1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2005 Jean-loup Gailly.
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
5
6 /*
7 * ALGORITHM
8 *
9 * The "deflation" process depends on being able to identify portions
10 * of the input text which are identical to earlier input (within a
11 * sliding window trailing behind the input currently being processed).
12 *
13 * The most straightforward technique turns out to be the fastest for
14 * most input files: try all possible matches and select the longest.
15 * The key feature of this algorithm is that insertions into the string
16 * dictionary are very simple and thus fast, and deletions are avoided
17 * completely. Insertions are performed at each input character, whereas
18 * string matches are performed only when the previous match ends. So it
19 * is preferable to spend more time in matches to allow very fast string
20 * insertions and avoid deletions. The matching algorithm for small
21 * strings is inspired from that of Rabin & Karp. A brute force approach
22 * is used to find longer strings when a small match has been found.
23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
24 * (by Leonid Broukhis).
25 * A previous version of this file used a more sophisticated algorithm
26 * (by Fiala and Greene) which is guaranteed to run in linear amortized
27 * time, but has a larger average cost, uses more memory and is patented.
28 * However the F&G algorithm may be faster for some highly redundant
29 * files if the parameter max_chain_length (described below) is too large.
30 *
31 * ACKNOWLEDGEMENTS
32 *
33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
34 * I found it in 'freeze' written by Leonid Broukhis.
35 * Thanks to many people for bug reports and testing.
36 *
37 * REFERENCES
38 *
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40 * Available in http://www.ietf.org/rfc/rfc1951.txt
41 *
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
44 *
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
47 *
48 */
49
50 /* @(#) $Id$ */
51
52 #include "zlib/deflate.h"
53
54 const char deflate_copyright[] =
55 " deflate 1.2.3 Copyright 1995-2005 Jean-loup Gailly ";
56 /*
57 If you use the zlib library in a product, an acknowledgment is welcome
58 in the documentation of your product. If for some reason you cannot
59 include such an acknowledgment, I would appreciate that you keep this
60 copyright string in the executable of your product.
61 */
62
63 /* ===========================================================================
64 * Function prototypes.
65 */
66 typedef enum {
67 need_more, /* block not completed, need more input or more output */
68 block_done, /* block flush performed */
69 finish_started, /* finish started, need only more output at next deflate */
70 finish_done /* finish done, accept no more input or output */
71 } block_state;
72
73 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
74 /* Compression function. Returns the block state after the call. */
75
76 local void fill_window OF((deflate_state *s));
77 local block_state deflate_stored OF((deflate_state *s, int flush));
78 local block_state deflate_fast OF((deflate_state *s, int flush));
79 #ifndef FASTEST
80 local block_state deflate_slow OF((deflate_state *s, int flush));
81 #endif
82 local void lm_init OF((deflate_state *s));
83 local void putShortMSB OF((deflate_state *s, uInt b));
84 local void flush_pending OF((z_streamp strm));
85 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
86 #ifndef FASTEST
87 #ifdef ASMV
88 void match_init OF((void)); /* asm code initialization */
89 uInt longest_match OF((deflate_state *s, IPos cur_match));
90 #else
91 local uInt longest_match OF((deflate_state *s, IPos cur_match));
92 #endif
93 #endif
94 local uInt longest_match_fast OF((deflate_state *s, IPos cur_match));
95
96 #ifdef DEBUG
97 local void check_match OF((deflate_state *s, IPos start, IPos match,
98 int length));
99 #endif
100
101 /* ===========================================================================
102 * Local data
103 */
104
105 #define NIL 0
106 /* Tail of hash chains */
107
108 #ifndef TOO_FAR
109 # define TOO_FAR 4096
110 #endif
111 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
112
113 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
114 /* Minimum amount of lookahead, except at the end of the input file.
115 * See deflate.c for comments about the MIN_MATCH+1.
116 */
117
118 /* Values for max_lazy_match, good_match and max_chain_length, depending on
119 * the desired pack level (0..9). The values given below have been tuned to
120 * exclude worst case performance for pathological files. Better values may be
121 * found for specific files.
122 */
123 typedef struct config_s {
124 ush good_length; /* reduce lazy search above this match length */
125 ush max_lazy; /* do not perform lazy search above this match length */
126 ush nice_length; /* quit search above this match length */
127 ush max_chain;
128 compress_func func;
129 } config;
130
131 #ifdef FASTEST
132 local const config configuration_table[2] = {
133 /* good lazy nice chain */
134 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
135 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
136 #else
137 local const config configuration_table[10] = {
138 /* good lazy nice chain */
139 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
140 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
141 /* 2 */ {4, 5, 16, 8, deflate_fast},
142 /* 3 */ {4, 6, 32, 32, deflate_fast},
143
144 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
145 /* 5 */ {8, 16, 32, 32, deflate_slow},
146 /* 6 */ {8, 16, 128, 128, deflate_slow},
147 /* 7 */ {8, 32, 128, 256, deflate_slow},
148 /* 8 */ {32, 128, 258, 1024, deflate_slow},
149 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
150 #endif
151
152 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
153 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
154 * meaning.
155 */
156
157 #define EQUAL 0
158 /* result of memcmp for equal strings */
159
160 #ifndef NO_DUMMY_DECL
161 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
162 #endif
163
164 /* ===========================================================================
165 * Update a hash value with the given input byte
166 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
167 * input characters, so that a running hash key can be computed from the
168 * previous key instead of complete recalculation each time.
169 */
170 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
171
172
173 /* ===========================================================================
174 * Insert string str in the dictionary and set match_head to the previous head
175 * of the hash chain (the most recent string with same hash key). Return
176 * the previous length of the hash chain.
177 * If this file is compiled with -DFASTEST, the compression level is forced
178 * to 1, and no hash chains are maintained.
179 * IN assertion: all calls to to INSERT_STRING are made with consecutive
180 * input characters and the first MIN_MATCH bytes of str are valid
181 * (except for the last MIN_MATCH-1 bytes of the input file).
182 */
183 #ifdef FASTEST
184 #define INSERT_STRING(s, str, match_head) \
185 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
186 match_head = s->head[s->ins_h], \
187 s->head[s->ins_h] = (Pos)(str))
188 #else
189 #define INSERT_STRING(s, str, match_head) \
190 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
191 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
192 s->head[s->ins_h] = (Pos)(str))
193 #endif
194
195 /* ===========================================================================
196 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
197 * prev[] will be initialized on the fly.
198 */
199 #define CLEAR_HASH(s) \
200 s->head[s->hash_size-1] = NIL; \
201 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
202
203 /* ========================================================================= */
deflateInit_(z_streamp strm,int level,const char * version,int stream_size)204 int ZEXPORT deflateInit_(z_streamp strm, int level, const char *version, int stream_size)
205 {
206 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
207 Z_DEFAULT_STRATEGY, version, stream_size);
208 /* To do: ignore strm->next_in if we use it as window */
209 }
210
211 /* ========================================================================= */
deflateInit2_(z_streamp strm,int level,int method,int windowBits,int memLevel,int strategy,const char * version,int stream_size)212 int ZEXPORT deflateInit2_(
213 z_streamp strm,
214 int level,
215 int method,
216 int windowBits,
217 int memLevel,
218 int strategy,
219 const char *version,
220 int stream_size)
221 {
222 deflate_state *s;
223 int wrap = 1;
224 static const char my_version[] = ZLIB_VERSION;
225
226 ushf *overlay;
227 /* We overlay pending_buf and d_buf+l_buf. This works since the average
228 * output size for (length,distance) codes is <= 24 bits.
229 */
230
231 if (version == Z_NULL || version[0] != my_version[0] ||
232 stream_size != sizeof(z_stream)) {
233 return Z_VERSION_ERROR;
234 }
235 if (strm == Z_NULL) return Z_STREAM_ERROR;
236
237 strm->msg = Z_NULL;
238 if (strm->zalloc == (alloc_func)0) {
239 strm->zalloc = zcalloc;
240 strm->opaque = (voidpf)0;
241 }
242 if (strm->zfree == (free_func)0) strm->zfree = zcfree;
243
244 #ifdef FASTEST
245 if (level != 0) level = 1;
246 #else
247 if (level == Z_DEFAULT_COMPRESSION) level = 6;
248 #endif
249
250 if (windowBits < 0) { /* suppress zlib wrapper */
251 wrap = 0;
252 windowBits = -windowBits;
253 }
254 #ifdef GZIP
255 else if (windowBits > 15) {
256 wrap = 2; /* write gzip wrapper instead */
257 windowBits -= 16;
258 }
259 #endif
260 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
261 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
262 strategy < 0 || strategy > Z_FIXED) {
263 return Z_STREAM_ERROR;
264 }
265 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
266 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
267 if (s == Z_NULL) return Z_MEM_ERROR;
268 strm->state = (struct internal_state FAR *)s;
269 s->strm = strm;
270
271 s->wrap = wrap;
272 s->gzhead = Z_NULL;
273 s->w_bits = windowBits;
274 s->w_size = 1 << s->w_bits;
275 s->w_mask = s->w_size - 1;
276
277 s->hash_bits = memLevel + 7;
278 s->hash_size = 1 << s->hash_bits;
279 s->hash_mask = s->hash_size - 1;
280 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
281
282 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
283 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
284 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
285
286 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
287
288 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
289 s->pending_buf = (uchf *) overlay;
290 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
291
292 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
293 s->pending_buf == Z_NULL) {
294 s->status = FINISH_STATE;
295 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
296 deflateEnd (strm);
297 return Z_MEM_ERROR;
298 }
299 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
300 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
301
302 s->level = level;
303 s->strategy = strategy;
304 s->method = (Byte)method;
305
306 return deflateReset(strm);
307 }
308
309 /* ========================================================================= */
deflateSetDictionary(z_streamp strm,const Bytef * dictionary,uInt dictLength)310 int ZEXPORT deflateSetDictionary (
311 z_streamp strm,
312 const Bytef *dictionary,
313 uInt dictLength
314 ){
315 deflate_state *s;
316 uInt length = dictLength;
317 uInt n;
318 IPos hash_head = 0;
319
320 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
321 strm->state->wrap == 2 ||
322 (strm->state->wrap == 1 && strm->state->status != INIT_STATE))
323 return Z_STREAM_ERROR;
324
325 s = strm->state;
326 if (s->wrap)
327 strm->adler = adler32(strm->adler, dictionary, dictLength);
328
329 if (length < MIN_MATCH) return Z_OK;
330 if (length > MAX_DIST(s)) {
331 length = MAX_DIST(s);
332 dictionary += dictLength - length; /* use the tail of the dictionary */
333 }
334 zmemcpy(s->window, dictionary, length);
335 s->strstart = length;
336 s->block_start = (long)length;
337
338 /* Insert all strings in the hash table (except for the last two bytes).
339 * s->lookahead stays null, so s->ins_h will be recomputed at the next
340 * call of fill_window.
341 */
342 s->ins_h = s->window[0];
343 UPDATE_HASH(s, s->ins_h, s->window[1]);
344 for (n = 0; n <= length - MIN_MATCH; n++) {
345 INSERT_STRING(s, n, hash_head);
346 }
347 if (hash_head) hash_head = 0; /* to make compiler happy */
348 return Z_OK;
349 }
350
351 /* ========================================================================= */
deflateReset(z_streamp strm)352 int ZEXPORT deflateReset (z_streamp strm)
353 {
354 deflate_state *s;
355
356 if (strm == Z_NULL || strm->state == Z_NULL ||
357 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
358 return Z_STREAM_ERROR;
359 }
360
361 strm->total_in = strm->total_out = 0;
362 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
363 strm->data_type = Z_UNKNOWN;
364
365 s = (deflate_state *)strm->state;
366 s->pending = 0;
367 s->pending_out = s->pending_buf;
368
369 if (s->wrap < 0) {
370 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
371 }
372 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
373 strm->adler =
374 #ifdef GZIP
375 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
376 #endif
377 adler32(0L, Z_NULL, 0);
378 s->last_flush = Z_NO_FLUSH;
379
380 _tr_init(s);
381 lm_init(s);
382
383 return Z_OK;
384 }
385
386 /* ========================================================================= */
deflateSetHeader(z_streamp strm,gz_headerp head)387 int ZEXPORT deflateSetHeader ( z_streamp strm, gz_headerp head)
388 {
389 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
390 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
391 strm->state->gzhead = head;
392 return Z_OK;
393 }
394
395 /* ========================================================================= */
deflatePrime(z_streamp strm,int bits,int value)396 int ZEXPORT deflatePrime (
397 z_streamp strm,
398 int bits,
399 int value
400 ){
401 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
402 strm->state->bi_valid = bits;
403 strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));
404 return Z_OK;
405 }
406
407 /* ========================================================================= */
deflateParams(z_streamp strm,int level,int strategy)408 int ZEXPORT deflateParams(
409 z_streamp strm,
410 int level,
411 int strategy
412 ){
413 deflate_state *s;
414 compress_func func;
415 int err = Z_OK;
416
417 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
418 s = strm->state;
419
420 #ifdef FASTEST
421 if (level != 0) level = 1;
422 #else
423 if (level == Z_DEFAULT_COMPRESSION) level = 6;
424 #endif
425 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
426 return Z_STREAM_ERROR;
427 }
428 func = configuration_table[s->level].func;
429
430 if (func != configuration_table[level].func && strm->total_in != 0) {
431 /* Flush the last buffer: */
432 err = deflate(strm, Z_PARTIAL_FLUSH);
433 }
434 if (s->level != level) {
435 s->level = level;
436 s->max_lazy_match = configuration_table[level].max_lazy;
437 s->good_match = configuration_table[level].good_length;
438 s->nice_match = configuration_table[level].nice_length;
439 s->max_chain_length = configuration_table[level].max_chain;
440 }
441 s->strategy = strategy;
442 return err;
443 }
444
445 /* ========================================================================= */
deflateTune(z_streamp strm,int good_length,int max_lazy,int nice_length,int max_chain)446 int ZEXPORT deflateTune(
447 z_streamp strm,
448 int good_length,
449 int max_lazy,
450 int nice_length,
451 int max_chain
452 ){
453 deflate_state *s;
454
455 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
456 s = strm->state;
457 s->good_match = good_length;
458 s->max_lazy_match = max_lazy;
459 s->nice_match = nice_length;
460 s->max_chain_length = max_chain;
461 return Z_OK;
462 }
463
464 /* =========================================================================
465 * For the default windowBits of 15 and memLevel of 8, this function returns
466 * a close to exact, as well as small, upper bound on the compressed size.
467 * They are coded as constants here for a reason--if the #define's are
468 * changed, then this function needs to be changed as well. The return
469 * value for 15 and 8 only works for those exact settings.
470 *
471 * For any setting other than those defaults for windowBits and memLevel,
472 * the value returned is a conservative worst case for the maximum expansion
473 * resulting from using fixed blocks instead of stored blocks, which deflate
474 * can emit on compressed data for some combinations of the parameters.
475 *
476 * This function could be more sophisticated to provide closer upper bounds
477 * for every combination of windowBits and memLevel, as well as wrap.
478 * But even the conservative upper bound of about 14% expansion does not
479 * seem onerous for output buffer allocation.
480 */
deflateBound(z_streamp strm,uLong sourceLen)481 uLong ZEXPORT deflateBound(z_streamp strm, uLong sourceLen)
482 {
483 deflate_state *s;
484 uLong destLen;
485
486 /* conservative upper bound */
487 destLen = sourceLen +
488 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 11;
489
490 /* if can't get parameters, return conservative bound */
491 if (strm == Z_NULL || strm->state == Z_NULL)
492 return destLen;
493
494 /* if not default parameters, return conservative bound */
495 s = strm->state;
496 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
497 return destLen;
498
499 /* default settings: return tight bound for that case */
500 return compressBound(sourceLen);
501 }
502
503 /* =========================================================================
504 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
505 * IN assertion: the stream state is correct and there is enough room in
506 * pending_buf.
507 */
putShortMSB(deflate_state * s,uInt b)508 local void putShortMSB (deflate_state *s, uInt b)
509 {
510 put_byte(s, (Byte)(b >> 8));
511 put_byte(s, (Byte)(b & 0xff));
512 }
513
514 /* =========================================================================
515 * Flush as much pending output as possible. All deflate() output goes
516 * through this function so some applications may wish to modify it
517 * to avoid allocating a large strm->next_out buffer and copying into it.
518 * (See also read_buf()).
519 */
flush_pending(z_streamp strm)520 local void flush_pending(z_streamp strm)
521 {
522 unsigned len = strm->state->pending;
523
524 if (len > strm->avail_out) len = strm->avail_out;
525 if (len == 0) return;
526
527 zmemcpy(strm->next_out, strm->state->pending_out, len);
528 strm->next_out += len;
529 strm->state->pending_out += len;
530 strm->total_out += len;
531 strm->avail_out -= len;
532 strm->state->pending -= len;
533 if (strm->state->pending == 0) {
534 strm->state->pending_out = strm->state->pending_buf;
535 }
536 }
537
538 /* ========================================================================= */
deflate(z_streamp strm,int flush)539 int ZEXPORT deflate (z_streamp strm, int flush)
540 {
541 int old_flush; /* value of flush param for previous deflate call */
542 deflate_state *s;
543
544 if (strm == Z_NULL || strm->state == Z_NULL ||
545 flush > Z_FINISH || flush < 0) {
546 return Z_STREAM_ERROR;
547 }
548 s = strm->state;
549
550 if (strm->next_out == Z_NULL ||
551 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
552 (s->status == FINISH_STATE && flush != Z_FINISH)) {
553 ERR_RETURN(strm, Z_STREAM_ERROR);
554 }
555 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
556
557 s->strm = strm; /* just in case */
558 old_flush = s->last_flush;
559 s->last_flush = flush;
560
561 /* Write the header */
562 if (s->status == INIT_STATE) {
563 #ifdef GZIP
564 if (s->wrap == 2) {
565 strm->adler = crc32(0L, Z_NULL, 0);
566 put_byte(s, 31);
567 put_byte(s, 139);
568 put_byte(s, 8);
569 if (s->gzhead == NULL) {
570 put_byte(s, 0);
571 put_byte(s, 0);
572 put_byte(s, 0);
573 put_byte(s, 0);
574 put_byte(s, 0);
575 put_byte(s, s->level == 9 ? 2 :
576 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
577 4 : 0));
578 put_byte(s, OS_CODE);
579 s->status = BUSY_STATE;
580 }
581 else {
582 put_byte(s, (s->gzhead->text ? 1 : 0) +
583 (s->gzhead->hcrc ? 2 : 0) +
584 (s->gzhead->extra == Z_NULL ? 0 : 4) +
585 (s->gzhead->name == Z_NULL ? 0 : 8) +
586 (s->gzhead->comment == Z_NULL ? 0 : 16)
587 );
588 put_byte(s, (Byte)(s->gzhead->time & 0xff));
589 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
590 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
591 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
592 put_byte(s, s->level == 9 ? 2 :
593 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
594 4 : 0));
595 put_byte(s, s->gzhead->os & 0xff);
596 if (s->gzhead->extra != NULL) {
597 put_byte(s, s->gzhead->extra_len & 0xff);
598 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
599 }
600 if (s->gzhead->hcrc)
601 strm->adler = crc32(strm->adler, s->pending_buf,
602 s->pending);
603 s->gzindex = 0;
604 s->status = EXTRA_STATE;
605 }
606 }
607 else
608 #endif
609 {
610 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
611 uInt level_flags;
612
613 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
614 level_flags = 0;
615 else if (s->level < 6)
616 level_flags = 1;
617 else if (s->level == 6)
618 level_flags = 2;
619 else
620 level_flags = 3;
621 header |= (level_flags << 6);
622 if (s->strstart != 0) header |= PRESET_DICT;
623 header += 31 - (header % 31);
624
625 s->status = BUSY_STATE;
626 putShortMSB(s, header);
627
628 /* Save the adler32 of the preset dictionary: */
629 if (s->strstart != 0) {
630 putShortMSB(s, (uInt)(strm->adler >> 16));
631 putShortMSB(s, (uInt)(strm->adler & 0xffff));
632 }
633 strm->adler = adler32(0L, Z_NULL, 0);
634 }
635 }
636 #ifdef GZIP
637 if (s->status == EXTRA_STATE) {
638 if (s->gzhead->extra != NULL) {
639 uInt beg = s->pending; /* start of bytes to update crc */
640
641 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
642 if (s->pending == s->pending_buf_size) {
643 if (s->gzhead->hcrc && s->pending > beg)
644 strm->adler = crc32(strm->adler, s->pending_buf + beg,
645 s->pending - beg);
646 flush_pending(strm);
647 beg = s->pending;
648 if (s->pending == s->pending_buf_size)
649 break;
650 }
651 put_byte(s, s->gzhead->extra[s->gzindex]);
652 s->gzindex++;
653 }
654 if (s->gzhead->hcrc && s->pending > beg)
655 strm->adler = crc32(strm->adler, s->pending_buf + beg,
656 s->pending - beg);
657 if (s->gzindex == s->gzhead->extra_len) {
658 s->gzindex = 0;
659 s->status = NAME_STATE;
660 }
661 }
662 else
663 s->status = NAME_STATE;
664 }
665 if (s->status == NAME_STATE) {
666 if (s->gzhead->name != NULL) {
667 uInt beg = s->pending; /* start of bytes to update crc */
668 int val;
669
670 do {
671 if (s->pending == s->pending_buf_size) {
672 if (s->gzhead->hcrc && s->pending > beg)
673 strm->adler = crc32(strm->adler, s->pending_buf + beg,
674 s->pending - beg);
675 flush_pending(strm);
676 beg = s->pending;
677 if (s->pending == s->pending_buf_size) {
678 val = 1;
679 break;
680 }
681 }
682 val = s->gzhead->name[s->gzindex++];
683 put_byte(s, val);
684 } while (val != 0);
685 if (s->gzhead->hcrc && s->pending > beg)
686 strm->adler = crc32(strm->adler, s->pending_buf + beg,
687 s->pending - beg);
688 if (val == 0) {
689 s->gzindex = 0;
690 s->status = COMMENT_STATE;
691 }
692 }
693 else
694 s->status = COMMENT_STATE;
695 }
696 if (s->status == COMMENT_STATE) {
697 if (s->gzhead->comment != NULL) {
698 uInt beg = s->pending; /* start of bytes to update crc */
699 int val;
700
701 do {
702 if (s->pending == s->pending_buf_size) {
703 if (s->gzhead->hcrc && s->pending > beg)
704 strm->adler = crc32(strm->adler, s->pending_buf + beg,
705 s->pending - beg);
706 flush_pending(strm);
707 beg = s->pending;
708 if (s->pending == s->pending_buf_size) {
709 val = 1;
710 break;
711 }
712 }
713 val = s->gzhead->comment[s->gzindex++];
714 put_byte(s, val);
715 } while (val != 0);
716 if (s->gzhead->hcrc && s->pending > beg)
717 strm->adler = crc32(strm->adler, s->pending_buf + beg,
718 s->pending - beg);
719 if (val == 0)
720 s->status = HCRC_STATE;
721 }
722 else
723 s->status = HCRC_STATE;
724 }
725 if (s->status == HCRC_STATE) {
726 if (s->gzhead->hcrc) {
727 if (s->pending + 2 > s->pending_buf_size)
728 flush_pending(strm);
729 if (s->pending + 2 <= s->pending_buf_size) {
730 put_byte(s, (Byte)(strm->adler & 0xff));
731 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
732 strm->adler = crc32(0L, Z_NULL, 0);
733 s->status = BUSY_STATE;
734 }
735 }
736 else
737 s->status = BUSY_STATE;
738 }
739 #endif
740
741 /* Flush as much pending output as possible */
742 if (s->pending != 0) {
743 flush_pending(strm);
744 if (strm->avail_out == 0) {
745 /* Since avail_out is 0, deflate will be called again with
746 * more output space, but possibly with both pending and
747 * avail_in equal to zero. There won't be anything to do,
748 * but this is not an error situation so make sure we
749 * return OK instead of BUF_ERROR at next call of deflate:
750 */
751 s->last_flush = -1;
752 return Z_OK;
753 }
754
755 /* Make sure there is something to do and avoid duplicate consecutive
756 * flushes. For repeated and useless calls with Z_FINISH, we keep
757 * returning Z_STREAM_END instead of Z_BUF_ERROR.
758 */
759 } else if (strm->avail_in == 0 && flush <= old_flush &&
760 flush != Z_FINISH) {
761 ERR_RETURN(strm, Z_BUF_ERROR);
762 }
763
764 /* User must not provide more input after the first FINISH: */
765 if (s->status == FINISH_STATE && strm->avail_in != 0) {
766 ERR_RETURN(strm, Z_BUF_ERROR);
767 }
768
769 /* Start a new block or continue the current one.
770 */
771 if (strm->avail_in != 0 || s->lookahead != 0 ||
772 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
773 block_state bstate;
774
775 bstate = (*(configuration_table[s->level].func))(s, flush);
776
777 if (bstate == finish_started || bstate == finish_done) {
778 s->status = FINISH_STATE;
779 }
780 if (bstate == need_more || bstate == finish_started) {
781 if (strm->avail_out == 0) {
782 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
783 }
784 return Z_OK;
785 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
786 * of deflate should use the same flush parameter to make sure
787 * that the flush is complete. So we don't have to output an
788 * empty block here, this will be done at next call. This also
789 * ensures that for a very small output buffer, we emit at most
790 * one empty block.
791 */
792 }
793 if (bstate == block_done) {
794 if (flush == Z_PARTIAL_FLUSH) {
795 _tr_align(s);
796 } else { /* FULL_FLUSH or SYNC_FLUSH */
797 _tr_stored_block(s, (char*)0, 0L, 0);
798 /* For a full flush, this empty block will be recognized
799 * as a special marker by inflate_sync().
800 */
801 if (flush == Z_FULL_FLUSH) {
802 CLEAR_HASH(s); /* forget history */
803 }
804 }
805 flush_pending(strm);
806 if (strm->avail_out == 0) {
807 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
808 return Z_OK;
809 }
810 }
811 }
812 Assert(strm->avail_out > 0, "bug2");
813
814 if (flush != Z_FINISH) return Z_OK;
815 if (s->wrap <= 0) return Z_STREAM_END;
816
817 /* Write the trailer */
818 #ifdef GZIP
819 if (s->wrap == 2) {
820 put_byte(s, (Byte)(strm->adler & 0xff));
821 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
822 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
823 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
824 put_byte(s, (Byte)(strm->total_in & 0xff));
825 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
826 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
827 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
828 }
829 else
830 #endif
831 {
832 putShortMSB(s, (uInt)(strm->adler >> 16));
833 putShortMSB(s, (uInt)(strm->adler & 0xffff));
834 }
835 flush_pending(strm);
836 /* If avail_out is zero, the application will call deflate again
837 * to flush the rest.
838 */
839 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
840 return s->pending != 0 ? Z_OK : Z_STREAM_END;
841 }
842
843 /* ========================================================================= */
deflateEnd(z_streamp strm)844 int ZEXPORT deflateEnd (z_streamp strm)
845 {
846 int status;
847
848 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
849
850 status = strm->state->status;
851 if (status != INIT_STATE &&
852 status != EXTRA_STATE &&
853 status != NAME_STATE &&
854 status != COMMENT_STATE &&
855 status != HCRC_STATE &&
856 status != BUSY_STATE &&
857 status != FINISH_STATE) {
858 return Z_STREAM_ERROR;
859 }
860
861 /* Deallocate in reverse order of allocations: */
862 TRY_FREE(strm, strm->state->pending_buf);
863 TRY_FREE(strm, strm->state->head);
864 TRY_FREE(strm, strm->state->prev);
865 TRY_FREE(strm, strm->state->window);
866
867 ZFREE(strm, strm->state);
868 strm->state = Z_NULL;
869
870 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
871 }
872
873 /* =========================================================================
874 * Copy the source state to the destination state.
875 * To simplify the source, this is not supported for 16-bit MSDOS (which
876 * doesn't have enough memory anyway to duplicate compression states).
877 */
deflateCopy(z_streamp dest,z_streamp source)878 int ZEXPORT deflateCopy (z_streamp dest, z_streamp source)
879 {
880 #ifdef MAXSEG_64K
881 return Z_STREAM_ERROR;
882 #else
883 deflate_state *ds;
884 deflate_state *ss;
885 ushf *overlay;
886
887
888 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
889 return Z_STREAM_ERROR;
890 }
891
892 ss = source->state;
893
894 zmemcpy(dest, source, sizeof(z_stream));
895
896 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
897 if (ds == Z_NULL) return Z_MEM_ERROR;
898 dest->state = (struct internal_state FAR *) ds;
899 zmemcpy(ds, ss, sizeof(deflate_state));
900 ds->strm = dest;
901
902 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
903 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
904 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
905 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
906 ds->pending_buf = (uchf *) overlay;
907
908 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
909 ds->pending_buf == Z_NULL) {
910 deflateEnd (dest);
911 return Z_MEM_ERROR;
912 }
913 /* following zmemcpy do not work for 16-bit MSDOS */
914 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
915 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
916 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
917 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
918
919 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
920 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
921 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
922
923 ds->l_desc.dyn_tree = ds->dyn_ltree;
924 ds->d_desc.dyn_tree = ds->dyn_dtree;
925 ds->bl_desc.dyn_tree = ds->bl_tree;
926
927 return Z_OK;
928 #endif /* MAXSEG_64K */
929 }
930
931 /* ===========================================================================
932 * Read a new buffer from the current input stream, update the adler32
933 * and total number of bytes read. All deflate() input goes through
934 * this function so some applications may wish to modify it to avoid
935 * allocating a large strm->next_in buffer and copying from it.
936 * (See also flush_pending()).
937 */
read_buf(z_streamp strm,Bytef * buf,unsigned size)938 local int read_buf(z_streamp strm, Bytef *buf, unsigned size)
939 {
940 unsigned len = strm->avail_in;
941
942 if (len > size) len = size;
943 if (len == 0) return 0;
944
945 strm->avail_in -= len;
946
947 if (strm->state->wrap == 1) {
948 strm->adler = adler32(strm->adler, strm->next_in, len);
949 }
950 #ifdef GZIP
951 else if (strm->state->wrap == 2) {
952 strm->adler = crc32(strm->adler, strm->next_in, len);
953 }
954 #endif
955 zmemcpy(buf, strm->next_in, len);
956 strm->next_in += len;
957 strm->total_in += len;
958
959 return (int)len;
960 }
961
962 /* ===========================================================================
963 * Initialize the "longest match" routines for a new zlib stream
964 */
lm_init(deflate_state * s)965 local void lm_init (deflate_state *s)
966 {
967 s->window_size = (ulg)2L*s->w_size;
968
969 CLEAR_HASH(s);
970
971 /* Set the default configuration parameters:
972 */
973 s->max_lazy_match = configuration_table[s->level].max_lazy;
974 s->good_match = configuration_table[s->level].good_length;
975 s->nice_match = configuration_table[s->level].nice_length;
976 s->max_chain_length = configuration_table[s->level].max_chain;
977
978 s->strstart = 0;
979 s->block_start = 0L;
980 s->lookahead = 0;
981 s->match_length = s->prev_length = MIN_MATCH-1;
982 s->match_available = 0;
983 s->ins_h = 0;
984 #ifndef FASTEST
985 #ifdef ASMV
986 match_init(); /* initialize the asm code */
987 #endif
988 #endif
989 }
990
991 #ifndef FASTEST
992 /* ===========================================================================
993 * Set match_start to the longest match starting at the given string and
994 * return its length. Matches shorter or equal to prev_length are discarded,
995 * in which case the result is equal to prev_length and match_start is
996 * garbage.
997 * IN assertions: cur_match is the head of the hash chain for the current
998 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
999 * OUT assertion: the match length is not greater than s->lookahead.
1000 */
1001 #ifndef ASMV
1002 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1003 * match.S. The code will be functionally equivalent.
1004 */
1005 local uInt longest_match(s, cur_match)
1006 deflate_state *s;
1007 IPos cur_match; /* current match */
1008 {
1009 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1010 register Bytef *scan = s->window + s->strstart; /* current string */
1011 register Bytef *match; /* matched string */
1012 register int len; /* length of current match */
1013 int best_len = s->prev_length; /* best match length so far */
1014 int nice_match = s->nice_match; /* stop if match long enough */
1015 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1016 s->strstart - (IPos)MAX_DIST(s) : NIL;
1017 /* Stop when cur_match becomes <= limit. To simplify the code,
1018 * we prevent matches with the string of window index 0.
1019 */
1020 Posf *prev = s->prev;
1021 uInt wmask = s->w_mask;
1022
1023 #ifdef UNALIGNED_OK
1024 /* Compare two bytes at a time. Note: this is not always beneficial.
1025 * Try with and without -DUNALIGNED_OK to check.
1026 */
1027 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1028 register ush scan_start = *(ushf*)scan;
1029 register ush scan_end = *(ushf*)(scan+best_len-1);
1030 #else
1031 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1032 register Byte scan_end1 = scan[best_len-1];
1033 register Byte scan_end = scan[best_len];
1034 #endif
1035
1036 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1037 * It is easy to get rid of this optimization if necessary.
1038 */
1039 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1040
1041 /* Do not waste too much time if we already have a good match: */
1042 if (s->prev_length >= s->good_match) {
1043 chain_length >>= 2;
1044 }
1045 /* Do not look for matches beyond the end of the input. This is necessary
1046 * to make deflate deterministic.
1047 */
1048 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1049
1050 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1051
1052 do {
1053 Assert(cur_match < s->strstart, "no future");
1054 match = s->window + cur_match;
1055
1056 /* Skip to next match if the match length cannot increase
1057 * or if the match length is less than 2. Note that the checks below
1058 * for insufficient lookahead only occur occasionally for performance
1059 * reasons. Therefore uninitialized memory will be accessed, and
1060 * conditional jumps will be made that depend on those values.
1061 * However the length of the match is limited to the lookahead, so
1062 * the output of deflate is not affected by the uninitialized values.
1063 */
1064 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1065 /* This code assumes sizeof(unsigned short) == 2. Do not use
1066 * UNALIGNED_OK if your compiler uses a different size.
1067 */
1068 if (*(ushf*)(match+best_len-1) != scan_end ||
1069 *(ushf*)match != scan_start) continue;
1070
1071 /* It is not necessary to compare scan[2] and match[2] since they are
1072 * always equal when the other bytes match, given that the hash keys
1073 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1074 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1075 * lookahead only every 4th comparison; the 128th check will be made
1076 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1077 * necessary to put more guard bytes at the end of the window, or
1078 * to check more often for insufficient lookahead.
1079 */
1080 Assert(scan[2] == match[2], "scan[2]?");
1081 scan++, match++;
1082 do {
1083 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1084 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1085 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1086 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1087 scan < strend);
1088 /* The funny "do {}" generates better code on most compilers */
1089
1090 /* Here, scan <= window+strstart+257 */
1091 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1092 if (*scan == *match) scan++;
1093
1094 len = (MAX_MATCH - 1) - (int)(strend-scan);
1095 scan = strend - (MAX_MATCH-1);
1096
1097 #else /* UNALIGNED_OK */
1098
1099 if (match[best_len] != scan_end ||
1100 match[best_len-1] != scan_end1 ||
1101 *match != *scan ||
1102 *++match != scan[1]) continue;
1103
1104 /* The check at best_len-1 can be removed because it will be made
1105 * again later. (This heuristic is not always a win.)
1106 * It is not necessary to compare scan[2] and match[2] since they
1107 * are always equal when the other bytes match, given that
1108 * the hash keys are equal and that HASH_BITS >= 8.
1109 */
1110 scan += 2, match++;
1111 Assert(*scan == *match, "match[2]?");
1112
1113 /* We check for insufficient lookahead only every 8th comparison;
1114 * the 256th check will be made at strstart+258.
1115 */
1116 do {
1117 } while (*++scan == *++match && *++scan == *++match &&
1118 *++scan == *++match && *++scan == *++match &&
1119 *++scan == *++match && *++scan == *++match &&
1120 *++scan == *++match && *++scan == *++match &&
1121 scan < strend);
1122
1123 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1124
1125 len = MAX_MATCH - (int)(strend - scan);
1126 scan = strend - MAX_MATCH;
1127
1128 #endif /* UNALIGNED_OK */
1129
1130 if (len > best_len) {
1131 s->match_start = cur_match;
1132 best_len = len;
1133 if (len >= nice_match) break;
1134 #ifdef UNALIGNED_OK
1135 scan_end = *(ushf*)(scan+best_len-1);
1136 #else
1137 scan_end1 = scan[best_len-1];
1138 scan_end = scan[best_len];
1139 #endif
1140 }
1141 } while ((cur_match = prev[cur_match & wmask]) > limit
1142 && --chain_length != 0);
1143
1144 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1145 return s->lookahead;
1146 }
1147 #endif /* ASMV */
1148 #endif /* FASTEST */
1149
1150 /* ---------------------------------------------------------------------------
1151 * Optimized version for level == 1 or strategy == Z_RLE only
1152 */
1153 local uInt longest_match_fast(s, cur_match)
1154 deflate_state *s;
1155 IPos cur_match; /* current match */
1156 {
1157 register Bytef *scan = s->window + s->strstart; /* current string */
1158 register Bytef *match; /* matched string */
1159 register int len; /* length of current match */
1160 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1161
1162 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1163 * It is easy to get rid of this optimization if necessary.
1164 */
1165 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1166
1167 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1168
1169 Assert(cur_match < s->strstart, "no future");
1170
1171 match = s->window + cur_match;
1172
1173 /* Return failure if the match length is less than 2:
1174 */
1175 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1176
1177 /* The check at best_len-1 can be removed because it will be made
1178 * again later. (This heuristic is not always a win.)
1179 * It is not necessary to compare scan[2] and match[2] since they
1180 * are always equal when the other bytes match, given that
1181 * the hash keys are equal and that HASH_BITS >= 8.
1182 */
1183 scan += 2, match += 2;
1184 Assert(*scan == *match, "match[2]?");
1185
1186 /* We check for insufficient lookahead only every 8th comparison;
1187 * the 256th check will be made at strstart+258.
1188 */
1189 do {
1190 } while (*++scan == *++match && *++scan == *++match &&
1191 *++scan == *++match && *++scan == *++match &&
1192 *++scan == *++match && *++scan == *++match &&
1193 *++scan == *++match && *++scan == *++match &&
1194 scan < strend);
1195
1196 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1197
1198 len = MAX_MATCH - (int)(strend - scan);
1199
1200 if (len < MIN_MATCH) return MIN_MATCH - 1;
1201
1202 s->match_start = cur_match;
1203 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1204 }
1205
1206 #ifdef DEBUG
1207 /* ===========================================================================
1208 * Check that the match at match_start is indeed a match.
1209 */
1210 local void check_match(s, start, match, length)
1211 deflate_state *s;
1212 IPos start, match;
1213 int length;
1214 {
1215 /* check that the match is indeed a match */
1216 if (zmemcmp(s->window + match,
1217 s->window + start, length) != EQUAL) {
1218 fprintf(stderr, " start %u, match %u, length %d\n",
1219 start, match, length);
1220 do {
1221 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1222 } while (--length != 0);
1223 z_error("invalid match");
1224 }
1225 if (z_verbose > 1) {
1226 fprintf(stderr,"\\[%d,%d]", start-match, length);
1227 do { putc(s->window[start++], stderr); } while (--length != 0);
1228 }
1229 }
1230 #else
1231 # define check_match(s, start, match, length)
1232 #endif /* DEBUG */
1233
1234 /* ===========================================================================
1235 * Fill the window when the lookahead becomes insufficient.
1236 * Updates strstart and lookahead.
1237 *
1238 * IN assertion: lookahead < MIN_LOOKAHEAD
1239 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1240 * At least one byte has been read, or avail_in == 0; reads are
1241 * performed for at least two bytes (required for the zip translate_eol
1242 * option -- not supported here).
1243 */
1244 local void fill_window(s)
1245 deflate_state *s;
1246 {
1247 register unsigned n, m;
1248 register Posf *p;
1249 unsigned more; /* Amount of free space at the end of the window. */
1250 uInt wsize = s->w_size;
1251
1252 do {
1253 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1254
1255 /* Deal with !@#$% 64K limit: */
1256 if (sizeof(int) <= 2) {
1257 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1258 more = wsize;
1259
1260 } else if (more == (unsigned)(-1)) {
1261 /* Very unlikely, but possible on 16 bit machine if
1262 * strstart == 0 && lookahead == 1 (input done a byte at time)
1263 */
1264 more--;
1265 }
1266 }
1267
1268 /* If the window is almost full and there is insufficient lookahead,
1269 * move the upper half to the lower one to make room in the upper half.
1270 */
1271 if (s->strstart >= wsize+MAX_DIST(s)) {
1272
1273 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1274 s->match_start -= wsize;
1275 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1276 s->block_start -= (long) wsize;
1277
1278 /* Slide the hash table (could be avoided with 32 bit values
1279 at the expense of memory usage). We slide even when level == 0
1280 to keep the hash table consistent if we switch back to level > 0
1281 later. (Using level 0 permanently is not an optimal usage of
1282 zlib, so we don't care about this pathological case.)
1283 */
1284 /* %%% avoid this when Z_RLE */
1285 n = s->hash_size;
1286 p = &s->head[n];
1287 do {
1288 m = *--p;
1289 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1290 } while (--n);
1291
1292 n = wsize;
1293 #ifndef FASTEST
1294 p = &s->prev[n];
1295 do {
1296 m = *--p;
1297 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1298 /* If n is not on any hash chain, prev[n] is garbage but
1299 * its value will never be used.
1300 */
1301 } while (--n);
1302 #endif
1303 more += wsize;
1304 }
1305 if (s->strm->avail_in == 0) return;
1306
1307 /* If there was no sliding:
1308 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1309 * more == window_size - lookahead - strstart
1310 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1311 * => more >= window_size - 2*WSIZE + 2
1312 * In the BIG_MEM or MMAP case (not yet supported),
1313 * window_size == input_size + MIN_LOOKAHEAD &&
1314 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1315 * Otherwise, window_size == 2*WSIZE so more >= 2.
1316 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1317 */
1318 Assert(more >= 2, "more < 2");
1319
1320 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1321 s->lookahead += n;
1322
1323 /* Initialize the hash value now that we have some input: */
1324 if (s->lookahead >= MIN_MATCH) {
1325 s->ins_h = s->window[s->strstart];
1326 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1327 #if MIN_MATCH != 3
1328 Call UPDATE_HASH() MIN_MATCH-3 more times
1329 #endif
1330 }
1331 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1332 * but this is not important since only literal bytes will be emitted.
1333 */
1334
1335 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1336 }
1337
1338 /* ===========================================================================
1339 * Flush the current block, with given end-of-file flag.
1340 * IN assertion: strstart is set to the end of the current match.
1341 */
1342 #define FLUSH_BLOCK_ONLY(s, eof) { \
1343 _tr_flush_block(s, (s->block_start >= 0L ? \
1344 (charf *)&s->window[(unsigned)s->block_start] : \
1345 (charf *)Z_NULL), \
1346 (ulg)((long)s->strstart - s->block_start), \
1347 (eof)); \
1348 s->block_start = s->strstart; \
1349 flush_pending(s->strm); \
1350 Tracev((stderr,"[FLUSH]")); \
1351 }
1352
1353 /* Same but force premature exit if necessary. */
1354 #define FLUSH_BLOCK(s, eof) { \
1355 FLUSH_BLOCK_ONLY(s, eof); \
1356 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
1357 }
1358
1359 /* ===========================================================================
1360 * Copy without compression as much as possible from the input stream, return
1361 * the current block state.
1362 * This function does not insert new strings in the dictionary since
1363 * uncompressible data is probably not useful. This function is used
1364 * only for the level=0 compression option.
1365 * NOTE: this function should be optimized to avoid extra copying from
1366 * window to pending_buf.
1367 */
1368 local block_state deflate_stored(s, flush)
1369 deflate_state *s;
1370 int flush;
1371 {
1372 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1373 * to pending_buf_size, and each stored block has a 5 byte header:
1374 */
1375 ulg max_block_size = 0xffff;
1376 ulg max_start;
1377
1378 if (max_block_size > s->pending_buf_size - 5) {
1379 max_block_size = s->pending_buf_size - 5;
1380 }
1381
1382 /* Copy as much as possible from input to output: */
1383 for (;;) {
1384 /* Fill the window as much as possible: */
1385 if (s->lookahead <= 1) {
1386
1387 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1388 s->block_start >= (long)s->w_size, "slide too late");
1389
1390 fill_window(s);
1391 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1392
1393 if (s->lookahead == 0) break; /* flush the current block */
1394 }
1395 Assert(s->block_start >= 0L, "block gone");
1396
1397 s->strstart += s->lookahead;
1398 s->lookahead = 0;
1399
1400 /* Emit a stored block if pending_buf will be full: */
1401 max_start = s->block_start + max_block_size;
1402 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1403 /* strstart == 0 is possible when wraparound on 16-bit machine */
1404 s->lookahead = (uInt)(s->strstart - max_start);
1405 s->strstart = (uInt)max_start;
1406 FLUSH_BLOCK(s, 0);
1407 }
1408 /* Flush if we may have to slide, otherwise block_start may become
1409 * negative and the data will be gone:
1410 */
1411 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1412 FLUSH_BLOCK(s, 0);
1413 }
1414 }
1415 FLUSH_BLOCK(s, flush == Z_FINISH);
1416 return flush == Z_FINISH ? finish_done : block_done;
1417 }
1418
1419 /* ===========================================================================
1420 * Compress as much as possible from the input stream, return the current
1421 * block state.
1422 * This function does not perform lazy evaluation of matches and inserts
1423 * new strings in the dictionary only for unmatched strings or for short
1424 * matches. It is used only for the fast compression options.
1425 */
1426 local block_state deflate_fast(s, flush)
1427 deflate_state *s;
1428 int flush;
1429 {
1430 IPos hash_head = NIL; /* head of the hash chain */
1431 int bflush; /* set if current block must be flushed */
1432
1433 for (;;) {
1434 /* Make sure that we always have enough lookahead, except
1435 * at the end of the input file. We need MAX_MATCH bytes
1436 * for the next match, plus MIN_MATCH bytes to insert the
1437 * string following the next match.
1438 */
1439 if (s->lookahead < MIN_LOOKAHEAD) {
1440 fill_window(s);
1441 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1442 return need_more;
1443 }
1444 if (s->lookahead == 0) break; /* flush the current block */
1445 }
1446
1447 /* Insert the string window[strstart .. strstart+2] in the
1448 * dictionary, and set hash_head to the head of the hash chain:
1449 */
1450 if (s->lookahead >= MIN_MATCH) {
1451 INSERT_STRING(s, s->strstart, hash_head);
1452 }
1453
1454 /* Find the longest match, discarding those <= prev_length.
1455 * At this point we have always match_length < MIN_MATCH
1456 */
1457 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1458 /* To simplify the code, we prevent matches with the string
1459 * of window index 0 (in particular we have to avoid a match
1460 * of the string with itself at the start of the input file).
1461 */
1462 #ifdef FASTEST
1463 if ((s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) ||
1464 (s->strategy == Z_RLE && s->strstart - hash_head == 1)) {
1465 s->match_length = longest_match_fast (s, hash_head);
1466 }
1467 #else
1468 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) {
1469 s->match_length = longest_match (s, hash_head);
1470 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) {
1471 s->match_length = longest_match_fast (s, hash_head);
1472 }
1473 #endif
1474 /* longest_match() or longest_match_fast() sets match_start */
1475 }
1476 if (s->match_length >= MIN_MATCH) {
1477 check_match(s, s->strstart, s->match_start, s->match_length);
1478
1479 _tr_tally_dist(s, s->strstart - s->match_start,
1480 s->match_length - MIN_MATCH, bflush);
1481
1482 s->lookahead -= s->match_length;
1483
1484 /* Insert new strings in the hash table only if the match length
1485 * is not too large. This saves time but degrades compression.
1486 */
1487 #ifndef FASTEST
1488 if (s->match_length <= s->max_insert_length &&
1489 s->lookahead >= MIN_MATCH) {
1490 s->match_length--; /* string at strstart already in table */
1491 do {
1492 s->strstart++;
1493 INSERT_STRING(s, s->strstart, hash_head);
1494 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1495 * always MIN_MATCH bytes ahead.
1496 */
1497 } while (--s->match_length != 0);
1498 s->strstart++;
1499 } else
1500 #endif
1501 {
1502 s->strstart += s->match_length;
1503 s->match_length = 0;
1504 s->ins_h = s->window[s->strstart];
1505 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1506 #if MIN_MATCH != 3
1507 Call UPDATE_HASH() MIN_MATCH-3 more times
1508 #endif
1509 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1510 * matter since it will be recomputed at next deflate call.
1511 */
1512 }
1513 } else {
1514 /* No match, output a literal byte */
1515 Tracevv((stderr,"%c", s->window[s->strstart]));
1516 _tr_tally_lit (s, s->window[s->strstart], bflush);
1517 s->lookahead--;
1518 s->strstart++;
1519 }
1520 if (bflush) FLUSH_BLOCK(s, 0);
1521 }
1522 FLUSH_BLOCK(s, flush == Z_FINISH);
1523 return flush == Z_FINISH ? finish_done : block_done;
1524 }
1525
1526 #ifndef FASTEST
1527 /* ===========================================================================
1528 * Same as above, but achieves better compression. We use a lazy
1529 * evaluation for matches: a match is finally adopted only if there is
1530 * no better match at the next window position.
1531 */
1532 local block_state deflate_slow(s, flush)
1533 deflate_state *s;
1534 int flush;
1535 {
1536 IPos hash_head = NIL; /* head of hash chain */
1537 int bflush; /* set if current block must be flushed */
1538
1539 /* Process the input block. */
1540 for (;;) {
1541 /* Make sure that we always have enough lookahead, except
1542 * at the end of the input file. We need MAX_MATCH bytes
1543 * for the next match, plus MIN_MATCH bytes to insert the
1544 * string following the next match.
1545 */
1546 if (s->lookahead < MIN_LOOKAHEAD) {
1547 fill_window(s);
1548 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1549 return need_more;
1550 }
1551 if (s->lookahead == 0) break; /* flush the current block */
1552 }
1553
1554 /* Insert the string window[strstart .. strstart+2] in the
1555 * dictionary, and set hash_head to the head of the hash chain:
1556 */
1557 if (s->lookahead >= MIN_MATCH) {
1558 INSERT_STRING(s, s->strstart, hash_head);
1559 }
1560
1561 /* Find the longest match, discarding those <= prev_length.
1562 */
1563 s->prev_length = s->match_length, s->prev_match = s->match_start;
1564 s->match_length = MIN_MATCH-1;
1565
1566 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1567 s->strstart - hash_head <= MAX_DIST(s)) {
1568 /* To simplify the code, we prevent matches with the string
1569 * of window index 0 (in particular we have to avoid a match
1570 * of the string with itself at the start of the input file).
1571 */
1572 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) {
1573 s->match_length = longest_match (s, hash_head);
1574 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) {
1575 s->match_length = longest_match_fast (s, hash_head);
1576 }
1577 /* longest_match() or longest_match_fast() sets match_start */
1578
1579 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1580 #if TOO_FAR <= 32767
1581 || (s->match_length == MIN_MATCH &&
1582 s->strstart - s->match_start > TOO_FAR)
1583 #endif
1584 )) {
1585
1586 /* If prev_match is also MIN_MATCH, match_start is garbage
1587 * but we will ignore the current match anyway.
1588 */
1589 s->match_length = MIN_MATCH-1;
1590 }
1591 }
1592 /* If there was a match at the previous step and the current
1593 * match is not better, output the previous match:
1594 */
1595 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1596 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1597 /* Do not insert strings in hash table beyond this. */
1598
1599 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1600
1601 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1602 s->prev_length - MIN_MATCH, bflush);
1603
1604 /* Insert in hash table all strings up to the end of the match.
1605 * strstart-1 and strstart are already inserted. If there is not
1606 * enough lookahead, the last two strings are not inserted in
1607 * the hash table.
1608 */
1609 s->lookahead -= s->prev_length-1;
1610 s->prev_length -= 2;
1611 do {
1612 if (++s->strstart <= max_insert) {
1613 INSERT_STRING(s, s->strstart, hash_head);
1614 }
1615 } while (--s->prev_length != 0);
1616 s->match_available = 0;
1617 s->match_length = MIN_MATCH-1;
1618 s->strstart++;
1619
1620 if (bflush) FLUSH_BLOCK(s, 0);
1621
1622 } else if (s->match_available) {
1623 /* If there was no match at the previous position, output a
1624 * single literal. If there was a match but the current match
1625 * is longer, truncate the previous match to a single literal.
1626 */
1627 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1628 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1629 if (bflush) {
1630 FLUSH_BLOCK_ONLY(s, 0);
1631 }
1632 s->strstart++;
1633 s->lookahead--;
1634 if (s->strm->avail_out == 0) return need_more;
1635 } else {
1636 /* There is no previous match to compare with, wait for
1637 * the next step to decide.
1638 */
1639 s->match_available = 1;
1640 s->strstart++;
1641 s->lookahead--;
1642 }
1643 }
1644 Assert (flush != Z_NO_FLUSH, "no flush?");
1645 if (s->match_available) {
1646 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1647 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1648 s->match_available = 0;
1649 }
1650 FLUSH_BLOCK(s, flush == Z_FINISH);
1651 return flush == Z_FINISH ? finish_done : block_done;
1652 }
1653 #endif /* FASTEST */
1654
1655 #if 0
1656 /* ===========================================================================
1657 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1658 * one. Do not maintain a hash table. (It will be regenerated if this run of
1659 * deflate switches away from Z_RLE.)
1660 */
1661 local block_state deflate_rle(s, flush)
1662 deflate_state *s;
1663 int flush;
1664 {
1665 int bflush; /* set if current block must be flushed */
1666 uInt run; /* length of run */
1667 uInt max; /* maximum length of run */
1668 uInt prev; /* byte at distance one to match */
1669 Bytef *scan; /* scan for end of run */
1670
1671 for (;;) {
1672 /* Make sure that we always have enough lookahead, except
1673 * at the end of the input file. We need MAX_MATCH bytes
1674 * for the longest encodable run.
1675 */
1676 if (s->lookahead < MAX_MATCH) {
1677 fill_window(s);
1678 if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {
1679 return need_more;
1680 }
1681 if (s->lookahead == 0) break; /* flush the current block */
1682 }
1683
1684 /* See how many times the previous byte repeats */
1685 run = 0;
1686 if (s->strstart > 0) { /* if there is a previous byte, that is */
1687 max = s->lookahead < MAX_MATCH ? s->lookahead : MAX_MATCH;
1688 scan = s->window + s->strstart - 1;
1689 prev = *scan++;
1690 do {
1691 if (*scan++ != prev)
1692 break;
1693 } while (++run < max);
1694 }
1695
1696 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1697 if (run >= MIN_MATCH) {
1698 check_match(s, s->strstart, s->strstart - 1, run);
1699 _tr_tally_dist(s, 1, run - MIN_MATCH, bflush);
1700 s->lookahead -= run;
1701 s->strstart += run;
1702 } else {
1703 /* No match, output a literal byte */
1704 Tracevv((stderr,"%c", s->window[s->strstart]));
1705 _tr_tally_lit (s, s->window[s->strstart], bflush);
1706 s->lookahead--;
1707 s->strstart++;
1708 }
1709 if (bflush) FLUSH_BLOCK(s, 0);
1710 }
1711 FLUSH_BLOCK(s, flush == Z_FINISH);
1712 return flush == Z_FINISH ? finish_done : block_done;
1713 }
1714 #endif
1715