1 /* deflate.c -- compress data using the deflation algorithm
3 Copyright (C) 1999, 2006, 2009-2013 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 static 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 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) */
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 static_unless_ASMV
242 # define static_unless_ASMV static
246 # define nice_match MAX_MATCH
248 /* Stop searching when current match exceeds this */
249 static_unless_ASMV int nice_match;
252 local config configuration_table[10] = {
253 /* good lazy nice chain */
254 /* 0 */ {0, 0, 0, 0}, /* store only */
255 /* 1 */ {4, 4, 8, 4}, /* maximum speed, no lazy matches */
256 /* 2 */ {4, 5, 16, 8},
257 /* 3 */ {4, 6, 32, 32},
259 /* 4 */ {4, 4, 16, 16}, /* lazy matches */
260 /* 5 */ {8, 16, 32, 32},
261 /* 6 */ {8, 16, 128, 128},
262 /* 7 */ {8, 32, 128, 256},
263 /* 8 */ {32, 128, 258, 1024},
264 /* 9 */ {32, 258, 258, 4096}}; /* maximum compression */
266 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
267 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
271 /* ===========================================================================
272 * Prototypes for local functions.
274 local void fill_window (void);
275 local off_t deflate_fast (void);
278 int longest_match (IPos cur_match);
279 void match_init (void); /* asm code initialization */
283 local void check_match (IPos start, IPos match, int length);
286 /* ===========================================================================
287 * Update a hash value with the given input byte
288 * IN assertion: all calls 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 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 */
348 rsync_chunk_end = 0xFFFFFFFFUL;
350 match_init(); /* initialize the asm code */
353 lookahead = read_buf((char*)window,
354 sizeof(int) <= 2 ? (unsigned)WSIZE : 2*WSIZE);
356 if (lookahead == 0 || lookahead == (unsigned)EOF) {
357 eofile = 1, lookahead = 0;
361 /* Make sure that we always have enough lookahead. This is important
362 * if input comes from a device such as a tty.
364 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
367 for (j=0; j<MIN_MATCH-1; j++) UPDATE_HASH(ins_h, window[j]);
368 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
369 * not important since only literal bytes will be emitted.
373 /* ===========================================================================
374 * Set match_start to the longest match starting at the given string and
375 * return its length. Matches shorter or equal to prev_length are discarded,
376 * in which case the result is equal to prev_length and match_start is
378 * IN assertions: cur_match is the head of the hash chain for the current
379 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
382 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
383 * match.s. The code is functionally equivalent, so you can use the C version
387 longest_match(IPos cur_match)
389 unsigned chain_length = max_chain_length; /* max hash chain length */
390 register uch *scan = window + strstart; /* current string */
391 register uch *match; /* matched string */
392 register int len; /* length of current match */
393 int best_len = prev_length; /* best match length so far */
394 IPos limit = strstart > (IPos)MAX_DIST ? strstart - (IPos)MAX_DIST : NIL;
395 /* Stop when cur_match becomes <= limit. To simplify the code,
396 * we prevent matches with the string of window index 0.
399 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
400 * It is easy to get rid of this optimization if necessary.
402 #if HASH_BITS < 8 || MAX_MATCH != 258
403 error: Code too clever
407 /* Compare two bytes at a time. Note: this is not always beneficial.
408 * Try with and without -DUNALIGNED_OK to check.
410 register uch *strend = window + strstart + MAX_MATCH - 1;
411 register ush scan_start = *(ush*)scan;
412 register ush scan_end = *(ush*)(scan+best_len-1);
414 register uch *strend = window + strstart + MAX_MATCH;
415 register uch scan_end1 = scan[best_len-1];
416 register uch scan_end = scan[best_len];
419 /* Do not waste too much time if we already have a good match: */
420 if (prev_length >= good_match) {
423 Assert(strstart <= window_size-MIN_LOOKAHEAD, "insufficient lookahead");
426 Assert(cur_match < strstart, "no future");
427 match = window + cur_match;
429 /* Skip to next match if the match length cannot increase
430 * or if the match length is less than 2:
432 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
433 /* This code assumes sizeof(unsigned short) == 2. Do not use
434 * UNALIGNED_OK if your compiler uses a different size.
436 if (*(ush*)(match+best_len-1) != scan_end ||
437 *(ush*)match != scan_start) continue;
439 /* It is not necessary to compare scan[2] and match[2] since they are
440 * always equal when the other bytes match, given that the hash keys
441 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
442 * strstart+3, +5, ... up to strstart+257. We check for insufficient
443 * lookahead only every 4th comparison; the 128th check will be made
444 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
445 * necessary to put more guard bytes at the end of the window, or
446 * to check more often for insufficient lookahead.
450 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
451 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
452 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
453 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
455 /* The funny "do {}" generates better code on most compilers */
457 /* Here, scan <= window+strstart+257 */
458 Assert(scan <= window+(unsigned)(window_size-1), "wild scan");
459 if (*scan == *match) scan++;
461 len = (MAX_MATCH - 1) - (int)(strend-scan);
462 scan = strend - (MAX_MATCH-1);
464 #else /* UNALIGNED_OK */
466 if (match[best_len] != scan_end ||
467 match[best_len-1] != scan_end1 ||
469 *++match != scan[1]) continue;
471 /* The check at best_len-1 can be removed because it will be made
472 * again later. (This heuristic is not always a win.)
473 * It is not necessary to compare scan[2] and match[2] since they
474 * are always equal when the other bytes match, given that
475 * the hash keys are equal and that HASH_BITS >= 8.
479 /* We check for insufficient lookahead only every 8th comparison;
480 * the 256th check will be made at strstart+258.
483 } while (*++scan == *++match && *++scan == *++match &&
484 *++scan == *++match && *++scan == *++match &&
485 *++scan == *++match && *++scan == *++match &&
486 *++scan == *++match && *++scan == *++match &&
489 len = MAX_MATCH - (int)(strend - scan);
490 scan = strend - MAX_MATCH;
492 #endif /* UNALIGNED_OK */
494 if (len > best_len) {
495 match_start = cur_match;
497 if (len >= nice_match) break;
499 scan_end = *(ush*)(scan+best_len-1);
501 scan_end1 = scan[best_len-1];
502 scan_end = scan[best_len];
505 } while ((cur_match = prev[cur_match & WMASK]) > limit
506 && --chain_length != 0);
513 /* ===========================================================================
514 * Check that the match at match_start is indeed a match.
516 local void check_match(start, match, length)
520 /* check that the match is indeed a match */
521 if (memcmp((char*)window + match,
522 (char*)window + start, length) != 0) {
524 " start %d, match %d, length %d\n",
525 start, match, length);
526 gzip_error ("invalid match");
529 fprintf(stderr,"\\[%d,%d]", start-match, length);
530 do { putc(window[start++], stderr); } while (--length != 0);
534 # define check_match(start, match, length)
537 /* ===========================================================================
538 * Fill the window when the lookahead becomes insufficient.
539 * Updates strstart and lookahead, and sets eofile if end of input file.
540 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
541 * OUT assertions: at least one byte has been read, or eofile is set;
542 * file reads are performed for at least two bytes (required for the
543 * translate_eol option).
545 local void fill_window()
547 register unsigned n, m;
548 unsigned more = (unsigned)(window_size - (ulg)lookahead - (ulg)strstart);
549 /* Amount of free space at the end of the window. */
551 /* If the window is almost full and there is insufficient lookahead,
552 * move the upper half to the lower one to make room in the upper half.
554 if (more == (unsigned)EOF) {
555 /* Very unlikely, but possible on 16 bit machine if strstart == 0
556 * and lookahead == 1 (input done one byte at time)
559 } else if (strstart >= WSIZE+MAX_DIST) {
560 /* By the IN assertion, the window is not empty so we can't confuse
561 * more == 0 with more == 64K on a 16 bit machine.
563 Assert(window_size == (ulg)2*WSIZE, "no sliding with BIG_MEM");
565 memcpy((char*)window, (char*)window+WSIZE, (unsigned)WSIZE);
566 match_start -= WSIZE;
567 strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
568 if (rsync_chunk_end != 0xFFFFFFFFUL)
569 rsync_chunk_end -= WSIZE;
571 block_start -= (long) WSIZE;
573 for (n = 0; n < HASH_SIZE; n++) {
575 head[n] = (Pos)(m >= WSIZE ? m-WSIZE : NIL);
577 for (n = 0; n < WSIZE; n++) {
579 prev[n] = (Pos)(m >= WSIZE ? m-WSIZE : NIL);
580 /* If n is not on any hash chain, prev[n] is garbage but
581 * its value will never be used.
586 /* At this point, more >= 2 */
588 n = read_buf((char*)window+strstart+lookahead, more);
589 if (n == 0 || n == (unsigned)EOF) {
591 /* Don't let garbage pollute the dictionary. */
592 memzero (window + strstart + lookahead, MIN_MATCH - 1);
599 local void rsync_roll(start, num)
605 if (start < RSYNC_WIN) {
606 /* before window fills. */
607 for (i = start; i < RSYNC_WIN; i++) {
608 if (i == start + num) return;
609 rsync_sum += (ulg)window[i];
611 num -= (RSYNC_WIN - start);
615 /* buffer after window full */
616 for (i = start; i < start+num; i++) {
617 /* New character in */
618 rsync_sum += (ulg)window[i];
619 /* Old character out */
620 rsync_sum -= (ulg)window[i - RSYNC_WIN];
621 if (rsync_chunk_end == 0xFFFFFFFFUL && RSYNC_SUM_MATCH(rsync_sum))
626 /* ===========================================================================
627 * Set rsync_chunk_end if window sum matches magic value.
629 #define RSYNC_ROLL(s, n) \
630 do { if (rsync) rsync_roll((s), (n)); } while(0)
632 /* ===========================================================================
633 * Flush the current block, with given end-of-file flag.
634 * IN assertion: strstart is set to the end of the current match.
636 #define FLUSH_BLOCK(eof) \
637 flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \
638 (char*)NULL, (long)strstart - block_start, (eof))
640 /* ===========================================================================
641 * Processes a new input file and return its compressed length. This
642 * function does not perform lazy evaluationof matches and inserts
643 * new strings in the dictionary only for unmatched strings or for short
644 * matches. It is used only for the fast compression options.
646 local off_t deflate_fast()
648 IPos hash_head; /* head of the hash chain */
649 int flush; /* set if current block must be flushed */
650 unsigned match_length = 0; /* length of best match */
652 prev_length = MIN_MATCH-1;
653 while (lookahead != 0) {
654 /* Insert the string window[strstart .. strstart+2] in the
655 * dictionary, and set hash_head to the head of the hash chain:
657 INSERT_STRING(strstart, hash_head);
659 /* Find the longest match, discarding those <= prev_length.
660 * At this point we have always match_length < MIN_MATCH
662 if (hash_head != NIL && strstart - hash_head <= MAX_DIST
663 && strstart <= window_size - MIN_LOOKAHEAD) {
664 /* To simplify the code, we prevent matches with the string
665 * of window index 0 (in particular we have to avoid a match
666 * of the string with itself at the start of the input file).
668 match_length = longest_match (hash_head);
669 /* longest_match() sets match_start */
670 if (match_length > lookahead) match_length = lookahead;
672 if (match_length >= MIN_MATCH) {
673 check_match(strstart, match_start, match_length);
675 flush = ct_tally(strstart-match_start, match_length - MIN_MATCH);
677 lookahead -= match_length;
679 RSYNC_ROLL(strstart, match_length);
680 /* Insert new strings in the hash table only if the match length
681 * is not too large. This saves time but degrades compression.
683 if (match_length <= max_insert_length) {
684 match_length--; /* string at strstart already in hash table */
687 INSERT_STRING(strstart, hash_head);
688 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
689 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
690 * these bytes are garbage, but it does not matter since
691 * the next lookahead bytes will be emitted as literals.
693 } while (--match_length != 0);
696 strstart += match_length;
698 ins_h = window[strstart];
699 UPDATE_HASH(ins_h, window[strstart+1]);
701 Call UPDATE_HASH() MIN_MATCH-3 more times
705 /* No match, output a literal byte */
706 Tracevv((stderr,"%c",window[strstart]));
707 flush = ct_tally (0, window[strstart]);
708 RSYNC_ROLL(strstart, 1);
712 if (rsync && strstart > rsync_chunk_end) {
713 ush attr = 0; /* ascii/binary flag */
716 /* Reset huffman tree */
717 ct_init(&attr, &method);
718 rsync_chunk_end = 0xFFFFFFFFUL;
720 if (flush) FLUSH_BLOCK(0), block_start = strstart;
722 /* Make sure that we always have enough lookahead, except
723 * at the end of the input file. We need MAX_MATCH bytes
724 * for the next match, plus MIN_MATCH bytes to insert the
725 * string following the next match.
727 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
730 return FLUSH_BLOCK(1); /* eof */
733 /* ===========================================================================
734 * Same as above, but achieves better compression. We use a lazy
735 * evaluation for matches: a match is finally adopted only if there is
736 * no better match at the next window position.
740 IPos hash_head; /* head of hash chain */
741 IPos prev_match; /* previous match */
742 int flush; /* set if current block must be flushed */
743 int match_available = 0; /* set if previous match exists */
744 register unsigned match_length = MIN_MATCH-1; /* length of best match */
746 if (compr_level <= 3) return deflate_fast(); /* optimized for speed */
748 /* Process the input block. */
749 while (lookahead != 0) {
750 /* Insert the string window[strstart .. strstart+2] in the
751 * dictionary, and set hash_head to the head of the hash chain:
753 INSERT_STRING(strstart, hash_head);
755 /* Find the longest match, discarding those <= prev_length.
757 prev_length = match_length, prev_match = match_start;
758 match_length = MIN_MATCH-1;
760 if (hash_head != NIL && prev_length < max_lazy_match &&
761 strstart - hash_head <= MAX_DIST &&
762 strstart <= window_size - MIN_LOOKAHEAD) {
763 /* To simplify the code, we prevent matches with the string
764 * of window index 0 (in particular we have to avoid a match
765 * of the string with itself at the start of the input file).
767 match_length = longest_match (hash_head);
768 /* longest_match() sets match_start */
769 if (match_length > lookahead) match_length = lookahead;
771 /* Ignore a length 3 match if it is too distant: */
772 if (match_length == MIN_MATCH && strstart-match_start > TOO_FAR){
773 /* If prev_match is also MIN_MATCH, match_start is garbage
774 * but we will ignore the current match anyway.
779 /* If there was a match at the previous step and the current
780 * match is not better, output the previous match:
782 if (prev_length >= MIN_MATCH && match_length <= prev_length) {
784 check_match(strstart-1, prev_match, prev_length);
786 flush = ct_tally(strstart-1-prev_match, prev_length - MIN_MATCH);
788 /* Insert in hash table all strings up to the end of the match.
789 * strstart-1 and strstart are already inserted.
791 lookahead -= prev_length-1;
793 RSYNC_ROLL(strstart, prev_length+1);
796 INSERT_STRING(strstart, hash_head);
797 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
798 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
799 * these bytes are garbage, but it does not matter since the
800 * next lookahead bytes will always be emitted as literals.
802 } while (--prev_length != 0);
804 match_length = MIN_MATCH-1;
807 if (rsync && strstart > rsync_chunk_end) {
808 ush attr = 0; /* ascii/binary flag */
810 /* Reset huffman tree */
811 ct_init(&attr, &method);
812 rsync_chunk_end = 0xFFFFFFFFUL;
815 if (flush) FLUSH_BLOCK(0), block_start = strstart;
816 } else if (match_available) {
817 /* If there was no match at the previous position, output a
818 * single literal. If there was a match but the current match
819 * is longer, truncate the previous match to a single literal.
821 Tracevv((stderr,"%c",window[strstart-1]));
822 flush = ct_tally (0, window[strstart-1]);
823 if (rsync && strstart > rsync_chunk_end) {
824 ush attr = 0; /* ascii/binary flag */
826 /* Reset huffman tree */
827 ct_init(&attr, &method);
828 rsync_chunk_end = 0xFFFFFFFFUL;
832 if (flush) FLUSH_BLOCK(0), block_start = strstart;
833 RSYNC_ROLL(strstart, 1);
837 /* There is no previous match to compare with, wait for
838 * the next step to decide.
840 if (rsync && strstart > rsync_chunk_end) {
841 ush attr = 0; /* ascii/binary flag */
843 /* Reset huffman tree */
844 ct_init(&attr, &method);
845 rsync_chunk_end = 0xFFFFFFFFUL;
847 FLUSH_BLOCK(0), block_start = strstart;
850 RSYNC_ROLL(strstart, 1);
854 Assert (strstart <= bytes_in && lookahead <= bytes_in, "a bit too far");
856 /* Make sure that we always have enough lookahead, except
857 * at the end of the input file. We need MAX_MATCH bytes
858 * for the next match, plus MIN_MATCH bytes to insert the
859 * string following the next match.
861 while (lookahead < MIN_LOOKAHEAD && !eofile) fill_window();
863 if (match_available) ct_tally (0, window[strstart-1]);
865 return FLUSH_BLOCK(1); /* eof */