1 /* deflate.c -- compress data using the deflation algorithm
3 Copyright (C) 1999, 2006 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 2, 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 */
85 static char rcsid[] = "$Id: deflate.c,v 1.5 2006/12/07 23:53:00 eggert Exp $";
88 /* ===========================================================================
89 * Configuration parameters
92 /* Compile with MEDIUM_MEM to reduce the memory requirements or
93 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
94 * entire input file can be held in memory (not possible on 16 bit systems).
95 * Warning: defining these symbols affects HASH_BITS (see below) and thus
96 * affects the compression ratio. The compressed output
97 * is still correct, and might even be smaller in some cases.
101 # define HASH_BITS 13 /* Number of bits used to hash strings */
104 # define HASH_BITS 14
107 # define HASH_BITS 15
108 /* For portability to 16 bit machines, do not use values above 15. */
111 /* To save space (see unlzw.c), we overlay prev+head with tab_prefix and
112 * window with tab_suffix. Check that we can do this:
114 #if (WSIZE<<1) > (1<<BITS)
115 error: cannot overlay window with tab_suffix and prev with tab_prefix0
117 #if HASH_BITS > BITS-1
118 error: cannot overlay head with tab_prefix1
121 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
122 #define HASH_MASK (HASH_SIZE-1)
123 #define WMASK (WSIZE-1)
124 /* HASH_SIZE and WSIZE must be powers of two */
127 /* Tail of hash chains */
131 /* speed options for the general purpose bit flag */
134 # define TOO_FAR 4096
136 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
139 # define RSYNC_WIN 4096
141 /* Size of rsync window, must be < MAX_DIST */
143 #define RSYNC_SUM_MATCH(sum) ((sum) % RSYNC_WIN == 0)
144 /* Whether window sum matches magic value */
146 /* ===========================================================================
147 * Local data used by the "longest match" routines.
151 typedef unsigned IPos;
152 /* A Pos is an index in the character window. We use short instead of int to
153 * save space in the various tables. IPos is used only for parameter passing.
156 /* DECLARE(uch, window, 2L*WSIZE); */
157 /* Sliding window. Input bytes are read into the second half of the window,
158 * and move to the first half later to keep a dictionary of at least WSIZE
159 * bytes. With this organization, matches are limited to a distance of
160 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
161 * performed with a length multiple of the block size. Also, it limits
162 * the window size to 64K, which is quite useful on MSDOS.
163 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
164 * be less efficient).
167 /* DECLARE(Pos, prev, WSIZE); */
168 /* Link to older string with same hash index. To limit the size of this
169 * array to 64K, this link is maintained only for the last 32K strings.
170 * An index in this array is thus a window index modulo 32K.
173 /* DECLARE(Pos, head, 1<<HASH_BITS); */
174 /* Heads of the hash chains or NIL. */
176 ulg window_size = (ulg)2*WSIZE;
177 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
178 * input file length plus MIN_LOOKAHEAD.
182 /* window position at the beginning of the current output block. Gets
183 * negative when the window is moved backwards.
186 local unsigned ins_h; /* hash index of string to be inserted */
188 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1)/MIN_MATCH)
189 /* Number of bits by which ins_h and del_h must be shifted at each
190 * input step. It must be such that after MIN_MATCH steps, the oldest
191 * byte no longer takes part in the hash key, that is:
192 * H_SHIFT * MIN_MATCH >= HASH_BITS
195 unsigned int near prev_length;
196 /* Length of the best match at previous step. Matches not greater than this
197 * are discarded. This is used in the lazy match evaluation.
200 unsigned near strstart; /* start of string to insert */
201 unsigned near match_start; /* start of matching string */
202 local int eofile; /* flag set at end of input file */
203 local unsigned lookahead; /* number of valid bytes ahead in window */
205 unsigned near max_chain_length;
206 /* To speed up deflation, hash chains are never searched beyond this length.
207 * A higher limit improves compression ratio but degrades the speed.
210 local unsigned int max_lazy_match;
211 /* Attempt to find a better match only when the current match is strictly
212 * smaller than this value. This mechanism is used only for compression
215 #define max_insert_length max_lazy_match
216 /* Insert new strings in the hash table only if the match length
217 * is not greater than this length. This saves time but degrades compression.
218 * max_insert_length is used only for compression levels <= 3.
221 local int compr_level;
222 /* compression level (1..9) */
224 unsigned near good_match;
225 /* Use a faster search when the previous match is longer than this */
227 local ulg rsync_sum; /* rolling sum of rsync window */
228 local ulg rsync_chunk_end; /* next rsync sequence point */
230 /* Values for max_lazy_match, good_match and max_chain_length, depending on
231 * the desired pack level (0..9). The values given below have been tuned to
232 * exclude worst case performance for pathological files. Better values may be
233 * found for specific files.
236 typedef struct config {
237 ush good_length; /* reduce lazy search above this match length */
238 ush max_lazy; /* do not perform lazy search above this match length */
239 ush nice_length; /* quit search above this match length */
244 # define nice_match MAX_MATCH
246 int near nice_match; /* Stop searching when current match exceeds this */
249 local config configuration_table[10] = {
250 /* good lazy nice chain */
251 /* 0 */ {0, 0, 0, 0}, /* store only */
252 /* 1 */ {4, 4, 8, 4}, /* maximum speed, no lazy matches */
253 /* 2 */ {4, 5, 16, 8},
254 /* 3 */ {4, 6, 32, 32},
256 /* 4 */ {4, 4, 16, 16}, /* lazy matches */
257 /* 5 */ {8, 16, 32, 32},
258 /* 6 */ {8, 16, 128, 128},
259 /* 7 */ {8, 32, 128, 256},
260 /* 8 */ {32, 128, 258, 1024},
261 /* 9 */ {32, 258, 258, 4096}}; /* maximum compression */
263 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
264 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
269 /* result of memcmp for equal strings */
271 /* ===========================================================================
272 * Prototypes for local functions.
274 local void fill_window OF((void));
275 local off_t deflate_fast OF((void));
277 int longest_match OF((IPos cur_match));
279 void match_init OF((void)); /* asm code initialization */
283 local void check_match OF((IPos start, IPos match, int length));
286 /* ===========================================================================
287 * Update a hash value with the given input byte
288 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
289 * input characters, so that a running hash key can be computed from the
290 * previous key instead of complete recalculation each time.
292 #define UPDATE_HASH(h,c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
294 /* ===========================================================================
295 * Insert string s in the dictionary and set match_head to the previous head
296 * of the hash chain (the most recent string with same hash key). Return
297 * the previous length of the hash chain.
298 * IN assertion: all calls to to INSERT_STRING are made with consecutive
299 * input characters and the first MIN_MATCH bytes of s are valid
300 * (except for the last MIN_MATCH-1 bytes of the input file).
302 #define INSERT_STRING(s, match_head) \
303 (UPDATE_HASH(ins_h, window[(s) + MIN_MATCH-1]), \
304 prev[(s) & WMASK] = match_head = head[ins_h], \
307 /* ===========================================================================
308 * Initialize the "longest match" routines for a new file
310 void lm_init (pack_level, flags)
311 int pack_level; /* 0: store, 1: best speed, 9: best compression */
312 ush *flags; /* general purpose bit flag */
316 if (pack_level < 1 || pack_level > 9) gzip_error ("bad pack level");
317 compr_level = pack_level;
319 /* Initialize the hash table. */
320 #if defined(MAXSEG_64K) && HASH_BITS == 15
321 for (j = 0; j < HASH_SIZE; j++) head[j] = NIL;
323 memzero((char*)head, HASH_SIZE*sizeof(*head));
325 /* prev will be initialized on the fly */
328 rsync_chunk_end = 0xFFFFFFFFUL;
331 /* Set the default configuration parameters:
333 max_lazy_match = configuration_table[pack_level].max_lazy;
334 good_match = configuration_table[pack_level].good_length;
336 nice_match = configuration_table[pack_level].nice_length;
338 max_chain_length = configuration_table[pack_level].max_chain;
339 if (pack_level == 1) {
341 } else if (pack_level == 9) {
344 /* ??? reduce max_chain_length for binary files */
349 match_init(); /* initialize the asm code */
352 lookahead = read_buf((char*)window,
353 sizeof(int) <= 2 ? (unsigned)WSIZE : 2*WSIZE);
355 if (lookahead == 0 || lookahead == (unsigned)EOF) {
356 eofile = 1, lookahead = 0;
360 /* Make sure that we always have enough lookahead. This is important
361 * if input comes from a device such as a tty.
363 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
366 for (j=0; j<MIN_MATCH-1; j++) UPDATE_HASH(ins_h, window[j]);
367 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
368 * not important since only literal bytes will be emitted.
372 /* ===========================================================================
373 * Set match_start to the longest match starting at the given string and
374 * return its length. Matches shorter or equal to prev_length are discarded,
375 * in which case the result is equal to prev_length and match_start is
377 * IN assertions: cur_match is the head of the hash chain for the current
378 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
381 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
382 * match.s. The code is functionally equivalent, so you can use the C version
385 int longest_match(cur_match)
386 IPos cur_match; /* current match */
388 unsigned chain_length = max_chain_length; /* max hash chain length */
389 register uch *scan = window + strstart; /* current string */
390 register uch *match; /* matched string */
391 register int len; /* length of current match */
392 int best_len = prev_length; /* best match length so far */
393 IPos limit = strstart > (IPos)MAX_DIST ? strstart - (IPos)MAX_DIST : NIL;
394 /* Stop when cur_match becomes <= limit. To simplify the code,
395 * we prevent matches with the string of window index 0.
398 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
399 * It is easy to get rid of this optimization if necessary.
401 #if HASH_BITS < 8 || MAX_MATCH != 258
402 error: Code too clever
406 /* Compare two bytes at a time. Note: this is not always beneficial.
407 * Try with and without -DUNALIGNED_OK to check.
409 register uch *strend = window + strstart + MAX_MATCH - 1;
410 register ush scan_start = *(ush*)scan;
411 register ush scan_end = *(ush*)(scan+best_len-1);
413 register uch *strend = window + strstart + MAX_MATCH;
414 register uch scan_end1 = scan[best_len-1];
415 register uch scan_end = scan[best_len];
418 /* Do not waste too much time if we already have a good match: */
419 if (prev_length >= good_match) {
422 Assert(strstart <= window_size-MIN_LOOKAHEAD, "insufficient lookahead");
425 Assert(cur_match < strstart, "no future");
426 match = window + cur_match;
428 /* Skip to next match if the match length cannot increase
429 * or if the match length is less than 2:
431 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
432 /* This code assumes sizeof(unsigned short) == 2. Do not use
433 * UNALIGNED_OK if your compiler uses a different size.
435 if (*(ush*)(match+best_len-1) != scan_end ||
436 *(ush*)match != scan_start) continue;
438 /* It is not necessary to compare scan[2] and match[2] since they are
439 * always equal when the other bytes match, given that the hash keys
440 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
441 * strstart+3, +5, ... up to strstart+257. We check for insufficient
442 * lookahead only every 4th comparison; the 128th check will be made
443 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
444 * necessary to put more guard bytes at the end of the window, or
445 * to check more often for insufficient lookahead.
449 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
450 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
451 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
452 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
454 /* The funny "do {}" generates better code on most compilers */
456 /* Here, scan <= window+strstart+257 */
457 Assert(scan <= window+(unsigned)(window_size-1), "wild scan");
458 if (*scan == *match) scan++;
460 len = (MAX_MATCH - 1) - (int)(strend-scan);
461 scan = strend - (MAX_MATCH-1);
463 #else /* UNALIGNED_OK */
465 if (match[best_len] != scan_end ||
466 match[best_len-1] != scan_end1 ||
468 *++match != scan[1]) continue;
470 /* The check at best_len-1 can be removed because it will be made
471 * again later. (This heuristic is not always a win.)
472 * It is not necessary to compare scan[2] and match[2] since they
473 * are always equal when the other bytes match, given that
474 * the hash keys are equal and that HASH_BITS >= 8.
478 /* We check for insufficient lookahead only every 8th comparison;
479 * the 256th check will be made at strstart+258.
482 } while (*++scan == *++match && *++scan == *++match &&
483 *++scan == *++match && *++scan == *++match &&
484 *++scan == *++match && *++scan == *++match &&
485 *++scan == *++match && *++scan == *++match &&
488 len = MAX_MATCH - (int)(strend - scan);
489 scan = strend - MAX_MATCH;
491 #endif /* UNALIGNED_OK */
493 if (len > best_len) {
494 match_start = cur_match;
496 if (len >= nice_match) break;
498 scan_end = *(ush*)(scan+best_len-1);
500 scan_end1 = scan[best_len-1];
501 scan_end = scan[best_len];
504 } while ((cur_match = prev[cur_match & WMASK]) > limit
505 && --chain_length != 0);
512 /* ===========================================================================
513 * Check that the match at match_start is indeed a match.
515 local void check_match(start, match, length)
519 /* check that the match is indeed a match */
520 if (memcmp((char*)window + match,
521 (char*)window + start, length) != EQUAL) {
523 " start %d, match %d, length %d\n",
524 start, match, length);
525 gzip_error ("invalid match");
528 fprintf(stderr,"\\[%d,%d]", start-match, length);
529 do { putc(window[start++], stderr); } while (--length != 0);
533 # define check_match(start, match, length)
536 /* ===========================================================================
537 * Fill the window when the lookahead becomes insufficient.
538 * Updates strstart and lookahead, and sets eofile if end of input file.
539 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
540 * OUT assertions: at least one byte has been read, or eofile is set;
541 * file reads are performed for at least two bytes (required for the
542 * translate_eol option).
544 local void fill_window()
546 register unsigned n, m;
547 unsigned more = (unsigned)(window_size - (ulg)lookahead - (ulg)strstart);
548 /* Amount of free space at the end of the window. */
550 /* If the window is almost full and there is insufficient lookahead,
551 * move the upper half to the lower one to make room in the upper half.
553 if (more == (unsigned)EOF) {
554 /* Very unlikely, but possible on 16 bit machine if strstart == 0
555 * and lookahead == 1 (input done one byte at time)
558 } else if (strstart >= WSIZE+MAX_DIST) {
559 /* By the IN assertion, the window is not empty so we can't confuse
560 * more == 0 with more == 64K on a 16 bit machine.
562 Assert(window_size == (ulg)2*WSIZE, "no sliding with BIG_MEM");
564 memcpy((char*)window, (char*)window+WSIZE, (unsigned)WSIZE);
565 match_start -= WSIZE;
566 strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
567 if (rsync_chunk_end != 0xFFFFFFFFUL)
568 rsync_chunk_end -= WSIZE;
570 block_start -= (long) WSIZE;
572 for (n = 0; n < HASH_SIZE; n++) {
574 head[n] = (Pos)(m >= WSIZE ? m-WSIZE : NIL);
576 for (n = 0; n < WSIZE; n++) {
578 prev[n] = (Pos)(m >= WSIZE ? m-WSIZE : NIL);
579 /* If n is not on any hash chain, prev[n] is garbage but
580 * its value will never be used.
585 /* At this point, more >= 2 */
587 n = read_buf((char*)window+strstart+lookahead, more);
588 if (n == 0 || n == (unsigned)EOF) {
596 local void rsync_roll(start, num)
602 if (start < RSYNC_WIN) {
603 /* before window fills. */
604 for (i = start; i < RSYNC_WIN; i++) {
605 if (i == start + num) return;
606 rsync_sum += (ulg)window[i];
608 num -= (RSYNC_WIN - start);
612 /* buffer after window full */
613 for (i = start; i < start+num; i++) {
614 /* New character in */
615 rsync_sum += (ulg)window[i];
616 /* Old character out */
617 rsync_sum -= (ulg)window[i - RSYNC_WIN];
618 if (rsync_chunk_end == 0xFFFFFFFFUL && RSYNC_SUM_MATCH(rsync_sum))
623 /* ===========================================================================
624 * Set rsync_chunk_end if window sum matches magic value.
626 #define RSYNC_ROLL(s, n) \
627 do { if (rsync) rsync_roll((s), (n)); } while(0)
629 /* ===========================================================================
630 * Flush the current block, with given end-of-file flag.
631 * IN assertion: strstart is set to the end of the current match.
633 #define FLUSH_BLOCK(eof) \
634 flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \
635 (char*)NULL, (long)strstart - block_start, flush-1, (eof))
637 /* ===========================================================================
638 * Processes a new input file and return its compressed length. This
639 * function does not perform lazy evaluationof matches and inserts
640 * new strings in the dictionary only for unmatched strings or for short
641 * matches. It is used only for the fast compression options.
643 local off_t deflate_fast()
645 IPos hash_head; /* head of the hash chain */
646 int flush; /* set if current block must be flushed, 2=>and padded */
647 unsigned match_length = 0; /* length of best match */
649 prev_length = MIN_MATCH-1;
650 while (lookahead != 0) {
651 /* Insert the string window[strstart .. strstart+2] in the
652 * dictionary, and set hash_head to the head of the hash chain:
654 INSERT_STRING(strstart, hash_head);
656 /* Find the longest match, discarding those <= prev_length.
657 * At this point we have always match_length < MIN_MATCH
659 if (hash_head != NIL && strstart - hash_head <= MAX_DIST
660 && strstart <= window_size - MIN_LOOKAHEAD) {
661 /* To simplify the code, we prevent matches with the string
662 * of window index 0 (in particular we have to avoid a match
663 * of the string with itself at the start of the input file).
665 match_length = longest_match (hash_head);
666 /* longest_match() sets match_start */
667 if (match_length > lookahead) match_length = lookahead;
669 if (match_length >= MIN_MATCH) {
670 check_match(strstart, match_start, match_length);
672 flush = ct_tally(strstart-match_start, match_length - MIN_MATCH);
674 lookahead -= match_length;
676 RSYNC_ROLL(strstart, match_length);
677 /* Insert new strings in the hash table only if the match length
678 * is not too large. This saves time but degrades compression.
680 if (match_length <= max_insert_length) {
681 match_length--; /* string at strstart already in hash table */
684 INSERT_STRING(strstart, hash_head);
685 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
686 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
687 * these bytes are garbage, but it does not matter since
688 * the next lookahead bytes will be emitted as literals.
690 } while (--match_length != 0);
693 strstart += match_length;
695 ins_h = window[strstart];
696 UPDATE_HASH(ins_h, window[strstart+1]);
698 Call UPDATE_HASH() MIN_MATCH-3 more times
702 /* No match, output a literal byte */
703 Tracevv((stderr,"%c",window[strstart]));
704 flush = ct_tally (0, window[strstart]);
705 RSYNC_ROLL(strstart, 1);
709 if (rsync && strstart > rsync_chunk_end) {
710 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 rsync_chunk_end = 0xFFFFFFFFUL;
804 if (flush) FLUSH_BLOCK(0), block_start = strstart;
805 } else if (match_available) {
806 /* If there was no match at the previous position, output a
807 * single literal. If there was a match but the current match
808 * is longer, truncate the previous match to a single literal.
810 Tracevv((stderr,"%c",window[strstart-1]));
811 flush = ct_tally (0, window[strstart-1]);
812 if (rsync && strstart > rsync_chunk_end) {
813 rsync_chunk_end = 0xFFFFFFFFUL;
816 if (flush) FLUSH_BLOCK(0), block_start = strstart;
817 RSYNC_ROLL(strstart, 1);
821 /* There is no previous match to compare with, wait for
822 * the next step to decide.
824 if (rsync && strstart > rsync_chunk_end) {
825 /* Reset huffman tree */
826 rsync_chunk_end = 0xFFFFFFFFUL;
828 FLUSH_BLOCK(0), block_start = strstart;
831 RSYNC_ROLL(strstart, 1);
835 Assert (strstart <= bytes_in && lookahead <= bytes_in, "a bit too far");
837 /* Make sure that we always have enough lookahead, except
838 * at the end of the input file. We need MAX_MATCH bytes
839 * for the next match, plus MIN_MATCH bytes to insert the
840 * string following the next match.
842 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
844 if (match_available) ct_tally (0, window[strstart-1]);
846 return FLUSH_BLOCK(1); /* eof */