1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-1998 Jean-loup Gailly.
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 ftp://ds.internic.net/rfc/rfc1951.txt
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 1.0.7 Copyright 1995-1998 Jean-loup Gailly ";
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));
79 local block_state deflate_slow OF((deflate_state *s, int flush));
80 local void lm_init OF((deflate_state *s));
81 local void putShortMSB OF((deflate_state *s, uInt b));
82 local void flush_pending OF((z_streamp strm));
83 local int read_buf OF((z_streamp strm, charf *buf, unsigned size));
85 void match_init OF((void)); /* asm code initialization */
86 uInt longest_match OF((deflate_state *s, IPos cur_match));
88 local uInt longest_match OF((deflate_state *s, IPos cur_match));
92 local void check_match OF((deflate_state *s, IPos start, IPos match,
96 /* ===========================================================================
101 /* Tail of hash chains */
104 # define TOO_FAR 4096
106 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
108 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
109 /* Minimum amount of lookahead, except at the end of the input file.
110 * See deflate.c for comments about the MIN_MATCH+1.
113 /* Values for max_lazy_match, good_match and max_chain_length, depending on
114 * the desired pack level (0..9). The values given below have been tuned to
115 * exclude worst case performance for pathological files. Better values may be
116 * found for specific files.
118 typedef struct config_s {
119 ush good_length; /* reduce lazy search above this match length */
120 ush max_lazy; /* do not perform lazy search above this match length */
121 ush nice_length; /* quit search above this match length */
126 local const config configuration_table[10] = {
127 /* good lazy nice chain */
128 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
129 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */
130 /* 2 */ {4, 5, 16, 8, deflate_fast},
131 /* 3 */ {4, 6, 32, 32, deflate_fast},
133 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
134 /* 5 */ {8, 16, 32, 32, deflate_slow},
135 /* 6 */ {8, 16, 128, 128, deflate_slow},
136 /* 7 */ {8, 32, 128, 256, deflate_slow},
137 /* 8 */ {32, 128, 258, 1024, deflate_slow},
138 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */
140 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
141 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
146 /* result of memcmp for equal strings */
148 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
150 /* ===========================================================================
151 * Update a hash value with the given input byte
152 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
153 * input characters, so that a running hash key can be computed from the
154 * previous key instead of complete recalculation each time.
156 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
159 /* ===========================================================================
160 * Insert string str in the dictionary and set match_head to the previous head
161 * of the hash chain (the most recent string with same hash key). Return
162 * the previous length of the hash chain.
163 * IN assertion: all calls to to INSERT_STRING are made with consecutive
164 * input characters and the first MIN_MATCH bytes of str are valid
165 * (except for the last MIN_MATCH-1 bytes of the input file).
167 #define INSERT_STRING(s, str, match_head) \
168 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
169 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
170 s->head[s->ins_h] = (Pos)(str))
172 /* ===========================================================================
173 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
174 * prev[] will be initialized on the fly.
176 #define CLEAR_HASH(s) \
177 s->head[s->hash_size-1] = NIL; \
178 zmemzero((charf *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
180 /* ========================================================================= */
181 int EXPORT deflateInit_(strm, level, version, stream_size)
187 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
188 Z_DEFAULT_STRATEGY, version, stream_size);
189 /* To do: ignore strm->next_in if we use it as window */
192 /* ========================================================================= */
193 int EXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
194 version, stream_size)
206 static const char* my_version = ZLIB_VERSION;
209 /* We overlay pending_buf and d_buf+l_buf. This works since the average
210 * output size for (length,distance) codes is <= 24 bits.
213 if (version == Z_NULL || version[0] != my_version[0] ||
214 stream_size != sizeof(z_stream)) {
215 return Z_VERSION_ERROR;
217 if (strm == Z_NULL) return Z_STREAM_ERROR;
220 if (strm->zalloc == Z_NULL) {
221 strm->zalloc = zcalloc;
222 strm->opaque = (voidpf)0;
224 if (strm->zfree == Z_NULL) strm->zfree = zcfree;
226 if (level == Z_DEFAULT_COMPRESSION) level = 6;
228 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
230 windowBits = -windowBits;
232 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
233 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
234 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
235 return Z_STREAM_ERROR;
237 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
238 if (s == Z_NULL) return Z_MEM_ERROR;
239 strm->state = (struct internal_state FAR *)s;
242 s->noheader = noheader;
243 s->w_bits = windowBits;
244 s->w_size = 1 << s->w_bits;
245 s->w_mask = s->w_size - 1;
247 s->hash_bits = memLevel + 7;
248 s->hash_size = 1 << s->hash_bits;
249 s->hash_mask = s->hash_size - 1;
250 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
252 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
253 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
254 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
256 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
258 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
259 s->pending_buf = (uchf *) overlay;
260 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
262 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
263 s->pending_buf == Z_NULL) {
264 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
268 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
269 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
272 s->strategy = strategy;
273 s->method = (Byte)method;
275 return deflateReset(strm);
278 /* ========================================================================= */
279 int EXPORT deflateSetDictionary (strm, dictionary, dictLength)
281 const Bytef *dictionary;
285 uInt length = dictLength;
289 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
290 strm->state->status != INIT_STATE) return Z_STREAM_ERROR;
293 strm->adler = adler32(strm->adler, dictionary, dictLength);
295 if (length < MIN_MATCH) return Z_OK;
296 if (length > MAX_DIST(s)) {
297 length = MAX_DIST(s);
298 #ifndef USE_DICT_HEAD
299 dictionary += dictLength - length; /* use the tail of the dictionary */
302 zmemcpy((charf *)s->window, dictionary, length);
303 s->strstart = length;
304 s->block_start = (long)length;
306 /* Insert all strings in the hash table (except for the last two bytes).
307 * s->lookahead stays null, so s->ins_h will be recomputed at the next
308 * call of fill_window.
310 s->ins_h = s->window[0];
311 UPDATE_HASH(s, s->ins_h, s->window[1]);
312 for (n = 0; n <= length - MIN_MATCH; n++) {
313 INSERT_STRING(s, n, hash_head);
315 if (hash_head) hash_head = 0; /* to make compiler happy */
319 /* ========================================================================= */
320 int EXPORT deflateReset (strm)
325 if (strm == Z_NULL || strm->state == Z_NULL ||
326 strm->zalloc == Z_NULL || strm->zfree == Z_NULL) return Z_STREAM_ERROR;
328 strm->total_in = strm->total_out = 0;
329 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
330 strm->data_type = Z_UNKNOWN;
332 s = (deflate_state *)strm->state;
334 s->pending_out = s->pending_buf;
336 if (s->noheader < 0) {
337 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
339 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
341 s->last_flush = Z_NO_FLUSH;
349 /* ========================================================================= */
350 int EXPORT deflateParams(strm, level, strategy)
359 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
362 if (level == Z_DEFAULT_COMPRESSION) {
365 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
366 return Z_STREAM_ERROR;
368 func = configuration_table[s->level].func;
370 if (func != configuration_table[level].func && strm->total_in != 0) {
371 /* Flush the last buffer: */
372 err = deflate(strm, Z_PARTIAL_FLUSH);
374 if (s->level != level) {
376 s->max_lazy_match = configuration_table[level].max_lazy;
377 s->good_match = configuration_table[level].good_length;
378 s->nice_match = configuration_table[level].nice_length;
379 s->max_chain_length = configuration_table[level].max_chain;
381 s->strategy = strategy;
385 /* =========================================================================
386 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
387 * IN assertion: the stream state is correct and there is enough room in
390 local void putShortMSB (s, b)
394 put_byte(s, (Byte)(b >> 8));
395 put_byte(s, (Byte)(b & 0xff));
398 /* =========================================================================
399 * Flush as much pending output as possible. All deflate() output goes
400 * through this function so some applications may wish to modify it
401 * to avoid allocating a large strm->next_out buffer and copying into it.
402 * (See also read_buf()).
404 local void flush_pending(strm)
407 unsigned len = strm->state->pending;
409 if (len > strm->avail_out) len = strm->avail_out;
410 if (len == 0) return;
412 zmemcpy(strm->next_out, strm->state->pending_out, len);
413 strm->next_out += len;
414 strm->state->pending_out += len;
415 strm->total_out += len;
416 strm->avail_out -= len;
417 strm->state->pending -= len;
418 if (strm->state->pending == 0) {
419 strm->state->pending_out = strm->state->pending_buf;
423 /* ========================================================================= */
424 int EXPORT deflate (strm, flush)
428 int old_flush; /* value of flush param for previous deflate call */
431 if (strm == Z_NULL || strm->state == Z_NULL ||
432 flush > Z_FINISH || flush < 0) {
433 return Z_STREAM_ERROR;
437 if (strm->next_out == Z_NULL ||
438 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
439 (s->status == FINISH_STATE && flush != Z_FINISH)) {
440 ERR_RETURN(strm, Z_STREAM_ERROR);
442 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
444 s->strm = strm; /* just in case */
445 old_flush = s->last_flush;
446 s->last_flush = flush;
448 /* Write the zlib header */
449 if (s->status == INIT_STATE) {
451 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
452 uInt level_flags = (s->level-1) >> 1;
454 if (level_flags > 3) level_flags = 3;
455 header |= (level_flags << 6);
456 if (s->strstart != 0) header |= PRESET_DICT;
457 header += 31 - (header % 31);
459 s->status = BUSY_STATE;
460 putShortMSB(s, header);
462 /* Save the adler32 of the preset dictionary: */
463 if (s->strstart != 0) {
464 putShortMSB(s, (uInt)(strm->adler >> 16));
465 putShortMSB(s, (uInt)(strm->adler & 0xffff));
470 /* Flush as much pending output as possible */
471 if (s->pending != 0) {
473 if (strm->avail_out == 0) {
474 /* Since avail_out is 0, deflate will be called again with
475 * more output space, but possibly with both pending and
476 * avail_in equal to zero. There won't be anything to do,
477 * but this is not an error situation so make sure we
478 * return OK instead of BUF_ERROR at next call of deflate:
484 /* Make sure there is something to do and avoid duplicate consecutive
485 * flushes. For repeated and useless calls with Z_FINISH, we keep
486 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
488 } else if (strm->avail_in == 0 && flush <= old_flush &&
490 ERR_RETURN(strm, Z_BUF_ERROR);
493 /* User must not provide more input after the first FINISH: */
494 if (s->status == FINISH_STATE && strm->avail_in != 0) {
495 ERR_RETURN(strm, Z_BUF_ERROR);
498 /* Start a new block or continue the current one.
500 if (strm->avail_in != 0 || s->lookahead != 0 ||
501 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
504 bstate = (*(configuration_table[s->level].func))(s, flush);
506 if (bstate == finish_started || bstate == finish_done) {
507 s->status = FINISH_STATE;
509 if (bstate == need_more || bstate == finish_started) {
510 if (strm->avail_out == 0) {
511 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
514 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
515 * of deflate should use the same flush parameter to make sure
516 * that the flush is complete. So we don't have to output an
517 * empty block here, this will be done at next call. This also
518 * ensures that for a very small output buffer, we emit at most
522 if (bstate == block_done) {
523 if (flush == Z_PARTIAL_FLUSH) {
525 } else { /* FULL_FLUSH or SYNC_FLUSH */
526 _tr_stored_block(s, (char*)0, 0L, 0);
527 /* For a full flush, this empty block will be recognized
528 * as a special marker by inflate_sync().
530 if (flush == Z_FULL_FLUSH) {
531 CLEAR_HASH(s); /* forget history */
535 if (strm->avail_out == 0) {
536 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
541 Assert(strm->avail_out > 0, "bug2");
543 if (flush != Z_FINISH) return Z_OK;
544 if (s->noheader) return Z_STREAM_END;
546 /* Write the zlib trailer (adler32) */
547 putShortMSB(s, (uInt)(strm->adler >> 16));
548 putShortMSB(s, (uInt)(strm->adler & 0xffff));
550 /* If avail_out is zero, the application will call deflate again
553 s->noheader = -1; /* write the trailer only once! */
554 return s->pending != 0 ? Z_OK : Z_STREAM_END;
557 /* ========================================================================= */
558 int EXPORT deflateEnd (strm)
563 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
565 status = strm->state->status;
566 if (status != INIT_STATE && status != BUSY_STATE &&
567 status != FINISH_STATE) {
568 return Z_STREAM_ERROR;
571 /* Deallocate in reverse order of allocations: */
572 TRY_FREE(strm, strm->state->pending_buf);
573 TRY_FREE(strm, strm->state->head);
574 TRY_FREE(strm, strm->state->prev);
575 TRY_FREE(strm, strm->state->window);
577 ZFREE(strm, strm->state);
578 strm->state = Z_NULL;
580 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
583 /* =========================================================================
584 * Copy the source state to the destination state.
585 * To simplify the source, this is not supported for 16-bit MSDOS (which
586 * doesn't have enough memory anyway to duplicate compression states).
588 int EXPORT deflateCopy (dest, source)
593 return Z_STREAM_ERROR;
601 if (source == Z_NULL || dest == Z_NULL || ss == Z_NULL) {
602 return Z_STREAM_ERROR;
606 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
607 if (ds == Z_NULL) return Z_MEM_ERROR;
608 dest->state = (struct internal_state FAR *) ds;
612 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
613 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
614 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
615 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
616 ds->pending_buf = (uchf *) overlay;
618 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
619 ds->pending_buf == Z_NULL) {
623 /* following zmemcpy do not work for 16-bit MSDOS */
624 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
625 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
626 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
627 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
629 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
630 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
631 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
633 ds->l_desc.dyn_tree = ds->dyn_ltree;
634 ds->d_desc.dyn_tree = ds->dyn_dtree;
635 ds->bl_desc.dyn_tree = ds->bl_tree;
641 /* ===========================================================================
642 * Read a new buffer from the current input stream, update the adler32
643 * and total number of bytes read. All deflate() input goes through
644 * this function so some applications may wish to modify it to avoid
645 * allocating a large strm->next_in buffer and copying from it.
646 * (See also flush_pending()).
648 local int read_buf(strm, buf, size)
653 unsigned len = strm->avail_in;
655 if (len > size) len = size;
656 if (len == 0) return 0;
658 strm->avail_in -= len;
660 if (!strm->state->noheader) {
661 strm->adler = adler32(strm->adler, strm->next_in, len);
663 zmemcpy(buf, strm->next_in, len);
664 strm->next_in += len;
665 strm->total_in += len;
670 /* ===========================================================================
671 * Initialize the "longest match" routines for a new zlib stream
673 local void lm_init (s)
676 s->window_size = (ulg)2L*s->w_size;
680 /* Set the default configuration parameters:
682 s->max_lazy_match = configuration_table[s->level].max_lazy;
683 s->good_match = configuration_table[s->level].good_length;
684 s->nice_match = configuration_table[s->level].nice_length;
685 s->max_chain_length = configuration_table[s->level].max_chain;
690 s->match_length = s->prev_length = MIN_MATCH-1;
691 s->match_available = 0;
694 match_init(); /* initialize the asm code */
698 /* ===========================================================================
699 * Set match_start to the longest match starting at the given string and
700 * return its length. Matches shorter or equal to prev_length are discarded,
701 * in which case the result is equal to prev_length and match_start is
703 * IN assertions: cur_match is the head of the hash chain for the current
704 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
705 * OUT assertion: the match length is not greater than s->lookahead.
708 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
709 * match.S. The code will be functionally equivalent.
711 local uInt longest_match(s, cur_match)
713 IPos cur_match; /* current match */
715 unsigned chain_length = s->max_chain_length;/* max hash chain length */
716 register Bytef *scan = s->window + s->strstart; /* current string */
717 register Bytef *match; /* matched string */
718 register int len; /* length of current match */
719 int best_len = s->prev_length; /* best match length so far */
720 int nice_match = s->nice_match; /* stop if match long enough */
721 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
722 s->strstart - (IPos)MAX_DIST(s) : NIL;
723 /* Stop when cur_match becomes <= limit. To simplify the code,
724 * we prevent matches with the string of window index 0.
726 Posf *prev = s->prev;
727 uInt wmask = s->w_mask;
730 /* Compare two bytes at a time. Note: this is not always beneficial.
731 * Try with and without -DUNALIGNED_OK to check.
733 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
734 register ush scan_start = *(ushf*)scan;
735 register ush scan_end = *(ushf*)(scan+best_len-1);
737 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
738 register Byte scan_end1 = scan[best_len-1];
739 register Byte scan_end = scan[best_len];
742 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
743 * It is easy to get rid of this optimization if necessary.
745 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
747 /* Do not waste too much time if we already have a good match: */
748 if (s->prev_length >= s->good_match) {
751 /* Do not look for matches beyond the end of the input. This is necessary
752 * to make deflate deterministic.
754 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
756 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
759 Assert(cur_match < s->strstart, "no future");
760 match = s->window + cur_match;
762 /* Skip to next match if the match length cannot increase
763 * or if the match length is less than 2:
765 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
766 /* This code assumes sizeof(unsigned short) == 2. Do not use
767 * UNALIGNED_OK if your compiler uses a different size.
769 if (*(ushf*)(match+best_len-1) != scan_end ||
770 *(ushf*)match != scan_start) continue;
772 /* It is not necessary to compare scan[2] and match[2] since they are
773 * always equal when the other bytes match, given that the hash keys
774 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
775 * strstart+3, +5, ... up to strstart+257. We check for insufficient
776 * lookahead only every 4th comparison; the 128th check will be made
777 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
778 * necessary to put more guard bytes at the end of the window, or
779 * to check more often for insufficient lookahead.
781 Assert(scan[2] == match[2], "scan[2]?");
784 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
785 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
786 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
787 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
789 /* The funny "do {}" generates better code on most compilers */
791 /* Here, scan <= window+strstart+257 */
792 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
793 if (*scan == *match) scan++;
795 len = (MAX_MATCH - 1) - (int)(strend-scan);
796 scan = strend - (MAX_MATCH-1);
798 #else /* UNALIGNED_OK */
800 if (match[best_len] != scan_end ||
801 match[best_len-1] != scan_end1 ||
803 *++match != scan[1]) continue;
805 /* The check at best_len-1 can be removed because it will be made
806 * again later. (This heuristic is not always a win.)
807 * It is not necessary to compare scan[2] and match[2] since they
808 * are always equal when the other bytes match, given that
809 * the hash keys are equal and that HASH_BITS >= 8.
812 Assert(*scan == *match, "match[2]?");
814 /* We check for insufficient lookahead only every 8th comparison;
815 * the 256th check will be made at strstart+258.
818 } while (*++scan == *++match && *++scan == *++match &&
819 *++scan == *++match && *++scan == *++match &&
820 *++scan == *++match && *++scan == *++match &&
821 *++scan == *++match && *++scan == *++match &&
824 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
826 len = MAX_MATCH - (int)(strend - scan);
827 scan = strend - MAX_MATCH;
829 #endif /* UNALIGNED_OK */
831 if (len > best_len) {
832 s->match_start = cur_match;
834 if (len >= nice_match) break;
836 scan_end = *(ushf*)(scan+best_len-1);
838 scan_end1 = scan[best_len-1];
839 scan_end = scan[best_len];
842 } while ((cur_match = prev[cur_match & wmask]) > limit
843 && --chain_length != 0);
845 if ((uInt)best_len <= s->lookahead) return best_len;
851 /* ===========================================================================
852 * Check that the match at match_start is indeed a match.
854 local void check_match(s, start, match, length)
859 /* check that the match is indeed a match */
860 if (zmemcmp((charf *)s->window + match,
861 (charf *)s->window + start, length) != EQUAL) {
862 fprintf(stderr, " start %u, match %u, length %d\n",
863 start, match, length);
865 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
866 } while (--length != 0);
867 z_error("invalid match");
870 fprintf(stderr,"\\[%d,%d]", start-match, length);
871 do { putc(s->window[start++], stderr); } while (--length != 0);
875 # define check_match(s, start, match, length)
878 /* ===========================================================================
879 * Fill the window when the lookahead becomes insufficient.
880 * Updates strstart and lookahead.
882 * IN assertion: lookahead < MIN_LOOKAHEAD
883 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
884 * At least one byte has been read, or avail_in == 0; reads are
885 * performed for at least two bytes (required for the zip translate_eol
886 * option -- not supported here).
888 local void fill_window(s)
891 register unsigned n, m;
893 unsigned more; /* Amount of free space at the end of the window. */
894 uInt wsize = s->w_size;
897 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
899 /* Deal with !@#$% 64K limit: */
900 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
903 } else if (more == (unsigned)(-1)) {
904 /* Very unlikely, but possible on 16 bit machine if strstart == 0
905 * and lookahead == 1 (input done one byte at time)
909 /* If the window is almost full and there is insufficient lookahead,
910 * move the upper half to the lower one to make room in the upper half.
912 } else if (s->strstart >= wsize+MAX_DIST(s)) {
914 zmemcpy((charf *)s->window, (charf *)s->window+wsize,
916 s->match_start -= wsize;
917 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
918 s->block_start -= (long) wsize;
920 /* Slide the hash table (could be avoided with 32 bit values
921 at the expense of memory usage). We slide even when level == 0
922 to keep the hash table consistent if we switch back to level > 0
923 later. (Using level 0 permanently is not an optimal usage of
924 zlib, so we don't care about this pathological case.)
930 *p = (Pos)(m >= wsize ? m-wsize : NIL);
937 *p = (Pos)(m >= wsize ? m-wsize : NIL);
938 /* If n is not on any hash chain, prev[n] is garbage but
939 * its value will never be used.
944 if (s->strm->avail_in == 0) return;
946 /* If there was no sliding:
947 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
948 * more == window_size - lookahead - strstart
949 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
950 * => more >= window_size - 2*WSIZE + 2
951 * In the BIG_MEM or MMAP case (not yet supported),
952 * window_size == input_size + MIN_LOOKAHEAD &&
953 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
954 * Otherwise, window_size == 2*WSIZE so more >= 2.
955 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
957 Assert(more >= 2, "more < 2");
959 n = read_buf(s->strm, (charf *)s->window + s->strstart + s->lookahead,
963 /* Initialize the hash value now that we have some input: */
964 if (s->lookahead >= MIN_MATCH) {
965 s->ins_h = s->window[s->strstart];
966 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
968 Call UPDATE_HASH() MIN_MATCH-3 more times
971 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
972 * but this is not important since only literal bytes will be emitted.
975 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
978 /* ===========================================================================
979 * Flush the current block, with given end-of-file flag.
980 * IN assertion: strstart is set to the end of the current match.
982 #define FLUSH_BLOCK_ONLY(s, eof) { \
983 _tr_flush_block(s, (s->block_start >= 0L ? \
984 (charf *)&s->window[(unsigned)s->block_start] : \
986 (ulg)((long)s->strstart - s->block_start), \
988 s->block_start = s->strstart; \
989 flush_pending(s->strm); \
990 Tracev((stderr,"[FLUSH]")); \
993 /* Same but force premature exit if necessary. */
994 #define FLUSH_BLOCK(s, eof) { \
995 FLUSH_BLOCK_ONLY(s, eof); \
996 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
999 /* ===========================================================================
1000 * Copy without compression as much as possible from the input stream, return
1001 * the current block state.
1002 * This function does not insert new strings in the dictionary since
1003 * uncompressible data is probably not useful. This function is used
1004 * only for the level=0 compression option.
1005 * NOTE: this function should be optimized to avoid extra copying from
1006 * window to pending_buf.
1008 local block_state deflate_stored(s, flush)
1012 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1013 * to pending_buf_size, and each stored block has a 5 byte header:
1015 ulg max_block_size = 0xffff;
1018 if (max_block_size > s->pending_buf_size - 5) {
1019 max_block_size = s->pending_buf_size - 5;
1022 /* Copy as much as possible from input to output: */
1024 /* Fill the window as much as possible: */
1025 if (s->lookahead <= 1) {
1027 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1028 s->block_start >= (long)s->w_size, "slide too late");
1031 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1033 if (s->lookahead == 0) break; /* flush the current block */
1035 Assert(s->block_start >= 0L, "block gone");
1037 s->strstart += s->lookahead;
1040 /* Emit a stored block if pending_buf will be full: */
1041 max_start = s->block_start + max_block_size;
1042 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1043 /* strstart == 0 is possible when wraparound on 16-bit machine */
1044 s->lookahead = (uInt)(s->strstart - max_start);
1045 s->strstart = (uInt)max_start;
1048 /* Flush if we may have to slide, otherwise block_start may become
1049 * negative and the data will be gone:
1051 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1055 FLUSH_BLOCK(s, flush == Z_FINISH);
1056 return flush == Z_FINISH ? finish_done : block_done;
1059 /* ===========================================================================
1060 * Compress as much as possible from the input stream, return the current
1062 * This function does not perform lazy evaluation of matches and inserts
1063 * new strings in the dictionary only for unmatched strings or for short
1064 * matches. It is used only for the fast compression options.
1066 local block_state deflate_fast(s, flush)
1070 IPos hash_head = NIL; /* head of the hash chain */
1071 int bflush; /* set if current block must be flushed */
1074 /* Make sure that we always have enough lookahead, except
1075 * at the end of the input file. We need MAX_MATCH bytes
1076 * for the next match, plus MIN_MATCH bytes to insert the
1077 * string following the next match.
1079 if (s->lookahead < MIN_LOOKAHEAD) {
1081 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1084 if (s->lookahead == 0) break; /* flush the current block */
1087 /* Insert the string window[strstart .. strstart+2] in the
1088 * dictionary, and set hash_head to the head of the hash chain:
1090 if (s->lookahead >= MIN_MATCH) {
1091 INSERT_STRING(s, s->strstart, hash_head);
1094 /* Find the longest match, discarding those <= prev_length.
1095 * At this point we have always match_length < MIN_MATCH
1097 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1098 /* To simplify the code, we prevent matches with the string
1099 * of window index 0 (in particular we have to avoid a match
1100 * of the string with itself at the start of the input file).
1102 if (s->strategy != Z_HUFFMAN_ONLY) {
1103 s->match_length = longest_match (s, hash_head);
1105 /* longest_match() sets match_start */
1107 if (s->match_length >= MIN_MATCH) {
1108 check_match(s, s->strstart, s->match_start, s->match_length);
1110 bflush = _tr_tally(s, s->strstart - s->match_start,
1111 s->match_length - MIN_MATCH);
1113 s->lookahead -= s->match_length;
1115 /* Insert new strings in the hash table only if the match length
1116 * is not too large. This saves time but degrades compression.
1118 if (s->match_length <= s->max_insert_length &&
1119 s->lookahead >= MIN_MATCH) {
1120 s->match_length--; /* string at strstart already in hash table */
1123 INSERT_STRING(s, s->strstart, hash_head);
1124 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1125 * always MIN_MATCH bytes ahead.
1127 } while (--s->match_length != 0);
1130 s->strstart += s->match_length;
1131 s->match_length = 0;
1132 s->ins_h = s->window[s->strstart];
1133 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1135 Call UPDATE_HASH() MIN_MATCH-3 more times
1137 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1138 * matter since it will be recomputed at next deflate call.
1142 /* No match, output a literal byte */
1143 Tracevv((stderr,"%c", s->window[s->strstart]));
1144 bflush = _tr_tally (s, 0, s->window[s->strstart]);
1148 if (bflush) FLUSH_BLOCK(s, 0);
1150 FLUSH_BLOCK(s, flush == Z_FINISH);
1151 return flush == Z_FINISH ? finish_done : block_done;
1154 /* ===========================================================================
1155 * Same as above, but achieves better compression. We use a lazy
1156 * evaluation for matches: a match is finally adopted only if there is
1157 * no better match at the next window position.
1159 local block_state deflate_slow(s, flush)
1163 IPos hash_head = NIL; /* head of hash chain */
1164 int bflush; /* set if current block must be flushed */
1166 /* Process the input block. */
1168 /* Make sure that we always have enough lookahead, except
1169 * at the end of the input file. We need MAX_MATCH bytes
1170 * for the next match, plus MIN_MATCH bytes to insert the
1171 * string following the next match.
1173 if (s->lookahead < MIN_LOOKAHEAD) {
1175 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1178 if (s->lookahead == 0) break; /* flush the current block */
1181 /* Insert the string window[strstart .. strstart+2] in the
1182 * dictionary, and set hash_head to the head of the hash chain:
1184 if (s->lookahead >= MIN_MATCH) {
1185 INSERT_STRING(s, s->strstart, hash_head);
1188 /* Find the longest match, discarding those <= prev_length.
1190 s->prev_length = s->match_length, s->prev_match = s->match_start;
1191 s->match_length = MIN_MATCH-1;
1193 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1194 s->strstart - hash_head <= MAX_DIST(s)) {
1195 /* To simplify the code, we prevent matches with the string
1196 * of window index 0 (in particular we have to avoid a match
1197 * of the string with itself at the start of the input file).
1199 if (s->strategy != Z_HUFFMAN_ONLY) {
1200 s->match_length = longest_match (s, hash_head);
1202 /* longest_match() sets match_start */
1204 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1205 (s->match_length == MIN_MATCH &&
1206 s->strstart - s->match_start > TOO_FAR))) {
1208 /* If prev_match is also MIN_MATCH, match_start is garbage
1209 * but we will ignore the current match anyway.
1211 s->match_length = MIN_MATCH-1;
1214 /* If there was a match at the previous step and the current
1215 * match is not better, output the previous match:
1217 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1218 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1219 /* Do not insert strings in hash table beyond this. */
1221 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1223 bflush = _tr_tally(s, s->strstart -1 - s->prev_match,
1224 s->prev_length - MIN_MATCH);
1226 /* Insert in hash table all strings up to the end of the match.
1227 * strstart-1 and strstart are already inserted. If there is not
1228 * enough lookahead, the last two strings are not inserted in
1231 s->lookahead -= s->prev_length-1;
1232 s->prev_length -= 2;
1234 if (++s->strstart <= max_insert) {
1235 INSERT_STRING(s, s->strstart, hash_head);
1237 } while (--s->prev_length != 0);
1238 s->match_available = 0;
1239 s->match_length = MIN_MATCH-1;
1242 if (bflush) FLUSH_BLOCK(s, 0);
1244 } else if (s->match_available) {
1245 /* If there was no match at the previous position, output a
1246 * single literal. If there was a match but the current match
1247 * is longer, truncate the previous match to a single literal.
1249 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1250 if (_tr_tally (s, 0, s->window[s->strstart-1])) {
1251 FLUSH_BLOCK_ONLY(s, 0);
1255 if (s->strm->avail_out == 0) return need_more;
1257 /* There is no previous match to compare with, wait for
1258 * the next step to decide.
1260 s->match_available = 1;
1265 Assert (flush != Z_NO_FLUSH, "no flush?");
1266 if (s->match_available) {
1267 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1268 _tr_tally (s, 0, s->window[s->strstart-1]);
1269 s->match_available = 0;
1271 FLUSH_BLOCK(s, flush == Z_FINISH);
1272 return flush == Z_FINISH ? finish_done : block_done;