++match && *++scan == *++match &&
1430 *++scan == *++match && *++scan == *++match &&
1433 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1435 len = MAX_MATCH - (int)(strend - scan);
1437 if (len < MIN_MATCH) return MIN_MATCH - 1;
1439 s->match_start = cur_match;
1440 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1443 #endif /* FASTEST */
1448 /* result of memcmp for equal strings */
1450 /* ===========================================================================
1451 * Check that the match at match_start is indeed a match.
1453 local void check_match(s, start, match, length)
1458 /* check that the match is indeed a match */
1459 if (zmemcmp(s->window + match,
1460 s->window + start, length) != EQUAL) {
1461 fprintf(stderr, " start %u, match %u, length %d\n",
1462 start, match, length);
1464 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1465 } while (--length != 0);
1466 z_error("invalid match");
1468 if (z_verbose > 1) {
1469 fprintf(stderr,"\\[%d,%d]", start-match, length);
1470 do { putc(s->window[start++], stderr); } while (--length != 0);
1474 # define check_match(s, start, match, length)
1475 #endif /* ZLIB_DEBUG */
1477 /* ===========================================================================
1478 * Fill the window when the lookahead becomes insufficient.
1479 * Updates strstart and lookahead.
1481 * IN assertion: lookahead < MIN_LOOKAHEAD
1482 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1483 * At least one byte has been read, or avail_in == 0; reads are
1484 * performed for at least two bytes (required for the zip translate_eol
1485 * option -- not supported here).
1487 local void fill_window(s)
1491 unsigned more; /* Amount of free space at the end of the window. */
1492 uInt wsize = s->w_size;
1494 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1497 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1499 /* Deal with !@#$% 64K limit: */
1500 if (sizeof(int) <= 2) {
1501 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1504 } else if (more == (unsigned)(-1)) {
1505 /* Very unlikely, but possible on 16 bit machine if
1506 * strstart == 0 && lookahead == 1 (input done a byte at time)
1512 /* If the window is almost full and there is insufficient lookahead,
1513 * move the upper half to the lower one to make room in the upper half.
1515 if (s->strstart >= wsize+MAX_DIST(s)) {
1517 zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1518 s->match_start -= wsize;
1519 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1520 s->block_start -= (long) wsize;
1521 if (s->insert > s->strstart)
1522 s->insert = s->strstart;
1526 if (s->strm->avail_in == 0) break;
1528 /* If there was no sliding:
1529 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1530 * more == window_size - lookahead - strstart
1531 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1532 * => more >= window_size - 2*WSIZE + 2
1533 * In the BIG_MEM or MMAP case (not yet supported),
1534 * window_size == input_size + MIN_LOOKAHEAD &&
1535 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1536 * Otherwise, window_size == 2*WSIZE so more >= 2.
1537 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1539 Assert(more >= 2, "more < 2");
1541 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1544 /* Initialize the hash value now that we have some input: */
1545 if (s->lookahead + s->insert >= MIN_MATCH) {
1546 uInt str = s->strstart - s->insert;
1547 s->ins_h = s->window[str];
1548 UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1550 Call UPDATE_HASH() MIN_MATCH-3 more times
1553 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1555 s->prev[str & s->w_mask] = s->head[s->ins_h];
1557 s->head[s->ins_h] = (Pos)str;
1560 if (s->lookahead + s->insert < MIN_MATCH)
1564 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1565 * but this is not important since only literal bytes will be emitted.
1568 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1570 /* If the WIN_INIT bytes after the end of the current data have never been
1571 * written, then zero those bytes in order to avoid memory check reports of
1572 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1573 * the longest match routines. Update the high water mark for the next
1574 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1575 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1577 if (s->high_water < s->window_size) {
1578 ulg curr = s->strstart + (ulg)(s->lookahead);
1581 if (s->high_water < curr) {
1582 /* Previous high water mark below current data -- zero WIN_INIT
1583 * bytes or up to end of window, whichever is less.
1585 init = s->window_size - curr;
1586 if (init > WIN_INIT)
1588 zmemzero(s->window + curr, (unsigned)init);
1589 s->high_water = curr + init;
1591 else if (s->high_water < (ulg)curr + WIN_INIT) {
1592 /* High water mark at or above current data, but below current data
1593 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1594 * to end of window, whichever is less.
1596 init = (ulg)curr + WIN_INIT - s->high_water;
1597 if (init > s->window_size - s->high_water)
1598 init = s->window_size - s->high_water;
1599 zmemzero(s->window + s->high_water, (unsigned)init);
1600 s->high_water += init;
1604 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1605 "not enough room for search");
1608 /* ===========================================================================
1609 * Flush the current block, with given end-of-file flag.
1610 * IN assertion: strstart is set to the end of the current match.
1612 #define FLUSH_BLOCK_ONLY(s, last) { \
1613 _tr_flush_block(s, (s->block_start >= 0L ? \
1614 (charf *)&s->window[(unsigned)s->block_start] : \
1616 (ulg)((long)s->strstart - s->block_start), \
1618 s->block_start = s->strstart; \
1619 flush_pending(s->strm); \
1620 Tracev((stderr,"[FLUSH]")); \
1623 /* Same but force premature exit if necessary. */
1624 #define FLUSH_BLOCK(s, last) { \
1625 FLUSH_BLOCK_ONLY(s, last); \
1626 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1629 /* Maximum stored block length in deflate format (not including header). */
1630 #define MAX_STORED 65535
1632 /* Minimum of a and b. */
1633 #define MIN(a, b) ((a) > (b) ? (b) : (a))
1635 /* ===========================================================================
1636 * Copy without compression as much as possible from the input stream, return
1637 * the current block state.
1639 * In case deflateParams() is used to later switch to a non-zero compression
1640 * level, s->matches (otherwise unused when storing) keeps track of the number
1641 * of hash table slides to perform. If s->matches is 1, then one hash table
1642 * slide will be done when switching. If s->matches is 2, the maximum value
1643 * allowed here, then the hash table will be cleared, since two or more slides
1644 * is the same as a clear.
1646 * deflate_stored() is written to minimize the number of times an input byte is
1647 * copied. It is most efficient with large input and output buffers, which
1648 * maximizes the opportunites to have a single copy from next_in to next_out.
1650 local block_state deflate_stored(s, flush)
1654 /* Smallest worthy block size when not flushing or finishing. By default
1655 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1656 * large input and output buffers, the stored block size will be larger.
1658 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1660 /* Copy as many min_block or larger stored blocks directly to next_out as
1661 * possible. If flushing, copy the remaining available input to next_out as
1662 * stored blocks, if there is enough space.
1664 unsigned len, left, have, last = 0;
1665 unsigned used = s->strm->avail_in;
1667 /* Set len to the maximum size block that we can copy directly with the
1668 * available input data and output space. Set left to how much of that
1669 * would be copied from what's left in the window.
1671 len = MAX_STORED; /* maximum deflate stored block length */
1672 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1673 if (s->strm->avail_out < have) /* need room for header */
1675 /* maximum stored block length that will fit in avail_out: */
1676 have = s->strm->avail_out - have;
1677 left = s->strstart - s->block_start; /* bytes left in window */
1678 if (len > (ulg)left + s->strm->avail_in)
1679 len = left + s->strm->avail_in; /* limit len to the input */
1681 len = have; /* limit len to the output */
1683 /* If the stored block would be less than min_block in length, or if
1684 * unable to copy all of the available input when flushing, then try
1685 * copying to the window and the pending buffer instead. Also don't
1686 * write an empty block when flushing -- deflate() does that.
1688 if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1689 flush == Z_NO_FLUSH ||
1690 len != left + s->strm->avail_in))
1693 /* Make a dummy stored block in pending to get the header bytes,
1694 * including any pending bits. This also updates the debugging counts.
1696 last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1697 _tr_stored_block(s, (char *)0, 0L, last);
1699 /* Replace the lengths in the dummy stored block with len. */
1700 s->pending_buf[s->pending - 4] = len;
1701 s->pending_buf[s->pending - 3] = len >> 8;
1702 s->pending_buf[s->pending - 2] = ~len;
1703 s->pending_buf[s->pending - 1] = ~len >> 8;
1705 /* Write the stored block header bytes. */
1706 flush_pending(s->strm);
1709 /* Update debugging counts for the data about to be copied. */
1710 s->compressed_len += len << 3;
1711 s->bits_sent += len << 3;
1714 /* Copy uncompressed bytes from the window to next_out. */
1718 zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1719 s->strm->next_out += left;
1720 s->strm->avail_out -= left;
1721 s->strm->total_out += left;
1722 s->block_start += left;
1726 /* Copy uncompressed bytes directly from next_in to next_out, updating
1730 read_buf(s->strm, s->strm->next_out, len);
1731 s->strm->next_out += len;
1732 s->strm->avail_out -= len;
1733 s->strm->total_out += len;
1735 } while (last == 0);
1737 /* Update the sliding window with the last s->w_size bytes of the copied
1738 * data, or append all of the copied data to the existing window if less
1739 * than s->w_size bytes were copied. Also update the number of bytes to
1740 * insert in the hash tables, in the event that deflateParams() switches to
1741 * a non-zero compression level.
1743 used -= s->strm->avail_in; /* number of input bytes directly copied */
1745 /* If any input was used, then no unused input remains in the window,
1746 * therefore s->block_start == s->strstart.
1748 if (used >= s->w_size) { /* supplant the previous history */
1749 s->matches = 2; /* clear hash */
1750 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1751 s->strstart = s->w_size;
1752 s->insert = s->strstart;
1755 if (s->window_size - s->strstart <= used) {
1756 /* Slide the window down. */
1757 s->strstart -= s->w_size;
1758 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1760 s->matches++; /* add a pending slide_hash() */
1761 if (s->insert > s->strstart)
1762 s->insert = s->strstart;
1764 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1765 s->strstart += used;
1766 s->insert += MIN(used, s->w_size - s->insert);
1768 s->block_start = s->strstart;
1770 if (s->high_water < s->strstart)
1771 s->high_water = s->strstart;
1773 /* If the last block was written to next_out, then done. */
1777 /* If flushing and all input has been consumed, then done. */
1778 if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1779 s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1782 /* Fill the window with any remaining input. */
1783 have = s->window_size - s->strstart;
1784 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1785 /* Slide the window down. */
1786 s->block_start -= s->w_size;
1787 s->strstart -= s->w_size;
1788 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1790 s->matches++; /* add a pending slide_hash() */
1791 have += s->w_size; /* more space now */
1792 if (s->insert > s->strstart)
1793 s->insert = s->strstart;
1795 if (have > s->strm->avail_in)
1796 have = s->strm->avail_in;
1798 read_buf(s->strm, s->window + s->strstart, have);
1799 s->strstart += have;
1800 s->insert += MIN(have, s->w_size - s->insert);
1802 if (s->high_water < s->strstart)
1803 s->high_water = s->strstart;
1805 /* There was not enough avail_out to write a complete worthy or flushed
1806 * stored block to next_out. Write a stored block to pending instead, if we
1807 * have enough input for a worthy block, or if flushing and there is enough
1808 * room for the remaining input as a stored block in the pending buffer.
1810 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1811 /* maximum stored block length that will fit in pending: */
1812 have = MIN(s->pending_buf_size - have, MAX_STORED);
1813 min_block = MIN(have, s->w_size);
1814 left = s->strstart - s->block_start;
1815 if (left >= min_block ||
1816 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1817 s->strm->avail_in == 0 && left <= have)) {
1818 len = MIN(left, have);
1819 last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1820 len == left ? 1 : 0;
1821 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1822 s->block_start += len;
1823 flush_pending(s->strm);
1826 /* We've done all we can with the available input and output. */
1827 return last ? finish_started : need_more;
1830 /* ===========================================================================
1831 * Compress as much as possible from the input stream, return the current
1833 * This function does not perform lazy evaluation of matches and inserts
1834 * new strings in the dictionary only for unmatched strings or for short
1835 * matches. It is used only for the fast compression options.
1837 local block_state deflate_fast(s, flush)
1841 IPos hash_head; /* head of the hash chain */
1842 int bflush; /* set if current block must be flushed */
1845 /* Make sure that we always have enough lookahead, except
1846 * at the end of the input file. We need MAX_MATCH bytes
1847 * for the next match, plus MIN_MATCH bytes to insert the
1848 * string following the next match.
1850 if (s->lookahead < MIN_LOOKAHEAD) {
1852 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1855 if (s->lookahead == 0) break; /* flush the current block */
1858 /* Insert the string window[strstart .. strstart+2] in the
1859 * dictionary, and set hash_head to the head of the hash chain:
1862 if (s->lookahead >= MIN_MATCH) {
1863 INSERT_STRING(s, s->strstart, hash_head);
1866 /* Find the longest match, discarding those <= prev_length.
1867 * At this point we have always match_length < MIN_MATCH
1869 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1870 /* To simplify the code, we prevent matches with the string
1871 * of window index 0 (in particular we have to avoid a match
1872 * of the string with itself at the start of the input file).
1874 s->match_length = longest_match (s, hash_head);
1875 /* longest_match() sets match_start */
1877 if (s->match_length >= MIN_MATCH) {
1878 check_match(s, s->strstart, s->match_start, s->match_length);
1880 _tr_tally_dist(s, s->strstart - s->match_start,
1881 s->match_length - MIN_MATCH, bflush);
1883 s->lookahead -= s->match_length;
1885 /* Insert new strings in the hash table only if the match length
1886 * is not too large. This saves time but degrades compression.
1889 if (s->match_length <= s->max_insert_length &&
1890 s->lookahead >= MIN_MATCH) {
1891 s->match_length--; /* string at strstart already in table */
1894 INSERT_STRING(s, s->strstart, hash_head);
1895 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1896 * always MIN_MATCH bytes ahead.
1898 } while (--s->match_length != 0);
1903 s->strstart += s->match_length;
1904 s->match_length = 0;
1905 s->ins_h = s->window[s->strstart];
1906 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1908 Call UPDATE_HASH() MIN_MATCH-3 more times
1910 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1911 * matter since it will be recomputed at next deflate call.
1915 /* No match, output a literal byte */
1916 Tracevv((stderr,"%c", s->window[s->strstart]));
1917 _tr_tally_lit (s, s->window[s->strstart], bflush);
1921 if (bflush) FLUSH_BLOCK(s, 0);
1923 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1924 if (flush == Z_FINISH) {
1934 /* ===========================================================================
1935 * Same as above, but achieves better compression. We use a lazy
1936 * evaluation for matches: a match is finally adopted only if there is
1937 * no better match at the next window position.
1939 local block_state deflate_slow(s, flush)
1943 IPos hash_head; /* head of hash chain */
1944 int bflush; /* set if current block must be flushed */
1946 /* Process the input block. */
1948 /* Make sure that we always have enough lookahead, except
1949 * at the end of the input file. We need MAX_MATCH bytes
1950 * for the next match, plus MIN_MATCH bytes to insert the
1951 * string following the next match.
1953 if (s->lookahead < MIN_LOOKAHEAD) {
1955 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1958 if (s->lookahead == 0) break; /* flush the current block */
1961 /* Insert the string window[strstart .. strstart+2] in the
1962 * dictionary, and set hash_head to the head of the hash chain:
1965 if (s->lookahead >= MIN_MATCH) {
1966 INSERT_STRING(s, s->strstart, hash_head);
1969 /* Find the longest match, discarding those <= prev_length.
1971 s->prev_length = s->match_length, s->prev_match = s->match_start;
1972 s->match_length = MIN_MATCH-1;
1974 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1975 s->strstart - hash_head <= MAX_DIST(s)) {
1976 /* To simplify the code, we prevent matches with the string
1977 * of window index 0 (in particular we have to avoid a match
1978 * of the string with itself at the start of the input file).
1980 s->match_length = longest_match (s, hash_head);
1981 /* longest_match() sets match_start */
1983 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1984 #if TOO_FAR <= 32767
1985 || (s->match_length == MIN_MATCH &&
1986 s->strstart - s->match_start > TOO_FAR)
1990 /* If prev_match is also MIN_MATCH, match_start is garbage
1991 * but we will ignore the current match anyway.
1993 s->match_length = MIN_MATCH-1;
1996 /* If there was a match at the previous step and the current
1997 * match is not better, output the previous match:
1999 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
2000 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
2001 /* Do not insert strings in hash table beyond this. */
2003 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
2005 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
2006 s->prev_length - MIN_MATCH, bflush);
2008 /* Insert in hash table all strings up to the end of the match.
2009 * strstart-1 and strstart are already inserted. If there is not
2010 * enough lookahead, the last two strings are not inserted in
2013 s->lookahead -= s->prev_length-1;
2014 s->prev_length -= 2;
2016 if (++s->strstart <= max_insert) {
2017 INSERT_STRING(s, s->strstart, hash_head);
2019 } while (--s->prev_length != 0);
2020 s->match_available = 0;
2021 s->match_length = MIN_MATCH-1;
2024 if (bflush) FLUSH_BLOCK(s, 0);
2026 } else if (s->match_available) {
2027 /* If there was no match at the previous position, output a
2028 * single literal. If there was a match but the current match
2029 * is longer, truncate the previous match to a single literal.
2031 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2032 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2034 FLUSH_BLOCK_ONLY(s, 0);
2038 if (s->strm->avail_out == 0) return need_more;
2040 /* There is no previous match to compare with, wait for
2041 * the next step to decide.
2043 s->match_available = 1;
2048 Assert (flush != Z_NO_FLUSH, "no flush?");
2049 if (s->match_available) {
2050 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2051 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2052 s->match_available = 0;
2054 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2055 if (flush == Z_FINISH) {
2063 #endif /* FASTEST */
2065 /* ===========================================================================
2066 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2067 * one. Do not maintain a hash table. (It will be regenerated if this run of
2068 * deflate switches away from Z_RLE.)
2070 local block_state deflate_rle(s, flush)
2074 int bflush; /* set if current block must be flushed */
2075 uInt prev; /* byte at distance one to match */
2076 Bytef *scan, *strend; /* scan goes up to strend for length of run */
2079 /* Make sure that we always have enough lookahead, except
2080 * at the end of the input file. We need MAX_MATCH bytes
2081 * for the longest run, plus one for the unrolled loop.
2083 if (s->lookahead <= MAX_MATCH) {
2085 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2088 if (s->lookahead == 0) break; /* flush the current block */
2091 /* See how many times the previous byte repeats */
2092 s->match_length = 0;
2093 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2094 scan = s->window + s->strstart - 1;
2096 if (prev == *++scan && prev == *++scan && prev == *++scan) {
2097 strend = s->window + s->strstart + MAX_MATCH;
2099 } while (prev == *++scan && prev == *++scan &&
2100 prev == *++scan && prev == *++scan &&
2101 prev == *++scan && prev == *++scan &&
2102 prev == *++scan && prev == *++scan &&
2104 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2105 if (s->match_length > s->lookahead)
2106 s->match_length = s->lookahead;
2108 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2111 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2112 if (s->match_length >= MIN_MATCH) {
2113 check_match(s, s->strstart, s->strstart - 1, s->match_length);
2115 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2117 s->lookahead -= s->match_length;
2118 s->strstart += s->match_length;
2119 s->match_length = 0;
2121 /* No match, output a literal byte */
2122 Tracevv((stderr,"%c", s->window[s->strstart]));
2123 _tr_tally_lit (s, s->window[s->strstart], bflush);
2127 if (bflush) FLUSH_BLOCK(s, 0);
2130 if (flush == Z_FINISH) {
2139 /* ===========================================================================
2140 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
2141 * (It will be regenerated if this run of deflate switches away from Huffman.)
2143 local block_state deflate_huff(s, flush)
2147 int bflush; /* set if current block must be flushed */
2150 /* Make sure that we have a literal to write. */
2151 if (s->lookahead == 0) {
2153 if (s->lookahead == 0) {
2154 if (flush == Z_NO_FLUSH)
2156 break; /* flush the current block */
2160 /* Output a literal byte */
2161 s->match_length = 0;
2162 Tracevv((stderr,"%c", s->window[s->strstart]));
2163 _tr_tally_lit (s, s->window[s->strstart], bflush);
2166 if (bflush) FLUSH_BLOCK(s, 0);
2169 if (flush == Z_FINISH) {