1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2010 Jean-loup Gailly and Mark Adler
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 /* \param (#) $Id$ */
51
52 #include "deflate.h"
53
54 const char deflate_copyright[] =
55 " deflate 1.2.5 Copyright 1995-2010 Jean-loup Gailly and Mark Adler ";
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 block_state deflate_rle OF((deflate_state *s, int flush));
83 local block_state deflate_huff OF((deflate_state *s, int flush));
84 local void lm_init OF((deflate_state *s));
85 local void putShortMSB OF((deflate_state *s, uInt b));
86 local void flush_pending OF((z_streamp strm));
87 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
88 #ifdef ASMV
89 void match_init OF((void)); /* asm code initialization */
90 uInt longest_match OF((deflate_state *s, IPos cur_match));
91 #else
92 local uInt longest_match OF((deflate_state *s, IPos cur_match));
93 #endif
94
95 #ifdef DEBUG
96 local void check_match OF((deflate_state *s, IPos start, IPos match,
97 int length));
98 #endif
99
100 /* ===========================================================================
101 * Local data
102 */
103
104 #define NIL 0
105 /* Tail of hash chains */
106
107 #ifndef TOO_FAR
108 # define TOO_FAR 4096
109 #endif
110 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
111
112 /* Values for max_lazy_match, good_match and max_chain_length, depending on
113 * the desired pack level (0..9). The values given below have been tuned to
114 * exclude worst case performance for pathological files. Better values may be
115 * found for specific files.
116 */
117 typedef struct config_s {
118 ush good_length; /* reduce lazy search above this match length */
119 ush max_lazy; /* do not perform lazy search above this match length */
120 ush nice_length; /* quit search above this match length */
121 ush max_chain;
122 compress_func func;
123 } config;
124
125 #ifdef FASTEST
126 local const config configuration_table[2] = {
127 /* good lazy nice chain */
128 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
129 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
130 #else
131 local const config configuration_table[10] = {
132 /* good lazy nice chain */
133 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
134 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
135 /* 2 */ {4, 5, 16, 8, deflate_fast},
136 /* 3 */ {4, 6, 32, 32, deflate_fast},
137
138 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
139 /* 5 */ {8, 16, 32, 32, deflate_slow},
140 /* 6 */ {8, 16, 128, 128, deflate_slow},
141 /* 7 */ {8, 32, 128, 256, deflate_slow},
142 /* 8 */ {32, 128, 258, 1024, deflate_slow},
143 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
144 #endif
145
146 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
147 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
148 * meaning.
149 */
150
151 #define EQUAL 0
152 /* result of memcmp for equal strings */
153
154 #ifndef NO_DUMMY_DECL
155 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
156 #endif
157
158 /* ===========================================================================
159 * Update a hash value with the given input byte
160 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
161 * input characters, so that a running hash key can be computed from the
162 * previous key instead of complete recalculation each time.
163 */
164 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
165
166
167 /* ===========================================================================
168 * Insert string str in the dictionary and set match_head to the previous head
169 * of the hash chain (the most recent string with same hash key). Return
170 * the previous length of the hash chain.
171 * If this file is compiled with -DFASTEST, the compression level is forced
172 * to 1, and no hash chains are maintained.
173 * IN assertion: all calls to to INSERT_STRING are made with consecutive
174 * input characters and the first MIN_MATCH bytes of str are valid
175 * (except for the last MIN_MATCH-1 bytes of the input file).
176 */
177 #ifdef FASTEST
178 #define INSERT_STRING(s, str, match_head) \
179 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
180 match_head = s->head[s->ins_h], \
181 s->head[s->ins_h] = (Pos)(str))
182 #else
183 #define INSERT_STRING(s, str, match_head) \
184 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
185 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
186 s->head[s->ins_h] = (Pos)(str))
187 #endif
188
189 /* ===========================================================================
190 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
191 * prev[] will be initialized on the fly.
192 */
193 #define CLEAR_HASH(s) \
194 s->head[s->hash_size-1] = NIL; \
195 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
196
197 /* ========================================================================= */
deflateInit_(strm,level,version,stream_size)198 int ZEXPORT deflateInit_(strm, level, version, stream_size)
199 z_streamp strm;
200 int level;
201 const char *version;
202 int stream_size;
203 {
204 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
205 Z_DEFAULT_STRATEGY, version, stream_size);
206 /* To do: ignore strm->next_in if we use it as window */
207 }
208
209 /* ========================================================================= */
deflateInit2_(strm,level,method,windowBits,memLevel,strategy,version,stream_size)210 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
211 version, stream_size)
212 z_streamp strm;
213 int level;
214 int method;
215 int windowBits;
216 int memLevel;
217 int strategy;
218 const char *version;
219 int stream_size;
220 {
221 deflate_state *s;
222 int wrap = 1;
223 static const char my_version[] = ZLIB_VERSION;
224
225 ushf *overlay;
226 /* We overlay pending_buf and d_buf+l_buf. This works since the average
227 * output size for (length,distance) codes is <= 24 bits.
228 */
229
230 if (version == Z_NULL || version[0] != my_version[0] ||
231 stream_size != sizeof(z_stream)) {
232 return Z_VERSION_ERROR;
233 }
234 if (strm == Z_NULL) return Z_STREAM_ERROR;
235
236 strm->msg = Z_NULL;
237 if (strm->zalloc == (alloc_func)0) {
238 strm->zalloc = zcalloc;
239 strm->opaque = (voidpf)0;
240 }
241 if (strm->zfree == (free_func)0) strm->zfree = zcfree;
242
243 #ifdef FASTEST
244 if (level != 0) level = 1;
245 #else
246 if (level == Z_DEFAULT_COMPRESSION) level = 6;
247 #endif
248
249 if (windowBits < 0) { /* suppress zlib wrapper */
250 wrap = 0;
251 windowBits = -windowBits;
252 }
253 #ifdef GZIP
254 else if (windowBits > 15) {
255 wrap = 2; /* write gzip wrapper instead */
256 windowBits -= 16;
257 }
258 #endif
259 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
260 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
261 strategy < 0 || strategy > Z_FIXED) {
262 return Z_STREAM_ERROR;
263 }
264 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
265 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
266 if (s == Z_NULL) return Z_MEM_ERROR;
267 strm->state = (struct internal_state FAR *)s;
268 s->strm = strm;
269
270 s->wrap = wrap;
271 s->gzhead = Z_NULL;
272 s->w_bits = windowBits;
273 s->w_size = 1 << s->w_bits;
274 s->w_mask = s->w_size - 1;
275
276 s->hash_bits = memLevel + 7;
277 s->hash_size = 1 << s->hash_bits;
278 s->hash_mask = s->hash_size - 1;
279 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
280
281 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
282 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
283 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
284
285 s->high_water = 0; /* nothing written to s->window yet */
286
287 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
288
289 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
290 s->pending_buf = (uchf *) overlay;
291 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
292
293 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
294 s->pending_buf == Z_NULL) {
295 s->status = FINISH_STATE;
296 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
297 deflateEnd (strm);
298 return Z_MEM_ERROR;
299 }
300 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
301 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
302
303 s->level = level;
304 s->strategy = strategy;
305 s->method = (Byte)method;
306
307 return deflateReset(strm);
308 }
309
310 /* ========================================================================= */
deflateSetDictionary(strm,dictionary,dictLength)311 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
312 z_streamp strm;
313 const Bytef *dictionary;
314 uInt dictLength;
315 {
316 deflate_state *s;
317 uInt length = dictLength;
318 uInt n;
319 IPos hash_head = 0;
320
321 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
322 strm->state->wrap == 2 ||
323 (strm->state->wrap == 1 && strm->state->status != INIT_STATE))
324 return Z_STREAM_ERROR;
325
326 s = strm->state;
327 if (s->wrap)
328 strm->adler = adler32(strm->adler, dictionary, dictLength);
329
330 if (length < MIN_MATCH) return Z_OK;
331 if (length > s->w_size) {
332 length = s->w_size;
333 dictionary += dictLength - length; /* use the tail of the dictionary */
334 }
335 zmemcpy(s->window, dictionary, length);
336 s->strstart = length;
337 s->block_start = (long)length;
338
339 /* Insert all strings in the hash table (except for the last two bytes).
340 * s->lookahead stays null, so s->ins_h will be recomputed at the next
341 * call of fill_window.
342 */
343 s->ins_h = s->window[0];
344 UPDATE_HASH(s, s->ins_h, s->window[1]);
345 for (n = 0; n <= length - MIN_MATCH; n++) {
346 INSERT_STRING(s, n, hash_head);
347 }
348 if (hash_head) hash_head = 0; /* to make compiler happy */
349 return Z_OK;
350 }
351
352 /* ========================================================================= */
deflateReset(strm)353 int ZEXPORT deflateReset (strm)
354 z_streamp strm;
355 {
356 deflate_state *s;
357
358 if (strm == Z_NULL || strm->state == Z_NULL ||
359 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
360 return Z_STREAM_ERROR;
361 }
362
363 strm->total_in = strm->total_out = 0;
364 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
365 strm->data_type = Z_UNKNOWN;
366
367 s = (deflate_state *)strm->state;
368 s->pending = 0;
369 s->pending_out = s->pending_buf;
370
371 if (s->wrap < 0) {
372 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
373 }
374 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
375 strm->adler =
376 #ifdef GZIP
377 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
378 #endif
379 adler32(0L, Z_NULL, 0);
380 s->last_flush = Z_NO_FLUSH;
381
382 _tr_init(s);
383 lm_init(s);
384
385 return Z_OK;
386 }
387
388 /* ========================================================================= */
deflateSetHeader(strm,head)389 int ZEXPORT deflateSetHeader (strm, head)
390 z_streamp strm;
391 gz_headerp head;
392 {
393 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
394 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
395 strm->state->gzhead = head;
396 return Z_OK;
397 }
398
399 /* ========================================================================= */
deflatePending(strm,pending,bits)400 int ZEXPORT deflatePending (strm, pending, bits)
401 unsigned *pending;
402 int *bits;
403 z_streamp strm;
404 {
405 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
406 *pending = strm->state->pending;
407 *bits = strm->state->bi_valid;
408 return Z_OK;
409 }
410
411 /* ========================================================================= */
deflatePrime(strm,bits,value)412 int ZEXPORT deflatePrime (strm, bits, value)
413 z_streamp strm;
414 int bits;
415 int value;
416 {
417 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
418 strm->state->bi_valid = bits;
419 strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));
420 return Z_OK;
421 }
422
423 /* ========================================================================= */
deflateParams(strm,level,strategy)424 int ZEXPORT deflateParams(strm, level, strategy)
425 z_streamp strm;
426 int level;
427 int strategy;
428 {
429 deflate_state *s;
430 compress_func func;
431 int err = Z_OK;
432
433 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
434 s = strm->state;
435
436 #ifdef FASTEST
437 if (level != 0) level = 1;
438 #else
439 if (level == Z_DEFAULT_COMPRESSION) level = 6;
440 #endif
441 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
442 return Z_STREAM_ERROR;
443 }
444 func = configuration_table[s->level].func;
445
446 if ((strategy != s->strategy || func != configuration_table[level].func) &&
447 strm->total_in != 0) {
448 /* Flush the last buffer: */
449 err = deflate(strm, Z_BLOCK);
450 }
451 if (s->level != level) {
452 s->level = level;
453 s->max_lazy_match = configuration_table[level].max_lazy;
454 s->good_match = configuration_table[level].good_length;
455 s->nice_match = configuration_table[level].nice_length;
456 s->max_chain_length = configuration_table[level].max_chain;
457 }
458 s->strategy = strategy;
459 return err;
460 }
461
462 /* ========================================================================= */
deflateTune(strm,good_length,max_lazy,nice_length,max_chain)463 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
464 z_streamp strm;
465 int good_length;
466 int max_lazy;
467 int nice_length;
468 int max_chain;
469 {
470 deflate_state *s;
471
472 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
473 s = strm->state;
474 s->good_match = good_length;
475 s->max_lazy_match = max_lazy;
476 s->nice_match = nice_length;
477 s->max_chain_length = max_chain;
478 return Z_OK;
479 }
480
481 /* =========================================================================
482 * For the default windowBits of 15 and memLevel of 8, this function returns
483 * a close to exact, as well as small, upper bound on the compressed size.
484 * They are coded as constants here for a reason--if the #define's are
485 * changed, then this function needs to be changed as well. The return
486 * value for 15 and 8 only works for those exact settings.
487 *
488 * For any setting other than those defaults for windowBits and memLevel,
489 * the value returned is a conservative worst case for the maximum expansion
490 * resulting from using fixed blocks instead of stored blocks, which deflate
491 * can emit on compressed data for some combinations of the parameters.
492 *
493 * This function could be more sophisticated to provide closer upper bounds for
494 * every combination of windowBits and memLevel. But even the conservative
495 * upper bound of about 14% expansion does not seem onerous for output buffer
496 * allocation.
497 */
deflateBound(strm,sourceLen)498 uLong ZEXPORT deflateBound(strm, sourceLen)
499 z_streamp strm;
500 uLong sourceLen;
501 {
502 deflate_state *s;
503 uLong complen, wraplen;
504 Bytef *str;
505
506 /* conservative upper bound for compressed data */
507 complen = sourceLen +
508 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
509
510 /* if can't get parameters, return conservative bound plus zlib wrapper */
511 if (strm == Z_NULL || strm->state == Z_NULL)
512 return complen + 6;
513
514 /* compute wrapper length */
515 s = strm->state;
516 switch (s->wrap) {
517 case 0: /* raw deflate */
518 wraplen = 0;
519 break;
520 case 1: /* zlib wrapper */
521 wraplen = 6 + (s->strstart ? 4 : 0);
522 break;
523 case 2: /* gzip wrapper */
524 wraplen = 18;
525 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
526 if (s->gzhead->extra != Z_NULL)
527 wraplen += 2 + s->gzhead->extra_len;
528 str = s->gzhead->name;
529 if (str != Z_NULL)
530 do {
531 wraplen++;
532 } while (*str++);
533 str = s->gzhead->comment;
534 if (str != Z_NULL)
535 do {
536 wraplen++;
537 } while (*str++);
538 if (s->gzhead->hcrc)
539 wraplen += 2;
540 }
541 break;
542 default: /* for compiler happiness */
543 wraplen = 6;
544 }
545
546 /* if not default parameters, return conservative bound */
547 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
548 return complen + wraplen;
549
550 /* default settings: return tight bound for that case */
551 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
552 (sourceLen >> 25) + 13 - 6 + wraplen;
553 }
554
555 /* =========================================================================
556 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
557 * IN assertion: the stream state is correct and there is enough room in
558 * pending_buf.
559 */
putShortMSB(s,b)560 local void putShortMSB (s, b)
561 deflate_state *s;
562 uInt b;
563 {
564 put_byte(s, (Byte)(b >> 8));
565 put_byte(s, (Byte)(b & 0xff));
566 }
567
568 /* =========================================================================
569 * Flush as much pending output as possible. All deflate() output goes
570 * through this function so some applications may wish to modify it
571 * to avoid allocating a large strm->next_out buffer and copying into it.
572 * (See also read_buf()).
573 */
flush_pending(strm)574 local void flush_pending(strm)
575 z_streamp strm;
576 {
577 unsigned len = strm->state->pending;
578
579 if (len > strm->avail_out) len = strm->avail_out;
580 if (len == 0) return;
581
582 zmemcpy(strm->next_out, strm->state->pending_out, len);
583 strm->next_out += len;
584 strm->state->pending_out += len;
585 strm->total_out += len;
586 strm->avail_out -= len;
587 strm->state->pending -= len;
588 if (strm->state->pending == 0) {
589 strm->state->pending_out = strm->state->pending_buf;
590 }
591 }
592
593 /* ========================================================================= */
deflate(strm,flush)594 int ZEXPORT deflate (strm, flush)
595 z_streamp strm;
596 int flush;
597 {
598 int old_flush; /* value of flush param for previous deflate call */
599 deflate_state *s;
600
601 if (strm == Z_NULL || strm->state == Z_NULL ||
602 flush > Z_BLOCK || flush < 0) {
603 return Z_STREAM_ERROR;
604 }
605 s = strm->state;
606
607 if (strm->next_out == Z_NULL ||
608 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
609 (s->status == FINISH_STATE && flush != Z_FINISH)) {
610 ERR_RETURN(strm, Z_STREAM_ERROR);
611 }
612 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
613
614 s->strm = strm; /* just in case */
615 old_flush = s->last_flush;
616 s->last_flush = flush;
617
618 /* Write the header */
619 if (s->status == INIT_STATE) {
620 #ifdef GZIP
621 if (s->wrap == 2) {
622 strm->adler = crc32(0L, Z_NULL, 0);
623 put_byte(s, 31);
624 put_byte(s, 139);
625 put_byte(s, 8);
626 if (s->gzhead == Z_NULL) {
627 put_byte(s, 0);
628 put_byte(s, 0);
629 put_byte(s, 0);
630 put_byte(s, 0);
631 put_byte(s, 0);
632 put_byte(s, s->level == 9 ? 2 :
633 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
634 4 : 0));
635 put_byte(s, OS_CODE);
636 s->status = BUSY_STATE;
637 }
638 else {
639 put_byte(s, (s->gzhead->text ? 1 : 0) +
640 (s->gzhead->hcrc ? 2 : 0) +
641 (s->gzhead->extra == Z_NULL ? 0 : 4) +
642 (s->gzhead->name == Z_NULL ? 0 : 8) +
643 (s->gzhead->comment == Z_NULL ? 0 : 16)
644 );
645 put_byte(s, (Byte)(s->gzhead->time & 0xff));
646 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
647 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
648 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
649 put_byte(s, s->level == 9 ? 2 :
650 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
651 4 : 0));
652 put_byte(s, s->gzhead->os & 0xff);
653 if (s->gzhead->extra != Z_NULL) {
654 put_byte(s, s->gzhead->extra_len & 0xff);
655 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
656 }
657 if (s->gzhead->hcrc)
658 strm->adler = crc32(strm->adler, s->pending_buf,
659 s->pending);
660 s->gzindex = 0;
661 s->status = EXTRA_STATE;
662 }
663 }
664 else
665 #endif
666 {
667 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
668 uInt level_flags;
669
670 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
671 level_flags = 0;
672 else if (s->level < 6)
673 level_flags = 1;
674 else if (s->level == 6)
675 level_flags = 2;
676 else
677 level_flags = 3;
678 header |= (level_flags << 6);
679 if (s->strstart != 0) header |= PRESET_DICT;
680 header += 31 - (header % 31);
681
682 s->status = BUSY_STATE;
683 putShortMSB(s, header);
684
685 /* Save the adler32 of the preset dictionary: */
686 if (s->strstart != 0) {
687 putShortMSB(s, (uInt)(strm->adler >> 16));
688 putShortMSB(s, (uInt)(strm->adler & 0xffff));
689 }
690 strm->adler = adler32(0L, Z_NULL, 0);
691 }
692 }
693 #ifdef GZIP
694 if (s->status == EXTRA_STATE) {
695 if (s->gzhead->extra != Z_NULL) {
696 uInt beg = s->pending; /* start of bytes to update crc */
697
698 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
699 if (s->pending == s->pending_buf_size) {
700 if (s->gzhead->hcrc && s->pending > beg)
701 strm->adler = crc32(strm->adler, s->pending_buf + beg,
702 s->pending - beg);
703 flush_pending(strm);
704 beg = s->pending;
705 if (s->pending == s->pending_buf_size)
706 break;
707 }
708 put_byte(s, s->gzhead->extra[s->gzindex]);
709 s->gzindex++;
710 }
711 if (s->gzhead->hcrc && s->pending > beg)
712 strm->adler = crc32(strm->adler, s->pending_buf + beg,
713 s->pending - beg);
714 if (s->gzindex == s->gzhead->extra_len) {
715 s->gzindex = 0;
716 s->status = NAME_STATE;
717 }
718 }
719 else
720 s->status = NAME_STATE;
721 }
722 if (s->status == NAME_STATE) {
723 if (s->gzhead->name != Z_NULL) {
724 uInt beg = s->pending; /* start of bytes to update crc */
725 int val;
726
727 do {
728 if (s->pending == s->pending_buf_size) {
729 if (s->gzhead->hcrc && s->pending > beg)
730 strm->adler = crc32(strm->adler, s->pending_buf + beg,
731 s->pending - beg);
732 flush_pending(strm);
733 beg = s->pending;
734 if (s->pending == s->pending_buf_size) {
735 val = 1;
736 break;
737 }
738 }
739 val = s->gzhead->name[s->gzindex++];
740 put_byte(s, val);
741 } while (val != 0);
742 if (s->gzhead->hcrc && s->pending > beg)
743 strm->adler = crc32(strm->adler, s->pending_buf + beg,
744 s->pending - beg);
745 if (val == 0) {
746 s->gzindex = 0;
747 s->status = COMMENT_STATE;
748 }
749 }
750 else
751 s->status = COMMENT_STATE;
752 }
753 if (s->status == COMMENT_STATE) {
754 if (s->gzhead->comment != Z_NULL) {
755 uInt beg = s->pending; /* start of bytes to update crc */
756 int val;
757
758 do {
759 if (s->pending == s->pending_buf_size) {
760 if (s->gzhead->hcrc && s->pending > beg)
761 strm->adler = crc32(strm->adler, s->pending_buf + beg,
762 s->pending - beg);
763 flush_pending(strm);
764 beg = s->pending;
765 if (s->pending == s->pending_buf_size) {
766 val = 1;
767 break;
768 }
769 }
770 val = s->gzhead->comment[s->gzindex++];
771 put_byte(s, val);
772 } while (val != 0);
773 if (s->gzhead->hcrc && s->pending > beg)
774 strm->adler = crc32(strm->adler, s->pending_buf + beg,
775 s->pending - beg);
776 if (val == 0)
777 s->status = HCRC_STATE;
778 }
779 else
780 s->status = HCRC_STATE;
781 }
782 if (s->status == HCRC_STATE) {
783 if (s->gzhead->hcrc) {
784 if (s->pending + 2 > s->pending_buf_size)
785 flush_pending(strm);
786 if (s->pending + 2 <= s->pending_buf_size) {
787 put_byte(s, (Byte)(strm->adler & 0xff));
788 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
789 strm->adler = crc32(0L, Z_NULL, 0);
790 s->status = BUSY_STATE;
791 }
792 }
793 else
794 s->status = BUSY_STATE;
795 }
796 #endif
797
798 /* Flush as much pending output as possible */
799 if (s->pending != 0) {
800 flush_pending(strm);
801 if (strm->avail_out == 0) {
802 /* Since avail_out is 0, deflate will be called again with
803 * more output space, but possibly with both pending and
804 * avail_in equal to zero. There won't be anything to do,
805 * but this is not an error situation so make sure we
806 * return OK instead of BUF_ERROR at next call of deflate:
807 */
808 s->last_flush = -1;
809 return Z_OK;
810 }
811
812 /* Make sure there is something to do and avoid duplicate consecutive
813 * flushes. For repeated and useless calls with Z_FINISH, we keep
814 * returning Z_STREAM_END instead of Z_BUF_ERROR.
815 */
816 } else if (strm->avail_in == 0 && flush <= old_flush &&
817 flush != Z_FINISH) {
818 ERR_RETURN(strm, Z_BUF_ERROR);
819 }
820
821 /* User must not provide more input after the first FINISH: */
822 if (s->status == FINISH_STATE && strm->avail_in != 0) {
823 ERR_RETURN(strm, Z_BUF_ERROR);
824 }
825
826 /* Start a new block or continue the current one.
827 */
828 if (strm->avail_in != 0 || s->lookahead != 0 ||
829 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
830 block_state bstate;
831
832 bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
833 (s->strategy == Z_RLE ? deflate_rle(s, flush) :
834 (*(configuration_table[s->level].func))(s, flush));
835
836 if (bstate == finish_started || bstate == finish_done) {
837 s->status = FINISH_STATE;
838 }
839 if (bstate == need_more || bstate == finish_started) {
840 if (strm->avail_out == 0) {
841 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
842 }
843 return Z_OK;
844 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
845 * of deflate should use the same flush parameter to make sure
846 * that the flush is complete. So we don't have to output an
847 * empty block here, this will be done at next call. This also
848 * ensures that for a very small output buffer, we emit at most
849 * one empty block.
850 */
851 }
852 if (bstate == block_done) {
853 if (flush == Z_PARTIAL_FLUSH) {
854 _tr_align(s);
855 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
856 _tr_stored_block(s, (char*)0, 0L, 0);
857 /* For a full flush, this empty block will be recognized
858 * as a special marker by inflate_sync().
859 */
860 if (flush == Z_FULL_FLUSH) {
861 CLEAR_HASH(s); /* forget history */
862 if (s->lookahead == 0) {
863 s->strstart = 0;
864 s->block_start = 0L;
865 }
866 }
867 }
868 flush_pending(strm);
869 if (strm->avail_out == 0) {
870 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
871 return Z_OK;
872 }
873 }
874 }
875 Assert(strm->avail_out > 0, "bug2");
876
877 if (flush != Z_FINISH) return Z_OK;
878 if (s->wrap <= 0) return Z_STREAM_END;
879
880 /* Write the trailer */
881 #ifdef GZIP
882 if (s->wrap == 2) {
883 put_byte(s, (Byte)(strm->adler & 0xff));
884 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
885 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
886 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
887 put_byte(s, (Byte)(strm->total_in & 0xff));
888 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
889 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
890 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
891 }
892 else
893 #endif
894 {
895 putShortMSB(s, (uInt)(strm->adler >> 16));
896 putShortMSB(s, (uInt)(strm->adler & 0xffff));
897 }
898 flush_pending(strm);
899 /* If avail_out is zero, the application will call deflate again
900 * to flush the rest.
901 */
902 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
903 return s->pending != 0 ? Z_OK : Z_STREAM_END;
904 }
905
906 /* ========================================================================= */
deflateEnd(strm)907 int ZEXPORT deflateEnd (strm)
908 z_streamp strm;
909 {
910 int status;
911
912 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
913
914 status = strm->state->status;
915 if (status != INIT_STATE &&
916 status != EXTRA_STATE &&
917 status != NAME_STATE &&
918 status != COMMENT_STATE &&
919 status != HCRC_STATE &&
920 status != BUSY_STATE &&
921 status != FINISH_STATE) {
922 return Z_STREAM_ERROR;
923 }
924
925 /* Deallocate in reverse order of allocations: */
926 TRY_FREE(strm, strm->state->pending_buf);
927 TRY_FREE(strm, strm->state->head);
928 TRY_FREE(strm, strm->state->prev);
929 TRY_FREE(strm, strm->state->window);
930
931 ZFREE(strm, strm->state);
932 strm->state = Z_NULL;
933
934 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
935 }
936
937 /* =========================================================================
938 * Copy the source state to the destination state.
939 * To simplify the source, this is not supported for 16-bit MSDOS (which
940 * doesn't have enough memory anyway to duplicate compression states).
941 */
deflateCopy(dest,source)942 int ZEXPORT deflateCopy (dest, source)
943 z_streamp dest;
944 z_streamp source;
945 {
946 #ifdef MAXSEG_64K
947 return Z_STREAM_ERROR;
948 #else
949 deflate_state *ds;
950 deflate_state *ss;
951 ushf *overlay;
952
953
954 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
955 return Z_STREAM_ERROR;
956 }
957
958 ss = source->state;
959
960 zmemcpy(dest, source, sizeof(z_stream));
961
962 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
963 if (ds == Z_NULL) return Z_MEM_ERROR;
964 dest->state = (struct internal_state FAR *) ds;
965 zmemcpy(ds, ss, sizeof(deflate_state));
966 ds->strm = dest;
967
968 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
969 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
970 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
971 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
972 ds->pending_buf = (uchf *) overlay;
973
974 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
975 ds->pending_buf == Z_NULL) {
976 deflateEnd (dest);
977 return Z_MEM_ERROR;
978 }
979 /* following zmemcpy do not work for 16-bit MSDOS */
980 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
981 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
982 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
983 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
984
985 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
986 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
987 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
988
989 ds->l_desc.dyn_tree = ds->dyn_ltree;
990 ds->d_desc.dyn_tree = ds->dyn_dtree;
991 ds->bl_desc.dyn_tree = ds->bl_tree;
992
993 return Z_OK;
994 #endif /* MAXSEG_64K */
995 }
996
997 /* ===========================================================================
998 * Read a new buffer from the current input stream, update the adler32
999 * and total number of bytes read. All deflate() input goes through
1000 * this function so some applications may wish to modify it to avoid
1001 * allocating a large strm->next_in buffer and copying from it.
1002 * (See also flush_pending()).
1003 */
read_buf(strm,buf,size)1004 local int read_buf(strm, buf, size)
1005 z_streamp strm;
1006 Bytef *buf;
1007 unsigned size;
1008 {
1009 unsigned len = strm->avail_in;
1010
1011 if (len > size) len = size;
1012 if (len == 0) return 0;
1013
1014 strm->avail_in -= len;
1015
1016 if (strm->state->wrap == 1) {
1017 strm->adler = adler32(strm->adler, strm->next_in, len);
1018 }
1019 #ifdef GZIP
1020 else if (strm->state->wrap == 2) {
1021 strm->adler = crc32(strm->adler, strm->next_in, len);
1022 }
1023 #endif
1024 zmemcpy(buf, strm->next_in, len);
1025 strm->next_in += len;
1026 strm->total_in += len;
1027
1028 return (int)len;
1029 }
1030
1031 /* ===========================================================================
1032 * Initialize the "longest match" routines for a new zlib stream
1033 */
lm_init(s)1034 local void lm_init (s)
1035 deflate_state *s;
1036 {
1037 s->window_size = (ulg)2L*s->w_size;
1038
1039 CLEAR_HASH(s);
1040
1041 /* Set the default configuration parameters:
1042 */
1043 s->max_lazy_match = configuration_table[s->level].max_lazy;
1044 s->good_match = configuration_table[s->level].good_length;
1045 s->nice_match = configuration_table[s->level].nice_length;
1046 s->max_chain_length = configuration_table[s->level].max_chain;
1047
1048 s->strstart = 0;
1049 s->block_start = 0L;
1050 s->lookahead = 0;
1051 s->match_length = s->prev_length = MIN_MATCH-1;
1052 s->match_available = 0;
1053 s->ins_h = 0;
1054 #ifndef FASTEST
1055 #ifdef ASMV
1056 match_init(); /* initialize the asm code */
1057 #endif
1058 #endif
1059 }
1060
1061 #ifndef FASTEST
1062 /* ===========================================================================
1063 * Set match_start to the longest match starting at the given string and
1064 * return its length. Matches shorter or equal to prev_length are discarded,
1065 * in which case the result is equal to prev_length and match_start is
1066 * garbage.
1067 * IN assertions: cur_match is the head of the hash chain for the current
1068 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1069 * OUT assertion: the match length is not greater than s->lookahead.
1070 */
1071 #ifndef ASMV
1072 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1073 * match.S. The code will be functionally equivalent.
1074 */
longest_match(s,cur_match)1075 local uInt longest_match(s, cur_match)
1076 deflate_state *s;
1077 IPos cur_match; /* current match */
1078 {
1079 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1080 register Bytef *scan = s->window + s->strstart; /* current string */
1081 register Bytef *match; /* matched string */
1082 register int len; /* length of current match */
1083 int best_len = s->prev_length; /* best match length so far */
1084 int nice_match = s->nice_match; /* stop if match long enough */
1085 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1086 s->strstart - (IPos)MAX_DIST(s) : NIL;
1087 /* Stop when cur_match becomes <= limit. To simplify the code,
1088 * we prevent matches with the string of window index 0.
1089 */
1090 Posf *prev = s->prev;
1091 uInt wmask = s->w_mask;
1092
1093 #ifdef UNALIGNED_OK
1094 /* Compare two bytes at a time. Note: this is not always beneficial.
1095 * Try with and without -DUNALIGNED_OK to check.
1096 */
1097 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1098 register ush scan_start = *(ushf*)scan;
1099 register ush scan_end = *(ushf*)(scan+best_len-1);
1100 #else
1101 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1102 register Byte scan_end1 = scan[best_len-1];
1103 register Byte scan_end = scan[best_len];
1104 #endif
1105
1106 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1107 * It is easy to get rid of this optimization if necessary.
1108 */
1109 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1110
1111 /* Do not waste too much time if we already have a good match: */
1112 if (s->prev_length >= s->good_match) {
1113 chain_length >>= 2;
1114 }
1115 /* Do not look for matches beyond the end of the input. This is necessary
1116 * to make deflate deterministic.
1117 */
1118 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1119
1120 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1121
1122 do {
1123 Assert(cur_match < s->strstart, "no future");
1124 match = s->window + cur_match;
1125
1126 /* Skip to next match if the match length cannot increase
1127 * or if the match length is less than 2. Note that the checks below
1128 * for insufficient lookahead only occur occasionally for performance
1129 * reasons. Therefore uninitialized memory will be accessed, and
1130 * conditional jumps will be made that depend on those values.
1131 * However the length of the match is limited to the lookahead, so
1132 * the output of deflate is not affected by the uninitialized values.
1133 */
1134 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1135 /* This code assumes sizeof(unsigned short) == 2. Do not use
1136 * UNALIGNED_OK if your compiler uses a different size.
1137 */
1138 if (*(ushf*)(match+best_len-1) != scan_end ||
1139 *(ushf*)match != scan_start) continue;
1140
1141 /* It is not necessary to compare scan[2] and match[2] since they are
1142 * always equal when the other bytes match, given that the hash keys
1143 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1144 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1145 * lookahead only every 4th comparison; the 128th check will be made
1146 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1147 * necessary to put more guard bytes at the end of the window, or
1148 * to check more often for insufficient lookahead.
1149 */
1150 Assert(scan[2] == match[2], "scan[2]?");
1151 scan++, match++;
1152 do {
1153 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1154 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1155 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1156 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1157 scan < strend);
1158 /* The funny "do {}" generates better code on most compilers */
1159
1160 /* Here, scan <= window+strstart+257 */
1161 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1162 if (*scan == *match) scan++;
1163
1164 len = (MAX_MATCH - 1) - (int)(strend-scan);
1165 scan = strend - (MAX_MATCH-1);
1166
1167 #else /* UNALIGNED_OK */
1168
1169 if (match[best_len] != scan_end ||
1170 match[best_len-1] != scan_end1 ||
1171 *match != *scan ||
1172 *++match != scan[1]) continue;
1173
1174 /* The check at best_len-1 can be removed because it will be made
1175 * again later. (This heuristic is not always a win.)
1176 * It is not necessary to compare scan[2] and match[2] since they
1177 * are always equal when the other bytes match, given that
1178 * the hash keys are equal and that HASH_BITS >= 8.
1179 */
1180 scan += 2, match++;
1181 Assert(*scan == *match, "match[2]?");
1182
1183 /* We check for insufficient lookahead only every 8th comparison;
1184 * the 256th check will be made at strstart+258.
1185 */
1186 do {
1187 } while (*++scan == *++match && *++scan == *++match &&
1188 *++scan == *++match && *++scan == *++match &&
1189 *++scan == *++match && *++scan == *++match &&
1190 *++scan == *++match && *++scan == *++match &&
1191 scan < strend);
1192
1193 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1194
1195 len = MAX_MATCH - (int)(strend - scan);
1196 scan = strend - MAX_MATCH;
1197
1198 #endif /* UNALIGNED_OK */
1199
1200 if (len > best_len) {
1201 s->match_start = cur_match;
1202 best_len = len;
1203 if (len >= nice_match) break;
1204 #ifdef UNALIGNED_OK
1205 scan_end = *(ushf*)(scan+best_len-1);
1206 #else
1207 scan_end1 = scan[best_len-1];
1208 scan_end = scan[best_len];
1209 #endif
1210 }
1211 } while ((cur_match = prev[cur_match & wmask]) > limit
1212 && --chain_length != 0);
1213
1214 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1215 return s->lookahead;
1216 }
1217 #endif /* ASMV */
1218
1219 #else /* FASTEST */
1220
1221 /* ---------------------------------------------------------------------------
1222 * Optimized version for FASTEST only
1223 */
longest_match(s,cur_match)1224 local uInt longest_match(s, cur_match)
1225 deflate_state *s;
1226 IPos cur_match; /* current match */
1227 {
1228 register Bytef *scan = s->window + s->strstart; /* current string */
1229 register Bytef *match; /* matched string */
1230 register int len; /* length of current match */
1231 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1232
1233 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1234 * It is easy to get rid of this optimization if necessary.
1235 */
1236 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1237
1238 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1239
1240 Assert(cur_match < s->strstart, "no future");
1241
1242 match = s->window + cur_match;
1243
1244 /* Return failure if the match length is less than 2:
1245 */
1246 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1247
1248 /* The check at best_len-1 can be removed because it will be made
1249 * again later. (This heuristic is not always a win.)
1250 * It is not necessary to compare scan[2] and match[2] since they
1251 * are always equal when the other bytes match, given that
1252 * the hash keys are equal and that HASH_BITS >= 8.
1253 */
1254 scan += 2, match += 2;
1255 Assert(*scan == *match, "match[2]?");
1256
1257 /* We check for insufficient lookahead only every 8th comparison;
1258 * the 256th check will be made at strstart+258.
1259 */
1260 do {
1261 } while (*++scan == *++match && *++scan == *++match &&
1262 *++scan == *++match && *++scan == *++match &&
1263 *++scan == *++match && *++scan == *++match &&
1264 *++scan == *++match && *++scan == *++match &&
1265 scan < strend);
1266
1267 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1268
1269 len = MAX_MATCH - (int)(strend - scan);
1270
1271 if (len < MIN_MATCH) return MIN_MATCH - 1;
1272
1273 s->match_start = cur_match;
1274 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1275 }
1276
1277 #endif /* FASTEST */
1278
1279 #ifdef DEBUG
1280 /* ===========================================================================
1281 * Check that the match at match_start is indeed a match.
1282 */
check_match(s,start,match,length)1283 local void check_match(s, start, match, length)
1284 deflate_state *s;
1285 IPos start, match;
1286 int length;
1287 {
1288 /* check that the match is indeed a match */
1289 if (zmemcmp(s->window + match,
1290 s->window + start, length) != EQUAL) {
1291 fprintf(stderr, " start %u, match %u, length %d\n",
1292 start, match, length);
1293 do {
1294 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1295 } while (--length != 0);
1296 z_error("invalid match");
1297 }
1298 if (z_verbose > 1) {
1299 fprintf(stderr,"\\[%d,%d]", start-match, length);
1300 do { putc(s->window[start++], stderr); } while (--length != 0);
1301 }
1302 }
1303 #else
1304 # define check_match(s, start, match, length)
1305 #endif /* DEBUG */
1306
1307 /* ===========================================================================
1308 * Fill the window when the lookahead becomes insufficient.
1309 * Updates strstart and lookahead.
1310 *
1311 * IN assertion: lookahead < MIN_LOOKAHEAD
1312 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1313 * At least one byte has been read, or avail_in == 0; reads are
1314 * performed for at least two bytes (required for the zip translate_eol
1315 * option -- not supported here).
1316 */
fill_window(s)1317 local void fill_window(s)
1318 deflate_state *s;
1319 {
1320 register unsigned n, m;
1321 register Posf *p;
1322 unsigned more; /* Amount of free space at the end of the window. */
1323 uInt wsize = s->w_size;
1324
1325 do {
1326 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1327
1328 /* Deal with !@#$% 64K limit: */
1329 if (sizeof(int) <= 2) {
1330 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1331 more = wsize;
1332
1333 } else if (more == (unsigned)(-1)) {
1334 /* Very unlikely, but possible on 16 bit machine if
1335 * strstart == 0 && lookahead == 1 (input done a byte at time)
1336 */
1337 more--;
1338 }
1339 }
1340
1341 /* If the window is almost full and there is insufficient lookahead,
1342 * move the upper half to the lower one to make room in the upper half.
1343 */
1344 if (s->strstart >= wsize+MAX_DIST(s)) {
1345
1346 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1347 s->match_start -= wsize;
1348 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1349 s->block_start -= (long) wsize;
1350
1351 /* Slide the hash table (could be avoided with 32 bit values
1352 at the expense of memory usage). We slide even when level == 0
1353 to keep the hash table consistent if we switch back to level > 0
1354 later. (Using level 0 permanently is not an optimal usage of
1355 zlib, so we don't care about this pathological case.)
1356 */
1357 n = s->hash_size;
1358 p = &s->head[n];
1359 do {
1360 m = *--p;
1361 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1362 } while (--n);
1363
1364 n = wsize;
1365 #ifndef FASTEST
1366 p = &s->prev[n];
1367 do {
1368 m = *--p;
1369 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1370 /* If n is not on any hash chain, prev[n] is garbage but
1371 * its value will never be used.
1372 */
1373 } while (--n);
1374 #endif
1375 more += wsize;
1376 }
1377 if (s->strm->avail_in == 0) return;
1378
1379 /* If there was no sliding:
1380 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1381 * more == window_size - lookahead - strstart
1382 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1383 * => more >= window_size - 2*WSIZE + 2
1384 * In the BIG_MEM or MMAP case (not yet supported),
1385 * window_size == input_size + MIN_LOOKAHEAD &&
1386 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1387 * Otherwise, window_size == 2*WSIZE so more >= 2.
1388 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1389 */
1390 Assert(more >= 2, "more < 2");
1391
1392 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1393 s->lookahead += n;
1394
1395 /* Initialize the hash value now that we have some input: */
1396 if (s->lookahead >= MIN_MATCH) {
1397 s->ins_h = s->window[s->strstart];
1398 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1399 #if MIN_MATCH != 3
1400 Call UPDATE_HASH() MIN_MATCH-3 more times
1401 #endif
1402 }
1403 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1404 * but this is not important since only literal bytes will be emitted.
1405 */
1406
1407 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1408
1409 /* If the WIN_INIT bytes after the end of the current data have never been
1410 * written, then zero those bytes in order to avoid memory check reports of
1411 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1412 * the longest match routines. Update the high water mark for the next
1413 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1414 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1415 */
1416 if (s->high_water < s->window_size) {
1417 ulg curr = s->strstart + (ulg)(s->lookahead);
1418 ulg init;
1419
1420 if (s->high_water < curr) {
1421 /* Previous high water mark below current data -- zero WIN_INIT
1422 * bytes or up to end of window, whichever is less.
1423 */
1424 init = s->window_size - curr;
1425 if (init > WIN_INIT)
1426 init = WIN_INIT;
1427 zmemzero(s->window + curr, (unsigned)init);
1428 s->high_water = curr + init;
1429 }
1430 else if (s->high_water < (ulg)curr + WIN_INIT) {
1431 /* High water mark at or above current data, but below current data
1432 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1433 * to end of window, whichever is less.
1434 */
1435 init = (ulg)curr + WIN_INIT - s->high_water;
1436 if (init > s->window_size - s->high_water)
1437 init = s->window_size - s->high_water;
1438 zmemzero(s->window + s->high_water, (unsigned)init);
1439 s->high_water += init;
1440 }
1441 }
1442 }
1443
1444 /* ===========================================================================
1445 * Flush the current block, with given end-of-file flag.
1446 * IN assertion: strstart is set to the end of the current match.
1447 */
1448 #define FLUSH_BLOCK_ONLY(s, last) { \
1449 _tr_flush_block(s, (s->block_start >= 0L ? \
1450 (charf *)&s->window[(unsigned)s->block_start] : \
1451 (charf *)Z_NULL), \
1452 (ulg)((long)s->strstart - s->block_start), \
1453 (last)); \
1454 s->block_start = s->strstart; \
1455 flush_pending(s->strm); \
1456 Tracev((stderr,"[FLUSH]")); \
1457 }
1458
1459 /* Same but force premature exit if necessary. */
1460 #define FLUSH_BLOCK(s, last) { \
1461 FLUSH_BLOCK_ONLY(s, last); \
1462 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1463 }
1464
1465 /* ===========================================================================
1466 * Copy without compression as much as possible from the input stream, return
1467 * the current block state.
1468 * This function does not insert new strings in the dictionary since
1469 * uncompressible data is probably not useful. This function is used
1470 * only for the level=0 compression option.
1471 * NOTE: this function should be optimized to avoid extra copying from
1472 * window to pending_buf.
1473 */
deflate_stored(s,flush)1474 local block_state deflate_stored(s, flush)
1475 deflate_state *s;
1476 int flush;
1477 {
1478 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1479 * to pending_buf_size, and each stored block has a 5 byte header:
1480 */
1481 ulg max_block_size = 0xffff;
1482 ulg max_start;
1483
1484 if (max_block_size > s->pending_buf_size - 5) {
1485 max_block_size = s->pending_buf_size - 5;
1486 }
1487
1488 /* Copy as much as possible from input to output: */
1489 for (;;) {
1490 /* Fill the window as much as possible: */
1491 if (s->lookahead <= 1) {
1492
1493 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1494 s->block_start >= (long)s->w_size, "slide too late");
1495
1496 fill_window(s);
1497 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1498
1499 if (s->lookahead == 0) break; /* flush the current block */
1500 }
1501 Assert(s->block_start >= 0L, "block gone");
1502
1503 s->strstart += s->lookahead;
1504 s->lookahead = 0;
1505
1506 /* Emit a stored block if pending_buf will be full: */
1507 max_start = s->block_start + max_block_size;
1508 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1509 /* strstart == 0 is possible when wraparound on 16-bit machine */
1510 s->lookahead = (uInt)(s->strstart - max_start);
1511 s->strstart = (uInt)max_start;
1512 FLUSH_BLOCK(s, 0);
1513 }
1514 /* Flush if we may have to slide, otherwise block_start may become
1515 * negative and the data will be gone:
1516 */
1517 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1518 FLUSH_BLOCK(s, 0);
1519 }
1520 }
1521 FLUSH_BLOCK(s, flush == Z_FINISH);
1522 return flush == Z_FINISH ? finish_done : block_done;
1523 }
1524
1525 /* ===========================================================================
1526 * Compress as much as possible from the input stream, return the current
1527 * block state.
1528 * This function does not perform lazy evaluation of matches and inserts
1529 * new strings in the dictionary only for unmatched strings or for short
1530 * matches. It is used only for the fast compression options.
1531 */
deflate_fast(s,flush)1532 local block_state deflate_fast(s, flush)
1533 deflate_state *s;
1534 int flush;
1535 {
1536 IPos hash_head; /* head of the hash chain */
1537 int bflush; /* set if current block must be flushed */
1538
1539 for (;;) {
1540 /* Make sure that we always have enough lookahead, except
1541 * at the end of the input file. We need MAX_MATCH bytes
1542 * for the next match, plus MIN_MATCH bytes to insert the
1543 * string following the next match.
1544 */
1545 if (s->lookahead < MIN_LOOKAHEAD) {
1546 fill_window(s);
1547 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1548 return need_more;
1549 }
1550 if (s->lookahead == 0) break; /* flush the current block */
1551 }
1552
1553 /* Insert the string window[strstart .. strstart+2] in the
1554 * dictionary, and set hash_head to the head of the hash chain:
1555 */
1556 hash_head = NIL;
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 * At this point we have always match_length < MIN_MATCH
1563 */
1564 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1565 /* To simplify the code, we prevent matches with the string
1566 * of window index 0 (in particular we have to avoid a match
1567 * of the string with itself at the start of the input file).
1568 */
1569 s->match_length = longest_match (s, hash_head);
1570 /* longest_match() sets match_start */
1571 }
1572 if (s->match_length >= MIN_MATCH) {
1573 check_match(s, s->strstart, s->match_start, s->match_length);
1574
1575 _tr_tally_dist(s, s->strstart - s->match_start,
1576 s->match_length - MIN_MATCH, bflush);
1577
1578 s->lookahead -= s->match_length;
1579
1580 /* Insert new strings in the hash table only if the match length
1581 * is not too large. This saves time but degrades compression.
1582 */
1583 #ifndef FASTEST
1584 if (s->match_length <= s->max_insert_length &&
1585 s->lookahead >= MIN_MATCH) {
1586 s->match_length--; /* string at strstart already in table */
1587 do {
1588 s->strstart++;
1589 INSERT_STRING(s, s->strstart, hash_head);
1590 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1591 * always MIN_MATCH bytes ahead.
1592 */
1593 } while (--s->match_length != 0);
1594 s->strstart++;
1595 } else
1596 #endif
1597 {
1598 s->strstart += s->match_length;
1599 s->match_length = 0;
1600 s->ins_h = s->window[s->strstart];
1601 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1602 #if MIN_MATCH != 3
1603 Call UPDATE_HASH() MIN_MATCH-3 more times
1604 #endif
1605 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1606 * matter since it will be recomputed at next deflate call.
1607 */
1608 }
1609 } else {
1610 /* No match, output a literal byte */
1611 Tracevv((stderr,"%c", s->window[s->strstart]));
1612 _tr_tally_lit (s, s->window[s->strstart], bflush);
1613 s->lookahead--;
1614 s->strstart++;
1615 }
1616 if (bflush) FLUSH_BLOCK(s, 0);
1617 }
1618 FLUSH_BLOCK(s, flush == Z_FINISH);
1619 return flush == Z_FINISH ? finish_done : block_done;
1620 }
1621
1622 #ifndef FASTEST
1623 /* ===========================================================================
1624 * Same as above, but achieves better compression. We use a lazy
1625 * evaluation for matches: a match is finally adopted only if there is
1626 * no better match at the next window position.
1627 */
deflate_slow(s,flush)1628 local block_state deflate_slow(s, flush)
1629 deflate_state *s;
1630 int flush;
1631 {
1632 IPos hash_head; /* head of hash chain */
1633 int bflush; /* set if current block must be flushed */
1634
1635 /* Process the input block. */
1636 for (;;) {
1637 /* Make sure that we always have enough lookahead, except
1638 * at the end of the input file. We need MAX_MATCH bytes
1639 * for the next match, plus MIN_MATCH bytes to insert the
1640 * string following the next match.
1641 */
1642 if (s->lookahead < MIN_LOOKAHEAD) {
1643 fill_window(s);
1644 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1645 return need_more;
1646 }
1647 if (s->lookahead == 0) break; /* flush the current block */
1648 }
1649
1650 /* Insert the string window[strstart .. strstart+2] in the
1651 * dictionary, and set hash_head to the head of the hash chain:
1652 */
1653 hash_head = NIL;
1654 if (s->lookahead >= MIN_MATCH) {
1655 INSERT_STRING(s, s->strstart, hash_head);
1656 }
1657
1658 /* Find the longest match, discarding those <= prev_length.
1659 */
1660 s->prev_length = s->match_length, s->prev_match = s->match_start;
1661 s->match_length = MIN_MATCH-1;
1662
1663 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1664 s->strstart - hash_head <= MAX_DIST(s)) {
1665 /* To simplify the code, we prevent matches with the string
1666 * of window index 0 (in particular we have to avoid a match
1667 * of the string with itself at the start of the input file).
1668 */
1669 s->match_length = longest_match (s, hash_head);
1670 /* longest_match() sets match_start */
1671
1672 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1673 #if TOO_FAR <= 32767
1674 || (s->match_length == MIN_MATCH &&
1675 s->strstart - s->match_start > TOO_FAR)
1676 #endif
1677 )) {
1678
1679 /* If prev_match is also MIN_MATCH, match_start is garbage
1680 * but we will ignore the current match anyway.
1681 */
1682 s->match_length = MIN_MATCH-1;
1683 }
1684 }
1685 /* If there was a match at the previous step and the current
1686 * match is not better, output the previous match:
1687 */
1688 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1689 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1690 /* Do not insert strings in hash table beyond this. */
1691
1692 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1693
1694 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1695 s->prev_length - MIN_MATCH, bflush);
1696
1697 /* Insert in hash table all strings up to the end of the match.
1698 * strstart-1 and strstart are already inserted. If there is not
1699 * enough lookahead, the last two strings are not inserted in
1700 * the hash table.
1701 */
1702 s->lookahead -= s->prev_length-1;
1703 s->prev_length -= 2;
1704 do {
1705 if (++s->strstart <= max_insert) {
1706 INSERT_STRING(s, s->strstart, hash_head);
1707 }
1708 } while (--s->prev_length != 0);
1709 s->match_available = 0;
1710 s->match_length = MIN_MATCH-1;
1711 s->strstart++;
1712
1713 if (bflush) FLUSH_BLOCK(s, 0);
1714
1715 } else if (s->match_available) {
1716 /* If there was no match at the previous position, output a
1717 * single literal. If there was a match but the current match
1718 * is longer, truncate the previous match to a single literal.
1719 */
1720 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1721 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1722 if (bflush) {
1723 FLUSH_BLOCK_ONLY(s, 0);
1724 }
1725 s->strstart++;
1726 s->lookahead--;
1727 if (s->strm->avail_out == 0) return need_more;
1728 } else {
1729 /* There is no previous match to compare with, wait for
1730 * the next step to decide.
1731 */
1732 s->match_available = 1;
1733 s->strstart++;
1734 s->lookahead--;
1735 }
1736 }
1737 Assert (flush != Z_NO_FLUSH, "no flush?");
1738 if (s->match_available) {
1739 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1740 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1741 s->match_available = 0;
1742 }
1743 FLUSH_BLOCK(s, flush == Z_FINISH);
1744 return flush == Z_FINISH ? finish_done : block_done;
1745 }
1746 #endif /* FASTEST */
1747
1748 /* ===========================================================================
1749 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1750 * one. Do not maintain a hash table. (It will be regenerated if this run of
1751 * deflate switches away from Z_RLE.)
1752 */
deflate_rle(s,flush)1753 local block_state deflate_rle(s, flush)
1754 deflate_state *s;
1755 int flush;
1756 {
1757 int bflush; /* set if current block must be flushed */
1758 uInt prev; /* byte at distance one to match */
1759 Bytef *scan, *strend; /* scan goes up to strend for length of run */
1760
1761 for (;;) {
1762 /* Make sure that we always have enough lookahead, except
1763 * at the end of the input file. We need MAX_MATCH bytes
1764 * for the longest encodable run.
1765 */
1766 if (s->lookahead < MAX_MATCH) {
1767 fill_window(s);
1768 if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {
1769 return need_more;
1770 }
1771 if (s->lookahead == 0) break; /* flush the current block */
1772 }
1773
1774 /* See how many times the previous byte repeats */
1775 s->match_length = 0;
1776 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
1777 scan = s->window + s->strstart - 1;
1778 prev = *scan;
1779 if (prev == *++scan && prev == *++scan && prev == *++scan) {
1780 strend = s->window + s->strstart + MAX_MATCH;
1781 do {
1782 } while (prev == *++scan && prev == *++scan &&
1783 prev == *++scan && prev == *++scan &&
1784 prev == *++scan && prev == *++scan &&
1785 prev == *++scan && prev == *++scan &&
1786 scan < strend);
1787 s->match_length = MAX_MATCH - (int)(strend - scan);
1788 if (s->match_length > s->lookahead)
1789 s->match_length = s->lookahead;
1790 }
1791 }
1792
1793 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1794 if (s->match_length >= MIN_MATCH) {
1795 check_match(s, s->strstart, s->strstart - 1, s->match_length);
1796
1797 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
1798
1799 s->lookahead -= s->match_length;
1800 s->strstart += s->match_length;
1801 s->match_length = 0;
1802 } else {
1803 /* No match, output a literal byte */
1804 Tracevv((stderr,"%c", s->window[s->strstart]));
1805 _tr_tally_lit (s, s->window[s->strstart], bflush);
1806 s->lookahead--;
1807 s->strstart++;
1808 }
1809 if (bflush) FLUSH_BLOCK(s, 0);
1810 }
1811 FLUSH_BLOCK(s, flush == Z_FINISH);
1812 return flush == Z_FINISH ? finish_done : block_done;
1813 }
1814
1815 /* ===========================================================================
1816 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
1817 * (It will be regenerated if this run of deflate switches away from Huffman.)
1818 */
deflate_huff(s,flush)1819 local block_state deflate_huff(s, flush)
1820 deflate_state *s;
1821 int flush;
1822 {
1823 int bflush; /* set if current block must be flushed */
1824
1825 for (;;) {
1826 /* Make sure that we have a literal to write. */
1827 if (s->lookahead == 0) {
1828 fill_window(s);
1829 if (s->lookahead == 0) {
1830 if (flush == Z_NO_FLUSH)
1831 return need_more;
1832 break; /* flush the current block */
1833 }
1834 }
1835
1836 /* Output a literal byte */
1837 s->match_length = 0;
1838 Tracevv((stderr,"%c", s->window[s->strstart]));
1839 _tr_tally_lit (s, s->window[s->strstart], bflush);
1840 s->lookahead--;
1841 s->strstart++;
1842 if (bflush) FLUSH_BLOCK(s, 0);
1843 }
1844 FLUSH_BLOCK(s, flush == Z_FINISH);
1845 return flush == Z_FINISH ? finish_done : block_done;
1846 }
1847