1 /*
2 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
3 *
4 * This code is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License version 2 only, as
6 * published by the Free Software Foundation. Oracle designates this
7 * particular file as subject to the "Classpath" exception as provided
8 * by Oracle in the LICENSE file that accompanied this code.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 */
24
25 /* deflate.c -- compress data using the deflation algorithm
26 * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler
27 * For conditions of distribution and use, see copyright notice in zlib.h
28 */
29
30 /*
31 * ALGORITHM
32 *
33 * The "deflation" process depends on being able to identify portions
34 * of the input text which are identical to earlier input (within a
35 * sliding window trailing behind the input currently being processed).
36 *
37 * The most straightforward technique turns out to be the fastest for
38 * most input files: try all possible matches and select the longest.
39 * The key feature of this algorithm is that insertions into the string
40 * dictionary are very simple and thus fast, and deletions are avoided
41 * completely. Insertions are performed at each input character, whereas
42 * string matches are performed only when the previous match ends. So it
43 * is preferable to spend more time in matches to allow very fast string
44 * insertions and avoid deletions. The matching algorithm for small
45 * strings is inspired from that of Rabin & Karp. A brute force approach
46 * is used to find longer strings when a small match has been found.
47 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
48 * (by Leonid Broukhis).
49 * A previous version of this file used a more sophisticated algorithm
50 * (by Fiala and Greene) which is guaranteed to run in linear amortized
51 * time, but has a larger average cost, uses more memory and is patented.
52 * However the F&G algorithm may be faster for some highly redundant
53 * files if the parameter max_chain_length (described below) is too large.
54 *
55 * ACKNOWLEDGEMENTS
56 *
57 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
58 * I found it in 'freeze' written by Leonid Broukhis.
59 * Thanks to many people for bug reports and testing.
60 *
61 * REFERENCES
62 *
63 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
64 * Available in http://tools.ietf.org/html/rfc1951
65 *
66 * A description of the Rabin and Karp algorithm is given in the book
67 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
68 *
69 * Fiala,E.R., and Greene,D.H.
70 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
71 *
72 */
73
74 /* @(#) $Id$ */
75
76 #include "deflate.h"
77
78 const char deflate_copyright[] =
79 " deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler ";
80 /*
81 If you use the zlib library in a product, an acknowledgment is welcome
82 in the documentation of your product. If for some reason you cannot
83 include such an acknowledgment, I would appreciate that you keep this
84 copyright string in the executable of your product.
85 */
86
87 /* ===========================================================================
88 * Function prototypes.
89 */
90 typedef enum {
91 need_more, /* block not completed, need more input or more output */
92 block_done, /* block flush performed */
93 finish_started, /* finish started, need only more output at next deflate */
94 finish_done /* finish done, accept no more input or output */
95 } block_state;
96
97 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
98 /* Compression function. Returns the block state after the call. */
99
100 local int deflateStateCheck OF((z_streamp strm));
101 local void slide_hash OF((deflate_state *s));
102 local void fill_window OF((deflate_state *s));
103 local block_state deflate_stored OF((deflate_state *s, int flush));
104 local block_state deflate_fast OF((deflate_state *s, int flush));
105 #ifndef FASTEST
106 local block_state deflate_slow OF((deflate_state *s, int flush));
107 #endif
108 local block_state deflate_rle OF((deflate_state *s, int flush));
109 local block_state deflate_huff OF((deflate_state *s, int flush));
110 local void lm_init OF((deflate_state *s));
111 local void putShortMSB OF((deflate_state *s, uInt b));
112 local void flush_pending OF((z_streamp strm));
113 local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
114 #ifdef ASMV
115 # pragma message("Assembler code may have bugs -- use at your own risk")
116 void match_init OF((void)); /* asm code initialization */
117 uInt longest_match OF((deflate_state *s, IPos cur_match));
118 #else
119 local uInt longest_match OF((deflate_state *s, IPos cur_match));
120 #endif
121
122 #ifdef ZLIB_DEBUG
123 local void check_match OF((deflate_state *s, IPos start, IPos match,
124 int length));
125 #endif
126
127 /* ===========================================================================
128 * Local data
129 */
130
131 #define NIL 0
132 /* Tail of hash chains */
133
134 #ifndef TOO_FAR
135 # define TOO_FAR 4096
136 #endif
137 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
138
139 /* Values for max_lazy_match, good_match and max_chain_length, depending on
140 * the desired pack level (0..9). The values given below have been tuned to
141 * exclude worst case performance for pathological files. Better values may be
142 * found for specific files.
143 */
144 typedef struct config_s {
145 ush good_length; /* reduce lazy search above this match length */
146 ush max_lazy; /* do not perform lazy search above this match length */
147 ush nice_length; /* quit search above this match length */
148 ush max_chain;
149 compress_func func;
150 } config;
151
152 #ifdef FASTEST
153 local const config configuration_table[2] = {
154 /* good lazy nice chain */
155 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
156 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
157 #else
158 local const config configuration_table[10] = {
159 /* good lazy nice chain */
160 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
161 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
162 /* 2 */ {4, 5, 16, 8, deflate_fast},
163 /* 3 */ {4, 6, 32, 32, deflate_fast},
164
165 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
166 /* 5 */ {8, 16, 32, 32, deflate_slow},
167 /* 6 */ {8, 16, 128, 128, deflate_slow},
168 /* 7 */ {8, 32, 128, 256, deflate_slow},
169 /* 8 */ {32, 128, 258, 1024, deflate_slow},
170 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
171 #endif
172
173 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
174 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
175 * meaning.
176 */
177
178 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
179 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
180
181 /* ===========================================================================
182 * Update a hash value with the given input byte
183 * IN assertion: all calls to UPDATE_HASH are made with consecutive input
184 * characters, so that a running hash key can be computed from the previous
185 * key instead of complete recalculation each time.
186 */
187 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
188
189
190 /* ===========================================================================
191 * Insert string str in the dictionary and set match_head to the previous head
192 * of the hash chain (the most recent string with same hash key). Return
193 * the previous length of the hash chain.
194 * If this file is compiled with -DFASTEST, the compression level is forced
195 * to 1, and no hash chains are maintained.
196 * IN assertion: all calls to INSERT_STRING are made with consecutive input
197 * characters and the first MIN_MATCH bytes of str are valid (except for
198 * the last MIN_MATCH-1 bytes of the input file).
199 */
200 #ifdef FASTEST
201 #define INSERT_STRING(s, str, match_head) \
202 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
203 match_head = s->head[s->ins_h], \
204 s->head[s->ins_h] = (Pos)(str))
205 #else
206 #define INSERT_STRING(s, str, match_head) \
207 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
208 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
209 s->head[s->ins_h] = (Pos)(str))
210 #endif
211
212 /* ===========================================================================
213 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
214 * prev[] will be initialized on the fly.
215 */
216 #define CLEAR_HASH(s) \
217 s->head[s->hash_size-1] = NIL; \
218 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
219
220 /* ===========================================================================
221 * Slide the hash table when sliding the window down (could be avoided with 32
222 * bit values at the expense of memory usage). We slide even when level == 0 to
223 * keep the hash table consistent if we switch back to level > 0 later.
224 */
225 local void slide_hash(s)
226 deflate_state *s;
227 {
228 unsigned n, m;
229 Posf *p;
230 uInt wsize = s->w_size;
231
232 n = s->hash_size;
233 p = &s->head[n];
234 do {
235 m = *--p;
236 *p = (Pos)(m >= wsize ? m - wsize : NIL);
237 } while (--n);
238 n = wsize;
239 #ifndef FASTEST
240 p = &s->prev[n];
241 do {
242 m = *--p;
243 *p = (Pos)(m >= wsize ? m - wsize : NIL);
244 /* If n is not on any hash chain, prev[n] is garbage but
245 * its value will never be used.
246 */
247 } while (--n);
248 #endif
249 }
250
251 /* ========================================================================= */
252 int ZEXPORT deflateInit_(strm, level, version, stream_size)
253 z_streamp strm;
254 int level;
255 const char *version;
256 int stream_size;
257 {
258 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
259 Z_DEFAULT_STRATEGY, version, stream_size);
260 /* To do: ignore strm->next_in if we use it as window */
261 }
262
263 /* ========================================================================= */
264 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
265 version, stream_size)
266 z_streamp strm;
267 int level;
268 int method;
269 int windowBits;
270 int memLevel;
271 int strategy;
272 const char *version;
273 int stream_size;
274 {
275 deflate_state *s;
276 int wrap = 1;
277 static const char my_version[] = ZLIB_VERSION;
278
279 ushf *overlay;
280 /* We overlay pending_buf and d_buf+l_buf. This works since the average
281 * output size for (length,distance) codes is <= 24 bits.
282 */
283
284 if (version == Z_NULL || version[0] != my_version[0] ||
285 stream_size != sizeof(z_stream)) {
286 return Z_VERSION_ERROR;
287 }
288 if (strm == Z_NULL) return Z_STREAM_ERROR;
289
290 strm->msg = Z_NULL;
291 if (strm->zalloc == (alloc_func)0) {
292 #ifdef Z_SOLO
293 return Z_STREAM_ERROR;
294 #else
295 strm->zalloc = zcalloc;
296 strm->opaque = (voidpf)0;
297 #endif
298 }
299 if (strm->zfree == (free_func)0)
300 #ifdef Z_SOLO
301 return Z_STREAM_ERROR;
302 #else
303 strm->zfree = zcfree;
304 #endif
305
306 #ifdef FASTEST
307 if (level != 0) level = 1;
308 #else
309 if (level == Z_DEFAULT_COMPRESSION) level = 6;
310 #endif
311
312 if (windowBits < 0) { /* suppress zlib wrapper */
313 wrap = 0;
314 windowBits = -windowBits;
315 }
316 #ifdef GZIP
317 else if (windowBits > 15) {
318 wrap = 2; /* write gzip wrapper instead */
319 windowBits -= 16;
320 }
321 #endif
322 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
323 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
324 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
325 return Z_STREAM_ERROR;
326 }
327 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
328 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
329 if (s == Z_NULL) return Z_MEM_ERROR;
330 strm->state = (struct internal_state FAR *)s;
331 s->strm = strm;
332 s->status = INIT_STATE; /* to pass state test in deflateReset() */
333
334 s->wrap = wrap;
335 s->gzhead = Z_NULL;
336 s->w_bits = (uInt)windowBits;
337 s->w_size = 1 << s->w_bits;
338 s->w_mask = s->w_size - 1;
339
340 s->hash_bits = (uInt)memLevel + 7;
341 s->hash_size = 1 << s->hash_bits;
342 s->hash_mask = s->hash_size - 1;
343 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
344
345 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
346 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
347 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
348
349 s->high_water = 0; /* nothing written to s->window yet */
350
351 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
352
353 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
354 s->pending_buf = (uchf *) overlay;
355 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
356
357 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
358 s->pending_buf == Z_NULL) {
359 s->status = FINISH_STATE;
360 strm->msg = ERR_MSG(Z_MEM_ERROR);
361 deflateEnd (strm);
362 return Z_MEM_ERROR;
363 }
364 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
365 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
366
367 s->level = level;
368 s->strategy = strategy;
369 s->method = (Byte)method;
370
371 return deflateReset(strm);
372 }
373
374 /* =========================================================================
375 * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
376 */
377 local int deflateStateCheck (strm)
378 z_streamp strm;
379 {
380 deflate_state *s;
381 if (strm == Z_NULL ||
382 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
383 return 1;
384 s = strm->state;
385 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
386 #ifdef GZIP
387 s->status != GZIP_STATE &&
388 #endif
389 s->status != EXTRA_STATE &&
390 s->status != NAME_STATE &&
391 s->status != COMMENT_STATE &&
392 s->status != HCRC_STATE &&
393 s->status != BUSY_STATE &&
394 s->status != FINISH_STATE))
395 return 1;
396 return 0;
397 }
398
399 /* ========================================================================= */
400 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
401 z_streamp strm;
402 const Bytef *dictionary;
403 uInt dictLength;
404 {
405 deflate_state *s;
406 uInt str, n;
407 int wrap;
408 unsigned avail;
409 z_const unsigned char *next;
410
411 if (deflateStateCheck(strm) || dictionary == Z_NULL)
412 return Z_STREAM_ERROR;
413 s = strm->state;
414 wrap = s->wrap;
415 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
416 return Z_STREAM_ERROR;
417
418 /* when using zlib wrappers, compute Adler-32 for provided dictionary */
419 if (wrap == 1)
420 strm->adler = adler32(strm->adler, dictionary, dictLength);
421 s->wrap = 0; /* avoid computing Adler-32 in read_buf */
422
423 /* if dictionary would fill window, just replace the history */
424 if (dictLength >= s->w_size) {
425 if (wrap == 0) { /* already empty otherwise */
426 CLEAR_HASH(s);
427 s->strstart = 0;
428 s->block_start = 0L;
429 s->insert = 0;
430 }
431 dictionary += dictLength - s->w_size; /* use the tail */
432 dictLength = s->w_size;
433 }
434
435 /* insert dictionary into window and hash */
436 avail = strm->avail_in;
437 next = strm->next_in;
438 strm->avail_in = dictLength;
439 strm->next_in = (z_const Bytef *)dictionary;
440 fill_window(s);
441 while (s->lookahead >= MIN_MATCH) {
442 str = s->strstart;
443 n = s->lookahead - (MIN_MATCH-1);
444 do {
445 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
446 #ifndef FASTEST
447 s->prev[str & s->w_mask] = s->head[s->ins_h];
448 #endif
449 s->head[s->ins_h] = (Pos)str;
450 str++;
451 } while (--n);
452 s->strstart = str;
453 s->lookahead = MIN_MATCH-1;
454 fill_window(s);
455 }
456 s->strstart += s->lookahead;
457 s->block_start = (long)s->strstart;
458 s->insert = s->lookahead;
459 s->lookahead = 0;
460 s->match_length = s->prev_length = MIN_MATCH-1;
461 s->match_available = 0;
462 strm->next_in = next;
463 strm->avail_in = avail;
464 s->wrap = wrap;
465 return Z_OK;
466 }
467
468 /* ========================================================================= */
469 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
470 z_streamp strm;
471 Bytef *dictionary;
472 uInt *dictLength;
473 {
474 deflate_state *s;
475 uInt len;
476
477 if (deflateStateCheck(strm))
478 return Z_STREAM_ERROR;
479 s = strm->state;
480 len = s->strstart + s->lookahead;
481 if (len > s->w_size)
482 len = s->w_size;
483 if (dictionary != Z_NULL && len)
484 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
485 if (dictLength != Z_NULL)
486 *dictLength = len;
487 return Z_OK;
488 }
489
490 /* ========================================================================= */
491 int ZEXPORT deflateResetKeep (strm)
492 z_streamp strm;
493 {
494 deflate_state *s;
495
496 if (deflateStateCheck(strm)) {
497 return Z_STREAM_ERROR;
498 }
499
500 strm->total_in = strm->total_out = 0;
501 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
502 strm->data_type = Z_UNKNOWN;
503
504 s = (deflate_state *)strm->state;
505 s->pending = 0;
506 s->pending_out = s->pending_buf;
507
508 if (s->wrap < 0) {
509 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
510 }
511 s->status =
512 #ifdef GZIP
513 s->wrap == 2 ? GZIP_STATE :
514 #endif
515 s->wrap ? INIT_STATE : BUSY_STATE;
516 strm->adler =
517 #ifdef GZIP
518 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
519 #endif
520 adler32(0L, Z_NULL, 0);
521 s->last_flush = -2;
522
523 _tr_init(s);
524
525 return Z_OK;
526 }
527
528 /* ========================================================================= */
529 int ZEXPORT deflateReset (strm)
530 z_streamp strm;
531 {
532 int ret;
533
534 ret = deflateResetKeep(strm);
535 if (ret == Z_OK)
536 lm_init(strm->state);
537 return ret;
538 }
539
540 /* ========================================================================= */
541 int ZEXPORT deflateSetHeader (strm, head)
542 z_streamp strm;
543 gz_headerp head;
544 {
545 if (deflateStateCheck(strm) || strm->state->wrap != 2)
546 return Z_STREAM_ERROR;
547 strm->state->gzhead = head;
548 return Z_OK;
549 }
550
551 /* ========================================================================= */
552 int ZEXPORT deflatePending (strm, pending, bits)
553 unsigned *pending;
554 int *bits;
555 z_streamp strm;
556 {
557 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
558 if (pending != Z_NULL)
559 *pending = strm->state->pending;
560 if (bits != Z_NULL)
561 *bits = strm->state->bi_valid;
562 return Z_OK;
563 }
564
565 /* ========================================================================= */
566 int ZEXPORT deflatePrime (strm, bits, value)
567 z_streamp strm;
568 int bits;
569 int value;
570 {
571 deflate_state *s;
572 int put;
573
574 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
575 s = strm->state;
576 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
577 return Z_BUF_ERROR;
578 do {
579 put = Buf_size - s->bi_valid;
580 if (put > bits)
581 put = bits;
582 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
583 s->bi_valid += put;
584 _tr_flush_bits(s);
585 value >>= put;
586 bits -= put;
587 } while (bits);
588 return Z_OK;
589 }
590
591 /* ========================================================================= */
592 int ZEXPORT deflateParams(strm, level, strategy)
593 z_streamp strm;
594 int level;
595 int strategy;
596 {
597 deflate_state *s;
598 compress_func func;
599
600 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
601 s = strm->state;
602
603 #ifdef FASTEST
604 if (level != 0) level = 1;
605 #else
606 if (level == Z_DEFAULT_COMPRESSION) level = 6;
607 #endif
608 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
609 return Z_STREAM_ERROR;
610 }
611 func = configuration_table[s->level].func;
612
613 if ((strategy != s->strategy || func != configuration_table[level].func) &&
614 s->last_flush != -2) {
615 /* Flush the last buffer: */
616 int err = deflate(strm, Z_BLOCK);
617 if (err == Z_STREAM_ERROR)
618 return err;
619 if (strm->avail_out == 0)
620 return Z_BUF_ERROR;
621 }
622 if (s->level != level) {
623 if (s->level == 0 && s->matches != 0) {
624 if (s->matches == 1)
625 slide_hash(s);
626 else
627 CLEAR_HASH(s);
628 s->matches = 0;
629 }
630 s->level = level;
631 s->max_lazy_match = configuration_table[level].max_lazy;
632 s->good_match = configuration_table[level].good_length;
633 s->nice_match = configuration_table[level].nice_length;
634 s->max_chain_length = configuration_table[level].max_chain;
635 }
636 s->strategy = strategy;
637 return Z_OK;
638 }
639
640 /* ========================================================================= */
641 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
642 z_streamp strm;
643 int good_length;
644 int max_lazy;
645 int nice_length;
646 int max_chain;
647 {
648 deflate_state *s;
649
650 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
651 s = strm->state;
652 s->good_match = (uInt)good_length;
653 s->max_lazy_match = (uInt)max_lazy;
654 s->nice_match = nice_length;
655 s->max_chain_length = (uInt)max_chain;
656 return Z_OK;
657 }
658
659 /* =========================================================================
660 * For the default windowBits of 15 and memLevel of 8, this function returns
661 * a close to exact, as well as small, upper bound on the compressed size.
662 * They are coded as constants here for a reason--if the #define's are
663 * changed, then this function needs to be changed as well. The return
664 * value for 15 and 8 only works for those exact settings.
665 *
666 * For any setting other than those defaults for windowBits and memLevel,
667 * the value returned is a conservative worst case for the maximum expansion
668 * resulting from using fixed blocks instead of stored blocks, which deflate
669 * can emit on compressed data for some combinations of the parameters.
670 *
671 * This function could be more sophisticated to provide closer upper bounds for
672 * every combination of windowBits and memLevel. But even the conservative
673 * upper bound of about 14% expansion does not seem onerous for output buffer
674 * allocation.
675 */
676 uLong ZEXPORT deflateBound(strm, sourceLen)
677 z_streamp strm;
678 uLong sourceLen;
679 {
680 deflate_state *s;
681 uLong complen, wraplen;
682
683 /* conservative upper bound for compressed data */
684 complen = sourceLen +
685 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
686
687 /* if can't get parameters, return conservative bound plus zlib wrapper */
688 if (deflateStateCheck(strm))
689 return complen + 6;
690
691 /* compute wrapper length */
692 s = strm->state;
693 switch (s->wrap) {
694 case 0: /* raw deflate */
695 wraplen = 0;
696 break;
697 case 1: /* zlib wrapper */
698 wraplen = 6 + (s->strstart ? 4 : 0);
699 break;
700 #ifdef GZIP
701 case 2: /* gzip wrapper */
702 wraplen = 18;
703 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
704 Bytef *str;
705 if (s->gzhead->extra != Z_NULL)
706 wraplen += 2 + s->gzhead->extra_len;
707 str = s->gzhead->name;
708 if (str != Z_NULL)
709 do {
710 wraplen++;
711 } while (*str++);
712 str = s->gzhead->comment;
713 if (str != Z_NULL)
714 do {
715 wraplen++;
716 } while (*str++);
717 if (s->gzhead->hcrc)
718 wraplen += 2;
719 }
720 break;
721 #endif
722 default: /* for compiler happiness */
723 wraplen = 6;
724 }
725
726 /* if not default parameters, return conservative bound */
727 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
728 return complen + wraplen;
729
730 /* default settings: return tight bound for that case */
731 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
732 (sourceLen >> 25) + 13 - 6 + wraplen;
733 }
734
735 /* =========================================================================
736 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
737 * IN assertion: the stream state is correct and there is enough room in
738 * pending_buf.
739 */
740 local void putShortMSB (s, b)
741 deflate_state *s;
742 uInt b;
743 {
744 put_byte(s, (Byte)(b >> 8));
745 put_byte(s, (Byte)(b & 0xff));
746 }
747
748 /* =========================================================================
749 * Flush as much pending output as possible. All deflate() output, except for
750 * some deflate_stored() output, goes through this function so some
751 * applications may wish to modify it to avoid allocating a large
752 * strm->next_out buffer and copying into it. (See also read_buf()).
753 */
754 local void flush_pending(strm)
755 z_streamp strm;
756 {
757 unsigned len;
758 deflate_state *s = strm->state;
759
760 _tr_flush_bits(s);
761 len = s->pending;
762 if (len > strm->avail_out) len = strm->avail_out;
763 if (len == 0) return;
764
765 zmemcpy(strm->next_out, s->pending_out, len);
766 strm->next_out += len;
767 s->pending_out += len;
768 strm->total_out += len;
769 strm->avail_out -= len;
770 s->pending -= len;
771 if (s->pending == 0) {
772 s->pending_out = s->pending_buf;
773 }
774 }
775
776 /* ===========================================================================
777 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
778 */
779 #define HCRC_UPDATE(beg) \
780 do { \
781 if (s->gzhead->hcrc && s->pending > (beg)) \
782 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
783 s->pending - (beg)); \
784 } while (0)
785
786 /* ========================================================================= */
787 int ZEXPORT deflate (strm, flush)
788 z_streamp strm;
789 int flush;
790 {
791 int old_flush; /* value of flush param for previous deflate call */
792 deflate_state *s;
793
794 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
795 return Z_STREAM_ERROR;
796 }
797 s = strm->state;
798
799 if (strm->next_out == Z_NULL ||
800 (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
801 (s->status == FINISH_STATE && flush != Z_FINISH)) {
802 ERR_RETURN(strm, Z_STREAM_ERROR);
803 }
804 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
805
806 old_flush = s->last_flush;
807 s->last_flush = flush;
808
809 /* Flush as much pending output as possible */
810 if (s->pending != 0) {
811 flush_pending(strm);
812 if (strm->avail_out == 0) {
813 /* Since avail_out is 0, deflate will be called again with
814 * more output space, but possibly with both pending and
815 * avail_in equal to zero. There won't be anything to do,
816 * but this is not an error situation so make sure we
817 * return OK instead of BUF_ERROR at next call of deflate:
818 */
819 s->last_flush = -1;
820 return Z_OK;
821 }
822
823 /* Make sure there is something to do and avoid duplicate consecutive
824 * flushes. For repeated and useless calls with Z_FINISH, we keep
825 * returning Z_STREAM_END instead of Z_BUF_ERROR.
826 */
827 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
828 flush != Z_FINISH) {
829 ERR_RETURN(strm, Z_BUF_ERROR);
830 }
831
832 /* User must not provide more input after the first FINISH: */
833 if (s->status == FINISH_STATE && strm->avail_in != 0) {
834 ERR_RETURN(strm, Z_BUF_ERROR);
835 }
836
837 /* Write the header */
838 if (s->status == INIT_STATE) {
839 /* zlib header */
840 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
841 uInt level_flags;
842
843 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
844 level_flags = 0;
845 else if (s->level < 6)
846 level_flags = 1;
847 else if (s->level == 6)
848 level_flags = 2;
849 else
850 level_flags = 3;
851 header |= (level_flags << 6);
852 if (s->strstart != 0) header |= PRESET_DICT;
853 header += 31 - (header % 31);
854
855 putShortMSB(s, header);
856
857 /* Save the adler32 of the preset dictionary: */
858 if (s->strstart != 0) {
859 putShortMSB(s, (uInt)(strm->adler >> 16));
860 putShortMSB(s, (uInt)(strm->adler & 0xffff));
861 }
862 strm->adler = adler32(0L, Z_NULL, 0);
863 s->status = BUSY_STATE;
864
865 /* Compression must start with an empty pending buffer */
866 flush_pending(strm);
867 if (s->pending != 0) {
868 s->last_flush = -1;
869 return Z_OK;
870 }
871 }
872 #ifdef GZIP
873 if (s->status == GZIP_STATE) {
874 /* gzip header */
875 strm->adler = crc32(0L, Z_NULL, 0);
876 put_byte(s, 31);
877 put_byte(s, 139);
878 put_byte(s, 8);
879 if (s->gzhead == Z_NULL) {
880 put_byte(s, 0);
881 put_byte(s, 0);
882 put_byte(s, 0);
883 put_byte(s, 0);
884 put_byte(s, 0);
885 put_byte(s, s->level == 9 ? 2 :
886 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
887 4 : 0));
888 put_byte(s, OS_CODE);
889 s->status = BUSY_STATE;
890
891 /* Compression must start with an empty pending buffer */
892 flush_pending(strm);
893 if (s->pending != 0) {
894 s->last_flush = -1;
895 return Z_OK;
896 }
897 }
898 else {
899 put_byte(s, (s->gzhead->text ? 1 : 0) +
900 (s->gzhead->hcrc ? 2 : 0) +
901 (s->gzhead->extra == Z_NULL ? 0 : 4) +
902 (s->gzhead->name == Z_NULL ? 0 : 8) +
903 (s->gzhead->comment == Z_NULL ? 0 : 16)
904 );
905 put_byte(s, (Byte)(s->gzhead->time & 0xff));
906 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
907 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
908 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
909 put_byte(s, s->level == 9 ? 2 :
910 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
911 4 : 0));
912 put_byte(s, s->gzhead->os & 0xff);
913 if (s->gzhead->extra != Z_NULL) {
914 put_byte(s, s->gzhead->extra_len & 0xff);
915 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
916 }
917 if (s->gzhead->hcrc)
918 strm->adler = crc32(strm->adler, s->pending_buf,
919 s->pending);
920 s->gzindex = 0;
921 s->status = EXTRA_STATE;
922 }
923 }
924 if (s->status == EXTRA_STATE) {
925 if (s->gzhead->extra != Z_NULL) {
926 ulg beg = s->pending; /* start of bytes to update crc */
927 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
928 while (s->pending + left > s->pending_buf_size) {
929 uInt copy = s->pending_buf_size - s->pending;
930 zmemcpy(s->pending_buf + s->pending,
931 s->gzhead->extra + s->gzindex, copy);
932 s->pending = s->pending_buf_size;
933 HCRC_UPDATE(beg);
934 s->gzindex += copy;
935 flush_pending(strm);
936 if (s->pending != 0) {
937 s->last_flush = -1;
938 return Z_OK;
939 }
940 beg = 0;
941 left -= copy;
942 }
943 zmemcpy(s->pending_buf + s->pending,
944 s->gzhead->extra + s->gzindex, left);
945 s->pending += left;
946 HCRC_UPDATE(beg);
947 s->gzindex = 0;
948 }
949 s->status = NAME_STATE;
950 }
951 if (s->status == NAME_STATE) {
952 if (s->gzhead->name != Z_NULL) {
953 ulg beg = s->pending; /* start of bytes to update crc */
954 int val;
955 do {
956 if (s->pending == s->pending_buf_size) {
957 HCRC_UPDATE(beg);
958 flush_pending(strm);
959 if (s->pending != 0) {
960 s->last_flush = -1;
961 return Z_OK;
962 }
963 beg = 0;
964 }
965 val = s->gzhead->name[s->gzindex++];
966 put_byte(s, val);
967 } while (val != 0);
968 HCRC_UPDATE(beg);
969 s->gzindex = 0;
970 }
971 s->status = COMMENT_STATE;
972 }
973 if (s->status == COMMENT_STATE) {
974 if (s->gzhead->comment != Z_NULL) {
975 ulg beg = s->pending; /* start of bytes to update crc */
976 int val;
977 do {
978 if (s->pending == s->pending_buf_size) {
979 HCRC_UPDATE(beg);
980 flush_pending(strm);
981 if (s->pending != 0) {
982 s->last_flush = -1;
983 return Z_OK;
984 }
985 beg = 0;
986 }
987 val = s->gzhead->comment[s->gzindex++];
988 put_byte(s, val);
989 } while (val != 0);
990 HCRC_UPDATE(beg);
991 }
992 s->status = HCRC_STATE;
993 }
994 if (s->status == HCRC_STATE) {
995 if (s->gzhead->hcrc) {
996 if (s->pending + 2 > s->pending_buf_size) {
997 flush_pending(strm);
998 if (s->pending != 0) {
999 s->last_flush = -1;
1000 return Z_OK;
1001 }
1002 }
1003 put_byte(s, (Byte)(strm->adler & 0xff));
1004 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1005 strm->adler = crc32(0L, Z_NULL, 0);
1006 }
1007 s->status = BUSY_STATE;
1008
1009 /* Compression must start with an empty pending buffer */
1010 flush_pending(strm);
1011 if (s->pending != 0) {
1012 s->last_flush = -1;
1013 return Z_OK;
1014 }
1015 }
1016 #endif
1017
1018 /* Start a new block or continue the current one.
1019 */
1020 if (strm->avail_in != 0 || s->lookahead != 0 ||
1021 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
1022 block_state bstate;
1023
1024 bstate = s->level == 0 ? deflate_stored(s, flush) :
1025 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
1026 s->strategy == Z_RLE ? deflate_rle(s, flush) :
1027 (*(configuration_table[s->level].func))(s, flush);
1028
1029 if (bstate == finish_started || bstate == finish_done) {
1030 s->status = FINISH_STATE;
1031 }
1032 if (bstate == need_more || bstate == finish_started) {
1033 if (strm->avail_out == 0) {
1034 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1035 }
1036 return Z_OK;
1037 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1038 * of deflate should use the same flush parameter to make sure
1039 * that the flush is complete. So we don't have to output an
1040 * empty block here, this will be done at next call. This also
1041 * ensures that for a very small output buffer, we emit at most
1042 * one empty block.
1043 */
1044 }
1045 if (bstate == block_done) {
1046 if (flush == Z_PARTIAL_FLUSH) {
1047 _tr_align(s);
1048 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1049 _tr_stored_block(s, (char*)0, 0L, 0);
1050 /* For a full flush, this empty block will be recognized
1051 * as a special marker by inflate_sync().
1052 */
1053 if (flush == Z_FULL_FLUSH) {
1054 CLEAR_HASH(s); /* forget history */
1055 if (s->lookahead == 0) {
1056 s->strstart = 0;
1057 s->block_start = 0L;
1058 s->insert = 0;
1059 }
1060 }
1061 }
1062 flush_pending(strm);
1063 if (strm->avail_out == 0) {
1064 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1065 return Z_OK;
1066 }
1067 }
1068 }
1069
1070 if (flush != Z_FINISH) return Z_OK;
1071 if (s->wrap <= 0) return Z_STREAM_END;
1072
1073 /* Write the trailer */
1074 #ifdef GZIP
1075 if (s->wrap == 2) {
1076 put_byte(s, (Byte)(strm->adler & 0xff));
1077 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1078 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1079 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1080 put_byte(s, (Byte)(strm->total_in & 0xff));
1081 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1082 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1083 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1084 }
1085 else
1086 #endif
1087 {
1088 putShortMSB(s, (uInt)(strm->adler >> 16));
1089 putShortMSB(s, (uInt)(strm->adler & 0xffff));
1090 }
1091 flush_pending(strm);
1092 /* If avail_out is zero, the application will call deflate again
1093 * to flush the rest.
1094 */
1095 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1096 return s->pending != 0 ? Z_OK : Z_STREAM_END;
1097 }
1098
1099 /* ========================================================================= */
1100 int ZEXPORT deflateEnd (strm)
1101 z_streamp strm;
1102 {
1103 int status;
1104
1105 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1106
1107 status = strm->state->status;
1108
1109 /* Deallocate in reverse order of allocations: */
1110 TRY_FREE(strm, strm->state->pending_buf);
1111 TRY_FREE(strm, strm->state->head);
1112 TRY_FREE(strm, strm->state->prev);
1113 TRY_FREE(strm, strm->state->window);
1114
1115 ZFREE(strm, strm->state);
1116 strm->state = Z_NULL;
1117
1118 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1119 }
1120
1121 /* =========================================================================
1122 * Copy the source state to the destination state.
1123 * To simplify the source, this is not supported for 16-bit MSDOS (which
1124 * doesn't have enough memory anyway to duplicate compression states).
1125 */
1126 int ZEXPORT deflateCopy (dest, source)
1127 z_streamp dest;
1128 z_streamp source;
1129 {
1130 #ifdef MAXSEG_64K
1131 return Z_STREAM_ERROR;
1132 #else
1133 deflate_state *ds;
1134 deflate_state *ss;
1135 ushf *overlay;
1136
1137
1138 if (deflateStateCheck(source) || dest == Z_NULL) {
1139 return Z_STREAM_ERROR;
1140 }
1141
1142 ss = source->state;
1143
1144 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1145
1146 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1147 if (ds == Z_NULL) return Z_MEM_ERROR;
1148 dest->state = (struct internal_state FAR *) ds;
1149 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1150 ds->strm = dest;
1151
1152 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1153 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
1154 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
1155 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1156 ds->pending_buf = (uchf *) overlay;
1157
1158 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1159 ds->pending_buf == Z_NULL) {
1160 deflateEnd (dest);
1161 return Z_MEM_ERROR;
1162 }
1163 /* following zmemcpy do not work for 16-bit MSDOS */
1164 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1165 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1166 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1167 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1168
1169 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1170 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1171 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1172
1173 ds->l_desc.dyn_tree = ds->dyn_ltree;
1174 ds->d_desc.dyn_tree = ds->dyn_dtree;
1175 ds->bl_desc.dyn_tree = ds->bl_tree;
1176
1177 return Z_OK;
1178 #endif /* MAXSEG_64K */
1179 }
1180
1181 /* ===========================================================================
1182 * Read a new buffer from the current input stream, update the adler32
1183 * and total number of bytes read. All deflate() input goes through
1184 * this function so some applications may wish to modify it to avoid
1185 * allocating a large strm->next_in buffer and copying from it.
1186 * (See also flush_pending()).
1187 */
1188 local unsigned read_buf(strm, buf, size)
1189 z_streamp strm;
1190 Bytef *buf;
1191 unsigned size;
1192 {
1193 unsigned len = strm->avail_in;
1194
1195 if (len > size) len = size;
1196 if (len == 0) return 0;
1197
1198 strm->avail_in -= len;
1199
1200 zmemcpy(buf, strm->next_in, len);
1201 if (strm->state->wrap == 1) {
1202 strm->adler = adler32(strm->adler, buf, len);
1203 }
1204 #ifdef GZIP
1205 else if (strm->state->wrap == 2) {
1206 strm->adler = crc32(strm->adler, buf, len);
1207 }
1208 #endif
1209 strm->next_in += len;
1210 strm->total_in += len;
1211
1212 return len;
1213 }
1214
1215 /* ===========================================================================
1216 * Initialize the "longest match" routines for a new zlib stream
1217 */
1218 local void lm_init (s)
1219 deflate_state *s;
1220 {
1221 s->window_size = (ulg)2L*s->w_size;
1222
1223 CLEAR_HASH(s);
1224
1225 /* Set the default configuration parameters:
1226 */
1227 s->max_lazy_match = configuration_table[s->level].max_lazy;
1228 s->good_match = configuration_table[s->level].good_length;
1229 s->nice_match = configuration_table[s->level].nice_length;
1230 s->max_chain_length = configuration_table[s->level].max_chain;
1231
1232 s->strstart = 0;
1233 s->block_start = 0L;
1234 s->lookahead = 0;
1235 s->insert = 0;
1236 s->match_length = s->prev_length = MIN_MATCH-1;
1237 s->match_available = 0;
1238 s->ins_h = 0;
1239 #ifndef FASTEST
1240 #ifdef ASMV
1241 match_init(); /* initialize the asm code */
1242 #endif
1243 #endif
1244 }
1245
1246 #ifndef FASTEST
1247 /* ===========================================================================
1248 * Set match_start to the longest match starting at the given string and
1249 * return its length. Matches shorter or equal to prev_length are discarded,
1250 * in which case the result is equal to prev_length and match_start is
1251 * garbage.
1252 * IN assertions: cur_match is the head of the hash chain for the current
1253 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1254 * OUT assertion: the match length is not greater than s->lookahead.
1255 */
1256 #ifndef ASMV
1257 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1258 * match.S. The code will be functionally equivalent.
1259 */
1260 local uInt longest_match(s, cur_match)
1261 deflate_state *s;
1262 IPos cur_match; /* current match */
1263 {
1264 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1265 register Bytef *scan = s->window + s->strstart; /* current string */
1266 register Bytef *match; /* matched string */
1267 register int len; /* length of current match */
1268 int best_len = (int)s->prev_length; /* best match length so far */
1269 int nice_match = s->nice_match; /* stop if match long enough */
1270 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1271 s->strstart - (IPos)MAX_DIST(s) : NIL;
1272 /* Stop when cur_match becomes <= limit. To simplify the code,
1273 * we prevent matches with the string of window index 0.
1274 */
1275 Posf *prev = s->prev;
1276 uInt wmask = s->w_mask;
1277
1278 #ifdef UNALIGNED_OK
1279 /* Compare two bytes at a time. Note: this is not always beneficial.
1280 * Try with and without -DUNALIGNED_OK to check.
1281 */
1282 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1283 register ush scan_start = *(ushf*)scan;
1284 register ush scan_end = *(ushf*)(scan+best_len-1);
1285 #else
1286 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1287 register Byte scan_end1 = scan[best_len-1];
1288 register Byte scan_end = scan[best_len];
1289 #endif
1290
1291 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1292 * It is easy to get rid of this optimization if necessary.
1293 */
1294 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1295
1296 /* Do not waste too much time if we already have a good match: */
1297 if (s->prev_length >= s->good_match) {
1298 chain_length >>= 2;
1299 }
1300 /* Do not look for matches beyond the end of the input. This is necessary
1301 * to make deflate deterministic.
1302 */
1303 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1304
1305 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1306
1307 do {
1308 Assert(cur_match < s->strstart, "no future");
1309 match = s->window + cur_match;
1310
1311 /* Skip to next match if the match length cannot increase
1312 * or if the match length is less than 2. Note that the checks below
1313 * for insufficient lookahead only occur occasionally for performance
1314 * reasons. Therefore uninitialized memory will be accessed, and
1315 * conditional jumps will be made that depend on those values.
1316 * However the length of the match is limited to the lookahead, so
1317 * the output of deflate is not affected by the uninitialized values.
1318 */
1319 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1320 /* This code assumes sizeof(unsigned short) == 2. Do not use
1321 * UNALIGNED_OK if your compiler uses a different size.
1322 */
1323 if (*(ushf*)(match+best_len-1) != scan_end ||
1324 *(ushf*)match != scan_start) continue;
1325
1326 /* It is not necessary to compare scan[2] and match[2] since they are
1327 * always equal when the other bytes match, given that the hash keys
1328 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1329 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1330 * lookahead only every 4th comparison; the 128th check will be made
1331 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1332 * necessary to put more guard bytes at the end of the window, or
1333 * to check more often for insufficient lookahead.
1334 */
1335 Assert(scan[2] == match[2], "scan[2]?");
1336 scan++, match++;
1337 do {
1338 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1339 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1340 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1341 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1342 scan < strend);
1343 /* The funny "do {}" generates better code on most compilers */
1344
1345 /* Here, scan <= window+strstart+257 */
1346 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1347 if (*scan == *match) scan++;
1348
1349 len = (MAX_MATCH - 1) - (int)(strend-scan);
1350 scan = strend - (MAX_MATCH-1);
1351
1352 #else /* UNALIGNED_OK */
1353
1354 if (match[best_len] != scan_end ||
1355 match[best_len-1] != scan_end1 ||
1356 *match != *scan ||
1357 *++match != scan[1]) continue;
1358
1359 /* The check at best_len-1 can be removed because it will be made
1360 * again later. (This heuristic is not always a win.)
1361 * It is not necessary to compare scan[2] and match[2] since they
1362 * are always equal when the other bytes match, given that
1363 * the hash keys are equal and that HASH_BITS >= 8.
1364 */
1365 scan += 2, match++;
1366 Assert(*scan == *match, "match[2]?");
1367
1368 /* We check for insufficient lookahead only every 8th comparison;
1369 * the 256th check will be made at strstart+258.
1370 */
1371 do {
1372 } while (*++scan == *++match && *++scan == *++match &&
1373 *++scan == *++match && *++scan == *++match &&
1374 *++scan == *++match && *++scan == *++match &&
1375 *++scan == *++match && *++scan == *++match &&
1376 scan < strend);
1377
1378 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1379
1380 len = MAX_MATCH - (int)(strend - scan);
1381 scan = strend - MAX_MATCH;
1382
1383 #endif /* UNALIGNED_OK */
1384
1385 if (len > best_len) {
1386 s->match_start = cur_match;
1387 best_len = len;
1388 if (len >= nice_match) break;
1389 #ifdef UNALIGNED_OK
1390 scan_end = *(ushf*)(scan+best_len-1);
1391 #else
1392 scan_end1 = scan[best_len-1];
1393 scan_end = scan[best_len];
1394 #endif
1395 }
1396 } while ((cur_match = prev[cur_match & wmask]) > limit
1397 && --chain_length != 0);
1398
1399 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1400 return s->lookahead;
1401 }
1402 #endif /* ASMV */
1403
1404 #else /* FASTEST */
1405
1406 /* ---------------------------------------------------------------------------
1407 * Optimized version for FASTEST only
1408 */
1409 local uInt longest_match(s, cur_match)
1410 deflate_state *s;
1411 IPos cur_match; /* current match */
1412 {
1413 register Bytef *scan = s->window + s->strstart; /* current string */
1414 register Bytef *match; /* matched string */
1415 register int len; /* length of current match */
1416 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1417
1418 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1419 * It is easy to get rid of this optimization if necessary.
1420 */
1421 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1422
1423 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1424
1425 Assert(cur_match < s->strstart, "no future");
1426
1427 match = s->window + cur_match;
1428
1429 /* Return failure if the match length is less than 2:
1430 */
1431 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1432
1433 /* The check at best_len-1 can be removed because it will be made
1434 * again later. (This heuristic is not always a win.)
1435 * It is not necessary to compare scan[2] and match[2] since they
1436 * are always equal when the other bytes match, given that
1437 * the hash keys are equal and that HASH_BITS >= 8.
1438 */
1439 scan += 2, match += 2;
1440 Assert(*scan == *match, "match[2]?");
1441
1442 /* We check for insufficient lookahead only every 8th comparison;
1443 * the 256th check will be made at strstart+258.
1444 */
1445 do {
1446 } while (*++scan == *++match && *++scan == *++match &&
1447 *++scan == *++match && *++scan == *++match &&
1448 *++scan == *++match && *++scan == *++match &&
1449 *++scan == *++match && *++scan == *++match &&
1450 scan < strend);
1451
1452 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1453
1454 len = MAX_MATCH - (int)(strend - scan);
1455
1456 if (len < MIN_MATCH) return MIN_MATCH - 1;
1457
1458 s->match_start = cur_match;
1459 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1460 }
1461
1462 #endif /* FASTEST */
1463
1464 #ifdef ZLIB_DEBUG
1465
1466 #define EQUAL 0
1467 /* result of memcmp for equal strings */
1468
1469 /* ===========================================================================
1470 * Check that the match at match_start is indeed a match.
1471 */
1472 local void check_match(s, start, match, length)
1473 deflate_state *s;
1474 IPos start, match;
1475 int length;
1476 {
1477 /* check that the match is indeed a match */
1478 if (zmemcmp(s->window + match,
1479 s->window + start, length) != EQUAL) {
1480 fprintf(stderr, " start %u, match %u, length %d\n",
1481 start, match, length);
1482 do {
1483 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1484 } while (--length != 0);
1485 z_error("invalid match");
1486 }
1487 if (z_verbose > 1) {
1488 fprintf(stderr,"\\[%d,%d]", start-match, length);
1489 do { putc(s->window[start++], stderr); } while (--length != 0);
1490 }
1491 }
1492 #else
1493 # define check_match(s, start, match, length)
1494 #endif /* ZLIB_DEBUG */
1495
1496 /* ===========================================================================
1497 * Fill the window when the lookahead becomes insufficient.
1498 * Updates strstart and lookahead.
1499 *
1500 * IN assertion: lookahead < MIN_LOOKAHEAD
1501 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1502 * At least one byte has been read, or avail_in == 0; reads are
1503 * performed for at least two bytes (required for the zip translate_eol
1504 * option -- not supported here).
1505 */
1506 local void fill_window(s)
1507 deflate_state *s;
1508 {
1509 unsigned n;
1510 unsigned more; /* Amount of free space at the end of the window. */
1511 uInt wsize = s->w_size;
1512
1513 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1514
1515 do {
1516 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1517
1518 /* Deal with !@#$% 64K limit: */
1519 if (sizeof(int) <= 2) {
1520 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1521 more = wsize;
1522
1523 } else if (more == (unsigned)(-1)) {
1524 /* Very unlikely, but possible on 16 bit machine if
1525 * strstart == 0 && lookahead == 1 (input done a byte at time)
1526 */
1527 more--;
1528 }
1529 }
1530
1531 /* If the window is almost full and there is insufficient lookahead,
1532 * move the upper half to the lower one to make room in the upper half.
1533 */
1534 if (s->strstart >= wsize+MAX_DIST(s)) {
1535
1536 zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1537 s->match_start -= wsize;
1538 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1539 s->block_start -= (long) wsize;
1540 slide_hash(s);
1541 more += wsize;
1542 }
1543 if (s->strm->avail_in == 0) break;
1544
1545 /* If there was no sliding:
1546 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1547 * more == window_size - lookahead - strstart
1548 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1549 * => more >= window_size - 2*WSIZE + 2
1550 * In the BIG_MEM or MMAP case (not yet supported),
1551 * window_size == input_size + MIN_LOOKAHEAD &&
1552 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1553 * Otherwise, window_size == 2*WSIZE so more >= 2.
1554 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1555 */
1556 Assert(more >= 2, "more < 2");
1557
1558 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1559 s->lookahead += n;
1560
1561 /* Initialize the hash value now that we have some input: */
1562 if (s->lookahead + s->insert >= MIN_MATCH) {
1563 uInt str = s->strstart - s->insert;
1564 s->ins_h = s->window[str];
1565 UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1566 #if MIN_MATCH != 3
1567 Call UPDATE_HASH() MIN_MATCH-3 more times
1568 #endif
1569 while (s->insert) {
1570 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1571 #ifndef FASTEST
1572 s->prev[str & s->w_mask] = s->head[s->ins_h];
1573 #endif
1574 s->head[s->ins_h] = (Pos)str;
1575 str++;
1576 s->insert--;
1577 if (s->lookahead + s->insert < MIN_MATCH)
1578 break;
1579 }
1580 }
1581 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1582 * but this is not important since only literal bytes will be emitted.
1583 */
1584
1585 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1586
1587 /* If the WIN_INIT bytes after the end of the current data have never been
1588 * written, then zero those bytes in order to avoid memory check reports of
1589 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1590 * the longest match routines. Update the high water mark for the next
1591 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1592 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1593 */
1594 if (s->high_water < s->window_size) {
1595 ulg curr = s->strstart + (ulg)(s->lookahead);
1596 ulg init;
1597
1598 if (s->high_water < curr) {
1599 /* Previous high water mark below current data -- zero WIN_INIT
1600 * bytes or up to end of window, whichever is less.
1601 */
1602 init = s->window_size - curr;
1603 if (init > WIN_INIT)
1604 init = WIN_INIT;
1605 zmemzero(s->window + curr, (unsigned)init);
1606 s->high_water = curr + init;
1607 }
1608 else if (s->high_water < (ulg)curr + WIN_INIT) {
1609 /* High water mark at or above current data, but below current data
1610 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1611 * to end of window, whichever is less.
1612 */
1613 init = (ulg)curr + WIN_INIT - s->high_water;
1614 if (init > s->window_size - s->high_water)
1615 init = s->window_size - s->high_water;
1616 zmemzero(s->window + s->high_water, (unsigned)init);
1617 s->high_water += init;
1618 }
1619 }
1620
1621 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1622 "not enough room for search");
1623 }
1624
1625 /* ===========================================================================
1626 * Flush the current block, with given end-of-file flag.
1627 * IN assertion: strstart is set to the end of the current match.
1628 */
1629 #define FLUSH_BLOCK_ONLY(s, last) { \
1630 _tr_flush_block(s, (s->block_start >= 0L ? \
1631 (charf *)&s->window[(unsigned)s->block_start] : \
1632 (charf *)Z_NULL), \
1633 (ulg)((long)s->strstart - s->block_start), \
1634 (last)); \
1635 s->block_start = s->strstart; \
1636 flush_pending(s->strm); \
1637 Tracev((stderr,"[FLUSH]")); \
1638 }
1639
1640 /* Same but force premature exit if necessary. */
1641 #define FLUSH_BLOCK(s, last) { \
1642 FLUSH_BLOCK_ONLY(s, last); \
1643 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1644 }
1645
1646 /* Maximum stored block length in deflate format (not including header). */
1647 #define MAX_STORED 65535
1648
1649 /* Minimum of a and b. */
1650 #define MIN(a, b) ((a) > (b) ? (b) : (a))
1651
1652 /* ===========================================================================
1653 * Copy without compression as much as possible from the input stream, return
1654 * the current block state.
1655 *
1656 * In case deflateParams() is used to later switch to a non-zero compression
1657 * level, s->matches (otherwise unused when storing) keeps track of the number
1658 * of hash table slides to perform. If s->matches is 1, then one hash table
1659 * slide will be done when switching. If s->matches is 2, the maximum value
1660 * allowed here, then the hash table will be cleared, since two or more slides
1661 * is the same as a clear.
1662 *
1663 * deflate_stored() is written to minimize the number of times an input byte is
1664 * copied. It is most efficient with large input and output buffers, which
1665 * maximizes the opportunites to have a single copy from next_in to next_out.
1666 */
1667 local block_state deflate_stored(s, flush)
1668 deflate_state *s;
1669 int flush;
1670 {
1671 /* Smallest worthy block size when not flushing or finishing. By default
1672 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1673 * large input and output buffers, the stored block size will be larger.
1674 */
1675 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1676
1677 /* Copy as many min_block or larger stored blocks directly to next_out as
1678 * possible. If flushing, copy the remaining available input to next_out as
1679 * stored blocks, if there is enough space.
1680 */
1681 unsigned len, left, have, last = 0;
1682 unsigned used = s->strm->avail_in;
1683 do {
1684 /* Set len to the maximum size block that we can copy directly with the
1685 * available input data and output space. Set left to how much of that
1686 * would be copied from what's left in the window.
1687 */
1688 len = MAX_STORED; /* maximum deflate stored block length */
1689 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1690 if (s->strm->avail_out < have) /* need room for header */
1691 break;
1692 /* maximum stored block length that will fit in avail_out: */
1693 have = s->strm->avail_out - have;
1694 left = s->strstart - s->block_start; /* bytes left in window */
1695 if (len > (ulg)left + s->strm->avail_in)
1696 len = left + s->strm->avail_in; /* limit len to the input */
1697 if (len > have)
1698 len = have; /* limit len to the output */
1699
1700 /* If the stored block would be less than min_block in length, or if
1701 * unable to copy all of the available input when flushing, then try
1702 * copying to the window and the pending buffer instead. Also don't
1703 * write an empty block when flushing -- deflate() does that.
1704 */
1705 if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1706 flush == Z_NO_FLUSH ||
1707 len != left + s->strm->avail_in))
1708 break;
1709
1710 /* Make a dummy stored block in pending to get the header bytes,
1711 * including any pending bits. This also updates the debugging counts.
1712 */
1713 last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1714 _tr_stored_block(s, (char *)0, 0L, last);
1715
1716 /* Replace the lengths in the dummy stored block with len. */
1717 s->pending_buf[s->pending - 4] = len;
1718 s->pending_buf[s->pending - 3] = len >> 8;
1719 s->pending_buf[s->pending - 2] = ~len;
1720 s->pending_buf[s->pending - 1] = ~len >> 8;
1721
1722 /* Write the stored block header bytes. */
1723 flush_pending(s->strm);
1724
1725 #ifdef ZLIB_DEBUG
1726 /* Update debugging counts for the data about to be copied. */
1727 s->compressed_len += len << 3;
1728 s->bits_sent += len << 3;
1729 #endif
1730
1731 /* Copy uncompressed bytes from the window to next_out. */
1732 if (left) {
1733 if (left > len)
1734 left = len;
1735 zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1736 s->strm->next_out += left;
1737 s->strm->avail_out -= left;
1738 s->strm->total_out += left;
1739 s->block_start += left;
1740 len -= left;
1741 }
1742
1743 /* Copy uncompressed bytes directly from next_in to next_out, updating
1744 * the check value.
1745 */
1746 if (len) {
1747 read_buf(s->strm, s->strm->next_out, len);
1748 s->strm->next_out += len;
1749 s->strm->avail_out -= len;
1750 s->strm->total_out += len;
1751 }
1752 } while (last == 0);
1753
1754 /* Update the sliding window with the last s->w_size bytes of the copied
1755 * data, or append all of the copied data to the existing window if less
1756 * than s->w_size bytes were copied. Also update the number of bytes to
1757 * insert in the hash tables, in the event that deflateParams() switches to
1758 * a non-zero compression level.
1759 */
1760 used -= s->strm->avail_in; /* number of input bytes directly copied */
1761 if (used) {
1762 /* If any input was used, then no unused input remains in the window,
1763 * therefore s->block_start == s->strstart.
1764 */
1765 if (used >= s->w_size) { /* supplant the previous history */
1766 s->matches = 2; /* clear hash */
1767 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1768 s->strstart = s->w_size;
1769 }
1770 else {
1771 if (s->window_size - s->strstart <= used) {
1772 /* Slide the window down. */
1773 s->strstart -= s->w_size;
1774 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1775 if (s->matches < 2)
1776 s->matches++; /* add a pending slide_hash() */
1777 }
1778 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1779 s->strstart += used;
1780 }
1781 s->block_start = s->strstart;
1782 s->insert += MIN(used, s->w_size - s->insert);
1783 }
1784 if (s->high_water < s->strstart)
1785 s->high_water = s->strstart;
1786
1787 /* If the last block was written to next_out, then done. */
1788 if (last)
1789 return finish_done;
1790
1791 /* If flushing and all input has been consumed, then done. */
1792 if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1793 s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1794 return block_done;
1795
1796 /* Fill the window with any remaining input. */
1797 have = s->window_size - s->strstart - 1;
1798 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1799 /* Slide the window down. */
1800 s->block_start -= s->w_size;
1801 s->strstart -= s->w_size;
1802 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1803 if (s->matches < 2)
1804 s->matches++; /* add a pending slide_hash() */
1805 have += s->w_size; /* more space now */
1806 }
1807 if (have > s->strm->avail_in)
1808 have = s->strm->avail_in;
1809 if (have) {
1810 read_buf(s->strm, s->window + s->strstart, have);
1811 s->strstart += have;
1812 }
1813 if (s->high_water < s->strstart)
1814 s->high_water = s->strstart;
1815
1816 /* There was not enough avail_out to write a complete worthy or flushed
1817 * stored block to next_out. Write a stored block to pending instead, if we
1818 * have enough input for a worthy block, or if flushing and there is enough
1819 * room for the remaining input as a stored block in the pending buffer.
1820 */
1821 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1822 /* maximum stored block length that will fit in pending: */
1823 have = MIN(s->pending_buf_size - have, MAX_STORED);
1824 min_block = MIN(have, s->w_size);
1825 left = s->strstart - s->block_start;
1826 if (left >= min_block ||
1827 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1828 s->strm->avail_in == 0 && left <= have)) {
1829 len = MIN(left, have);
1830 last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1831 len == left ? 1 : 0;
1832 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1833 s->block_start += len;
1834 flush_pending(s->strm);
1835 }
1836
1837 /* We've done all we can with the available input and output. */
1838 return last ? finish_started : need_more;
1839 }
1840
1841 /* ===========================================================================
1842 * Compress as much as possible from the input stream, return the current
1843 * block state.
1844 * This function does not perform lazy evaluation of matches and inserts
1845 * new strings in the dictionary only for unmatched strings or for short
1846 * matches. It is used only for the fast compression options.
1847 */
1848 local block_state deflate_fast(s, flush)
1849 deflate_state *s;
1850 int flush;
1851 {
1852 IPos hash_head; /* head of the hash chain */
1853 int bflush; /* set if current block must be flushed */
1854
1855 for (;;) {
1856 /* Make sure that we always have enough lookahead, except
1857 * at the end of the input file. We need MAX_MATCH bytes
1858 * for the next match, plus MIN_MATCH bytes to insert the
1859 * string following the next match.
1860 */
1861 if (s->lookahead < MIN_LOOKAHEAD) {
1862 fill_window(s);
1863 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1864 return need_more;
1865 }
1866 if (s->lookahead == 0) break; /* flush the current block */
1867 }
1868
1869 /* Insert the string window[strstart .. strstart+2] in the
1870 * dictionary, and set hash_head to the head of the hash chain:
1871 */
1872 hash_head = NIL;
1873 if (s->lookahead >= MIN_MATCH) {
1874 INSERT_STRING(s, s->strstart, hash_head);
1875 }
1876
1877 /* Find the longest match, discarding those <= prev_length.
1878 * At this point we have always match_length < MIN_MATCH
1879 */
1880 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1881 /* To simplify the code, we prevent matches with the string
1882 * of window index 0 (in particular we have to avoid a match
1883 * of the string with itself at the start of the input file).
1884 */
1885 s->match_length = longest_match (s, hash_head);
1886 /* longest_match() sets match_start */
1887 }
1888 if (s->match_length >= MIN_MATCH) {
1889 check_match(s, s->strstart, s->match_start, s->match_length);
1890
1891 _tr_tally_dist(s, s->strstart - s->match_start,
1892 s->match_length - MIN_MATCH, bflush);
1893
1894 s->lookahead -= s->match_length;
1895
1896 /* Insert new strings in the hash table only if the match length
1897 * is not too large. This saves time but degrades compression.
1898 */
1899 #ifndef FASTEST
1900 if (s->match_length <= s->max_insert_length &&
1901 s->lookahead >= MIN_MATCH) {
1902 s->match_length--; /* string at strstart already in table */
1903 do {
1904 s->strstart++;
1905 INSERT_STRING(s, s->strstart, hash_head);
1906 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1907 * always MIN_MATCH bytes ahead.
1908 */
1909 } while (--s->match_length != 0);
1910 s->strstart++;
1911 } else
1912 #endif
1913 {
1914 s->strstart += s->match_length;
1915 s->match_length = 0;
1916 s->ins_h = s->window[s->strstart];
1917 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1918 #if MIN_MATCH != 3
1919 Call UPDATE_HASH() MIN_MATCH-3 more times
1920 #endif
1921 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1922 * matter since it will be recomputed at next deflate call.
1923 */
1924 }
1925 } else {
1926 /* No match, output a literal byte */
1927 Tracevv((stderr,"%c", s->window[s->strstart]));
1928 _tr_tally_lit (s, s->window[s->strstart], bflush);
1929 s->lookahead--;
1930 s->strstart++;
1931 }
1932 if (bflush) FLUSH_BLOCK(s, 0);
1933 }
1934 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1935 if (flush == Z_FINISH) {
1936 FLUSH_BLOCK(s, 1);
1937 return finish_done;
1938 }
1939 if (s->last_lit)
1940 FLUSH_BLOCK(s, 0);
1941 return block_done;
1942 }
1943
1944 #ifndef FASTEST
1945 /* ===========================================================================
1946 * Same as above, but achieves better compression. We use a lazy
1947 * evaluation for matches: a match is finally adopted only if there is
1948 * no better match at the next window position.
1949 */
1950 local block_state deflate_slow(s, flush)
1951 deflate_state *s;
1952 int flush;
1953 {
1954 IPos hash_head; /* head of hash chain */
1955 int bflush; /* set if current block must be flushed */
1956
1957 /* Process the input block. */
1958 for (;;) {
1959 /* Make sure that we always have enough lookahead, except
1960 * at the end of the input file. We need MAX_MATCH bytes
1961 * for the next match, plus MIN_MATCH bytes to insert the
1962 * string following the next match.
1963 */
1964 if (s->lookahead < MIN_LOOKAHEAD) {
1965 fill_window(s);
1966 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1967 return need_more;
1968 }
1969 if (s->lookahead == 0) break; /* flush the current block */
1970 }
1971
1972 /* Insert the string window[strstart .. strstart+2] in the
1973 * dictionary, and set hash_head to the head of the hash chain:
1974 */
1975 hash_head = NIL;
1976 if (s->lookahead >= MIN_MATCH) {
1977 INSERT_STRING(s, s->strstart, hash_head);
1978 }
1979
1980 /* Find the longest match, discarding those <= prev_length.
1981 */
1982 s->prev_length = s->match_length, s->prev_match = s->match_start;
1983 s->match_length = MIN_MATCH-1;
1984
1985 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1986 s->strstart - hash_head <= MAX_DIST(s)) {
1987 /* To simplify the code, we prevent matches with the string
1988 * of window index 0 (in particular we have to avoid a match
1989 * of the string with itself at the start of the input file).
1990 */
1991 s->match_length = longest_match (s, hash_head);
1992 /* longest_match() sets match_start */
1993
1994 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1995 #if TOO_FAR <= 32767
1996 || (s->match_length == MIN_MATCH &&
1997 s->strstart - s->match_start > TOO_FAR)
1998 #endif
1999 )) {
2000
2001 /* If prev_match is also MIN_MATCH, match_start is garbage
2002 * but we will ignore the current match anyway.
2003 */
2004 s->match_length = MIN_MATCH-1;
2005 }
2006 }
2007 /* If there was a match at the previous step and the current
2008 * match is not better, output the previous match:
2009 */
2010 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
2011 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
2012 /* Do not insert strings in hash table beyond this. */
2013
2014 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
2015
2016 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
2017 s->prev_length - MIN_MATCH, bflush);
2018
2019 /* Insert in hash table all strings up to the end of the match.
2020 * strstart-1 and strstart are already inserted. If there is not
2021 * enough lookahead, the last two strings are not inserted in
2022 * the hash table.
2023 */
2024 s->lookahead -= s->prev_length-1;
2025 s->prev_length -= 2;
2026 do {
2027 if (++s->strstart <= max_insert) {
2028 INSERT_STRING(s, s->strstart, hash_head);
2029 }
2030 } while (--s->prev_length != 0);
2031 s->match_available = 0;
2032 s->match_length = MIN_MATCH-1;
2033 s->strstart++;
2034
2035 if (bflush) FLUSH_BLOCK(s, 0);
2036
2037 } else if (s->match_available) {
2038 /* If there was no match at the previous position, output a
2039 * single literal. If there was a match but the current match
2040 * is longer, truncate the previous match to a single literal.
2041 */
2042 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2043 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2044 if (bflush) {
2045 FLUSH_BLOCK_ONLY(s, 0);
2046 }
2047 s->strstart++;
2048 s->lookahead--;
2049 if (s->strm->avail_out == 0) return need_more;
2050 } else {
2051 /* There is no previous match to compare with, wait for
2052 * the next step to decide.
2053 */
2054 s->match_available = 1;
2055 s->strstart++;
2056 s->lookahead--;
2057 }
2058 }
2059 Assert (flush != Z_NO_FLUSH, "no flush?");
2060 if (s->match_available) {
2061 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2062 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2063 s->match_available = 0;
2064 }
2065 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2066 if (flush == Z_FINISH) {
2067 FLUSH_BLOCK(s, 1);
2068 return finish_done;
2069 }
2070 if (s->last_lit)
2071 FLUSH_BLOCK(s, 0);
2072 return block_done;
2073 }
2074 #endif /* FASTEST */
2075
2076 /* ===========================================================================
2077 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2078 * one. Do not maintain a hash table. (It will be regenerated if this run of
2079 * deflate switches away from Z_RLE.)
2080 */
2081 local block_state deflate_rle(s, flush)
2082 deflate_state *s;
2083 int flush;
2084 {
2085 int bflush; /* set if current block must be flushed */
2086 uInt prev; /* byte at distance one to match */
2087 Bytef *scan, *strend; /* scan goes up to strend for length of run */
2088
2089 for (;;) {
2090 /* Make sure that we always have enough lookahead, except
2091 * at the end of the input file. We need MAX_MATCH bytes
2092 * for the longest run, plus one for the unrolled loop.
2093 */
2094 if (s->lookahead <= MAX_MATCH) {
2095 fill_window(s);
2096 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2097 return need_more;
2098 }
2099 if (s->lookahead == 0) break; /* flush the current block */
2100 }
2101
2102 /* See how many times the previous byte repeats */
2103 s->match_length = 0;
2104 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2105 scan = s->window + s->strstart - 1;
2106 prev = *scan;
2107 if (prev == *++scan && prev == *++scan && prev == *++scan) {
2108 strend = s->window + s->strstart + MAX_MATCH;
2109 do {
2110 } while (prev == *++scan && prev == *++scan &&
2111 prev == *++scan && prev == *++scan &&
2112 prev == *++scan && prev == *++scan &&
2113 prev == *++scan && prev == *++scan &&
2114 scan < strend);
2115 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2116 if (s->match_length > s->lookahead)
2117 s->match_length = s->lookahead;
2118 }
2119 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2120 }
2121
2122 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2123 if (s->match_length >= MIN_MATCH) {
2124 check_match(s, s->strstart, s->strstart - 1, s->match_length);
2125
2126 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2127
2128 s->lookahead -= s->match_length;
2129 s->strstart += s->match_length;
2130 s->match_length = 0;
2131 } else {
2132 /* No match, output a literal byte */
2133 Tracevv((stderr,"%c", s->window[s->strstart]));
2134 _tr_tally_lit (s, s->window[s->strstart], bflush);
2135 s->lookahead--;
2136 s->strstart++;
2137 }
2138 if (bflush) FLUSH_BLOCK(s, 0);
2139 }
2140 s->insert = 0;
2141 if (flush == Z_FINISH) {
2142 FLUSH_BLOCK(s, 1);
2143 return finish_done;
2144 }
2145 if (s->last_lit)
2146 FLUSH_BLOCK(s, 0);
2147 return block_done;
2148 }
2149
2150 /* ===========================================================================
2151 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
2152 * (It will be regenerated if this run of deflate switches away from Huffman.)
2153 */
2154 local block_state deflate_huff(s, flush)
2155 deflate_state *s;
2156 int flush;
2157 {
2158 int bflush; /* set if current block must be flushed */
2159
2160 for (;;) {
2161 /* Make sure that we have a literal to write. */
2162 if (s->lookahead == 0) {
2163 fill_window(s);
2164 if (s->lookahead == 0) {
2165 if (flush == Z_NO_FLUSH)
2166 return need_more;
2167 break; /* flush the current block */
2168 }
2169 }
2170
2171 /* Output a literal byte */
2172 s->match_length = 0;
2173 Tracevv((stderr,"%c", s->window[s->strstart]));
2174 _tr_tally_lit (s, s->window[s->strstart], bflush);
2175 s->lookahead--;
2176 s->strstart++;
2177 if (bflush) FLUSH_BLOCK(s, 0);
2178 }
2179 s->insert = 0;
2180 if (flush == Z_FINISH) {
2181 FLUSH_BLOCK(s, 1);
2182 return finish_done;
2183 }
2184 if (s->last_lit)
2185 FLUSH_BLOCK(s, 0);
2186 return block_done;
2187 }