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 /* @(#) $Id$ */
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 memset(s->prev, 0, s->w_size*sizeof(Pos)); /* XXX Zero out for emscripten, to avoid SAFE_HEAP warnings */
284 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
285
286 s->high_water = 0; /* nothing written to s->window yet */
287
288 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
289
290 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
291 s->pending_buf = (uchf *) overlay;
292 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
293
294 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
295 s->pending_buf == Z_NULL) {
296 s->status = FINISH_STATE;
297 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
298 deflateEnd (strm);
299 return Z_MEM_ERROR;
300 }
301 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
302 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
303
304 s->level = level;
305 s->strategy = strategy;
306 s->method = (Byte)method;
307
308 return deflateReset(strm);
309 }
310
311 /* ========================================================================= */
deflateSetDictionary(strm,dictionary,dictLength)312 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
313 z_streamp strm;
314 const Bytef *dictionary;
315 uInt dictLength;
316 {
317 deflate_state *s;
318 uInt length = dictLength;
319 uInt n;
320 IPos hash_head = 0;
321
322 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
323 strm->state->wrap == 2 ||
324 (strm->state->wrap == 1 && strm->state->status != INIT_STATE))
325 return Z_STREAM_ERROR;
326
327 s = strm->state;
328 if (s->wrap)
329 strm->adler = adler32(strm->adler, dictionary, dictLength);
330
331 if (length < MIN_MATCH) return Z_OK;
332 if (length > s->w_size) {
333 length = s->w_size;
334 dictionary += dictLength - length; /* use the tail of the dictionary */
335 }
336 zmemcpy(s->window, dictionary, length);
337 s->strstart = length;
338 s->block_start = (long)length;
339
340 /* Insert all strings in the hash table (except for the last two bytes).
341 * s->lookahead stays null, so s->ins_h will be recomputed at the next
342 * call of fill_window.
343 */
344 s->ins_h = s->window[0];
345 UPDATE_HASH(s, s->ins_h, s->window[1]);
346 for (n = 0; n <= length - MIN_MATCH; n++) {
347 INSERT_STRING(s, n, hash_head);
348 }
349 if (hash_head) hash_head = 0; /* to make compiler happy */
350 return Z_OK;
351 }
352
353 /* ========================================================================= */
deflateReset(strm)354 int ZEXPORT deflateReset (strm)
355 z_streamp strm;
356 {
357 deflate_state *s;
358
359 if (strm == Z_NULL || strm->state == Z_NULL ||
360 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
361 return Z_STREAM_ERROR;
362 }
363
364 strm->total_in = strm->total_out = 0;
365 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
366 strm->data_type = Z_UNKNOWN;
367
368 s = (deflate_state *)strm->state;
369 s->pending = 0;
370 s->pending_out = s->pending_buf;
371
372 if (s->wrap < 0) {
373 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
374 }
375 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
376 strm->adler =
377 #ifdef GZIP
378 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
379 #endif
380 adler32(0L, Z_NULL, 0);
381 s->last_flush = Z_NO_FLUSH;
382
383 _tr_init(s);
384 lm_init(s);
385
386 return Z_OK;
387 }
388
389 /* ========================================================================= */
deflateSetHeader(strm,head)390 int ZEXPORT deflateSetHeader (strm, head)
391 z_streamp strm;
392 gz_headerp head;
393 {
394 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
395 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
396 strm->state->gzhead = head;
397 return Z_OK;
398 }
399
400 /* ========================================================================= */
deflatePrime(strm,bits,value)401 int ZEXPORT deflatePrime (strm, bits, value)
402 z_streamp strm;
403 int bits;
404 int value;
405 {
406 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
407 strm->state->bi_valid = bits;
408 strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));
409 return Z_OK;
410 }
411
412 /* ========================================================================= */
deflateParams(strm,level,strategy)413 int ZEXPORT deflateParams(strm, level, strategy)
414 z_streamp strm;
415 int level;
416 int strategy;
417 {
418 deflate_state *s;
419 compress_func func;
420 int err = Z_OK;
421
422 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
423 s = strm->state;
424
425 #ifdef FASTEST
426 if (level != 0) level = 1;
427 #else
428 if (level == Z_DEFAULT_COMPRESSION) level = 6;
429 #endif
430 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
431 return Z_STREAM_ERROR;
432 }
433 func = configuration_table[s->level].func;
434
435 if ((strategy != s->strategy || func != configuration_table[level].func) &&
436 strm->total_in != 0) {
437 /* Flush the last buffer: */
438 err = deflate(strm, Z_BLOCK);
439 }
440 if (s->level != level) {
441 s->level = level;
442 s->max_lazy_match = configuration_table[level].max_lazy;
443 s->good_match = configuration_table[level].good_length;
444 s->nice_match = configuration_table[level].nice_length;
445 s->max_chain_length = configuration_table[level].max_chain;
446 }
447 s->strategy = strategy;
448 return err;
449 }
450
451 /* ========================================================================= */
deflateTune(strm,good_length,max_lazy,nice_length,max_chain)452 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
453 z_streamp strm;
454 int good_length;
455 int max_lazy;
456 int nice_length;
457 int max_chain;
458 {
459 deflate_state *s;
460
461 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
462 s = strm->state;
463 s->good_match = good_length;
464 s->max_lazy_match = max_lazy;
465 s->nice_match = nice_length;
466 s->max_chain_length = max_chain;
467 return Z_OK;
468 }
469
470 /* =========================================================================
471 * For the default windowBits of 15 and memLevel of 8, this function returns
472 * a close to exact, as well as small, upper bound on the compressed size.
473 * They are coded as constants here for a reason--if the #define's are
474 * changed, then this function needs to be changed as well. The return
475 * value for 15 and 8 only works for those exact settings.
476 *
477 * For any setting other than those defaults for windowBits and memLevel,
478 * the value returned is a conservative worst case for the maximum expansion
479 * resulting from using fixed blocks instead of stored blocks, which deflate
480 * can emit on compressed data for some combinations of the parameters.
481 *
482 * This function could be more sophisticated to provide closer upper bounds for
483 * every combination of windowBits and memLevel. But even the conservative
484 * upper bound of about 14% expansion does not seem onerous for output buffer
485 * allocation.
486 */
deflateBound(strm,sourceLen)487 uLong ZEXPORT deflateBound(strm, sourceLen)
488 z_streamp strm;
489 uLong sourceLen;
490 {
491 deflate_state *s;
492 uLong complen, wraplen;
493 Bytef *str;
494
495 /* conservative upper bound for compressed data */
496 complen = sourceLen +
497 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
498
499 /* if can't get parameters, return conservative bound plus zlib wrapper */
500 if (strm == Z_NULL || strm->state == Z_NULL)
501 return complen + 6;
502
503 /* compute wrapper length */
504 s = strm->state;
505 switch (s->wrap) {
506 case 0: /* raw deflate */
507 wraplen = 0;
508 break;
509 case 1: /* zlib wrapper */
510 wraplen = 6 + (s->strstart ? 4 : 0);
511 break;
512 case 2: /* gzip wrapper */
513 wraplen = 18;
514 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
515 if (s->gzhead->extra != Z_NULL)
516 wraplen += 2 + s->gzhead->extra_len;
517 str = s->gzhead->name;
518 if (str != Z_NULL)
519 do {
520 wraplen++;
521 } while (*str++);
522 str = s->gzhead->comment;
523 if (str != Z_NULL)
524 do {
525 wraplen++;
526 } while (*str++);
527 if (s->gzhead->hcrc)
528 wraplen += 2;
529 }
530 break;
531 default: /* for compiler happiness */
532 wraplen = 6;
533 }
534
535 /* if not default parameters, return conservative bound */
536 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
537 return complen + wraplen;
538
539 /* default settings: return tight bound for that case */
540 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
541 (sourceLen >> 25) + 13 - 6 + wraplen;
542 }
543
544 /* =========================================================================
545 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
546 * IN assertion: the stream state is correct and there is enough room in
547 * pending_buf.
548 */
putShortMSB(s,b)549 local void putShortMSB (s, b)
550 deflate_state *s;
551 uInt b;
552 {
553 put_byte(s, (Byte)(b >> 8));
554 put_byte(s, (Byte)(b & 0xff));
555 }
556
557 /* =========================================================================
558 * Flush as much pending output as possible. All deflate() output goes
559 * through this function so some applications may wish to modify it
560 * to avoid allocating a large strm->next_out buffer and copying into it.
561 * (See also read_buf()).
562 */
flush_pending(strm)563 local void flush_pending(strm)
564 z_streamp strm;
565 {
566 unsigned len = strm->state->pending;
567
568 if (len > strm->avail_out) len = strm->avail_out;
569 if (len == 0) return;
570
571 zmemcpy(strm->next_out, strm->state->pending_out, len);
572 strm->next_out += len;
573 strm->state->pending_out += len;
574 strm->total_out += len;
575 strm->avail_out -= len;
576 strm->state->pending -= len;
577 if (strm->state->pending == 0) {
578 strm->state->pending_out = strm->state->pending_buf;
579 }
580 }
581
582 /* ========================================================================= */
deflate(strm,flush)583 int ZEXPORT deflate (strm, flush)
584 z_streamp strm;
585 int flush;
586 {
587 int old_flush; /* value of flush param for previous deflate call */
588 deflate_state *s;
589
590 if (strm == Z_NULL || strm->state == Z_NULL ||
591 flush > Z_BLOCK || flush < 0) {
592 return Z_STREAM_ERROR;
593 }
594 s = strm->state;
595
596 if (strm->next_out == Z_NULL ||
597 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
598 (s->status == FINISH_STATE && flush != Z_FINISH)) {
599 ERR_RETURN(strm, Z_STREAM_ERROR);
600 }
601 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
602
603 s->strm = strm; /* just in case */
604 old_flush = s->last_flush;
605 s->last_flush = flush;
606
607 /* Write the header */
608 if (s->status == INIT_STATE) {
609 #ifdef GZIP
610 if (s->wrap == 2) {
611 strm->adler = crc32(0L, Z_NULL, 0);
612 put_byte(s, 31);
613 put_byte(s, 139);
614 put_byte(s, 8);
615 if (s->gzhead == Z_NULL) {
616 put_byte(s, 0);
617 put_byte(s, 0);
618 put_byte(s, 0);
619 put_byte(s, 0);
620 put_byte(s, 0);
621 put_byte(s, s->level == 9 ? 2 :
622 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
623 4 : 0));
624 put_byte(s, OS_CODE);
625 s->status = BUSY_STATE;
626 }
627 else {
628 put_byte(s, (s->gzhead->text ? 1 : 0) +
629 (s->gzhead->hcrc ? 2 : 0) +
630 (s->gzhead->extra == Z_NULL ? 0 : 4) +
631 (s->gzhead->name == Z_NULL ? 0 : 8) +
632 (s->gzhead->comment == Z_NULL ? 0 : 16)
633 );
634 put_byte(s, (Byte)(s->gzhead->time & 0xff));
635 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
636 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
637 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
638 put_byte(s, s->level == 9 ? 2 :
639 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
640 4 : 0));
641 put_byte(s, s->gzhead->os & 0xff);
642 if (s->gzhead->extra != Z_NULL) {
643 put_byte(s, s->gzhead->extra_len & 0xff);
644 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
645 }
646 if (s->gzhead->hcrc)
647 strm->adler = crc32(strm->adler, s->pending_buf,
648 s->pending);
649 s->gzindex = 0;
650 s->status = EXTRA_STATE;
651 }
652 }
653 else
654 #endif
655 {
656 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
657 uInt level_flags;
658
659 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
660 level_flags = 0;
661 else if (s->level < 6)
662 level_flags = 1;
663 else if (s->level == 6)
664 level_flags = 2;
665 else
666 level_flags = 3;
667 header |= (level_flags << 6);
668 if (s->strstart != 0) header |= PRESET_DICT;
669 header += 31 - (header % 31);
670
671 s->status = BUSY_STATE;
672 putShortMSB(s, header);
673
674 /* Save the adler32 of the preset dictionary: */
675 if (s->strstart != 0) {
676 putShortMSB(s, (uInt)(strm->adler >> 16));
677 putShortMSB(s, (uInt)(strm->adler & 0xffff));
678 }
679 strm->adler = adler32(0L, Z_NULL, 0);
680 }
681 }
682 #ifdef GZIP
683 if (s->status == EXTRA_STATE) {
684 if (s->gzhead->extra != Z_NULL) {
685 uInt beg = s->pending; /* start of bytes to update crc */
686
687 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
688 if (s->pending == s->pending_buf_size) {
689 if (s->gzhead->hcrc && s->pending > beg)
690 strm->adler = crc32(strm->adler, s->pending_buf + beg,
691 s->pending - beg);
692 flush_pending(strm);
693 beg = s->pending;
694 if (s->pending == s->pending_buf_size)
695 break;
696 }
697 put_byte(s, s->gzhead->extra[s->gzindex]);
698 s->gzindex++;
699 }
700 if (s->gzhead->hcrc && s->pending > beg)
701 strm->adler = crc32(strm->adler, s->pending_buf + beg,
702 s->pending - beg);
703 if (s->gzindex == s->gzhead->extra_len) {
704 s->gzindex = 0;
705 s->status = NAME_STATE;
706 }
707 }
708 else
709 s->status = NAME_STATE;
710 }
711 if (s->status == NAME_STATE) {
712 if (s->gzhead->name != Z_NULL) {
713 uInt beg = s->pending; /* start of bytes to update crc */
714 int val;
715
716 do {
717 if (s->pending == s->pending_buf_size) {
718 if (s->gzhead->hcrc && s->pending > beg)
719 strm->adler = crc32(strm->adler, s->pending_buf + beg,
720 s->pending - beg);
721 flush_pending(strm);
722 beg = s->pending;
723 if (s->pending == s->pending_buf_size) {
724 val = 1;
725 break;
726 }
727 }
728 val = s->gzhead->name[s->gzindex++];
729 put_byte(s, val);
730 } while (val != 0);
731 if (s->gzhead->hcrc && s->pending > beg)
732 strm->adler = crc32(strm->adler, s->pending_buf + beg,
733 s->pending - beg);
734 if (val == 0) {
735 s->gzindex = 0;
736 s->status = COMMENT_STATE;
737 }
738 }
739 else
740 s->status = COMMENT_STATE;
741 }
742 if (s->status == COMMENT_STATE) {
743 if (s->gzhead->comment != Z_NULL) {
744 uInt beg = s->pending; /* start of bytes to update crc */
745 int val;
746
747 do {
748 if (s->pending == s->pending_buf_size) {
749 if (s->gzhead->hcrc && s->pending > beg)
750 strm->adler = crc32(strm->adler, s->pending_buf + beg,
751 s->pending - beg);
752 flush_pending(strm);
753 beg = s->pending;
754 if (s->pending == s->pending_buf_size) {
755 val = 1;
756 break;
757 }
758 }
759 val = s->gzhead->comment[s->gzindex++];
760 put_byte(s, val);
761 } while (val != 0);
762 if (s->gzhead->hcrc && s->pending > beg)
763 strm->adler = crc32(strm->adler, s->pending_buf + beg,
764 s->pending - beg);
765 if (val == 0)
766 s->status = HCRC_STATE;
767 }
768 else
769 s->status = HCRC_STATE;
770 }
771 if (s->status == HCRC_STATE) {
772 if (s->gzhead->hcrc) {
773 if (s->pending + 2 > s->pending_buf_size)
774 flush_pending(strm);
775 if (s->pending + 2 <= s->pending_buf_size) {
776 put_byte(s, (Byte)(strm->adler & 0xff));
777 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
778 strm->adler = crc32(0L, Z_NULL, 0);
779 s->status = BUSY_STATE;
780 }
781 }
782 else
783 s->status = BUSY_STATE;
784 }
785 #endif
786
787 /* Flush as much pending output as possible */
788 if (s->pending != 0) {
789 flush_pending(strm);
790 if (strm->avail_out == 0) {
791 /* Since avail_out is 0, deflate will be called again with
792 * more output space, but possibly with both pending and
793 * avail_in equal to zero. There won't be anything to do,
794 * but this is not an error situation so make sure we
795 * return OK instead of BUF_ERROR at next call of deflate:
796 */
797 s->last_flush = -1;
798 return Z_OK;
799 }
800
801 /* Make sure there is something to do and avoid duplicate consecutive
802 * flushes. For repeated and useless calls with Z_FINISH, we keep
803 * returning Z_STREAM_END instead of Z_BUF_ERROR.
804 */
805 } else if (strm->avail_in == 0 && flush <= old_flush &&
806 flush != Z_FINISH) {
807 ERR_RETURN(strm, Z_BUF_ERROR);
808 }
809
810 /* User must not provide more input after the first FINISH: */
811 if (s->status == FINISH_STATE && strm->avail_in != 0) {
812 ERR_RETURN(strm, Z_BUF_ERROR);
813 }
814
815 /* Start a new block or continue the current one.
816 */
817 if (strm->avail_in != 0 || s->lookahead != 0 ||
818 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
819 block_state bstate;
820
821 bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
822 (s->strategy == Z_RLE ? deflate_rle(s, flush) :
823 (*(configuration_table[s->level].func))(s, flush));
824
825 if (bstate == finish_started || bstate == finish_done) {
826 s->status = FINISH_STATE;
827 }
828 if (bstate == need_more || bstate == finish_started) {
829 if (strm->avail_out == 0) {
830 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
831 }
832 return Z_OK;
833 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
834 * of deflate should use the same flush parameter to make sure
835 * that the flush is complete. So we don't have to output an
836 * empty block here, this will be done at next call. This also
837 * ensures that for a very small output buffer, we emit at most
838 * one empty block.
839 */
840 }
841 if (bstate == block_done) {
842 if (flush == Z_PARTIAL_FLUSH) {
843 _tr_align(s);
844 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
845 _tr_stored_block(s, (char*)0, 0L, 0);
846 /* For a full flush, this empty block will be recognized
847 * as a special marker by inflate_sync().
848 */
849 if (flush == Z_FULL_FLUSH) {
850 CLEAR_HASH(s); /* forget history */
851 if (s->lookahead == 0) {
852 s->strstart = 0;
853 s->block_start = 0L;
854 }
855 }
856 }
857 flush_pending(strm);
858 if (strm->avail_out == 0) {
859 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
860 return Z_OK;
861 }
862 }
863 }
864 Assert(strm->avail_out > 0, "bug2");
865
866 if (flush != Z_FINISH) return Z_OK;
867 if (s->wrap <= 0) return Z_STREAM_END;
868
869 /* Write the trailer */
870 #ifdef GZIP
871 if (s->wrap == 2) {
872 put_byte(s, (Byte)(strm->adler & 0xff));
873 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
874 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
875 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
876 put_byte(s, (Byte)(strm->total_in & 0xff));
877 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
878 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
879 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
880 }
881 else
882 #endif
883 {
884 putShortMSB(s, (uInt)(strm->adler >> 16));
885 putShortMSB(s, (uInt)(strm->adler & 0xffff));
886 }
887 flush_pending(strm);
888 /* If avail_out is zero, the application will call deflate again
889 * to flush the rest.
890 */
891 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
892 return s->pending != 0 ? Z_OK : Z_STREAM_END;
893 }
894
895 /* ========================================================================= */
deflateEnd(strm)896 int ZEXPORT deflateEnd (strm)
897 z_streamp strm;
898 {
899 int status;
900
901 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
902
903 status = strm->state->status;
904 if (status != INIT_STATE &&
905 status != EXTRA_STATE &&
906 status != NAME_STATE &&
907 status != COMMENT_STATE &&
908 status != HCRC_STATE &&
909 status != BUSY_STATE &&
910 status != FINISH_STATE) {
911 return Z_STREAM_ERROR;
912 }
913
914 /* Deallocate in reverse order of allocations: */
915 TRY_FREE(strm, strm->state->pending_buf);
916 TRY_FREE(strm, strm->state->head);
917 TRY_FREE(strm, strm->state->prev);
918 TRY_FREE(strm, strm->state->window);
919
920 ZFREE(strm, strm->state);
921 strm->state = Z_NULL;
922
923 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
924 }
925
926 /* =========================================================================
927 * Copy the source state to the destination state.
928 * To simplify the source, this is not supported for 16-bit MSDOS (which
929 * doesn't have enough memory anyway to duplicate compression states).
930 */
deflateCopy(dest,source)931 int ZEXPORT deflateCopy (dest, source)
932 z_streamp dest;
933 z_streamp source;
934 {
935 #ifdef MAXSEG_64K
936 return Z_STREAM_ERROR;
937 #else
938 deflate_state *ds;
939 deflate_state *ss;
940 ushf *overlay;
941
942
943 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
944 return Z_STREAM_ERROR;
945 }
946
947 ss = source->state;
948
949 zmemcpy(dest, source, sizeof(z_stream));
950
951 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
952 if (ds == Z_NULL) return Z_MEM_ERROR;
953 dest->state = (struct internal_state FAR *) ds;
954 zmemcpy(ds, ss, sizeof(deflate_state));
955 ds->strm = dest;
956
957 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
958 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
959 memset(ds->prev, 0, ds->w_size*sizeof(Pos)); /* XXX Zero out for emscripten, to avoid SAFE_HEAP warnings */
960 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
961 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
962 ds->pending_buf = (uchf *) overlay;
963
964 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
965 ds->pending_buf == Z_NULL) {
966 deflateEnd (dest);
967 return Z_MEM_ERROR;
968 }
969 /* following zmemcpy do not work for 16-bit MSDOS */
970 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
971 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
972 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
973 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
974
975 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
976 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
977 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
978
979 ds->l_desc.dyn_tree = ds->dyn_ltree;
980 ds->d_desc.dyn_tree = ds->dyn_dtree;
981 ds->bl_desc.dyn_tree = ds->bl_tree;
982
983 return Z_OK;
984 #endif /* MAXSEG_64K */
985 }
986
987 /* ===========================================================================
988 * Read a new buffer from the current input stream, update the adler32
989 * and total number of bytes read. All deflate() input goes through
990 * this function so some applications may wish to modify it to avoid
991 * allocating a large strm->next_in buffer and copying from it.
992 * (See also flush_pending()).
993 */
read_buf(strm,buf,size)994 local int read_buf(strm, buf, size)
995 z_streamp strm;
996 Bytef *buf;
997 unsigned size;
998 {
999 unsigned len = strm->avail_in;
1000
1001 if (len > size) len = size;
1002 if (len == 0) return 0;
1003
1004 strm->avail_in -= len;
1005
1006 if (strm->state->wrap == 1) {
1007 strm->adler = adler32(strm->adler, strm->next_in, len);
1008 }
1009 #ifdef GZIP
1010 else if (strm->state->wrap == 2) {
1011 strm->adler = crc32(strm->adler, strm->next_in, len);
1012 }
1013 #endif
1014 zmemcpy(buf, strm->next_in, len);
1015 strm->next_in += len;
1016 strm->total_in += len;
1017
1018 return (int)len;
1019 }
1020
1021 /* ===========================================================================
1022 * Initialize the "longest match" routines for a new zlib stream
1023 */
lm_init(s)1024 local void lm_init (s)
1025 deflate_state *s;
1026 {
1027 s->window_size = (ulg)2L*s->w_size;
1028
1029 CLEAR_HASH(s);
1030
1031 /* Set the default configuration parameters:
1032 */
1033 s->max_lazy_match = configuration_table[s->level].max_lazy;
1034 s->good_match = configuration_table[s->level].good_length;
1035 s->nice_match = configuration_table[s->level].nice_length;
1036 s->max_chain_length = configuration_table[s->level].max_chain;
1037
1038 s->strstart = 0;
1039 s->block_start = 0L;
1040 s->lookahead = 0;
1041 s->match_length = s->prev_length = MIN_MATCH-1;
1042 s->match_start = 0; /* XXX Emscripten: initialize to 0, to prevent a SAFE_HEAP notification */
1043 s->match_available = 0;
1044 s->ins_h = 0;
1045 #ifndef FASTEST
1046 #ifdef ASMV
1047 match_init(); /* initialize the asm code */
1048 #endif
1049 #endif
1050 }
1051
1052 #ifndef FASTEST
1053 /* ===========================================================================
1054 * Set match_start to the longest match starting at the given string and
1055 * return its length. Matches shorter or equal to prev_length are discarded,
1056 * in which case the result is equal to prev_length and match_start is
1057 * garbage.
1058 * IN assertions: cur_match is the head of the hash chain for the current
1059 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1060 * OUT assertion: the match length is not greater than s->lookahead.
1061 */
1062 #ifndef ASMV
1063 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1064 * match.S. The code will be functionally equivalent.
1065 */
longest_match(s,cur_match)1066 local uInt longest_match(s, cur_match)
1067 deflate_state *s;
1068 IPos cur_match; /* current match */
1069 {
1070 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1071 register Bytef *scan = s->window + s->strstart; /* current string */
1072 register Bytef *match; /* matched string */
1073 register int len; /* length of current match */
1074 int best_len = s->prev_length; /* best match length so far */
1075 int nice_match = s->nice_match; /* stop if match long enough */
1076 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1077 s->strstart - (IPos)MAX_DIST(s) : NIL;
1078 /* Stop when cur_match becomes <= limit. To simplify the code,
1079 * we prevent matches with the string of window index 0.
1080 */
1081 Posf *prev = s->prev;
1082 uInt wmask = s->w_mask;
1083
1084 #ifdef UNALIGNED_OK
1085 /* Compare two bytes at a time. Note: this is not always beneficial.
1086 * Try with and without -DUNALIGNED_OK to check.
1087 */
1088 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1089 register ush scan_start = *(ushf*)scan;
1090 register ush scan_end = *(ushf*)(scan+best_len-1);
1091 #else
1092 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1093 register Byte scan_end1 = scan[best_len-1];
1094 register Byte scan_end = scan[best_len];
1095 #endif
1096
1097 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1098 * It is easy to get rid of this optimization if necessary.
1099 */
1100 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1101
1102 /* Do not waste too much time if we already have a good match: */
1103 if (s->prev_length >= s->good_match) {
1104 chain_length >>= 2;
1105 }
1106 /* Do not look for matches beyond the end of the input. This is necessary
1107 * to make deflate deterministic.
1108 */
1109 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1110
1111 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1112
1113 do {
1114 Assert(cur_match < s->strstart, "no future");
1115 match = s->window + cur_match;
1116
1117 /* Skip to next match if the match length cannot increase
1118 * or if the match length is less than 2. Note that the checks below
1119 * for insufficient lookahead only occur occasionally for performance
1120 * reasons. Therefore uninitialized memory will be accessed, and
1121 * conditional jumps will be made that depend on those values.
1122 * However the length of the match is limited to the lookahead, so
1123 * the output of deflate is not affected by the uninitialized values.
1124 */
1125 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1126 /* This code assumes sizeof(unsigned short) == 2. Do not use
1127 * UNALIGNED_OK if your compiler uses a different size.
1128 */
1129 if (*(ushf*)(match+best_len-1) != scan_end ||
1130 *(ushf*)match != scan_start) continue;
1131
1132 /* It is not necessary to compare scan[2] and match[2] since they are
1133 * always equal when the other bytes match, given that the hash keys
1134 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1135 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1136 * lookahead only every 4th comparison; the 128th check will be made
1137 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1138 * necessary to put more guard bytes at the end of the window, or
1139 * to check more often for insufficient lookahead.
1140 */
1141 Assert(scan[2] == match[2], "scan[2]?");
1142 scan++, match++;
1143 do {
1144 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1145 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1146 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1147 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1148 scan < strend);
1149 /* The funny "do {}" generates better code on most compilers */
1150
1151 /* Here, scan <= window+strstart+257 */
1152 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1153 if (*scan == *match) scan++;
1154
1155 len = (MAX_MATCH - 1) - (int)(strend-scan);
1156 scan = strend - (MAX_MATCH-1);
1157
1158 #else /* UNALIGNED_OK */
1159
1160 if (match[best_len] != scan_end ||
1161 match[best_len-1] != scan_end1 ||
1162 *match != *scan ||
1163 *++match != scan[1]) continue;
1164
1165 /* The check at best_len-1 can be removed because it will be made
1166 * again later. (This heuristic is not always a win.)
1167 * It is not necessary to compare scan[2] and match[2] since they
1168 * are always equal when the other bytes match, given that
1169 * the hash keys are equal and that HASH_BITS >= 8.
1170 */
1171 scan += 2, match++;
1172 Assert(*scan == *match, "match[2]?");
1173
1174 /* We check for insufficient lookahead only every 8th comparison;
1175 * the 256th check will be made at strstart+258.
1176 */
1177 do {
1178 } while (*++scan == *++match && *++scan == *++match &&
1179 *++scan == *++match && *++scan == *++match &&
1180 *++scan == *++match && *++scan == *++match &&
1181 *++scan == *++match && *++scan == *++match &&
1182 scan < strend);
1183
1184 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1185
1186 len = MAX_MATCH - (int)(strend - scan);
1187 scan = strend - MAX_MATCH;
1188
1189 #endif /* UNALIGNED_OK */
1190
1191 if (len > best_len) {
1192 s->match_start = cur_match;
1193 best_len = len;
1194 if (len >= nice_match) break;
1195 #ifdef UNALIGNED_OK
1196 scan_end = *(ushf*)(scan+best_len-1);
1197 #else
1198 scan_end1 = scan[best_len-1];
1199 scan_end = scan[best_len];
1200 #endif
1201 }
1202 } while ((cur_match = prev[cur_match & wmask]) > limit
1203 && --chain_length != 0);
1204
1205 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1206 return s->lookahead;
1207 }
1208 #endif /* ASMV */
1209
1210 #else /* FASTEST */
1211
1212 /* ---------------------------------------------------------------------------
1213 * Optimized version for FASTEST only
1214 */
longest_match(s,cur_match)1215 local uInt longest_match(s, cur_match)
1216 deflate_state *s;
1217 IPos cur_match; /* current match */
1218 {
1219 register Bytef *scan = s->window + s->strstart; /* current string */
1220 register Bytef *match; /* matched string */
1221 register int len; /* length of current match */
1222 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1223
1224 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1225 * It is easy to get rid of this optimization if necessary.
1226 */
1227 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1228
1229 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1230
1231 Assert(cur_match < s->strstart, "no future");
1232
1233 match = s->window + cur_match;
1234
1235 /* Return failure if the match length is less than 2:
1236 */
1237 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1238
1239 /* The check at best_len-1 can be removed because it will be made
1240 * again later. (This heuristic is not always a win.)
1241 * It is not necessary to compare scan[2] and match[2] since they
1242 * are always equal when the other bytes match, given that
1243 * the hash keys are equal and that HASH_BITS >= 8.
1244 */
1245 scan += 2, match += 2;
1246 Assert(*scan == *match, "match[2]?");
1247
1248 /* We check for insufficient lookahead only every 8th comparison;
1249 * the 256th check will be made at strstart+258.
1250 */
1251 do {
1252 } while (*++scan == *++match && *++scan == *++match &&
1253 *++scan == *++match && *++scan == *++match &&
1254 *++scan == *++match && *++scan == *++match &&
1255 *++scan == *++match && *++scan == *++match &&
1256 scan < strend);
1257
1258 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1259
1260 len = MAX_MATCH - (int)(strend - scan);
1261
1262 if (len < MIN_MATCH) return MIN_MATCH - 1;
1263
1264 s->match_start = cur_match;
1265 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1266 }
1267
1268 #endif /* FASTEST */
1269
1270 #ifdef DEBUG
1271 /* ===========================================================================
1272 * Check that the match at match_start is indeed a match.
1273 */
check_match(s,start,match,length)1274 local void check_match(s, start, match, length)
1275 deflate_state *s;
1276 IPos start, match;
1277 int length;
1278 {
1279 /* check that the match is indeed a match */
1280 if (zmemcmp(s->window + match,
1281 s->window + start, length) != EQUAL) {
1282 fprintf(stderr, " start %u, match %u, length %d\n",
1283 start, match, length);
1284 do {
1285 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1286 } while (--length != 0);
1287 z_error("invalid match");
1288 }
1289 if (z_verbose > 1) {
1290 fprintf(stderr,"\\[%d,%d]", start-match, length);
1291 do { putc(s->window[start++], stderr); } while (--length != 0);
1292 }
1293 }
1294 #else
1295 # define check_match(s, start, match, length)
1296 #endif /* DEBUG */
1297
1298 /* ===========================================================================
1299 * Fill the window when the lookahead becomes insufficient.
1300 * Updates strstart and lookahead.
1301 *
1302 * IN assertion: lookahead < MIN_LOOKAHEAD
1303 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1304 * At least one byte has been read, or avail_in == 0; reads are
1305 * performed for at least two bytes (required for the zip translate_eol
1306 * option -- not supported here).
1307 */
fill_window(s)1308 local void fill_window(s)
1309 deflate_state *s;
1310 {
1311 register unsigned n, m;
1312 register Posf *p;
1313 unsigned more; /* Amount of free space at the end of the window. */
1314 uInt wsize = s->w_size;
1315
1316 do {
1317 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1318
1319 /* Deal with !@#$% 64K limit: */
1320 if (sizeof(int) <= 2) {
1321 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1322 more = wsize;
1323
1324 } else if (more == (unsigned)(-1)) {
1325 /* Very unlikely, but possible on 16 bit machine if
1326 * strstart == 0 && lookahead == 1 (input done a byte at time)
1327 */
1328 more--;
1329 }
1330 }
1331
1332 /* If the window is almost full and there is insufficient lookahead,
1333 * move the upper half to the lower one to make room in the upper half.
1334 */
1335 if (s->strstart >= wsize+MAX_DIST(s)) {
1336
1337 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1338 s->match_start -= wsize;
1339 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1340 s->block_start -= (long) wsize;
1341
1342 /* Slide the hash table (could be avoided with 32 bit values
1343 at the expense of memory usage). We slide even when level == 0
1344 to keep the hash table consistent if we switch back to level > 0
1345 later. (Using level 0 permanently is not an optimal usage of
1346 zlib, so we don't care about this pathological case.)
1347 */
1348 n = s->hash_size;
1349 p = &s->head[n];
1350 do {
1351 m = *--p;
1352 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1353 } while (--n);
1354
1355 n = wsize;
1356 #ifndef FASTEST
1357 p = &s->prev[n];
1358 do {
1359 m = *--p;
1360 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1361 /* If n is not on any hash chain, prev[n] is garbage but
1362 * its value will never be used.
1363 */
1364 } while (--n);
1365 #endif
1366 more += wsize;
1367 }
1368 if (s->strm->avail_in == 0) return;
1369
1370 /* If there was no sliding:
1371 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1372 * more == window_size - lookahead - strstart
1373 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1374 * => more >= window_size - 2*WSIZE + 2
1375 * In the BIG_MEM or MMAP case (not yet supported),
1376 * window_size == input_size + MIN_LOOKAHEAD &&
1377 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1378 * Otherwise, window_size == 2*WSIZE so more >= 2.
1379 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1380 */
1381 Assert(more >= 2, "more < 2");
1382
1383 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1384 s->lookahead += n;
1385
1386 /* Initialize the hash value now that we have some input: */
1387 if (s->lookahead >= MIN_MATCH) {
1388 s->ins_h = s->window[s->strstart];
1389 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1390 #if MIN_MATCH != 3
1391 Call UPDATE_HASH() MIN_MATCH-3 more times
1392 #endif
1393 }
1394 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1395 * but this is not important since only literal bytes will be emitted.
1396 */
1397
1398 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1399
1400 /* If the WIN_INIT bytes after the end of the current data have never been
1401 * written, then zero those bytes in order to avoid memory check reports of
1402 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1403 * the longest match routines. Update the high water mark for the next
1404 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1405 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1406 */
1407 if (s->high_water < s->window_size) {
1408 ulg curr = s->strstart + (ulg)(s->lookahead);
1409 ulg init;
1410
1411 if (s->high_water < curr) {
1412 /* Previous high water mark below current data -- zero WIN_INIT
1413 * bytes or up to end of window, whichever is less.
1414 */
1415 init = s->window_size - curr;
1416 if (init > WIN_INIT)
1417 init = WIN_INIT;
1418 zmemzero(s->window + curr, (unsigned)init);
1419 s->high_water = curr + init;
1420 }
1421 else if (s->high_water < (ulg)curr + WIN_INIT) {
1422 /* High water mark at or above current data, but below current data
1423 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1424 * to end of window, whichever is less.
1425 */
1426 init = (ulg)curr + WIN_INIT - s->high_water;
1427 if (init > s->window_size - s->high_water)
1428 init = s->window_size - s->high_water;
1429 zmemzero(s->window + s->high_water, (unsigned)init);
1430 s->high_water += init;
1431 }
1432 }
1433 }
1434
1435 /* ===========================================================================
1436 * Flush the current block, with given end-of-file flag.
1437 * IN assertion: strstart is set to the end of the current match.
1438 */
1439 #define FLUSH_BLOCK_ONLY(s, last) { \
1440 _tr_flush_block(s, (s->block_start >= 0L ? \
1441 (charf *)&s->window[(unsigned)s->block_start] : \
1442 (charf *)Z_NULL), \
1443 (ulg)((long)s->strstart - s->block_start), \
1444 (last)); \
1445 s->block_start = s->strstart; \
1446 flush_pending(s->strm); \
1447 Tracev((stderr,"[FLUSH]")); \
1448 }
1449
1450 /* Same but force premature exit if necessary. */
1451 #define FLUSH_BLOCK(s, last) { \
1452 FLUSH_BLOCK_ONLY(s, last); \
1453 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1454 }
1455
1456 /* ===========================================================================
1457 * Copy without compression as much as possible from the input stream, return
1458 * the current block state.
1459 * This function does not insert new strings in the dictionary since
1460 * uncompressible data is probably not useful. This function is used
1461 * only for the level=0 compression option.
1462 * NOTE: this function should be optimized to avoid extra copying from
1463 * window to pending_buf.
1464 */
deflate_stored(s,flush)1465 local block_state deflate_stored(s, flush)
1466 deflate_state *s;
1467 int flush;
1468 {
1469 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1470 * to pending_buf_size, and each stored block has a 5 byte header:
1471 */
1472 ulg max_block_size = 0xffff;
1473 ulg max_start;
1474
1475 if (max_block_size > s->pending_buf_size - 5) {
1476 max_block_size = s->pending_buf_size - 5;
1477 }
1478
1479 /* Copy as much as possible from input to output: */
1480 for (;;) {
1481 /* Fill the window as much as possible: */
1482 if (s->lookahead <= 1) {
1483
1484 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1485 s->block_start >= (long)s->w_size, "slide too late");
1486
1487 fill_window(s);
1488 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1489
1490 if (s->lookahead == 0) break; /* flush the current block */
1491 }
1492 Assert(s->block_start >= 0L, "block gone");
1493
1494 s->strstart += s->lookahead;
1495 s->lookahead = 0;
1496
1497 /* Emit a stored block if pending_buf will be full: */
1498 max_start = s->block_start + max_block_size;
1499 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1500 /* strstart == 0 is possible when wraparound on 16-bit machine */
1501 s->lookahead = (uInt)(s->strstart - max_start);
1502 s->strstart = (uInt)max_start;
1503 FLUSH_BLOCK(s, 0);
1504 }
1505 /* Flush if we may have to slide, otherwise block_start may become
1506 * negative and the data will be gone:
1507 */
1508 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1509 FLUSH_BLOCK(s, 0);
1510 }
1511 }
1512 FLUSH_BLOCK(s, flush == Z_FINISH);
1513 return flush == Z_FINISH ? finish_done : block_done;
1514 }
1515
1516 /* ===========================================================================
1517 * Compress as much as possible from the input stream, return the current
1518 * block state.
1519 * This function does not perform lazy evaluation of matches and inserts
1520 * new strings in the dictionary only for unmatched strings or for short
1521 * matches. It is used only for the fast compression options.
1522 */
deflate_fast(s,flush)1523 local block_state deflate_fast(s, flush)
1524 deflate_state *s;
1525 int flush;
1526 {
1527 IPos hash_head; /* head of the hash chain */
1528 int bflush; /* set if current block must be flushed */
1529
1530 for (;;) {
1531 /* Make sure that we always have enough lookahead, except
1532 * at the end of the input file. We need MAX_MATCH bytes
1533 * for the next match, plus MIN_MATCH bytes to insert the
1534 * string following the next match.
1535 */
1536 if (s->lookahead < MIN_LOOKAHEAD) {
1537 fill_window(s);
1538 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1539 return need_more;
1540 }
1541 if (s->lookahead == 0) break; /* flush the current block */
1542 }
1543
1544 /* Insert the string window[strstart .. strstart+2] in the
1545 * dictionary, and set hash_head to the head of the hash chain:
1546 */
1547 hash_head = NIL;
1548 if (s->lookahead >= MIN_MATCH) {
1549 INSERT_STRING(s, s->strstart, hash_head);
1550 }
1551
1552 /* Find the longest match, discarding those <= prev_length.
1553 * At this point we have always match_length < MIN_MATCH
1554 */
1555 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1556 /* To simplify the code, we prevent matches with the string
1557 * of window index 0 (in particular we have to avoid a match
1558 * of the string with itself at the start of the input file).
1559 */
1560 s->match_length = longest_match (s, hash_head);
1561 /* longest_match() sets match_start */
1562 }
1563 if (s->match_length >= MIN_MATCH) {
1564 check_match(s, s->strstart, s->match_start, s->match_length);
1565
1566 _tr_tally_dist(s, s->strstart - s->match_start,
1567 s->match_length - MIN_MATCH, bflush);
1568
1569 s->lookahead -= s->match_length;
1570
1571 /* Insert new strings in the hash table only if the match length
1572 * is not too large. This saves time but degrades compression.
1573 */
1574 #ifndef FASTEST
1575 if (s->match_length <= s->max_insert_length &&
1576 s->lookahead >= MIN_MATCH) {
1577 s->match_length--; /* string at strstart already in table */
1578 do {
1579 s->strstart++;
1580 INSERT_STRING(s, s->strstart, hash_head);
1581 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1582 * always MIN_MATCH bytes ahead.
1583 */
1584 } while (--s->match_length != 0);
1585 s->strstart++;
1586 } else
1587 #endif
1588 {
1589 s->strstart += s->match_length;
1590 s->match_length = 0;
1591 s->ins_h = s->window[s->strstart];
1592 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1593 #if MIN_MATCH != 3
1594 Call UPDATE_HASH() MIN_MATCH-3 more times
1595 #endif
1596 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1597 * matter since it will be recomputed at next deflate call.
1598 */
1599 }
1600 } else {
1601 /* No match, output a literal byte */
1602 Tracevv((stderr,"%c", s->window[s->strstart]));
1603 _tr_tally_lit (s, s->window[s->strstart], bflush);
1604 s->lookahead--;
1605 s->strstart++;
1606 }
1607 if (bflush) FLUSH_BLOCK(s, 0);
1608 }
1609 FLUSH_BLOCK(s, flush == Z_FINISH);
1610 return flush == Z_FINISH ? finish_done : block_done;
1611 }
1612
1613 #ifndef FASTEST
1614 /* ===========================================================================
1615 * Same as above, but achieves better compression. We use a lazy
1616 * evaluation for matches: a match is finally adopted only if there is
1617 * no better match at the next window position.
1618 */
deflate_slow(s,flush)1619 local block_state deflate_slow(s, flush)
1620 deflate_state *s;
1621 int flush;
1622 {
1623 IPos hash_head; /* head of hash chain */
1624 int bflush; /* set if current block must be flushed */
1625
1626 /* Process the input block. */
1627 for (;;) {
1628 /* Make sure that we always have enough lookahead, except
1629 * at the end of the input file. We need MAX_MATCH bytes
1630 * for the next match, plus MIN_MATCH bytes to insert the
1631 * string following the next match.
1632 */
1633 if (s->lookahead < MIN_LOOKAHEAD) {
1634 fill_window(s);
1635 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1636 return need_more;
1637 }
1638 if (s->lookahead == 0) break; /* flush the current block */
1639 }
1640
1641 /* Insert the string window[strstart .. strstart+2] in the
1642 * dictionary, and set hash_head to the head of the hash chain:
1643 */
1644 hash_head = NIL;
1645 if (s->lookahead >= MIN_MATCH) {
1646 INSERT_STRING(s, s->strstart, hash_head);
1647 }
1648
1649 /* Find the longest match, discarding those <= prev_length.
1650 */
1651 s->prev_length = s->match_length, s->prev_match = s->match_start;
1652 s->match_length = MIN_MATCH-1;
1653
1654 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1655 s->strstart - hash_head <= MAX_DIST(s)) {
1656 /* To simplify the code, we prevent matches with the string
1657 * of window index 0 (in particular we have to avoid a match
1658 * of the string with itself at the start of the input file).
1659 */
1660 s->match_length = longest_match (s, hash_head);
1661 /* longest_match() sets match_start */
1662
1663 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1664 #if TOO_FAR <= 32767
1665 || (s->match_length == MIN_MATCH &&
1666 s->strstart - s->match_start > TOO_FAR)
1667 #endif
1668 )) {
1669
1670 /* If prev_match is also MIN_MATCH, match_start is garbage
1671 * but we will ignore the current match anyway.
1672 */
1673 s->match_length = MIN_MATCH-1;
1674 }
1675 }
1676 /* If there was a match at the previous step and the current
1677 * match is not better, output the previous match:
1678 */
1679 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1680 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1681 /* Do not insert strings in hash table beyond this. */
1682
1683 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1684
1685 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1686 s->prev_length - MIN_MATCH, bflush);
1687
1688 /* Insert in hash table all strings up to the end of the match.
1689 * strstart-1 and strstart are already inserted. If there is not
1690 * enough lookahead, the last two strings are not inserted in
1691 * the hash table.
1692 */
1693 s->lookahead -= s->prev_length-1;
1694 s->prev_length -= 2;
1695 do {
1696 if (++s->strstart <= max_insert) {
1697 INSERT_STRING(s, s->strstart, hash_head);
1698 }
1699 } while (--s->prev_length != 0);
1700 s->match_available = 0;
1701 s->match_length = MIN_MATCH-1;
1702 s->strstart++;
1703
1704 if (bflush) FLUSH_BLOCK(s, 0);
1705
1706 } else if (s->match_available) {
1707 /* If there was no match at the previous position, output a
1708 * single literal. If there was a match but the current match
1709 * is longer, truncate the previous match to a single literal.
1710 */
1711 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1712 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1713 if (bflush) {
1714 FLUSH_BLOCK_ONLY(s, 0);
1715 }
1716 s->strstart++;
1717 s->lookahead--;
1718 if (s->strm->avail_out == 0) return need_more;
1719 } else {
1720 /* There is no previous match to compare with, wait for
1721 * the next step to decide.
1722 */
1723 s->match_available = 1;
1724 s->strstart++;
1725 s->lookahead--;
1726 }
1727 }
1728 Assert (flush != Z_NO_FLUSH, "no flush?");
1729 if (s->match_available) {
1730 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1731 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1732 s->match_available = 0;
1733 }
1734 FLUSH_BLOCK(s, flush == Z_FINISH);
1735 return flush == Z_FINISH ? finish_done : block_done;
1736 }
1737 #endif /* FASTEST */
1738
1739 /* ===========================================================================
1740 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1741 * one. Do not maintain a hash table. (It will be regenerated if this run of
1742 * deflate switches away from Z_RLE.)
1743 */
deflate_rle(s,flush)1744 local block_state deflate_rle(s, flush)
1745 deflate_state *s;
1746 int flush;
1747 {
1748 int bflush; /* set if current block must be flushed */
1749 uInt prev; /* byte at distance one to match */
1750 Bytef *scan, *strend; /* scan goes up to strend for length of run */
1751
1752 for (;;) {
1753 /* Make sure that we always have enough lookahead, except
1754 * at the end of the input file. We need MAX_MATCH bytes
1755 * for the longest encodable run.
1756 */
1757 if (s->lookahead < MAX_MATCH) {
1758 fill_window(s);
1759 if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {
1760 return need_more;
1761 }
1762 if (s->lookahead == 0) break; /* flush the current block */
1763 }
1764
1765 /* See how many times the previous byte repeats */
1766 s->match_length = 0;
1767 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
1768 scan = s->window + s->strstart - 1;
1769 prev = *scan;
1770 if (prev == *++scan && prev == *++scan && prev == *++scan) {
1771 strend = s->window + s->strstart + MAX_MATCH;
1772 do {
1773 } while (prev == *++scan && prev == *++scan &&
1774 prev == *++scan && prev == *++scan &&
1775 prev == *++scan && prev == *++scan &&
1776 prev == *++scan && prev == *++scan &&
1777 scan < strend);
1778 s->match_length = MAX_MATCH - (int)(strend - scan);
1779 if (s->match_length > s->lookahead)
1780 s->match_length = s->lookahead;
1781 }
1782 }
1783
1784 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1785 if (s->match_length >= MIN_MATCH) {
1786 check_match(s, s->strstart, s->strstart - 1, s->match_length);
1787
1788 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
1789
1790 s->lookahead -= s->match_length;
1791 s->strstart += s->match_length;
1792 s->match_length = 0;
1793 } else {
1794 /* No match, output a literal byte */
1795 Tracevv((stderr,"%c", s->window[s->strstart]));
1796 _tr_tally_lit (s, s->window[s->strstart], bflush);
1797 s->lookahead--;
1798 s->strstart++;
1799 }
1800 if (bflush) FLUSH_BLOCK(s, 0);
1801 }
1802 FLUSH_BLOCK(s, flush == Z_FINISH);
1803 return flush == Z_FINISH ? finish_done : block_done;
1804 }
1805
1806 /* ===========================================================================
1807 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
1808 * (It will be regenerated if this run of deflate switches away from Huffman.)
1809 */
deflate_huff(s,flush)1810 local block_state deflate_huff(s, flush)
1811 deflate_state *s;
1812 int flush;
1813 {
1814 int bflush; /* set if current block must be flushed */
1815
1816 for (;;) {
1817 /* Make sure that we have a literal to write. */
1818 if (s->lookahead == 0) {
1819 fill_window(s);
1820 if (s->lookahead == 0) {
1821 if (flush == Z_NO_FLUSH)
1822 return need_more;
1823 break; /* flush the current block */
1824 }
1825 }
1826
1827 /* Output a literal byte */
1828 s->match_length = 0;
1829 Tracevv((stderr,"%c", s->window[s->strstart]));
1830 _tr_tally_lit (s, s->window[s->strstart], bflush);
1831 s->lookahead--;
1832 s->strstart++;
1833 if (bflush) FLUSH_BLOCK(s, 0);
1834 }
1835 FLUSH_BLOCK(s, flush == Z_FINISH);
1836 return flush == Z_FINISH ? finish_done : block_done;
1837 }
1838