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