1 /* deflate.c -- compress data using the deflation algorithm
3 Copyright (C) 1999, 2006, 2009-2010 Free Software Foundation, Inc.
4 Copyright (C) 1992-1993 Jean-loup Gailly
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3, or (at your option)
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software Foundation,
18 Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */
23 * Identify new text as repetitions of old text within a fixed-
24 * length sliding window trailing behind the new text.
28 * The "deflation" process depends on being able to identify portions
29 * of the input text which are identical to earlier input (within a
30 * sliding window trailing behind the input currently being processed).
32 * The most straightforward technique turns out to be the fastest for
33 * most input files: try all possible matches and select the longest.
34 * The key feature of this algorithm is that insertions into the string
35 * dictionary are very simple and thus fast, and deletions are avoided
36 * completely. Insertions are performed at each input character, whereas
37 * string matches are performed only when the previous match ends. So it
38 * is preferable to spend more time in matches to allow very fast string
39 * insertions and avoid deletions. The matching algorithm for small
40 * strings is inspired from that of Rabin & Karp. A brute force approach
41 * is used to find longer strings when a small match has been found.
42 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
43 * (by Leonid Broukhis).
44 * A previous version of this file used a more sophisticated algorithm
45 * (by Fiala and Greene) which is guaranteed to run in linear amortized
46 * time, but has a larger average cost, uses more memory and is patented.
47 * However the F&G algorithm may be faster for some highly redundant
48 * files if the parameter max_chain_length (described below) is too large.
52 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
53 * I found it in 'freeze' written by Leonid Broukhis.
54 * Thanks to many info-zippers for bug reports and testing.
58 * APPNOTE.TXT documentation file in PKZIP 1.93a distribution.
60 * A description of the Rabin and Karp algorithm is given in the book
61 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
63 * Fiala,E.R., and Greene,D.H.
64 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
68 * void lm_init (int pack_level, ush *flags)
69 * Initialize the "longest match" routines for a new file
71 * off_t deflate (void)
72 * Processes a new input file and return its compressed length. Sets
73 * the compressed length, crc, deflate flags and internal file
82 #include "lzw.h" /* just for consistency checking */
84 /* ===========================================================================
85 * Configuration parameters
88 /* Compile with MEDIUM_MEM to reduce the memory requirements or
89 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
90 * entire input file can be held in memory (not possible on 16 bit systems).
91 * Warning: defining these symbols affects HASH_BITS (see below) and thus
92 * affects the compression ratio. The compressed output
93 * is still correct, and might even be smaller in some cases.
97 # define HASH_BITS 13 /* Number of bits used to hash strings */
100 # define HASH_BITS 14
103 # define HASH_BITS 15
104 /* For portability to 16 bit machines, do not use values above 15. */
107 /* To save space (see unlzw.c), we overlay prev+head with tab_prefix and
108 * window with tab_suffix. Check that we can do this:
110 #if (WSIZE<<1) > (1<<BITS)
111 error: cannot overlay window with tab_suffix and prev with tab_prefix0
113 #if HASH_BITS > BITS-1
114 error: cannot overlay head with tab_prefix1
117 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
118 #define HASH_MASK (HASH_SIZE-1)
119 #define WMASK (WSIZE-1)
120 /* HASH_SIZE and WSIZE must be powers of two */
123 /* Tail of hash chains */
127 /* speed options for the general purpose bit flag */
130 # define TOO_FAR 4096
132 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
135 # define RSYNC_WIN 8192
137 /* Size of rsync window, must be < MAX_DIST */
139 #define RSYNC_SUM_MATCH(sum) (((sum) & (RSYNC_WIN - 1)) == 0)
140 /* Whether window sum matches magic value */
142 /* ===========================================================================
143 * Local data used by the "longest match" routines.
147 typedef unsigned IPos;
148 /* A Pos is an index in the character window. We use short instead of int to
149 * save space in the various tables. IPos is used only for parameter passing.
152 /* DECLARE(uch, window, 2L*WSIZE); */
153 /* Sliding window. Input bytes are read into the second half of the window,
154 * and move to the first half later to keep a dictionary of at least WSIZE
155 * bytes. With this organization, matches are limited to a distance of
156 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
157 * performed with a length multiple of the block size. Also, it limits
158 * the window size to 64K, which is quite useful on MSDOS.
159 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
160 * be less efficient).
163 /* DECLARE(Pos, prev, WSIZE); */
164 /* Link to older string with same hash index. To limit the size of this
165 * array to 64K, this link is maintained only for the last 32K strings.
166 * An index in this array is thus a window index modulo 32K.
169 /* DECLARE(Pos, head, 1<<HASH_BITS); */
170 /* Heads of the hash chains or NIL. */
172 ulg window_size = (ulg)2*WSIZE;
173 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
174 * input file length plus MIN_LOOKAHEAD.
178 /* window position at the beginning of the current output block. Gets
179 * negative when the window is moved backwards.
182 local unsigned ins_h; /* hash index of string to be inserted */
184 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1)/MIN_MATCH)
185 /* Number of bits by which ins_h and del_h must be shifted at each
186 * input step. It must be such that after MIN_MATCH steps, the oldest
187 * byte no longer takes part in the hash key, that is:
188 * H_SHIFT * MIN_MATCH >= HASH_BITS
191 unsigned int near prev_length;
192 /* Length of the best match at previous step. Matches not greater than this
193 * are discarded. This is used in the lazy match evaluation.
196 unsigned near strstart; /* start of string to insert */
197 unsigned near match_start; /* start of matching string */
198 local int eofile; /* flag set at end of input file */
199 local unsigned lookahead; /* number of valid bytes ahead in window */
201 unsigned near max_chain_length;
202 /* To speed up deflation, hash chains are never searched beyond this length.
203 * A higher limit improves compression ratio but degrades the speed.
206 local unsigned int max_lazy_match;
207 /* Attempt to find a better match only when the current match is strictly
208 * smaller than this value. This mechanism is used only for compression
211 #define max_insert_length max_lazy_match
212 /* Insert new strings in the hash table only if the match length
213 * is not greater than this length. This saves time but degrades compression.
214 * max_insert_length is used only for compression levels <= 3.
217 local int compr_level;
218 /* compression level (1..9) */
220 unsigned near good_match;
221 /* Use a faster search when the previous match is longer than this */
223 local ulg rsync_sum; /* rolling sum of rsync window */
224 local ulg rsync_chunk_end; /* next rsync sequence point */
226 /* Values for max_lazy_match, good_match and max_chain_length, depending on
227 * the desired pack level (0..9). The values given below have been tuned to
228 * exclude worst case performance for pathological files. Better values may be
229 * found for specific files.
232 typedef struct config {
233 ush good_length; /* reduce lazy search above this match length */
234 ush max_lazy; /* do not perform lazy search above this match length */
235 ush nice_length; /* quit search above this match length */
240 # define nice_match MAX_MATCH
242 int near nice_match; /* Stop searching when current match exceeds this */
245 local config configuration_table[10] = {
246 /* good lazy nice chain */
247 /* 0 */ {0, 0, 0, 0}, /* store only */
248 /* 1 */ {4, 4, 8, 4}, /* maximum speed, no lazy matches */
249 /* 2 */ {4, 5, 16, 8},
250 /* 3 */ {4, 6, 32, 32},
252 /* 4 */ {4, 4, 16, 16}, /* lazy matches */
253 /* 5 */ {8, 16, 32, 32},
254 /* 6 */ {8, 16, 128, 128},
255 /* 7 */ {8, 32, 128, 256},
256 /* 8 */ {32, 128, 258, 1024},
257 /* 9 */ {32, 258, 258, 4096}}; /* maximum compression */
259 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
260 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
264 /* ===========================================================================
265 * Prototypes for local functions.
267 local void fill_window OF((void));
268 local off_t deflate_fast OF((void));
270 int longest_match OF((IPos cur_match));
272 void match_init OF((void)); /* asm code initialization */
276 local void check_match OF((IPos start, IPos match, int length));
279 /* ===========================================================================
280 * Update a hash value with the given input byte
281 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
282 * input characters, so that a running hash key can be computed from the
283 * previous key instead of complete recalculation each time.
285 #define UPDATE_HASH(h,c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
287 /* ===========================================================================
288 * Insert string s in the dictionary and set match_head to the previous head
289 * of the hash chain (the most recent string with same hash key). Return
290 * the previous length of the hash chain.
291 * IN assertion: all calls to to INSERT_STRING are made with consecutive
292 * input characters and the first MIN_MATCH bytes of s are valid
293 * (except for the last MIN_MATCH-1 bytes of the input file).
295 #define INSERT_STRING(s, match_head) \
296 (UPDATE_HASH(ins_h, window[(s) + MIN_MATCH-1]), \
297 prev[(s) & WMASK] = match_head = head[ins_h], \
300 /* ===========================================================================
301 * Initialize the "longest match" routines for a new file
303 void lm_init (pack_level, flags)
304 int pack_level; /* 0: store, 1: best speed, 9: best compression */
305 ush *flags; /* general purpose bit flag */
309 if (pack_level < 1 || pack_level > 9) gzip_error ("bad pack level");
310 compr_level = pack_level;
312 /* Initialize the hash table. */
313 #if defined(MAXSEG_64K) && HASH_BITS == 15
314 for (j = 0; j < HASH_SIZE; j++) head[j] = NIL;
316 memzero((char*)head, HASH_SIZE*sizeof(*head));
318 /* prev will be initialized on the fly */
321 rsync_chunk_end = 0xFFFFFFFFUL;
324 /* Set the default configuration parameters:
326 max_lazy_match = configuration_table[pack_level].max_lazy;
327 good_match = configuration_table[pack_level].good_length;
329 nice_match = configuration_table[pack_level].nice_length;
331 max_chain_length = configuration_table[pack_level].max_chain;
332 if (pack_level == 1) {
334 } else if (pack_level == 9) {
337 /* ??? reduce max_chain_length for binary files */
341 rsync_chunk_end = 0xFFFFFFFFUL;
343 match_init(); /* initialize the asm code */
346 lookahead = read_buf((char*)window,
347 sizeof(int) <= 2 ? (unsigned)WSIZE : 2*WSIZE);
349 if (lookahead == 0 || lookahead == (unsigned)EOF) {
350 eofile = 1, lookahead = 0;
354 /* Make sure that we always have enough lookahead. This is important
355 * if input comes from a device such as a tty.
357 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
360 for (j=0; j<MIN_MATCH-1; j++) UPDATE_HASH(ins_h, window[j]);
361 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
362 * not important since only literal bytes will be emitted.
366 /* ===========================================================================
367 * Set match_start to the longest match starting at the given string and
368 * return its length. Matches shorter or equal to prev_length are discarded,
369 * in which case the result is equal to prev_length and match_start is
371 * IN assertions: cur_match is the head of the hash chain for the current
372 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
375 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
376 * match.s. The code is functionally equivalent, so you can use the C version
379 int longest_match(cur_match)
380 IPos cur_match; /* current match */
382 unsigned chain_length = max_chain_length; /* max hash chain length */
383 register uch *scan = window + strstart; /* current string */
384 register uch *match; /* matched string */
385 register int len; /* length of current match */
386 int best_len = prev_length; /* best match length so far */
387 IPos limit = strstart > (IPos)MAX_DIST ? strstart - (IPos)MAX_DIST : NIL;
388 /* Stop when cur_match becomes <= limit. To simplify the code,
389 * we prevent matches with the string of window index 0.
392 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
393 * It is easy to get rid of this optimization if necessary.
395 #if HASH_BITS < 8 || MAX_MATCH != 258
396 error: Code too clever
400 /* Compare two bytes at a time. Note: this is not always beneficial.
401 * Try with and without -DUNALIGNED_OK to check.
403 register uch *strend = window + strstart + MAX_MATCH - 1;
404 register ush scan_start = *(ush*)scan;
405 register ush scan_end = *(ush*)(scan+best_len-1);
407 register uch *strend = window + strstart + MAX_MATCH;
408 register uch scan_end1 = scan[best_len-1];
409 register uch scan_end = scan[best_len];
412 /* Do not waste too much time if we already have a good match: */
413 if (prev_length >= good_match) {
416 Assert(strstart <= window_size-MIN_LOOKAHEAD, "insufficient lookahead");
419 Assert(cur_match < strstart, "no future");
420 match = window + cur_match;
422 /* Skip to next match if the match length cannot increase
423 * or if the match length is less than 2:
425 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
426 /* This code assumes sizeof(unsigned short) == 2. Do not use
427 * UNALIGNED_OK if your compiler uses a different size.
429 if (*(ush*)(match+best_len-1) != scan_end ||
430 *(ush*)match != scan_start) continue;
432 /* It is not necessary to compare scan[2] and match[2] since they are
433 * always equal when the other bytes match, given that the hash keys
434 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
435 * strstart+3, +5, ... up to strstart+257. We check for insufficient
436 * lookahead only every 4th comparison; the 128th check will be made
437 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
438 * necessary to put more guard bytes at the end of the window, or
439 * to check more often for insufficient lookahead.
443 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
444 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
445 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
446 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
448 /* The funny "do {}" generates better code on most compilers */
450 /* Here, scan <= window+strstart+257 */
451 Assert(scan <= window+(unsigned)(window_size-1), "wild scan");
452 if (*scan == *match) scan++;
454 len = (MAX_MATCH - 1) - (int)(strend-scan);
455 scan = strend - (MAX_MATCH-1);
457 #else /* UNALIGNED_OK */
459 if (match[best_len] != scan_end ||
460 match[best_len-1] != scan_end1 ||
462 *++match != scan[1]) continue;
464 /* The check at best_len-1 can be removed because it will be made
465 * again later. (This heuristic is not always a win.)
466 * It is not necessary to compare scan[2] and match[2] since they
467 * are always equal when the other bytes match, given that
468 * the hash keys are equal and that HASH_BITS >= 8.
472 /* We check for insufficient lookahead only every 8th comparison;
473 * the 256th check will be made at strstart+258.
476 } while (*++scan == *++match && *++scan == *++match &&
477 *++scan == *++match && *++scan == *++match &&
478 *++scan == *++match && *++scan == *++match &&
479 *++scan == *++match && *++scan == *++match &&
482 len = MAX_MATCH - (int)(strend - scan);
483 scan = strend - MAX_MATCH;
485 #endif /* UNALIGNED_OK */
487 if (len > best_len) {
488 match_start = cur_match;
490 if (len >= nice_match) break;
492 scan_end = *(ush*)(scan+best_len-1);
494 scan_end1 = scan[best_len-1];
495 scan_end = scan[best_len];
498 } while ((cur_match = prev[cur_match & WMASK]) > limit
499 && --chain_length != 0);
506 /* ===========================================================================
507 * Check that the match at match_start is indeed a match.
509 local void check_match(start, match, length)
513 /* check that the match is indeed a match */
514 if (memcmp((char*)window + match,
515 (char*)window + start, length) != 0) {
517 " start %d, match %d, length %d\n",
518 start, match, length);
519 gzip_error ("invalid match");
522 fprintf(stderr,"\\[%d,%d]", start-match, length);
523 do { putc(window[start++], stderr); } while (--length != 0);
527 # define check_match(start, match, length)
530 /* ===========================================================================
531 * Fill the window when the lookahead becomes insufficient.
532 * Updates strstart and lookahead, and sets eofile if end of input file.
533 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
534 * OUT assertions: at least one byte has been read, or eofile is set;
535 * file reads are performed for at least two bytes (required for the
536 * translate_eol option).
538 local void fill_window()
540 register unsigned n, m;
541 unsigned more = (unsigned)(window_size - (ulg)lookahead - (ulg)strstart);
542 /* Amount of free space at the end of the window. */
544 /* If the window is almost full and there is insufficient lookahead,
545 * move the upper half to the lower one to make room in the upper half.
547 if (more == (unsigned)EOF) {
548 /* Very unlikely, but possible on 16 bit machine if strstart == 0
549 * and lookahead == 1 (input done one byte at time)
552 } else if (strstart >= WSIZE+MAX_DIST) {
553 /* By the IN assertion, the window is not empty so we can't confuse
554 * more == 0 with more == 64K on a 16 bit machine.
556 Assert(window_size == (ulg)2*WSIZE, "no sliding with BIG_MEM");
558 memcpy((char*)window, (char*)window+WSIZE, (unsigned)WSIZE);
559 match_start -= WSIZE;
560 strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
561 if (rsync_chunk_end != 0xFFFFFFFFUL)
562 rsync_chunk_end -= WSIZE;
564 block_start -= (long) WSIZE;
566 for (n = 0; n < HASH_SIZE; n++) {
568 head[n] = (Pos)(m >= WSIZE ? m-WSIZE : NIL);
570 for (n = 0; n < WSIZE; n++) {
572 prev[n] = (Pos)(m >= WSIZE ? m-WSIZE : NIL);
573 /* If n is not on any hash chain, prev[n] is garbage but
574 * its value will never be used.
579 /* At this point, more >= 2 */
581 n = read_buf((char*)window+strstart+lookahead, more);
582 if (n == 0 || n == (unsigned)EOF) {
584 /* Don't let garbage pollute the dictionary. */
585 memzero (window + strstart + lookahead, MIN_MATCH - 1);
592 local void rsync_roll(start, num)
598 if (start < RSYNC_WIN) {
599 /* before window fills. */
600 for (i = start; i < RSYNC_WIN; i++) {
601 if (i == start + num) return;
602 rsync_sum += (ulg)window[i];
604 num -= (RSYNC_WIN - start);
608 /* buffer after window full */
609 for (i = start; i < start+num; i++) {
610 /* New character in */
611 rsync_sum += (ulg)window[i];
612 /* Old character out */
613 rsync_sum -= (ulg)window[i - RSYNC_WIN];
614 if (rsync_chunk_end == 0xFFFFFFFFUL && RSYNC_SUM_MATCH(rsync_sum))
619 /* ===========================================================================
620 * Set rsync_chunk_end if window sum matches magic value.
622 #define RSYNC_ROLL(s, n) \
623 do { if (rsync) rsync_roll((s), (n)); } while(0)
625 /* ===========================================================================
626 * Flush the current block, with given end-of-file flag.
627 * IN assertion: strstart is set to the end of the current match.
629 #define FLUSH_BLOCK(eof) \
630 flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \
631 (char*)NULL, (long)strstart - block_start, (eof))
633 /* ===========================================================================
634 * Processes a new input file and return its compressed length. This
635 * function does not perform lazy evaluationof matches and inserts
636 * new strings in the dictionary only for unmatched strings or for short
637 * matches. It is used only for the fast compression options.
639 local off_t deflate_fast()
641 IPos hash_head; /* head of the hash chain */
642 int flush; /* set if current block must be flushed */
643 unsigned match_length = 0; /* length of best match */
645 prev_length = MIN_MATCH-1;
646 while (lookahead != 0) {
647 /* Insert the string window[strstart .. strstart+2] in the
648 * dictionary, and set hash_head to the head of the hash chain:
650 INSERT_STRING(strstart, hash_head);
652 /* Find the longest match, discarding those <= prev_length.
653 * At this point we have always match_length < MIN_MATCH
655 if (hash_head != NIL && strstart - hash_head <= MAX_DIST
656 && strstart <= window_size - MIN_LOOKAHEAD) {
657 /* To simplify the code, we prevent matches with the string
658 * of window index 0 (in particular we have to avoid a match
659 * of the string with itself at the start of the input file).
661 match_length = longest_match (hash_head);
662 /* longest_match() sets match_start */
663 if (match_length > lookahead) match_length = lookahead;
665 if (match_length >= MIN_MATCH) {
666 check_match(strstart, match_start, match_length);
668 flush = ct_tally(strstart-match_start, match_length - MIN_MATCH);
670 lookahead -= match_length;
672 RSYNC_ROLL(strstart, match_length);
673 /* Insert new strings in the hash table only if the match length
674 * is not too large. This saves time but degrades compression.
676 if (match_length <= max_insert_length) {
677 match_length--; /* string at strstart already in hash table */
680 INSERT_STRING(strstart, hash_head);
681 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
682 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
683 * these bytes are garbage, but it does not matter since
684 * the next lookahead bytes will be emitted as literals.
686 } while (--match_length != 0);
689 strstart += match_length;
691 ins_h = window[strstart];
692 UPDATE_HASH(ins_h, window[strstart+1]);
694 Call UPDATE_HASH() MIN_MATCH-3 more times
698 /* No match, output a literal byte */
699 Tracevv((stderr,"%c",window[strstart]));
700 flush = ct_tally (0, window[strstart]);
701 RSYNC_ROLL(strstart, 1);
705 if (rsync && strstart > rsync_chunk_end) {
706 ush attr = 0; /* ascii/binary flag */
709 /* Reset huffman tree */
710 ct_init(&attr, &method);
711 rsync_chunk_end = 0xFFFFFFFFUL;
713 if (flush) FLUSH_BLOCK(0), block_start = strstart;
715 /* Make sure that we always have enough lookahead, except
716 * at the end of the input file. We need MAX_MATCH bytes
717 * for the next match, plus MIN_MATCH bytes to insert the
718 * string following the next match.
720 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
723 return FLUSH_BLOCK(1); /* eof */
726 /* ===========================================================================
727 * Same as above, but achieves better compression. We use a lazy
728 * evaluation for matches: a match is finally adopted only if there is
729 * no better match at the next window position.
733 IPos hash_head; /* head of hash chain */
734 IPos prev_match; /* previous match */
735 int flush; /* set if current block must be flushed */
736 int match_available = 0; /* set if previous match exists */
737 register unsigned match_length = MIN_MATCH-1; /* length of best match */
739 if (compr_level <= 3) return deflate_fast(); /* optimized for speed */
741 /* Process the input block. */
742 while (lookahead != 0) {
743 /* Insert the string window[strstart .. strstart+2] in the
744 * dictionary, and set hash_head to the head of the hash chain:
746 INSERT_STRING(strstart, hash_head);
748 /* Find the longest match, discarding those <= prev_length.
750 prev_length = match_length, prev_match = match_start;
751 match_length = MIN_MATCH-1;
753 if (hash_head != NIL && prev_length < max_lazy_match &&
754 strstart - hash_head <= MAX_DIST &&
755 strstart <= window_size - MIN_LOOKAHEAD) {
756 /* To simplify the code, we prevent matches with the string
757 * of window index 0 (in particular we have to avoid a match
758 * of the string with itself at the start of the input file).
760 match_length = longest_match (hash_head);
761 /* longest_match() sets match_start */
762 if (match_length > lookahead) match_length = lookahead;
764 /* Ignore a length 3 match if it is too distant: */
765 if (match_length == MIN_MATCH && strstart-match_start > TOO_FAR){
766 /* If prev_match is also MIN_MATCH, match_start is garbage
767 * but we will ignore the current match anyway.
772 /* If there was a match at the previous step and the current
773 * match is not better, output the previous match:
775 if (prev_length >= MIN_MATCH && match_length <= prev_length) {
777 check_match(strstart-1, prev_match, prev_length);
779 flush = ct_tally(strstart-1-prev_match, prev_length - MIN_MATCH);
781 /* Insert in hash table all strings up to the end of the match.
782 * strstart-1 and strstart are already inserted.
784 lookahead -= prev_length-1;
786 RSYNC_ROLL(strstart, prev_length+1);
789 INSERT_STRING(strstart, hash_head);
790 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
791 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
792 * these bytes are garbage, but it does not matter since the
793 * next lookahead bytes will always be emitted as literals.
795 } while (--prev_length != 0);
797 match_length = MIN_MATCH-1;
800 if (rsync && strstart > rsync_chunk_end) {
801 ush attr = 0; /* ascii/binary flag */
803 /* Reset huffman tree */
804 ct_init(&attr, &method);
805 rsync_chunk_end = 0xFFFFFFFFUL;
808 if (flush) FLUSH_BLOCK(0), block_start = strstart;
809 } else if (match_available) {
810 /* If there was no match at the previous position, output a
811 * single literal. If there was a match but the current match
812 * is longer, truncate the previous match to a single literal.
814 Tracevv((stderr,"%c",window[strstart-1]));
815 flush = ct_tally (0, window[strstart-1]);
816 if (rsync && strstart > rsync_chunk_end) {
817 ush attr = 0; /* ascii/binary flag */
819 /* Reset huffman tree */
820 ct_init(&attr, &method);
821 rsync_chunk_end = 0xFFFFFFFFUL;
825 if (flush) FLUSH_BLOCK(0), block_start = strstart;
826 RSYNC_ROLL(strstart, 1);
830 /* There is no previous match to compare with, wait for
831 * the next step to decide.
833 if (rsync && strstart > rsync_chunk_end) {
834 ush attr = 0; /* ascii/binary flag */
836 /* Reset huffman tree */
837 ct_init(&attr, &method);
838 rsync_chunk_end = 0xFFFFFFFFUL;
840 FLUSH_BLOCK(0), block_start = strstart;
843 RSYNC_ROLL(strstart, 1);
847 Assert (strstart <= bytes_in && lookahead <= bytes_in, "a bit too far");
849 /* Make sure that we always have enough lookahead, except
850 * at the end of the input file. We need MAX_MATCH bytes
851 * for the next match, plus MIN_MATCH bytes to insert the
852 * string following the next match.
854 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
856 if (match_available) ct_tally (0, window[strstart-1]);
858 return FLUSH_BLOCK(1); /* eof */