1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2005 Jean-loup Gailly.
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
5
6 /*
7 * ALGORITHM
8 *
9 * The "deflation" process depends on being able to identify portions
10 * of the input text which are identical to earlier input (within a
11 * sliding window trailing behind the input currently being processed).
12 *
13 * The most straightforward technique turns out to be the fastest for
14 * most input files: try all possible matches and select the longest.
15 * The key feature of this algorithm is that insertions into the string
16 * dictionary are very simple and thus fast, and deletions are avoided
17 * completely. Insertions are performed at each input character, whereas
18 * string matches are performed only when the previous match ends. So it
19 * is preferable to spend more time in matches to allow very fast string
20 * insertions and avoid deletions. The matching algorithm for small
21 * strings is inspired from that of Rabin & Karp. A brute force approach
22 * is used to find longer strings when a small match has been found.
23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
24 * (by Leonid Broukhis).
25 * A previous version of this file used a more sophisticated algorithm
26 * (by Fiala and Greene) which is guaranteed to run in linear amortized
27 * time, but has a larger average cost, uses more memory and is patented.
28 * However the F&G algorithm may be faster for some highly redundant
29 * files if the parameter max_chain_length (described below) is too large.
30 *
31 * ACKNOWLEDGEMENTS
32 *
33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
34 * I found it in 'freeze' written by Leonid Broukhis.
35 * Thanks to many people for bug reports and testing.
36 *
37 * REFERENCES
38 *
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40 * Available in http://www.ietf.org/rfc/rfc1951.txt
41 *
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
44 *
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
47 *
48 */
49
50 /* @(#) $Id: deflate.c,v 1.1 2007/06/07 17:54:37 jules_rms Exp $ */
51
52 #include "deflate.h"
53
54 const char deflate_copyright[] =
55 " deflate 1.2.3 Copyright 1995-2005 Jean-loup Gailly ";
56 /*
57 If you use the zlib library in a product, an acknowledgment is welcome
58 in the documentation of your product. If for some reason you cannot
59 include such an acknowledgment, I would appreciate that you keep this
60 copyright string in the executable of your product.
61 */
62
63 /* ===========================================================================
64 * Function prototypes.
65 */
66 typedef enum {
67 need_more, /* block not completed, need more input or more output */
68 block_done, /* block flush performed */
69 finish_started, /* finish started, need only more output at next deflate */
70 finish_done /* finish done, accept no more input or output */
71 } block_state;
72
73 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
74 /* Compression function. Returns the block state after the call. */
75
76 local void fill_window OF((deflate_state *s));
77 local block_state deflate_stored OF((deflate_state *s, int flush));
78 local block_state deflate_fast OF((deflate_state *s, int flush));
79 #ifndef FASTEST
80 local block_state deflate_slow OF((deflate_state *s, int flush));
81 #endif
82 local void lm_init OF((deflate_state *s));
83 local void putShortMSB OF((deflate_state *s, uInt b));
84 local void flush_pending OF((z_streamp strm));
85 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
86 #ifndef FASTEST
87 #ifdef ASMV
88 void match_init OF((void)); /* asm code initialization */
89 uInt longest_match OF((deflate_state *s, IPos cur_match));
90 #else
91 local uInt longest_match OF((deflate_state *s, IPos cur_match));
92 #endif
93 #endif
94 local uInt longest_match_fast OF((deflate_state *s, IPos cur_match));
95
96 #ifdef DEBUG
97 local void check_match OF((deflate_state *s, IPos start, IPos match,
98 int length));
99 #endif
100
101 /* ===========================================================================
102 * Local data
103 */
104
105 #define NIL 0
106 /* Tail of hash chains */
107
108 #ifndef TOO_FAR
109 # define TOO_FAR 4096
110 #endif
111 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
112
113 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
114 /* Minimum amount of lookahead, except at the end of the input file.
115 * See deflate.c for comments about the MIN_MATCH+1.
116 */
117
118 /* Values for max_lazy_match, good_match and max_chain_length, depending on
119 * the desired pack level (0..9). The values given below have been tuned to
120 * exclude worst case performance for pathological files. Better values may be
121 * found for specific files.
122 */
123 typedef struct config_s {
124 ush good_length; /* reduce lazy search above this match length */
125 ush max_lazy; /* do not perform lazy search above this match length */
126 ush nice_length; /* quit search above this match length */
127 ush max_chain;
128 compress_func func;
129 } config;
130
131 #ifdef FASTEST
132 local const config configuration_table[2] = {
133 /* good lazy nice chain */
134 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
135 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
136 #else
137 local const config configuration_table[10] = {
138 /* good lazy nice chain */
139 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
140 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
141 /* 2 */ {4, 5, 16, 8, deflate_fast},
142 /* 3 */ {4, 6, 32, 32, deflate_fast},
143
144 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
145 /* 5 */ {8, 16, 32, 32, deflate_slow},
146 /* 6 */ {8, 16, 128, 128, deflate_slow},
147 /* 7 */ {8, 32, 128, 256, deflate_slow},
148 /* 8 */ {32, 128, 258, 1024, deflate_slow},
149 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
150 #endif
151
152 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
153 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
154 * meaning.
155 */
156
157 #define EQUAL 0
158 /* result of memcmp for equal strings */
159
160 #ifndef NO_DUMMY_DECL
161 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
162 #endif
163
164 /* ===========================================================================
165 * Update a hash value with the given input byte
166 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
167 * input characters, so that a running hash key can be computed from the
168 * previous key instead of complete recalculation each time.
169 */
170 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
171
172
173 /* ===========================================================================
174 * Insert string str in the dictionary and set match_head to the previous head
175 * of the hash chain (the most recent string with same hash key). Return
176 * the previous length of the hash chain.
177 * If this file is compiled with -DFASTEST, the compression level is forced
178 * to 1, and no hash chains are maintained.
179 * IN assertion: all calls to to INSERT_STRING are made with consecutive
180 * input characters and the first MIN_MATCH bytes of str are valid
181 * (except for the last MIN_MATCH-1 bytes of the input file).
182 */
183 #ifdef FASTEST
184 #define INSERT_STRING(s, str, match_head) \
185 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
186 match_head = s->head[s->ins_h], \
187 s->head[s->ins_h] = (Pos)(str))
188 #else
189 #define INSERT_STRING(s, str, match_head) \
190 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
191 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
192 s->head[s->ins_h] = (Pos)(str))
193 #endif
194
195 /* ===========================================================================
196 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
197 * prev[] will be initialized on the fly.
198 */
199 #define CLEAR_HASH(s) \
200 s->head[s->hash_size-1] = NIL; \
201 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
202
203 /* ========================================================================= */
deflateInit_(z_streamp strm,int level,const char * version,int stream_size)204 int ZEXPORT deflateInit_(z_streamp strm, int level, const char *version, int stream_size)
205 {
206 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
207 Z_DEFAULT_STRATEGY, version, stream_size);
208 /* To do: ignore strm->next_in if we use it as window */
209 }
210
211 /* ========================================================================= */
deflateInit2_(z_streamp strm,int level,int method,int windowBits,int memLevel,int strategy,const char * version,int stream_size)212 int ZEXPORT deflateInit2_ (z_streamp strm, int level, int method, int windowBits, int memLevel, int strategy, const char *version, int stream_size)
213 {
214 deflate_state *s;
215 int wrap = 1;
216 static const char my_version[] = ZLIB_VERSION;
217
218 ushf *overlay;
219 /* We overlay pending_buf and d_buf+l_buf. This works since the average
220 * output size for (length,distance) codes is <= 24 bits.
221 */
222
223 if (version == Z_NULL || version[0] != my_version[0] ||
224 stream_size != sizeof(z_stream)) {
225 return Z_VERSION_ERROR;
226 }
227 if (strm == Z_NULL) return Z_STREAM_ERROR;
228
229 strm->msg = Z_NULL;
230 if (strm->zalloc == (alloc_func)0) {
231 strm->zalloc = zcalloc;
232 strm->opaque = (voidpf)0;
233 }
234 if (strm->zfree == (free_func)0) strm->zfree = zcfree;
235
236 #ifdef FASTEST
237 if (level != 0) level = 1;
238 #else
239 if (level == Z_DEFAULT_COMPRESSION) level = 6;
240 #endif
241
242 if (windowBits < 0) { /* suppress zlib wrapper */
243 wrap = 0;
244 windowBits = -windowBits;
245 }
246 #ifdef GZIP
247 else if (windowBits > 15) {
248 wrap = 2; /* write gzip wrapper instead */
249 windowBits -= 16;
250 }
251 #endif
252 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
253 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
254 strategy < 0 || strategy > Z_FIXED) {
255 return Z_STREAM_ERROR;
256 }
257 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
258 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
259 if (s == Z_NULL) return Z_MEM_ERROR;
260 strm->state = (struct internal_state FAR *)s;
261 s->strm = strm;
262
263 s->wrap = wrap;
264 s->gzhead = Z_NULL;
265 s->w_bits = windowBits;
266 s->w_size = 1 << s->w_bits;
267 s->w_mask = s->w_size - 1;
268
269 s->hash_bits = memLevel + 7;
270 s->hash_size = 1 << s->hash_bits;
271 s->hash_mask = s->hash_size - 1;
272 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
273
274 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
275 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
276 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
277
278 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
279
280 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
281 s->pending_buf = (uchf *) overlay;
282 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
283
284 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
285 s->pending_buf == Z_NULL) {
286 s->status = FINISH_STATE;
287 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
288 deflateEnd (strm);
289 return Z_MEM_ERROR;
290 }
291 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
292 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
293
294 s->level = level;
295 s->strategy = strategy;
296 s->method = (Byte)method;
297
298 return deflateReset(strm);
299 }
300
301 /* ========================================================================= */
deflateSetDictionary(z_streamp strm,const Bytef * dictionary,uInt dictLength)302 int ZEXPORT deflateSetDictionary (z_streamp strm, const Bytef *dictionary, uInt dictLength)
303 {
304 deflate_state *s;
305 uInt length = dictLength;
306 uInt n;
307 IPos hash_head = 0;
308
309 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
310 strm->state->wrap == 2 ||
311 (strm->state->wrap == 1 && strm->state->status != INIT_STATE))
312 return Z_STREAM_ERROR;
313
314 s = strm->state;
315 if (s->wrap)
316 strm->adler = adler32(strm->adler, dictionary, dictLength);
317
318 if (length < MIN_MATCH) return Z_OK;
319 if (length > MAX_DIST(s)) {
320 length = MAX_DIST(s);
321 dictionary += dictLength - length; /* use the tail of the dictionary */
322 }
323 zmemcpy(s->window, dictionary, length);
324 s->strstart = length;
325 s->block_start = (long)length;
326
327 /* Insert all strings in the hash table (except for the last two bytes).
328 * s->lookahead stays null, so s->ins_h will be recomputed at the next
329 * call of fill_window.
330 */
331 s->ins_h = s->window[0];
332 UPDATE_HASH(s, s->ins_h, s->window[1]);
333 for (n = 0; n <= length - MIN_MATCH; n++) {
334 INSERT_STRING(s, n, hash_head);
335 }
336
337 (void) hash_head; /* to make compiler happy */
338 return Z_OK;
339 }
340
341 /* ========================================================================= */
deflateReset(z_streamp strm)342 int ZEXPORT deflateReset (z_streamp strm)
343 {
344 deflate_state *s;
345
346 if (strm == Z_NULL || strm->state == Z_NULL ||
347 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
348 return Z_STREAM_ERROR;
349 }
350
351 strm->total_in = strm->total_out = 0;
352 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
353 strm->data_type = Z_UNKNOWN;
354
355 s = (deflate_state *)strm->state;
356 s->pending = 0;
357 s->pending_out = s->pending_buf;
358
359 if (s->wrap < 0) {
360 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
361 }
362 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
363 strm->adler =
364 #ifdef GZIP
365 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
366 #endif
367 adler32(0L, Z_NULL, 0);
368 s->last_flush = Z_NO_FLUSH;
369
370 _tr_init(s);
371 lm_init(s);
372
373 return Z_OK;
374 }
375
376 /* ========================================================================= */
deflateSetHeader(z_streamp strm,gz_headerp head)377 int ZEXPORT deflateSetHeader (z_streamp strm, gz_headerp head)
378 {
379 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
380 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
381 strm->state->gzhead = head;
382 return Z_OK;
383 }
384
385 /* ========================================================================= */
deflatePrime(z_streamp strm,int bits,int value)386 int ZEXPORT deflatePrime (z_streamp strm, int bits, int value)
387 {
388 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
389 strm->state->bi_valid = bits;
390 strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));
391 return Z_OK;
392 }
393
394 /* ========================================================================= */
deflateParams(z_streamp strm,int level,int strategy)395 int ZEXPORT deflateParams (z_streamp strm, int level, int strategy)
396 {
397 deflate_state *s;
398 compress_func func;
399 int err = Z_OK;
400
401 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
402 s = strm->state;
403
404 #ifdef FASTEST
405 if (level != 0) level = 1;
406 #else
407 if (level == Z_DEFAULT_COMPRESSION) level = 6;
408 #endif
409 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
410 return Z_STREAM_ERROR;
411 }
412 func = configuration_table[s->level].func;
413
414 if (func != configuration_table[level].func && strm->total_in != 0) {
415 /* Flush the last buffer: */
416 err = deflate(strm, Z_PARTIAL_FLUSH);
417 }
418 if (s->level != level) {
419 s->level = level;
420 s->max_lazy_match = configuration_table[level].max_lazy;
421 s->good_match = configuration_table[level].good_length;
422 s->nice_match = configuration_table[level].nice_length;
423 s->max_chain_length = configuration_table[level].max_chain;
424 }
425 s->strategy = strategy;
426 return err;
427 }
428
429 /* ========================================================================= */
deflateTune(z_streamp strm,int good_length,int max_lazy,int nice_length,int max_chain)430 int ZEXPORT deflateTune (z_streamp strm, int good_length, int max_lazy, int nice_length, int max_chain)
431 {
432 deflate_state *s;
433
434 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
435 s = strm->state;
436 s->good_match = good_length;
437 s->max_lazy_match = max_lazy;
438 s->nice_match = nice_length;
439 s->max_chain_length = max_chain;
440 return Z_OK;
441 }
442
443 /* =========================================================================
444 * For the default windowBits of 15 and memLevel of 8, this function returns
445 * a close to exact, as well as small, upper bound on the compressed size.
446 * They are coded as constants here for a reason--if the #define's are
447 * changed, then this function needs to be changed as well. The return
448 * value for 15 and 8 only works for those exact settings.
449 *
450 * For any setting other than those defaults for windowBits and memLevel,
451 * the value returned is a conservative worst case for the maximum expansion
452 * resulting from using fixed blocks instead of stored blocks, which deflate
453 * can emit on compressed data for some combinations of the parameters.
454 *
455 * This function could be more sophisticated to provide closer upper bounds
456 * for every combination of windowBits and memLevel, as well as wrap.
457 * But even the conservative upper bound of about 14% expansion does not
458 * seem onerous for output buffer allocation.
459 */
deflateBound(z_streamp strm,uLong sourceLen)460 uLong ZEXPORT deflateBound (z_streamp strm, uLong sourceLen)
461 {
462 deflate_state *s;
463 uLong destLen;
464
465 /* conservative upper bound */
466 destLen = sourceLen +
467 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 11;
468
469 /* if can't get parameters, return conservative bound */
470 if (strm == Z_NULL || strm->state == Z_NULL)
471 return destLen;
472
473 /* if not default parameters, return conservative bound */
474 s = strm->state;
475 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
476 return destLen;
477
478 /* default settings: return tight bound for that case */
479 return compressBound(sourceLen);
480 }
481
482 /* =========================================================================
483 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
484 * IN assertion: the stream state is correct and there is enough room in
485 * pending_buf.
486 */
putShortMSB(deflate_state * s,uInt b)487 local void putShortMSB (deflate_state *s, uInt b)
488 {
489 put_byte(s, (Byte)(b >> 8));
490 put_byte(s, (Byte)(b & 0xff));
491 }
492
493 /* =========================================================================
494 * Flush as much pending output as possible. All deflate() output goes
495 * through this function so some applications may wish to modify it
496 * to avoid allocating a large strm->next_out buffer and copying into it.
497 * (See also read_buf()).
498 */
flush_pending(z_streamp strm)499 local void flush_pending (z_streamp strm)
500 {
501 unsigned len = strm->state->pending;
502
503 if (len > strm->avail_out) len = strm->avail_out;
504 if (len == 0) return;
505
506 zmemcpy(strm->next_out, strm->state->pending_out, len);
507 strm->next_out += len;
508 strm->state->pending_out += len;
509 strm->total_out += len;
510 strm->avail_out -= len;
511 strm->state->pending -= len;
512 if (strm->state->pending == 0) {
513 strm->state->pending_out = strm->state->pending_buf;
514 }
515 }
516
517 /* ========================================================================= */
deflate(z_streamp strm,int flush)518 int ZEXPORT deflate (z_streamp strm, int flush)
519 {
520 int old_flush; /* value of flush param for previous deflate call */
521 deflate_state *s;
522
523 if (strm == Z_NULL || strm->state == Z_NULL ||
524 flush > Z_FINISH || flush < 0) {
525 return Z_STREAM_ERROR;
526 }
527 s = strm->state;
528
529 if (strm->next_out == Z_NULL ||
530 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
531 (s->status == FINISH_STATE && flush != Z_FINISH)) {
532 ERR_RETURN(strm, Z_STREAM_ERROR);
533 }
534 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
535
536 s->strm = strm; /* just in case */
537 old_flush = s->last_flush;
538 s->last_flush = flush;
539
540 /* Write the header */
541 if (s->status == INIT_STATE) {
542 #ifdef GZIP
543 if (s->wrap == 2) {
544 strm->adler = crc32(0L, Z_NULL, 0);
545 put_byte(s, 31);
546 put_byte(s, 139);
547 put_byte(s, 8);
548 if (s->gzhead == NULL) {
549 put_byte(s, 0);
550 put_byte(s, 0);
551 put_byte(s, 0);
552 put_byte(s, 0);
553 put_byte(s, 0);
554 put_byte(s, s->level == 9 ? 2 :
555 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
556 4 : 0));
557 put_byte(s, OS_CODE);
558 s->status = BUSY_STATE;
559 }
560 else {
561 put_byte(s, (s->gzhead->text ? 1 : 0) +
562 (s->gzhead->hcrc ? 2 : 0) +
563 (s->gzhead->extra == Z_NULL ? 0 : 4) +
564 (s->gzhead->name == Z_NULL ? 0 : 8) +
565 (s->gzhead->comment == Z_NULL ? 0 : 16)
566 );
567 put_byte(s, (Byte)(s->gzhead->time & 0xff));
568 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
569 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
570 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
571 put_byte(s, s->level == 9 ? 2 :
572 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
573 4 : 0));
574 put_byte(s, s->gzhead->os & 0xff);
575 if (s->gzhead->extra != NULL) {
576 put_byte(s, s->gzhead->extra_len & 0xff);
577 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
578 }
579 if (s->gzhead->hcrc)
580 strm->adler = crc32(strm->adler, s->pending_buf,
581 s->pending);
582 s->gzindex = 0;
583 s->status = EXTRA_STATE;
584 }
585 }
586 else
587 #endif
588 {
589 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
590 uInt level_flags;
591
592 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
593 level_flags = 0;
594 else if (s->level < 6)
595 level_flags = 1;
596 else if (s->level == 6)
597 level_flags = 2;
598 else
599 level_flags = 3;
600 header |= (level_flags << 6);
601 if (s->strstart != 0) header |= PRESET_DICT;
602 header += 31 - (header % 31);
603
604 s->status = BUSY_STATE;
605 putShortMSB(s, header);
606
607 /* Save the adler32 of the preset dictionary: */
608 if (s->strstart != 0) {
609 putShortMSB(s, (uInt)(strm->adler >> 16));
610 putShortMSB(s, (uInt)(strm->adler & 0xffff));
611 }
612 strm->adler = adler32(0L, Z_NULL, 0);
613 }
614 }
615 #ifdef GZIP
616 if (s->status == EXTRA_STATE) {
617 if (s->gzhead->extra != NULL) {
618 uInt beg = s->pending; /* start of bytes to update crc */
619
620 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
621 if (s->pending == s->pending_buf_size) {
622 if (s->gzhead->hcrc && s->pending > beg)
623 strm->adler = crc32(strm->adler, s->pending_buf + beg,
624 s->pending - beg);
625 flush_pending(strm);
626 beg = s->pending;
627 if (s->pending == s->pending_buf_size)
628 break;
629 }
630 put_byte(s, s->gzhead->extra[s->gzindex]);
631 s->gzindex++;
632 }
633 if (s->gzhead->hcrc && s->pending > beg)
634 strm->adler = crc32(strm->adler, s->pending_buf + beg,
635 s->pending - beg);
636 if (s->gzindex == s->gzhead->extra_len) {
637 s->gzindex = 0;
638 s->status = NAME_STATE;
639 }
640 }
641 else
642 s->status = NAME_STATE;
643 }
644 if (s->status == NAME_STATE) {
645 if (s->gzhead->name != NULL) {
646 uInt beg = s->pending; /* start of bytes to update crc */
647 int val;
648
649 do {
650 if (s->pending == s->pending_buf_size) {
651 if (s->gzhead->hcrc && s->pending > beg)
652 strm->adler = crc32(strm->adler, s->pending_buf + beg,
653 s->pending - beg);
654 flush_pending(strm);
655 beg = s->pending;
656 if (s->pending == s->pending_buf_size) {
657 val = 1;
658 break;
659 }
660 }
661 val = s->gzhead->name[s->gzindex++];
662 put_byte(s, val);
663 } while (val != 0);
664 if (s->gzhead->hcrc && s->pending > beg)
665 strm->adler = crc32(strm->adler, s->pending_buf + beg,
666 s->pending - beg);
667 if (val == 0) {
668 s->gzindex = 0;
669 s->status = COMMENT_STATE;
670 }
671 }
672 else
673 s->status = COMMENT_STATE;
674 }
675 if (s->status == COMMENT_STATE) {
676 if (s->gzhead->comment != NULL) {
677 uInt beg = s->pending; /* start of bytes to update crc */
678 int val;
679
680 do {
681 if (s->pending == s->pending_buf_size) {
682 if (s->gzhead->hcrc && s->pending > beg)
683 strm->adler = crc32(strm->adler, s->pending_buf + beg,
684 s->pending - beg);
685 flush_pending(strm);
686 beg = s->pending;
687 if (s->pending == s->pending_buf_size) {
688 val = 1;
689 break;
690 }
691 }
692 val = s->gzhead->comment[s->gzindex++];
693 put_byte(s, val);
694 } while (val != 0);
695 if (s->gzhead->hcrc && s->pending > beg)
696 strm->adler = crc32(strm->adler, s->pending_buf + beg,
697 s->pending - beg);
698 if (val == 0)
699 s->status = HCRC_STATE;
700 }
701 else
702 s->status = HCRC_STATE;
703 }
704 if (s->status == HCRC_STATE) {
705 if (s->gzhead->hcrc) {
706 if (s->pending + 2 > s->pending_buf_size)
707 flush_pending(strm);
708 if (s->pending + 2 <= s->pending_buf_size) {
709 put_byte(s, (Byte)(strm->adler & 0xff));
710 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
711 strm->adler = crc32(0L, Z_NULL, 0);
712 s->status = BUSY_STATE;
713 }
714 }
715 else
716 s->status = BUSY_STATE;
717 }
718 #endif
719
720 /* Flush as much pending output as possible */
721 if (s->pending != 0) {
722 flush_pending(strm);
723 if (strm->avail_out == 0) {
724 /* Since avail_out is 0, deflate will be called again with
725 * more output space, but possibly with both pending and
726 * avail_in equal to zero. There won't be anything to do,
727 * but this is not an error situation so make sure we
728 * return OK instead of BUF_ERROR at next call of deflate:
729 */
730 s->last_flush = -1;
731 return Z_OK;
732 }
733
734 /* Make sure there is something to do and avoid duplicate consecutive
735 * flushes. For repeated and useless calls with Z_FINISH, we keep
736 * returning Z_STREAM_END instead of Z_BUF_ERROR.
737 */
738 } else if (strm->avail_in == 0 && flush <= old_flush &&
739 flush != Z_FINISH) {
740 ERR_RETURN(strm, Z_BUF_ERROR);
741 }
742
743 /* User must not provide more input after the first FINISH: */
744 if (s->status == FINISH_STATE && strm->avail_in != 0) {
745 ERR_RETURN(strm, Z_BUF_ERROR);
746 }
747
748 /* Start a new block or continue the current one.
749 */
750 if (strm->avail_in != 0 || s->lookahead != 0 ||
751 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
752 block_state bstate;
753
754 bstate = (*(configuration_table[s->level].func))(s, flush);
755
756 if (bstate == finish_started || bstate == finish_done) {
757 s->status = FINISH_STATE;
758 }
759 if (bstate == need_more || bstate == finish_started) {
760 if (strm->avail_out == 0) {
761 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
762 }
763 return Z_OK;
764 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
765 * of deflate should use the same flush parameter to make sure
766 * that the flush is complete. So we don't have to output an
767 * empty block here, this will be done at next call. This also
768 * ensures that for a very small output buffer, we emit at most
769 * one empty block.
770 */
771 }
772 if (bstate == block_done) {
773 if (flush == Z_PARTIAL_FLUSH) {
774 _tr_align(s);
775 } else { /* FULL_FLUSH or SYNC_FLUSH */
776 _tr_stored_block(s, (char*)0, 0L, 0);
777 /* For a full flush, this empty block will be recognized
778 * as a special marker by inflate_sync().
779 */
780 if (flush == Z_FULL_FLUSH) {
781 CLEAR_HASH(s); /* forget history */
782 }
783 }
784 flush_pending(strm);
785 if (strm->avail_out == 0) {
786 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
787 return Z_OK;
788 }
789 }
790 }
791 Assert(strm->avail_out > 0, "bug2");
792
793 if (flush != Z_FINISH) return Z_OK;
794 if (s->wrap <= 0) return Z_STREAM_END;
795
796 /* Write the trailer */
797 #ifdef GZIP
798 if (s->wrap == 2) {
799 put_byte(s, (Byte)(strm->adler & 0xff));
800 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
801 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
802 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
803 put_byte(s, (Byte)(strm->total_in & 0xff));
804 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
805 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
806 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
807 }
808 else
809 #endif
810 {
811 putShortMSB(s, (uInt)(strm->adler >> 16));
812 putShortMSB(s, (uInt)(strm->adler & 0xffff));
813 }
814 flush_pending(strm);
815 /* If avail_out is zero, the application will call deflate again
816 * to flush the rest.
817 */
818 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
819 return s->pending != 0 ? Z_OK : Z_STREAM_END;
820 }
821
822 /* ========================================================================= */
deflateEnd(z_streamp strm)823 int ZEXPORT deflateEnd (z_streamp strm)
824 {
825 int status;
826
827 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
828
829 status = strm->state->status;
830 if (status != INIT_STATE &&
831 status != EXTRA_STATE &&
832 status != NAME_STATE &&
833 status != COMMENT_STATE &&
834 status != HCRC_STATE &&
835 status != BUSY_STATE &&
836 status != FINISH_STATE) {
837 return Z_STREAM_ERROR;
838 }
839
840 /* Deallocate in reverse order of allocations: */
841 TRY_FREE(strm, strm->state->pending_buf);
842 TRY_FREE(strm, strm->state->head);
843 TRY_FREE(strm, strm->state->prev);
844 TRY_FREE(strm, strm->state->window);
845
846 ZFREE(strm, strm->state);
847 strm->state = Z_NULL;
848
849 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
850 }
851
852 /* =========================================================================
853 * Copy the source state to the destination state.
854 * To simplify the source, this is not supported for 16-bit MSDOS (which
855 * doesn't have enough memory anyway to duplicate compression states).
856 */
deflateCopy(z_streamp dest,z_streamp source)857 int ZEXPORT deflateCopy (z_streamp dest, z_streamp source)
858 {
859 #ifdef MAXSEG_64K
860 return Z_STREAM_ERROR;
861 #else
862 deflate_state *ds;
863 deflate_state *ss;
864 ushf *overlay;
865
866
867 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
868 return Z_STREAM_ERROR;
869 }
870
871 ss = source->state;
872
873 zmemcpy(dest, source, sizeof(z_stream));
874
875 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
876 if (ds == Z_NULL) return Z_MEM_ERROR;
877 dest->state = (struct internal_state FAR *) ds;
878 zmemcpy(ds, ss, sizeof(deflate_state));
879 ds->strm = dest;
880
881 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
882 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
883 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
884 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
885 ds->pending_buf = (uchf *) overlay;
886
887 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
888 ds->pending_buf == Z_NULL) {
889 deflateEnd (dest);
890 return Z_MEM_ERROR;
891 }
892 /* following zmemcpy do not work for 16-bit MSDOS */
893 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
894 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
895 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
896 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
897
898 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
899 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
900 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
901
902 ds->l_desc.dyn_tree = ds->dyn_ltree;
903 ds->d_desc.dyn_tree = ds->dyn_dtree;
904 ds->bl_desc.dyn_tree = ds->bl_tree;
905
906 return Z_OK;
907 #endif /* MAXSEG_64K */
908 }
909
910 /* ===========================================================================
911 * Read a new buffer from the current input stream, update the adler32
912 * and total number of bytes read. All deflate() input goes through
913 * this function so some applications may wish to modify it to avoid
914 * allocating a large strm->next_in buffer and copying from it.
915 * (See also flush_pending()).
916 */
read_buf(z_streamp strm,Bytef * buf,unsigned size)917 local int read_buf (z_streamp strm, Bytef *buf, unsigned size)
918 {
919 unsigned len = strm->avail_in;
920
921 if (len > size) len = size;
922 if (len == 0) return 0;
923
924 strm->avail_in -= len;
925
926 if (strm->state->wrap == 1) {
927 strm->adler = adler32(strm->adler, strm->next_in, len);
928 }
929 #ifdef GZIP
930 else if (strm->state->wrap == 2) {
931 strm->adler = crc32(strm->adler, strm->next_in, len);
932 }
933 #endif
934 zmemcpy(buf, strm->next_in, len);
935 strm->next_in += len;
936 strm->total_in += len;
937
938 return (int)len;
939 }
940
941 /* ===========================================================================
942 * Initialize the "longest match" routines for a new zlib stream
943 */
lm_init(deflate_state * s)944 local void lm_init (deflate_state *s)
945 {
946 s->window_size = (ulg)2L*s->w_size;
947
948 CLEAR_HASH(s);
949
950 /* Set the default configuration parameters:
951 */
952 s->max_lazy_match = configuration_table[s->level].max_lazy;
953 s->good_match = configuration_table[s->level].good_length;
954 s->nice_match = configuration_table[s->level].nice_length;
955 s->max_chain_length = configuration_table[s->level].max_chain;
956
957 s->strstart = 0;
958 s->block_start = 0L;
959 s->lookahead = 0;
960 s->match_length = s->prev_length = MIN_MATCH-1;
961 s->match_available = 0;
962 s->ins_h = 0;
963 #ifndef FASTEST
964 #ifdef ASMV
965 match_init(); /* initialize the asm code */
966 #endif
967 #endif
968 }
969
970 #ifndef FASTEST
971 /* ===========================================================================
972 * Set match_start to the longest match starting at the given string and
973 * return its length. Matches shorter or equal to prev_length are discarded,
974 * in which case the result is equal to prev_length and match_start is
975 * garbage.
976 * IN assertions: cur_match is the head of the hash chain for the current
977 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
978 * OUT assertion: the match length is not greater than s->lookahead.
979 */
980 #ifndef ASMV
981 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
982 * match.S. The code will be functionally equivalent.
983 */
longest_match(deflate_state * s,IPos cur_match)984 local uInt longest_match(deflate_state *s, IPos cur_match)
985 {
986 unsigned chain_length = s->max_chain_length;/* max hash chain length */
987 Bytef *scan = s->window + s->strstart; /* current string */
988 Bytef *match; /* matched string */
989 int len; /* length of current match */
990 int best_len = s->prev_length; /* best match length so far */
991 int nice_match = s->nice_match; /* stop if match long enough */
992 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
993 s->strstart - (IPos)MAX_DIST(s) : NIL;
994 /* Stop when cur_match becomes <= limit. To simplify the code,
995 * we prevent matches with the string of window index 0.
996 */
997 Posf *prev = s->prev;
998 uInt wmask = s->w_mask;
999
1000 #ifdef UNALIGNED_OK
1001 /* Compare two bytes at a time. Note: this is not always beneficial.
1002 * Try with and without -DUNALIGNED_OK to check.
1003 */
1004 Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1005 ush scan_start = *(ushf*)scan;
1006 ush scan_end = *(ushf*)(scan+best_len-1);
1007 #else
1008 Bytef *strend = s->window + s->strstart + MAX_MATCH;
1009 Byte scan_end1 = scan[best_len-1];
1010 Byte scan_end = scan[best_len];
1011 #endif
1012
1013 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1014 * It is easy to get rid of this optimization if necessary.
1015 */
1016 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1017
1018 /* Do not waste too much time if we already have a good match: */
1019 if (s->prev_length >= s->good_match) {
1020 chain_length >>= 2;
1021 }
1022 /* Do not look for matches beyond the end of the input. This is necessary
1023 * to make deflate deterministic.
1024 */
1025 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1026
1027 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1028
1029 do {
1030 Assert(cur_match < s->strstart, "no future");
1031 match = s->window + cur_match;
1032
1033 /* Skip to next match if the match length cannot increase
1034 * or if the match length is less than 2. Note that the checks below
1035 * for insufficient lookahead only occur occasionally for performance
1036 * reasons. Therefore uninitialized memory will be accessed, and
1037 * conditional jumps will be made that depend on those values.
1038 * However the length of the match is limited to the lookahead, so
1039 * the output of deflate is not affected by the uninitialized values.
1040 */
1041 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1042 /* This code assumes sizeof(unsigned short) == 2. Do not use
1043 * UNALIGNED_OK if your compiler uses a different size.
1044 */
1045 if (*(ushf*)(match+best_len-1) != scan_end ||
1046 *(ushf*)match != scan_start) continue;
1047
1048 /* It is not necessary to compare scan[2] and match[2] since they are
1049 * always equal when the other bytes match, given that the hash keys
1050 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1051 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1052 * lookahead only every 4th comparison; the 128th check will be made
1053 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1054 * necessary to put more guard bytes at the end of the window, or
1055 * to check more often for insufficient lookahead.
1056 */
1057 Assert(scan[2] == match[2], "scan[2]?");
1058 scan++, match++;
1059 do {
1060 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1061 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1062 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1063 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1064 scan < strend);
1065 /* The funny "do {}" generates better code on most compilers */
1066
1067 /* Here, scan <= window+strstart+257 */
1068 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1069 if (*scan == *match) scan++;
1070
1071 len = (MAX_MATCH - 1) - (int)(strend-scan);
1072 scan = strend - (MAX_MATCH-1);
1073
1074 #else /* UNALIGNED_OK */
1075
1076 if (match[best_len] != scan_end ||
1077 match[best_len-1] != scan_end1 ||
1078 *match != *scan ||
1079 *++match != scan[1]) continue;
1080
1081 /* The check at best_len-1 can be removed because it will be made
1082 * again later. (This heuristic is not always a win.)
1083 * It is not necessary to compare scan[2] and match[2] since they
1084 * are always equal when the other bytes match, given that
1085 * the hash keys are equal and that HASH_BITS >= 8.
1086 */
1087 scan += 2, match++;
1088 Assert(*scan == *match, "match[2]?");
1089
1090 /* We check for insufficient lookahead only every 8th comparison;
1091 * the 256th check will be made at strstart+258.
1092 */
1093 do {
1094 } while (*++scan == *++match && *++scan == *++match &&
1095 *++scan == *++match && *++scan == *++match &&
1096 *++scan == *++match && *++scan == *++match &&
1097 *++scan == *++match && *++scan == *++match &&
1098 scan < strend);
1099
1100 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1101
1102 len = MAX_MATCH - (int)(strend - scan);
1103 scan = strend - MAX_MATCH;
1104
1105 #endif /* UNALIGNED_OK */
1106
1107 if (len > best_len) {
1108 s->match_start = cur_match;
1109 best_len = len;
1110 if (len >= nice_match) break;
1111 #ifdef UNALIGNED_OK
1112 scan_end = *(ushf*)(scan+best_len-1);
1113 #else
1114 scan_end1 = scan[best_len-1];
1115 scan_end = scan[best_len];
1116 #endif
1117 }
1118 } while ((cur_match = prev[cur_match & wmask]) > limit
1119 && --chain_length != 0);
1120
1121 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1122 return s->lookahead;
1123 }
1124 #endif /* ASMV */
1125 #endif /* FASTEST */
1126
1127 /* ---------------------------------------------------------------------------
1128 * Optimized version for level == 1 or strategy == Z_RLE only
1129 */
longest_match_fast(deflate_state * s,IPos cur_match)1130 local uInt longest_match_fast (deflate_state *s, IPos cur_match)
1131 {
1132 Bytef *scan = s->window + s->strstart; /* current string */
1133 Bytef *match; /* matched string */
1134 int len; /* length of current match */
1135 Bytef *strend = s->window + s->strstart + MAX_MATCH;
1136
1137 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1138 * It is easy to get rid of this optimization if necessary.
1139 */
1140 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1141
1142 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1143
1144 Assert(cur_match < s->strstart, "no future");
1145
1146 match = s->window + cur_match;
1147
1148 /* Return failure if the match length is less than 2:
1149 */
1150 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1151
1152 /* The check at best_len-1 can be removed because it will be made
1153 * again later. (This heuristic is not always a win.)
1154 * It is not necessary to compare scan[2] and match[2] since they
1155 * are always equal when the other bytes match, given that
1156 * the hash keys are equal and that HASH_BITS >= 8.
1157 */
1158 scan += 2, match += 2;
1159 Assert(*scan == *match, "match[2]?");
1160
1161 /* We check for insufficient lookahead only every 8th comparison;
1162 * the 256th check will be made at strstart+258.
1163 */
1164 do {
1165 } while (*++scan == *++match && *++scan == *++match &&
1166 *++scan == *++match && *++scan == *++match &&
1167 *++scan == *++match && *++scan == *++match &&
1168 *++scan == *++match && *++scan == *++match &&
1169 scan < strend);
1170
1171 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1172
1173 len = MAX_MATCH - (int)(strend - scan);
1174
1175 if (len < MIN_MATCH) return MIN_MATCH - 1;
1176
1177 s->match_start = cur_match;
1178 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1179 }
1180
1181 #ifdef DEBUG
1182 /* ===========================================================================
1183 * Check that the match at match_start is indeed a match.
1184 */
check_match(deflate_state * s,IPos start,IPos match,int length)1185 local void check_match(deflate_state *s, IPos start, IPos match, int length)
1186 {
1187 /* check that the match is indeed a match */
1188 if (zmemcmp(s->window + match,
1189 s->window + start, length) != EQUAL) {
1190 fprintf(stderr, " start %u, match %u, length %d\n",
1191 start, match, length);
1192 do {
1193 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1194 } while (--length != 0);
1195 z_error("invalid match");
1196 }
1197 if (z_verbose > 1) {
1198 fprintf(stderr,"\\[%d,%d]", start-match, length);
1199 do { putc(s->window[start++], stderr); } while (--length != 0);
1200 }
1201 }
1202 #else
1203 # define check_match(s, start, match, length)
1204 #endif /* DEBUG */
1205
1206 /* ===========================================================================
1207 * Fill the window when the lookahead becomes insufficient.
1208 * Updates strstart and lookahead.
1209 *
1210 * IN assertion: lookahead < MIN_LOOKAHEAD
1211 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1212 * At least one byte has been read, or avail_in == 0; reads are
1213 * performed for at least two bytes (required for the zip translate_eol
1214 * option -- not supported here).
1215 */
fill_window(deflate_state * s)1216 local void fill_window (deflate_state *s)
1217 {
1218 unsigned n, m;
1219 Posf *p;
1220 unsigned more; /* Amount of free space at the end of the window. */
1221 uInt wsize = s->w_size;
1222
1223 do {
1224 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1225
1226 /* Deal with !@#$% 64K limit: */
1227 if (sizeof(int) <= 2) {
1228 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1229 more = wsize;
1230
1231 } else if (more == (unsigned)(-1)) {
1232 /* Very unlikely, but possible on 16 bit machine if
1233 * strstart == 0 && lookahead == 1 (input done a byte at time)
1234 */
1235 more--;
1236 }
1237 }
1238
1239 /* If the window is almost full and there is insufficient lookahead,
1240 * move the upper half to the lower one to make room in the upper half.
1241 */
1242 if (s->strstart >= wsize+MAX_DIST(s)) {
1243
1244 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1245 s->match_start -= wsize;
1246 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1247 s->block_start -= (long) wsize;
1248
1249 /* Slide the hash table (could be avoided with 32 bit values
1250 at the expense of memory usage). We slide even when level == 0
1251 to keep the hash table consistent if we switch back to level > 0
1252 later. (Using level 0 permanently is not an optimal usage of
1253 zlib, so we don't care about this pathological case.)
1254 */
1255 /* %%% avoid this when Z_RLE */
1256 n = s->hash_size;
1257 p = &s->head[n];
1258 do {
1259 m = *--p;
1260 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1261 } while (--n);
1262
1263 n = wsize;
1264 #ifndef FASTEST
1265 p = &s->prev[n];
1266 do {
1267 m = *--p;
1268 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1269 /* If n is not on any hash chain, prev[n] is garbage but
1270 * its value will never be used.
1271 */
1272 } while (--n);
1273 #endif
1274 more += wsize;
1275 }
1276 if (s->strm->avail_in == 0) return;
1277
1278 /* If there was no sliding:
1279 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1280 * more == window_size - lookahead - strstart
1281 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1282 * => more >= window_size - 2*WSIZE + 2
1283 * In the BIG_MEM or MMAP case (not yet supported),
1284 * window_size == input_size + MIN_LOOKAHEAD &&
1285 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1286 * Otherwise, window_size == 2*WSIZE so more >= 2.
1287 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1288 */
1289 Assert(more >= 2, "more < 2");
1290
1291 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1292 s->lookahead += n;
1293
1294 /* Initialize the hash value now that we have some input: */
1295 if (s->lookahead >= MIN_MATCH) {
1296 s->ins_h = s->window[s->strstart];
1297 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1298 #if MIN_MATCH != 3
1299 Call UPDATE_HASH() MIN_MATCH-3 more times
1300 #endif
1301 }
1302 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1303 * but this is not important since only literal bytes will be emitted.
1304 */
1305
1306 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1307 }
1308
1309 /* ===========================================================================
1310 * Flush the current block, with given end-of-file flag.
1311 * IN assertion: strstart is set to the end of the current match.
1312 */
1313 #define FLUSH_BLOCK_ONLY(s, eof) { \
1314 _tr_flush_block(s, (s->block_start >= 0L ? \
1315 (charf *)&s->window[(unsigned)s->block_start] : \
1316 (charf *)Z_NULL), \
1317 (ulg)((long)s->strstart - s->block_start), \
1318 (eof)); \
1319 s->block_start = s->strstart; \
1320 flush_pending(s->strm); \
1321 Tracev((stderr,"[FLUSH]")); \
1322 }
1323
1324 /* Same but force premature exit if necessary. */
1325 #define FLUSH_BLOCK(s, eof) { \
1326 FLUSH_BLOCK_ONLY(s, eof); \
1327 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
1328 }
1329
1330 /* ===========================================================================
1331 * Copy without compression as much as possible from the input stream, return
1332 * the current block state.
1333 * This function does not insert new strings in the dictionary since
1334 * uncompressible data is probably not useful. This function is used
1335 * only for the level=0 compression option.
1336 * NOTE: this function should be optimized to avoid extra copying from
1337 * window to pending_buf.
1338 */
deflate_stored(deflate_state * s,int flush)1339 local block_state deflate_stored(deflate_state *s, int flush)
1340 {
1341 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1342 * to pending_buf_size, and each stored block has a 5 byte header:
1343 */
1344 ulg max_block_size = 0xffff;
1345 ulg max_start;
1346
1347 if (max_block_size > s->pending_buf_size - 5) {
1348 max_block_size = s->pending_buf_size - 5;
1349 }
1350
1351 /* Copy as much as possible from input to output: */
1352 for (;;) {
1353 /* Fill the window as much as possible: */
1354 if (s->lookahead <= 1) {
1355
1356 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1357 s->block_start >= (long)s->w_size, "slide too late");
1358
1359 fill_window(s);
1360 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1361
1362 if (s->lookahead == 0) break; /* flush the current block */
1363 }
1364 Assert(s->block_start >= 0L, "block gone");
1365
1366 s->strstart += s->lookahead;
1367 s->lookahead = 0;
1368
1369 /* Emit a stored block if pending_buf will be full: */
1370 max_start = s->block_start + max_block_size;
1371 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1372 /* strstart == 0 is possible when wraparound on 16-bit machine */
1373 s->lookahead = (uInt)(s->strstart - max_start);
1374 s->strstart = (uInt)max_start;
1375 FLUSH_BLOCK(s, 0);
1376 }
1377 /* Flush if we may have to slide, otherwise block_start may become
1378 * negative and the data will be gone:
1379 */
1380 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1381 FLUSH_BLOCK(s, 0);
1382 }
1383 }
1384 FLUSH_BLOCK(s, flush == Z_FINISH);
1385 return flush == Z_FINISH ? finish_done : block_done;
1386 }
1387
1388 /* ===========================================================================
1389 * Compress as much as possible from the input stream, return the current
1390 * block state.
1391 * This function does not perform lazy evaluation of matches and inserts
1392 * new strings in the dictionary only for unmatched strings or for short
1393 * matches. It is used only for the fast compression options.
1394 */
deflate_fast(deflate_state * s,int flush)1395 local block_state deflate_fast(deflate_state *s, int flush)
1396 {
1397 IPos hash_head = NIL; /* head of the hash chain */
1398 int bflush; /* set if current block must be flushed */
1399
1400 for (;;) {
1401 /* Make sure that we always have enough lookahead, except
1402 * at the end of the input file. We need MAX_MATCH bytes
1403 * for the next match, plus MIN_MATCH bytes to insert the
1404 * string following the next match.
1405 */
1406 if (s->lookahead < MIN_LOOKAHEAD) {
1407 fill_window(s);
1408 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1409 return need_more;
1410 }
1411 if (s->lookahead == 0) break; /* flush the current block */
1412 }
1413
1414 /* Insert the string window[strstart .. strstart+2] in the
1415 * dictionary, and set hash_head to the head of the hash chain:
1416 */
1417 if (s->lookahead >= MIN_MATCH) {
1418 INSERT_STRING(s, s->strstart, hash_head);
1419 }
1420
1421 /* Find the longest match, discarding those <= prev_length.
1422 * At this point we have always match_length < MIN_MATCH
1423 */
1424 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1425 /* To simplify the code, we prevent matches with the string
1426 * of window index 0 (in particular we have to avoid a match
1427 * of the string with itself at the start of the input file).
1428 */
1429 #ifdef FASTEST
1430 if ((s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) ||
1431 (s->strategy == Z_RLE && s->strstart - hash_head == 1)) {
1432 s->match_length = longest_match_fast (s, hash_head);
1433 }
1434 #else
1435 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) {
1436 s->match_length = longest_match (s, hash_head);
1437 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) {
1438 s->match_length = longest_match_fast (s, hash_head);
1439 }
1440 #endif
1441 /* longest_match() or longest_match_fast() sets match_start */
1442 }
1443 if (s->match_length >= MIN_MATCH) {
1444 check_match(s, s->strstart, s->match_start, s->match_length);
1445
1446 _tr_tally_dist(s, s->strstart - s->match_start,
1447 s->match_length - MIN_MATCH, bflush);
1448
1449 s->lookahead -= s->match_length;
1450
1451 /* Insert new strings in the hash table only if the match length
1452 * is not too large. This saves time but degrades compression.
1453 */
1454 #ifndef FASTEST
1455 if (s->match_length <= s->max_insert_length &&
1456 s->lookahead >= MIN_MATCH) {
1457 s->match_length--; /* string at strstart already in table */
1458 do {
1459 s->strstart++;
1460 INSERT_STRING(s, s->strstart, hash_head);
1461 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1462 * always MIN_MATCH bytes ahead.
1463 */
1464 } while (--s->match_length != 0);
1465 s->strstart++;
1466 } else
1467 #endif
1468 {
1469 s->strstart += s->match_length;
1470 s->match_length = 0;
1471 s->ins_h = s->window[s->strstart];
1472 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1473 #if MIN_MATCH != 3
1474 Call UPDATE_HASH() MIN_MATCH-3 more times
1475 #endif
1476 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1477 * matter since it will be recomputed at next deflate call.
1478 */
1479 }
1480 } else {
1481 /* No match, output a literal byte */
1482 Tracevv((stderr,"%c", s->window[s->strstart]));
1483 _tr_tally_lit (s, s->window[s->strstart], bflush);
1484 s->lookahead--;
1485 s->strstart++;
1486 }
1487 if (bflush) FLUSH_BLOCK(s, 0);
1488 }
1489 FLUSH_BLOCK(s, flush == Z_FINISH);
1490 return flush == Z_FINISH ? finish_done : block_done;
1491 }
1492
1493 #ifndef FASTEST
1494 /* ===========================================================================
1495 * Same as above, but achieves better compression. We use a lazy
1496 * evaluation for matches: a match is finally adopted only if there is
1497 * no better match at the next window position.
1498 */
deflate_slow(deflate_state * s,int flush)1499 local block_state deflate_slow(deflate_state *s, int flush)
1500 {
1501 IPos hash_head = NIL; /* head of hash chain */
1502 int bflush; /* set if current block must be flushed */
1503
1504 /* Process the input block. */
1505 for (;;) {
1506 /* Make sure that we always have enough lookahead, except
1507 * at the end of the input file. We need MAX_MATCH bytes
1508 * for the next match, plus MIN_MATCH bytes to insert the
1509 * string following the next match.
1510 */
1511 if (s->lookahead < MIN_LOOKAHEAD) {
1512 fill_window(s);
1513 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1514 return need_more;
1515 }
1516 if (s->lookahead == 0) break; /* flush the current block */
1517 }
1518
1519 /* Insert the string window[strstart .. strstart+2] in the
1520 * dictionary, and set hash_head to the head of the hash chain:
1521 */
1522 if (s->lookahead >= MIN_MATCH) {
1523 INSERT_STRING(s, s->strstart, hash_head);
1524 }
1525
1526 /* Find the longest match, discarding those <= prev_length.
1527 */
1528 s->prev_length = s->match_length, s->prev_match = s->match_start;
1529 s->match_length = MIN_MATCH-1;
1530
1531 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1532 s->strstart - hash_head <= MAX_DIST(s)) {
1533 /* To simplify the code, we prevent matches with the string
1534 * of window index 0 (in particular we have to avoid a match
1535 * of the string with itself at the start of the input file).
1536 */
1537 if (s->strategy != Z_HUFFMAN_ONLY && s->strategy != Z_RLE) {
1538 s->match_length = longest_match (s, hash_head);
1539 } else if (s->strategy == Z_RLE && s->strstart - hash_head == 1) {
1540 s->match_length = longest_match_fast (s, hash_head);
1541 }
1542 /* longest_match() or longest_match_fast() sets match_start */
1543
1544 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1545 #if TOO_FAR <= 32767
1546 || (s->match_length == MIN_MATCH &&
1547 s->strstart - s->match_start > TOO_FAR)
1548 #endif
1549 )) {
1550
1551 /* If prev_match is also MIN_MATCH, match_start is garbage
1552 * but we will ignore the current match anyway.
1553 */
1554 s->match_length = MIN_MATCH-1;
1555 }
1556 }
1557 /* If there was a match at the previous step and the current
1558 * match is not better, output the previous match:
1559 */
1560 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1561 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1562 /* Do not insert strings in hash table beyond this. */
1563
1564 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1565
1566 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1567 s->prev_length - MIN_MATCH, bflush);
1568
1569 /* Insert in hash table all strings up to the end of the match.
1570 * strstart-1 and strstart are already inserted. If there is not
1571 * enough lookahead, the last two strings are not inserted in
1572 * the hash table.
1573 */
1574 s->lookahead -= s->prev_length-1;
1575 s->prev_length -= 2;
1576 do {
1577 if (++s->strstart <= max_insert) {
1578 INSERT_STRING(s, s->strstart, hash_head);
1579 }
1580 } while (--s->prev_length != 0);
1581 s->match_available = 0;
1582 s->match_length = MIN_MATCH-1;
1583 s->strstart++;
1584
1585 if (bflush) FLUSH_BLOCK(s, 0);
1586
1587 } else if (s->match_available) {
1588 /* If there was no match at the previous position, output a
1589 * single literal. If there was a match but the current match
1590 * is longer, truncate the previous match to a single literal.
1591 */
1592 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1593 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1594 if (bflush) {
1595 FLUSH_BLOCK_ONLY(s, 0);
1596 }
1597 s->strstart++;
1598 s->lookahead--;
1599 if (s->strm->avail_out == 0) return need_more;
1600 } else {
1601 /* There is no previous match to compare with, wait for
1602 * the next step to decide.
1603 */
1604 s->match_available = 1;
1605 s->strstart++;
1606 s->lookahead--;
1607 }
1608 }
1609 Assert (flush != Z_NO_FLUSH, "no flush?");
1610 if (s->match_available) {
1611 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1612 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1613 s->match_available = 0;
1614 }
1615 FLUSH_BLOCK(s, flush == Z_FINISH);
1616 return flush == Z_FINISH ? finish_done : block_done;
1617 }
1618 #endif /* FASTEST */
1619
1620 #if 0
1621 /* ===========================================================================
1622 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1623 * one. Do not maintain a hash table. (It will be regenerated if this run of
1624 * deflate switches away from Z_RLE.)
1625 */
1626 local block_state deflate_rle(s, flush)
1627 deflate_state *s;
1628 int flush;
1629 {
1630 int bflush; /* set if current block must be flushed */
1631 uInt run; /* length of run */
1632 uInt max; /* maximum length of run */
1633 uInt prev; /* byte at distance one to match */
1634 Bytef *scan; /* scan for end of run */
1635
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 longest encodable run.
1640 */
1641 if (s->lookahead < MAX_MATCH) {
1642 fill_window(s);
1643 if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {
1644 return need_more;
1645 }
1646 if (s->lookahead == 0) break; /* flush the current block */
1647 }
1648
1649 /* See how many times the previous byte repeats */
1650 run = 0;
1651 if (s->strstart > 0) { /* if there is a previous byte, that is */
1652 max = s->lookahead < MAX_MATCH ? s->lookahead : MAX_MATCH;
1653 scan = s->window + s->strstart - 1;
1654 prev = *scan++;
1655 do {
1656 if (*scan++ != prev)
1657 break;
1658 } while (++run < max);
1659 }
1660
1661 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1662 if (run >= MIN_MATCH) {
1663 check_match(s, s->strstart, s->strstart - 1, run);
1664 _tr_tally_dist(s, 1, run - MIN_MATCH, bflush);
1665 s->lookahead -= run;
1666 s->strstart += run;
1667 } else {
1668 /* No match, output a literal byte */
1669 Tracevv((stderr,"%c", s->window[s->strstart]));
1670 _tr_tally_lit (s, s->window[s->strstart], bflush);
1671 s->lookahead--;
1672 s->strstart++;
1673 }
1674 if (bflush) FLUSH_BLOCK(s, 0);
1675 }
1676 FLUSH_BLOCK(s, flush == Z_FINISH);
1677 return flush == Z_FINISH ? finish_done : block_done;
1678 }
1679 #endif
1680