1
/* deflate.c -- compress data using the deflation algorithm
2
* Copyright (C) 1995-2010 Jean-loup Gailly and Mark Adler
3
* For conditions of distribution and use, see copyright notice in zlib.h
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).
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.
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.
39
* Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40
* Available in http://www.ietf.org/rfc/rfc1951.txt
42
* A description of the Rabin and Karp algorithm is given in the book
43
* "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
45
* Fiala,E.R., and Greene,D.H.
46
* Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
54
const char deflate_copyright[] =
55
" deflate 1.2.5 Copyright 1995-2010 Jean-loup Gailly and Mark Adler ";
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.
63
/* ===========================================================================
64
* Function prototypes.
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 */
73
typedef block_state (*compress_func) OF((deflate_state *s, int flush));
74
/* Compression function. Returns the block state after the call. */
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));
80
local block_state deflate_slow OF((deflate_state *s, int flush));
82
local block_state deflate_rle OF((deflate_state *s, int flush));
83
local block_state deflate_huff OF((deflate_state *s, int flush));
84
local void lm_init OF((deflate_state *s));
85
local void putShortMSB OF((deflate_state *s, uInt b));
86
local void flush_pending OF((z_streamp strm));
87
local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
89
void match_init OF((void)); /* asm code initialization */
90
uInt longest_match OF((deflate_state *s, IPos cur_match));
92
local uInt longest_match OF((deflate_state *s, IPos cur_match));
96
local void check_match OF((deflate_state *s, IPos start, IPos match,
100
/* ===========================================================================
105
/* Tail of hash chains */
108
# define TOO_FAR 4096
110
/* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
112
/* Values for max_lazy_match, good_match and max_chain_length, depending on
113
* the desired pack level (0..9). The values given below have been tuned to
114
* exclude worst case performance for pathological files. Better values may be
115
* found for specific files.
117
typedef struct config_s {
118
ush good_length; /* reduce lazy search above this match length */
119
ush max_lazy; /* do not perform lazy search above this match length */
120
ush nice_length; /* quit search above this match length */
126
local const config configuration_table[2] = {
127
/* good lazy nice chain */
128
/* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
129
/* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
131
local const config configuration_table[10] = {
132
/* good lazy nice chain */
133
/* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
134
/* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
135
/* 2 */ {4, 5, 16, 8, deflate_fast},
136
/* 3 */ {4, 6, 32, 32, deflate_fast},
138
/* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
139
/* 5 */ {8, 16, 32, 32, deflate_slow},
140
/* 6 */ {8, 16, 128, 128, deflate_slow},
141
/* 7 */ {8, 32, 128, 256, deflate_slow},
142
/* 8 */ {32, 128, 258, 1024, deflate_slow},
143
/* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
146
/* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
147
* For deflate_fast() (levels <= 3) good is ignored and lazy has a different
152
/* result of memcmp for equal strings */
154
#ifndef NO_DUMMY_DECL
155
struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
158
/* ===========================================================================
159
* Update a hash value with the given input byte
160
* IN assertion: all calls to to UPDATE_HASH are made with consecutive
161
* input characters, so that a running hash key can be computed from the
162
* previous key instead of complete recalculation each time.
164
#define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
167
/* ===========================================================================
168
* Insert string str in the dictionary and set match_head to the previous head
169
* of the hash chain (the most recent string with same hash key). Return
170
* the previous length of the hash chain.
171
* If this file is compiled with -DFASTEST, the compression level is forced
172
* to 1, and no hash chains are maintained.
173
* IN assertion: all calls to to INSERT_STRING are made with consecutive
174
* input characters and the first MIN_MATCH bytes of str are valid
175
* (except for the last MIN_MATCH-1 bytes of the input file).
178
#define INSERT_STRING(s, str, match_head) \
179
(UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
180
match_head = s->head[s->ins_h], \
181
s->head[s->ins_h] = (Pos)(str))
183
#define INSERT_STRING(s, str, match_head) \
184
(UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
185
match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
186
s->head[s->ins_h] = (Pos)(str))
189
/* ===========================================================================
190
* Initialize the hash table (avoiding 64K overflow for 16 bit systems).
191
* prev[] will be initialized on the fly.
193
#define CLEAR_HASH(s) \
194
s->head[s->hash_size-1] = NIL; \
195
zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
197
/* ========================================================================= */
198
int ZEXPORT deflateInit_(strm, level, version, stream_size)
204
return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
205
Z_DEFAULT_STRATEGY, version, stream_size);
206
/* To do: ignore strm->next_in if we use it as window */
209
/* ========================================================================= */
210
int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
211
version, stream_size)
223
static const char my_version[] = ZLIB_VERSION;
226
/* We overlay pending_buf and d_buf+l_buf. This works since the average
227
* output size for (length,distance) codes is <= 24 bits.
230
if (version == Z_NULL || version[0] != my_version[0] ||
231
stream_size != sizeof(z_stream)) {
232
return Z_VERSION_ERROR;
234
if (strm == Z_NULL) return Z_STREAM_ERROR;
237
if (strm->zalloc == (alloc_func)0) {
238
strm->zalloc = zcalloc;
239
strm->opaque = (voidpf)0;
241
if (strm->zfree == (free_func)0) strm->zfree = zcfree;
244
if (level != 0) level = 1;
246
if (level == Z_DEFAULT_COMPRESSION) level = 6;
249
if (windowBits < 0) { /* suppress zlib wrapper */
251
windowBits = -windowBits;
254
else if (windowBits > 15) {
255
wrap = 2; /* write gzip wrapper instead */
259
if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
260
windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
261
strategy < 0 || strategy > Z_FIXED) {
262
return Z_STREAM_ERROR;
264
if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
265
s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
266
if (s == Z_NULL) return Z_MEM_ERROR;
267
strm->state = (struct internal_state FAR *)s;
272
s->w_bits = windowBits;
273
s->w_size = 1 << s->w_bits;
274
s->w_mask = s->w_size - 1;
276
s->hash_bits = memLevel + 7;
277
s->hash_size = 1 << s->hash_bits;
278
s->hash_mask = s->hash_size - 1;
279
s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
281
s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
282
s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
283
memset(s->prev, 0, s->w_size*sizeof(Pos)); /* XXX Zero out for emscripten, to avoid SAFE_HEAP warnings */
284
s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
286
s->high_water = 0; /* nothing written to s->window yet */
288
s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
290
overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
291
s->pending_buf = (uchf *) overlay;
292
s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
294
if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
295
s->pending_buf == Z_NULL) {
296
s->status = FINISH_STATE;
297
strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
301
s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
302
s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
305
s->strategy = strategy;
306
s->method = (Byte)method;
308
return deflateReset(strm);
311
/* ========================================================================= */
312
int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
314
const Bytef *dictionary;
318
uInt length = dictLength;
322
if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
323
strm->state->wrap == 2 ||
324
(strm->state->wrap == 1 && strm->state->status != INIT_STATE))
325
return Z_STREAM_ERROR;
329
strm->adler = adler32(strm->adler, dictionary, dictLength);
331
if (length < MIN_MATCH) return Z_OK;
332
if (length > s->w_size) {
334
dictionary += dictLength - length; /* use the tail of the dictionary */
336
zmemcpy(s->window, dictionary, length);
337
s->strstart = length;
338
s->block_start = (long)length;
340
/* Insert all strings in the hash table (except for the last two bytes).
341
* s->lookahead stays null, so s->ins_h will be recomputed at the next
342
* call of fill_window.
344
s->ins_h = s->window[0];
345
UPDATE_HASH(s, s->ins_h, s->window[1]);
346
for (n = 0; n <= length - MIN_MATCH; n++) {
347
INSERT_STRING(s, n, hash_head);
349
if (hash_head) hash_head = 0; /* to make compiler happy */
353
/* ========================================================================= */
354
int ZEXPORT deflateReset (strm)
359
if (strm == Z_NULL || strm->state == Z_NULL ||
360
strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
361
return Z_STREAM_ERROR;
364
strm->total_in = strm->total_out = 0;
365
strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
366
strm->data_type = Z_UNKNOWN;
368
s = (deflate_state *)strm->state;
370
s->pending_out = s->pending_buf;
373
s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
375
s->status = s->wrap ? INIT_STATE : BUSY_STATE;
378
s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
380
adler32(0L, Z_NULL, 0);
381
s->last_flush = Z_NO_FLUSH;
389
/* ========================================================================= */
390
int ZEXPORT deflateSetHeader (strm, head)
394
if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
395
if (strm->state->wrap != 2) return Z_STREAM_ERROR;
396
strm->state->gzhead = head;
400
/* ========================================================================= */
401
int ZEXPORT deflatePrime (strm, bits, value)
406
if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
407
strm->state->bi_valid = bits;
408
strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));
412
/* ========================================================================= */
413
int ZEXPORT deflateParams(strm, level, strategy)
422
if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
426
if (level != 0) level = 1;
428
if (level == Z_DEFAULT_COMPRESSION) level = 6;
430
if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
431
return Z_STREAM_ERROR;
433
func = configuration_table[s->level].func;
435
if ((strategy != s->strategy || func != configuration_table[level].func) &&
436
strm->total_in != 0) {
437
/* Flush the last buffer: */
438
err = deflate(strm, Z_BLOCK);
440
if (s->level != level) {
442
s->max_lazy_match = configuration_table[level].max_lazy;
443
s->good_match = configuration_table[level].good_length;
444
s->nice_match = configuration_table[level].nice_length;
445
s->max_chain_length = configuration_table[level].max_chain;
447
s->strategy = strategy;
451
/* ========================================================================= */
452
int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
461
if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
463
s->good_match = good_length;
464
s->max_lazy_match = max_lazy;
465
s->nice_match = nice_length;
466
s->max_chain_length = max_chain;
470
/* =========================================================================
471
* For the default windowBits of 15 and memLevel of 8, this function returns
472
* a close to exact, as well as small, upper bound on the compressed size.
473
* They are coded as constants here for a reason--if the #define's are
474
* changed, then this function needs to be changed as well. The return
475
* value for 15 and 8 only works for those exact settings.
477
* For any setting other than those defaults for windowBits and memLevel,
478
* the value returned is a conservative worst case for the maximum expansion
479
* resulting from using fixed blocks instead of stored blocks, which deflate
480
* can emit on compressed data for some combinations of the parameters.
482
* This function could be more sophisticated to provide closer upper bounds for
483
* every combination of windowBits and memLevel. But even the conservative
484
* upper bound of about 14% expansion does not seem onerous for output buffer
487
uLong ZEXPORT deflateBound(strm, sourceLen)
492
uLong complen, wraplen;
495
/* conservative upper bound for compressed data */
496
complen = sourceLen +
497
((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
499
/* if can't get parameters, return conservative bound plus zlib wrapper */
500
if (strm == Z_NULL || strm->state == Z_NULL)
503
/* compute wrapper length */
506
case 0: /* raw deflate */
509
case 1: /* zlib wrapper */
510
wraplen = 6 + (s->strstart ? 4 : 0);
512
case 2: /* gzip wrapper */
514
if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
515
if (s->gzhead->extra != Z_NULL)
516
wraplen += 2 + s->gzhead->extra_len;
517
str = s->gzhead->name;
522
str = s->gzhead->comment;
531
default: /* for compiler happiness */
535
/* if not default parameters, return conservative bound */
536
if (s->w_bits != 15 || s->hash_bits != 8 + 7)
537
return complen + wraplen;
539
/* default settings: return tight bound for that case */
540
return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
541
(sourceLen >> 25) + 13 - 6 + wraplen;
544
/* =========================================================================
545
* Put a short in the pending buffer. The 16-bit value is put in MSB order.
546
* IN assertion: the stream state is correct and there is enough room in
549
local void putShortMSB (s, b)
553
put_byte(s, (Byte)(b >> 8));
554
put_byte(s, (Byte)(b & 0xff));
557
/* =========================================================================
558
* Flush as much pending output as possible. All deflate() output goes
559
* through this function so some applications may wish to modify it
560
* to avoid allocating a large strm->next_out buffer and copying into it.
561
* (See also read_buf()).
563
local void flush_pending(strm)
566
unsigned len = strm->state->pending;
568
if (len > strm->avail_out) len = strm->avail_out;
569
if (len == 0) return;
571
zmemcpy(strm->next_out, strm->state->pending_out, len);
572
strm->next_out += len;
573
strm->state->pending_out += len;
574
strm->total_out += len;
575
strm->avail_out -= len;
576
strm->state->pending -= len;
577
if (strm->state->pending == 0) {
578
strm->state->pending_out = strm->state->pending_buf;
582
/* ========================================================================= */
583
int ZEXPORT deflate (strm, flush)
587
int old_flush; /* value of flush param for previous deflate call */
590
if (strm == Z_NULL || strm->state == Z_NULL ||
591
flush > Z_BLOCK || flush < 0) {
592
return Z_STREAM_ERROR;
596
if (strm->next_out == Z_NULL ||
597
(strm->next_in == Z_NULL && strm->avail_in != 0) ||
598
(s->status == FINISH_STATE && flush != Z_FINISH)) {
599
ERR_RETURN(strm, Z_STREAM_ERROR);
601
if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
603
s->strm = strm; /* just in case */
604
old_flush = s->last_flush;
605
s->last_flush = flush;
607
/* Write the header */
608
if (s->status == INIT_STATE) {
611
strm->adler = crc32(0L, Z_NULL, 0);
615
if (s->gzhead == Z_NULL) {
621
put_byte(s, s->level == 9 ? 2 :
622
(s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
624
put_byte(s, OS_CODE);
625
s->status = BUSY_STATE;
628
put_byte(s, (s->gzhead->text ? 1 : 0) +
629
(s->gzhead->hcrc ? 2 : 0) +
630
(s->gzhead->extra == Z_NULL ? 0 : 4) +
631
(s->gzhead->name == Z_NULL ? 0 : 8) +
632
(s->gzhead->comment == Z_NULL ? 0 : 16)
634
put_byte(s, (Byte)(s->gzhead->time & 0xff));
635
put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
636
put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
637
put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
638
put_byte(s, s->level == 9 ? 2 :
639
(s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
641
put_byte(s, s->gzhead->os & 0xff);
642
if (s->gzhead->extra != Z_NULL) {
643
put_byte(s, s->gzhead->extra_len & 0xff);
644
put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
647
strm->adler = crc32(strm->adler, s->pending_buf,
650
s->status = EXTRA_STATE;
656
uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
659
if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
661
else if (s->level < 6)
663
else if (s->level == 6)
667
header |= (level_flags << 6);
668
if (s->strstart != 0) header |= PRESET_DICT;
669
header += 31 - (header % 31);
671
s->status = BUSY_STATE;
672
putShortMSB(s, header);
674
/* Save the adler32 of the preset dictionary: */
675
if (s->strstart != 0) {
676
putShortMSB(s, (uInt)(strm->adler >> 16));
677
putShortMSB(s, (uInt)(strm->adler & 0xffff));
679
strm->adler = adler32(0L, Z_NULL, 0);
683
if (s->status == EXTRA_STATE) {
684
if (s->gzhead->extra != Z_NULL) {
685
uInt beg = s->pending; /* start of bytes to update crc */
687
while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
688
if (s->pending == s->pending_buf_size) {
689
if (s->gzhead->hcrc && s->pending > beg)
690
strm->adler = crc32(strm->adler, s->pending_buf + beg,
694
if (s->pending == s->pending_buf_size)
697
put_byte(s, s->gzhead->extra[s->gzindex]);
700
if (s->gzhead->hcrc && s->pending > beg)
701
strm->adler = crc32(strm->adler, s->pending_buf + beg,
703
if (s->gzindex == s->gzhead->extra_len) {
705
s->status = NAME_STATE;
709
s->status = NAME_STATE;
711
if (s->status == NAME_STATE) {
712
if (s->gzhead->name != Z_NULL) {
713
uInt beg = s->pending; /* start of bytes to update crc */
717
if (s->pending == s->pending_buf_size) {
718
if (s->gzhead->hcrc && s->pending > beg)
719
strm->adler = crc32(strm->adler, s->pending_buf + beg,
723
if (s->pending == s->pending_buf_size) {
728
val = s->gzhead->name[s->gzindex++];
731
if (s->gzhead->hcrc && s->pending > beg)
732
strm->adler = crc32(strm->adler, s->pending_buf + beg,
736
s->status = COMMENT_STATE;
740
s->status = COMMENT_STATE;
742
if (s->status == COMMENT_STATE) {
743
if (s->gzhead->comment != Z_NULL) {
744
uInt beg = s->pending; /* start of bytes to update crc */
748
if (s->pending == s->pending_buf_size) {
749
if (s->gzhead->hcrc && s->pending > beg)
750
strm->adler = crc32(strm->adler, s->pending_buf + beg,
754
if (s->pending == s->pending_buf_size) {
759
val = s->gzhead->comment[s->gzindex++];
762
if (s->gzhead->hcrc && s->pending > beg)
763
strm->adler = crc32(strm->adler, s->pending_buf + beg,
766
s->status = HCRC_STATE;
769
s->status = HCRC_STATE;
771
if (s->status == HCRC_STATE) {
772
if (s->gzhead->hcrc) {
773
if (s->pending + 2 > s->pending_buf_size)
775
if (s->pending + 2 <= s->pending_buf_size) {
776
put_byte(s, (Byte)(strm->adler & 0xff));
777
put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
778
strm->adler = crc32(0L, Z_NULL, 0);
779
s->status = BUSY_STATE;
783
s->status = BUSY_STATE;
787
/* Flush as much pending output as possible */
788
if (s->pending != 0) {
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:
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.
805
} else if (strm->avail_in == 0 && flush <= old_flush &&
807
ERR_RETURN(strm, Z_BUF_ERROR);
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);
815
/* Start a new block or continue the current one.
817
if (strm->avail_in != 0 || s->lookahead != 0 ||
818
(flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
821
bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
822
(s->strategy == Z_RLE ? deflate_rle(s, flush) :
823
(*(configuration_table[s->level].func))(s, flush));
825
if (bstate == finish_started || bstate == finish_done) {
826
s->status = FINISH_STATE;
828
if (bstate == need_more || bstate == finish_started) {
829
if (strm->avail_out == 0) {
830
s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
833
/* If flush != Z_NO_FLUSH && avail_out == 0, the next call
834
* of deflate should use the same flush parameter to make sure
835
* that the flush is complete. So we don't have to output an
836
* empty block here, this will be done at next call. This also
837
* ensures that for a very small output buffer, we emit at most
841
if (bstate == block_done) {
842
if (flush == Z_PARTIAL_FLUSH) {
844
} else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
845
_tr_stored_block(s, (char*)0, 0L, 0);
846
/* For a full flush, this empty block will be recognized
847
* as a special marker by inflate_sync().
849
if (flush == Z_FULL_FLUSH) {
850
CLEAR_HASH(s); /* forget history */
851
if (s->lookahead == 0) {
858
if (strm->avail_out == 0) {
859
s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
864
Assert(strm->avail_out > 0, "bug2");
866
if (flush != Z_FINISH) return Z_OK;
867
if (s->wrap <= 0) return Z_STREAM_END;
869
/* Write the trailer */
872
put_byte(s, (Byte)(strm->adler & 0xff));
873
put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
874
put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
875
put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
876
put_byte(s, (Byte)(strm->total_in & 0xff));
877
put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
878
put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
879
put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
884
putShortMSB(s, (uInt)(strm->adler >> 16));
885
putShortMSB(s, (uInt)(strm->adler & 0xffff));
888
/* If avail_out is zero, the application will call deflate again
891
if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
892
return s->pending != 0 ? Z_OK : Z_STREAM_END;
895
/* ========================================================================= */
896
int ZEXPORT deflateEnd (strm)
901
if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
903
status = strm->state->status;
904
if (status != INIT_STATE &&
905
status != EXTRA_STATE &&
906
status != NAME_STATE &&
907
status != COMMENT_STATE &&
908
status != HCRC_STATE &&
909
status != BUSY_STATE &&
910
status != FINISH_STATE) {
911
return Z_STREAM_ERROR;
914
/* Deallocate in reverse order of allocations: */
915
TRY_FREE(strm, strm->state->pending_buf);
916
TRY_FREE(strm, strm->state->head);
917
TRY_FREE(strm, strm->state->prev);
918
TRY_FREE(strm, strm->state->window);
920
ZFREE(strm, strm->state);
921
strm->state = Z_NULL;
923
return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
926
/* =========================================================================
927
* Copy the source state to the destination state.
928
* To simplify the source, this is not supported for 16-bit MSDOS (which
929
* doesn't have enough memory anyway to duplicate compression states).
931
int ZEXPORT deflateCopy (dest, source)
936
return Z_STREAM_ERROR;
943
if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
944
return Z_STREAM_ERROR;
949
zmemcpy(dest, source, sizeof(z_stream));
951
ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
952
if (ds == Z_NULL) return Z_MEM_ERROR;
953
dest->state = (struct internal_state FAR *) ds;
954
zmemcpy(ds, ss, sizeof(deflate_state));
957
ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
958
ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
959
memset(ds->prev, 0, ds->w_size*sizeof(Pos)); /* XXX Zero out for emscripten, to avoid SAFE_HEAP warnings */
960
ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
961
overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
962
ds->pending_buf = (uchf *) overlay;
964
if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
965
ds->pending_buf == Z_NULL) {
969
/* following zmemcpy do not work for 16-bit MSDOS */
970
zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
971
zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
972
zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
973
zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
975
ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
976
ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
977
ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
979
ds->l_desc.dyn_tree = ds->dyn_ltree;
980
ds->d_desc.dyn_tree = ds->dyn_dtree;
981
ds->bl_desc.dyn_tree = ds->bl_tree;
984
#endif /* MAXSEG_64K */
987
/* ===========================================================================
988
* Read a new buffer from the current input stream, update the adler32
989
* and total number of bytes read. All deflate() input goes through
990
* this function so some applications may wish to modify it to avoid
991
* allocating a large strm->next_in buffer and copying from it.
992
* (See also flush_pending()).
994
local int read_buf(strm, buf, size)
999
unsigned len = strm->avail_in;
1001
if (len > size) len = size;
1002
if (len == 0) return 0;
1004
strm->avail_in -= len;
1006
if (strm->state->wrap == 1) {
1007
strm->adler = adler32(strm->adler, strm->next_in, len);
1010
else if (strm->state->wrap == 2) {
1011
strm->adler = crc32(strm->adler, strm->next_in, len);
1014
zmemcpy(buf, strm->next_in, len);
1015
strm->next_in += len;
1016
strm->total_in += len;
1021
/* ===========================================================================
1022
* Initialize the "longest match" routines for a new zlib stream
1024
local void lm_init (s)
1027
s->window_size = (ulg)2L*s->w_size;
1031
/* Set the default configuration parameters:
1033
s->max_lazy_match = configuration_table[s->level].max_lazy;
1034
s->good_match = configuration_table[s->level].good_length;
1035
s->nice_match = configuration_table[s->level].nice_length;
1036
s->max_chain_length = configuration_table[s->level].max_chain;
1039
s->block_start = 0L;
1041
s->match_length = s->prev_length = MIN_MATCH-1;
1042
s->match_start = 0; /* XXX Emscripten: initialize to 0, to prevent a SAFE_HEAP notification */
1043
s->match_available = 0;
1047
match_init(); /* initialize the asm code */
1053
/* ===========================================================================
1054
* Set match_start to the longest match starting at the given string and
1055
* return its length. Matches shorter or equal to prev_length are discarded,
1056
* in which case the result is equal to prev_length and match_start is
1058
* IN assertions: cur_match is the head of the hash chain for the current
1059
* string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1060
* OUT assertion: the match length is not greater than s->lookahead.
1063
/* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1064
* match.S. The code will be functionally equivalent.
1066
local uInt longest_match(s, cur_match)
1068
IPos cur_match; /* current match */
1070
unsigned chain_length = s->max_chain_length;/* max hash chain length */
1071
register Bytef *scan = s->window + s->strstart; /* current string */
1072
register Bytef *match; /* matched string */
1073
register int len; /* length of current match */
1074
int best_len = s->prev_length; /* best match length so far */
1075
int nice_match = s->nice_match; /* stop if match long enough */
1076
IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1077
s->strstart - (IPos)MAX_DIST(s) : NIL;
1078
/* Stop when cur_match becomes <= limit. To simplify the code,
1079
* we prevent matches with the string of window index 0.
1081
Posf *prev = s->prev;
1082
uInt wmask = s->w_mask;
1085
/* Compare two bytes at a time. Note: this is not always beneficial.
1086
* Try with and without -DUNALIGNED_OK to check.
1088
register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1089
register ush scan_start = *(ushf*)scan;
1090
register ush scan_end = *(ushf*)(scan+best_len-1);
1092
register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1093
register Byte scan_end1 = scan[best_len-1];
1094
register Byte scan_end = scan[best_len];
1097
/* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1098
* It is easy to get rid of this optimization if necessary.
1100
Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1102
/* Do not waste too much time if we already have a good match: */
1103
if (s->prev_length >= s->good_match) {
1106
/* Do not look for matches beyond the end of the input. This is necessary
1107
* to make deflate deterministic.
1109
if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1111
Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1114
Assert(cur_match < s->strstart, "no future");
1115
match = s->window + cur_match;
1117
/* Skip to next match if the match length cannot increase
1118
* or if the match length is less than 2. Note that the checks below
1119
* for insufficient lookahead only occur occasionally for performance
1120
* reasons. Therefore uninitialized memory will be accessed, and
1121
* conditional jumps will be made that depend on those values.
1122
* However the length of the match is limited to the lookahead, so
1123
* the output of deflate is not affected by the uninitialized values.
1125
#if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1126
/* This code assumes sizeof(unsigned short) == 2. Do not use
1127
* UNALIGNED_OK if your compiler uses a different size.
1129
if (*(ushf*)(match+best_len-1) != scan_end ||
1130
*(ushf*)match != scan_start) continue;
1132
/* It is not necessary to compare scan[2] and match[2] since they are
1133
* always equal when the other bytes match, given that the hash keys
1134
* are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1135
* strstart+3, +5, ... up to strstart+257. We check for insufficient
1136
* lookahead only every 4th comparison; the 128th check will be made
1137
* at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1138
* necessary to put more guard bytes at the end of the window, or
1139
* to check more often for insufficient lookahead.
1141
Assert(scan[2] == match[2], "scan[2]?");
1144
} while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1145
*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1146
*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1147
*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1149
/* The funny "do {}" generates better code on most compilers */
1151
/* Here, scan <= window+strstart+257 */
1152
Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1153
if (*scan == *match) scan++;
1155
len = (MAX_MATCH - 1) - (int)(strend-scan);
1156
scan = strend - (MAX_MATCH-1);
1158
#else /* UNALIGNED_OK */
1160
if (match[best_len] != scan_end ||
1161
match[best_len-1] != scan_end1 ||
1163
*++match != scan[1]) continue;
1165
/* The check at best_len-1 can be removed because it will be made
1166
* again later. (This heuristic is not always a win.)
1167
* It is not necessary to compare scan[2] and match[2] since they
1168
* are always equal when the other bytes match, given that
1169
* the hash keys are equal and that HASH_BITS >= 8.
1172
Assert(*scan == *match, "match[2]?");
1174
/* We check for insufficient lookahead only every 8th comparison;
1175
* the 256th check will be made at strstart+258.
1178
} while (*++scan == *++match && *++scan == *++match &&
1179
*++scan == *++match && *++scan == *++match &&
1180
*++scan == *++match && *++scan == *++match &&
1181
*++scan == *++match && *++scan == *++match &&
1184
Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1186
len = MAX_MATCH - (int)(strend - scan);
1187
scan = strend - MAX_MATCH;
1189
#endif /* UNALIGNED_OK */
1191
if (len > best_len) {
1192
s->match_start = cur_match;
1194
if (len >= nice_match) break;
1196
scan_end = *(ushf*)(scan+best_len-1);
1198
scan_end1 = scan[best_len-1];
1199
scan_end = scan[best_len];
1202
} while ((cur_match = prev[cur_match & wmask]) > limit
1203
&& --chain_length != 0);
1205
if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1206
return s->lookahead;
1212
/* ---------------------------------------------------------------------------
1213
* Optimized version for FASTEST only
1215
local uInt longest_match(s, cur_match)
1217
IPos cur_match; /* current match */
1219
register Bytef *scan = s->window + s->strstart; /* current string */
1220
register Bytef *match; /* matched string */
1221
register int len; /* length of current match */
1222
register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1224
/* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1225
* It is easy to get rid of this optimization if necessary.
1227
Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1229
Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1231
Assert(cur_match < s->strstart, "no future");
1233
match = s->window + cur_match;
1235
/* Return failure if the match length is less than 2:
1237
if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1239
/* The check at best_len-1 can be removed because it will be made
1240
* again later. (This heuristic is not always a win.)
1241
* It is not necessary to compare scan[2] and match[2] since they
1242
* are always equal when the other bytes match, given that
1243
* the hash keys are equal and that HASH_BITS >= 8.
1245
scan += 2, match += 2;
1246
Assert(*scan == *match, "match[2]?");
1248
/* We check for insufficient lookahead only every 8th comparison;
1249
* the 256th check will be made at strstart+258.
1252
} while (*++scan == *++match && *++scan == *++match &&
1253
*++scan == *++match && *++scan == *++match &&
1254
*++scan == *++match && *++scan == *++match &&
1255
*++scan == *++match && *++scan == *++match &&
1258
Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1260
len = MAX_MATCH - (int)(strend - scan);
1262
if (len < MIN_MATCH) return MIN_MATCH - 1;
1264
s->match_start = cur_match;
1265
return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1268
#endif /* FASTEST */
1271
/* ===========================================================================
1272
* Check that the match at match_start is indeed a match.
1274
local void check_match(s, start, match, length)
1279
/* check that the match is indeed a match */
1280
if (zmemcmp(s->window + match,
1281
s->window + start, length) != EQUAL) {
1282
fprintf(stderr, " start %u, match %u, length %d\n",
1283
start, match, length);
1285
fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1286
} while (--length != 0);
1287
z_error("invalid match");
1289
if (z_verbose > 1) {
1290
fprintf(stderr,"\\[%d,%d]", start-match, length);
1291
do { putc(s->window[start++], stderr); } while (--length != 0);
1295
# define check_match(s, start, match, length)
1298
/* ===========================================================================
1299
* Fill the window when the lookahead becomes insufficient.
1300
* Updates strstart and lookahead.
1302
* IN assertion: lookahead < MIN_LOOKAHEAD
1303
* OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1304
* At least one byte has been read, or avail_in == 0; reads are
1305
* performed for at least two bytes (required for the zip translate_eol
1306
* option -- not supported here).
1308
local void fill_window(s)
1311
register unsigned n, m;
1313
unsigned more; /* Amount of free space at the end of the window. */
1314
uInt wsize = s->w_size;
1317
more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1319
/* Deal with !@#$% 64K limit: */
1320
if (sizeof(int) <= 2) {
1321
if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1324
} else if (more == (unsigned)(-1)) {
1325
/* Very unlikely, but possible on 16 bit machine if
1326
* strstart == 0 && lookahead == 1 (input done a byte at time)
1332
/* If the window is almost full and there is insufficient lookahead,
1333
* move the upper half to the lower one to make room in the upper half.
1335
if (s->strstart >= wsize+MAX_DIST(s)) {
1337
zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1338
s->match_start -= wsize;
1339
s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1340
s->block_start -= (long) wsize;
1342
/* Slide the hash table (could be avoided with 32 bit values
1343
at the expense of memory usage). We slide even when level == 0
1344
to keep the hash table consistent if we switch back to level > 0
1345
later. (Using level 0 permanently is not an optimal usage of
1346
zlib, so we don't care about this pathological case.)
1352
*p = (Pos)(m >= wsize ? m-wsize : NIL);
1360
*p = (Pos)(m >= wsize ? m-wsize : NIL);
1361
/* If n is not on any hash chain, prev[n] is garbage but
1362
* its value will never be used.
1368
if (s->strm->avail_in == 0) return;
1370
/* If there was no sliding:
1371
* strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1372
* more == window_size - lookahead - strstart
1373
* => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1374
* => more >= window_size - 2*WSIZE + 2
1375
* In the BIG_MEM or MMAP case (not yet supported),
1376
* window_size == input_size + MIN_LOOKAHEAD &&
1377
* strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1378
* Otherwise, window_size == 2*WSIZE so more >= 2.
1379
* If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1381
Assert(more >= 2, "more < 2");
1383
n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1386
/* Initialize the hash value now that we have some input: */
1387
if (s->lookahead >= MIN_MATCH) {
1388
s->ins_h = s->window[s->strstart];
1389
UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1391
Call UPDATE_HASH() MIN_MATCH-3 more times
1394
/* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1395
* but this is not important since only literal bytes will be emitted.
1398
} while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1400
/* If the WIN_INIT bytes after the end of the current data have never been
1401
* written, then zero those bytes in order to avoid memory check reports of
1402
* the use of uninitialized (or uninitialised as Julian writes) bytes by
1403
* the longest match routines. Update the high water mark for the next
1404
* time through here. WIN_INIT is set to MAX_MATCH since the longest match
1405
* routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1407
if (s->high_water < s->window_size) {
1408
ulg curr = s->strstart + (ulg)(s->lookahead);
1411
if (s->high_water < curr) {
1412
/* Previous high water mark below current data -- zero WIN_INIT
1413
* bytes or up to end of window, whichever is less.
1415
init = s->window_size - curr;
1416
if (init > WIN_INIT)
1418
zmemzero(s->window + curr, (unsigned)init);
1419
s->high_water = curr + init;
1421
else if (s->high_water < (ulg)curr + WIN_INIT) {
1422
/* High water mark at or above current data, but below current data
1423
* plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1424
* to end of window, whichever is less.
1426
init = (ulg)curr + WIN_INIT - s->high_water;
1427
if (init > s->window_size - s->high_water)
1428
init = s->window_size - s->high_water;
1429
zmemzero(s->window + s->high_water, (unsigned)init);
1430
s->high_water += init;
1435
/* ===========================================================================
1436
* Flush the current block, with given end-of-file flag.
1437
* IN assertion: strstart is set to the end of the current match.
1439
#define FLUSH_BLOCK_ONLY(s, last) { \
1440
_tr_flush_block(s, (s->block_start >= 0L ? \
1441
(charf *)&s->window[(unsigned)s->block_start] : \
1443
(ulg)((long)s->strstart - s->block_start), \
1445
s->block_start = s->strstart; \
1446
flush_pending(s->strm); \
1447
Tracev((stderr,"[FLUSH]")); \
1450
/* Same but force premature exit if necessary. */
1451
#define FLUSH_BLOCK(s, last) { \
1452
FLUSH_BLOCK_ONLY(s, last); \
1453
if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1456
/* ===========================================================================
1457
* Copy without compression as much as possible from the input stream, return
1458
* the current block state.
1459
* This function does not insert new strings in the dictionary since
1460
* uncompressible data is probably not useful. This function is used
1461
* only for the level=0 compression option.
1462
* NOTE: this function should be optimized to avoid extra copying from
1463
* window to pending_buf.
1465
local block_state deflate_stored(s, flush)
1469
/* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1470
* to pending_buf_size, and each stored block has a 5 byte header:
1472
ulg max_block_size = 0xffff;
1475
if (max_block_size > s->pending_buf_size - 5) {
1476
max_block_size = s->pending_buf_size - 5;
1479
/* Copy as much as possible from input to output: */
1481
/* Fill the window as much as possible: */
1482
if (s->lookahead <= 1) {
1484
Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1485
s->block_start >= (long)s->w_size, "slide too late");
1488
if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1490
if (s->lookahead == 0) break; /* flush the current block */
1492
Assert(s->block_start >= 0L, "block gone");
1494
s->strstart += s->lookahead;
1497
/* Emit a stored block if pending_buf will be full: */
1498
max_start = s->block_start + max_block_size;
1499
if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1500
/* strstart == 0 is possible when wraparound on 16-bit machine */
1501
s->lookahead = (uInt)(s->strstart - max_start);
1502
s->strstart = (uInt)max_start;
1505
/* Flush if we may have to slide, otherwise block_start may become
1506
* negative and the data will be gone:
1508
if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1512
FLUSH_BLOCK(s, flush == Z_FINISH);
1513
return flush == Z_FINISH ? finish_done : block_done;
1516
/* ===========================================================================
1517
* Compress as much as possible from the input stream, return the current
1519
* This function does not perform lazy evaluation of matches and inserts
1520
* new strings in the dictionary only for unmatched strings or for short
1521
* matches. It is used only for the fast compression options.
1523
local block_state deflate_fast(s, flush)
1527
IPos hash_head; /* head of the hash chain */
1528
int bflush; /* set if current block must be flushed */
1531
/* Make sure that we always have enough lookahead, except
1532
* at the end of the input file. We need MAX_MATCH bytes
1533
* for the next match, plus MIN_MATCH bytes to insert the
1534
* string following the next match.
1536
if (s->lookahead < MIN_LOOKAHEAD) {
1538
if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1541
if (s->lookahead == 0) break; /* flush the current block */
1544
/* Insert the string window[strstart .. strstart+2] in the
1545
* dictionary, and set hash_head to the head of the hash chain:
1548
if (s->lookahead >= MIN_MATCH) {
1549
INSERT_STRING(s, s->strstart, hash_head);
1552
/* Find the longest match, discarding those <= prev_length.
1553
* At this point we have always match_length < MIN_MATCH
1555
if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1556
/* To simplify the code, we prevent matches with the string
1557
* of window index 0 (in particular we have to avoid a match
1558
* of the string with itself at the start of the input file).
1560
s->match_length = longest_match (s, hash_head);
1561
/* longest_match() sets match_start */
1563
if (s->match_length >= MIN_MATCH) {
1564
check_match(s, s->strstart, s->match_start, s->match_length);
1566
_tr_tally_dist(s, s->strstart - s->match_start,
1567
s->match_length - MIN_MATCH, bflush);
1569
s->lookahead -= s->match_length;
1571
/* Insert new strings in the hash table only if the match length
1572
* is not too large. This saves time but degrades compression.
1575
if (s->match_length <= s->max_insert_length &&
1576
s->lookahead >= MIN_MATCH) {
1577
s->match_length--; /* string at strstart already in table */
1580
INSERT_STRING(s, s->strstart, hash_head);
1581
/* strstart never exceeds WSIZE-MAX_MATCH, so there are
1582
* always MIN_MATCH bytes ahead.
1584
} while (--s->match_length != 0);
1589
s->strstart += s->match_length;
1590
s->match_length = 0;
1591
s->ins_h = s->window[s->strstart];
1592
UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1594
Call UPDATE_HASH() MIN_MATCH-3 more times
1596
/* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1597
* matter since it will be recomputed at next deflate call.
1601
/* No match, output a literal byte */
1602
Tracevv((stderr,"%c", s->window[s->strstart]));
1603
_tr_tally_lit (s, s->window[s->strstart], bflush);
1607
if (bflush) FLUSH_BLOCK(s, 0);
1609
FLUSH_BLOCK(s, flush == Z_FINISH);
1610
return flush == Z_FINISH ? finish_done : block_done;
1614
/* ===========================================================================
1615
* Same as above, but achieves better compression. We use a lazy
1616
* evaluation for matches: a match is finally adopted only if there is
1617
* no better match at the next window position.
1619
local block_state deflate_slow(s, flush)
1623
IPos hash_head; /* head of hash chain */
1624
int bflush; /* set if current block must be flushed */
1626
/* Process the input block. */
1628
/* Make sure that we always have enough lookahead, except
1629
* at the end of the input file. We need MAX_MATCH bytes
1630
* for the next match, plus MIN_MATCH bytes to insert the
1631
* string following the next match.
1633
if (s->lookahead < MIN_LOOKAHEAD) {
1635
if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1638
if (s->lookahead == 0) break; /* flush the current block */
1641
/* Insert the string window[strstart .. strstart+2] in the
1642
* dictionary, and set hash_head to the head of the hash chain:
1645
if (s->lookahead >= MIN_MATCH) {
1646
INSERT_STRING(s, s->strstart, hash_head);
1649
/* Find the longest match, discarding those <= prev_length.
1651
s->prev_length = s->match_length, s->prev_match = s->match_start;
1652
s->match_length = MIN_MATCH-1;
1654
if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1655
s->strstart - hash_head <= MAX_DIST(s)) {
1656
/* To simplify the code, we prevent matches with the string
1657
* of window index 0 (in particular we have to avoid a match
1658
* of the string with itself at the start of the input file).
1660
s->match_length = longest_match (s, hash_head);
1661
/* longest_match() sets match_start */
1663
if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1664
#if TOO_FAR <= 32767
1665
|| (s->match_length == MIN_MATCH &&
1666
s->strstart - s->match_start > TOO_FAR)
1670
/* If prev_match is also MIN_MATCH, match_start is garbage
1671
* but we will ignore the current match anyway.
1673
s->match_length = MIN_MATCH-1;
1676
/* If there was a match at the previous step and the current
1677
* match is not better, output the previous match:
1679
if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1680
uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1681
/* Do not insert strings in hash table beyond this. */
1683
check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1685
_tr_tally_dist(s, s->strstart -1 - s->prev_match,
1686
s->prev_length - MIN_MATCH, bflush);
1688
/* Insert in hash table all strings up to the end of the match.
1689
* strstart-1 and strstart are already inserted. If there is not
1690
* enough lookahead, the last two strings are not inserted in
1693
s->lookahead -= s->prev_length-1;
1694
s->prev_length -= 2;
1696
if (++s->strstart <= max_insert) {
1697
INSERT_STRING(s, s->strstart, hash_head);
1699
} while (--s->prev_length != 0);
1700
s->match_available = 0;
1701
s->match_length = MIN_MATCH-1;
1704
if (bflush) FLUSH_BLOCK(s, 0);
1706
} else if (s->match_available) {
1707
/* If there was no match at the previous position, output a
1708
* single literal. If there was a match but the current match
1709
* is longer, truncate the previous match to a single literal.
1711
Tracevv((stderr,"%c", s->window[s->strstart-1]));
1712
_tr_tally_lit(s, s->window[s->strstart-1], bflush);
1714
FLUSH_BLOCK_ONLY(s, 0);
1718
if (s->strm->avail_out == 0) return need_more;
1720
/* There is no previous match to compare with, wait for
1721
* the next step to decide.
1723
s->match_available = 1;
1728
Assert (flush != Z_NO_FLUSH, "no flush?");
1729
if (s->match_available) {
1730
Tracevv((stderr,"%c", s->window[s->strstart-1]));
1731
_tr_tally_lit(s, s->window[s->strstart-1], bflush);
1732
s->match_available = 0;
1734
FLUSH_BLOCK(s, flush == Z_FINISH);
1735
return flush == Z_FINISH ? finish_done : block_done;
1737
#endif /* FASTEST */
1739
/* ===========================================================================
1740
* For Z_RLE, simply look for runs of bytes, generate matches only of distance
1741
* one. Do not maintain a hash table. (It will be regenerated if this run of
1742
* deflate switches away from Z_RLE.)
1744
local block_state deflate_rle(s, flush)
1748
int bflush; /* set if current block must be flushed */
1749
uInt prev; /* byte at distance one to match */
1750
Bytef *scan, *strend; /* scan goes up to strend for length of run */
1753
/* Make sure that we always have enough lookahead, except
1754
* at the end of the input file. We need MAX_MATCH bytes
1755
* for the longest encodable run.
1757
if (s->lookahead < MAX_MATCH) {
1759
if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {
1762
if (s->lookahead == 0) break; /* flush the current block */
1765
/* See how many times the previous byte repeats */
1766
s->match_length = 0;
1767
if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
1768
scan = s->window + s->strstart - 1;
1770
if (prev == *++scan && prev == *++scan && prev == *++scan) {
1771
strend = s->window + s->strstart + MAX_MATCH;
1773
} while (prev == *++scan && prev == *++scan &&
1774
prev == *++scan && prev == *++scan &&
1775
prev == *++scan && prev == *++scan &&
1776
prev == *++scan && prev == *++scan &&
1778
s->match_length = MAX_MATCH - (int)(strend - scan);
1779
if (s->match_length > s->lookahead)
1780
s->match_length = s->lookahead;
1784
/* Emit match if have run of MIN_MATCH or longer, else emit literal */
1785
if (s->match_length >= MIN_MATCH) {
1786
check_match(s, s->strstart, s->strstart - 1, s->match_length);
1788
_tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
1790
s->lookahead -= s->match_length;
1791
s->strstart += s->match_length;
1792
s->match_length = 0;
1794
/* No match, output a literal byte */
1795
Tracevv((stderr,"%c", s->window[s->strstart]));
1796
_tr_tally_lit (s, s->window[s->strstart], bflush);
1800
if (bflush) FLUSH_BLOCK(s, 0);
1802
FLUSH_BLOCK(s, flush == Z_FINISH);
1803
return flush == Z_FINISH ? finish_done : block_done;
1806
/* ===========================================================================
1807
* For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
1808
* (It will be regenerated if this run of deflate switches away from Huffman.)
1810
local block_state deflate_huff(s, flush)
1814
int bflush; /* set if current block must be flushed */
1817
/* Make sure that we have a literal to write. */
1818
if (s->lookahead == 0) {
1820
if (s->lookahead == 0) {
1821
if (flush == Z_NO_FLUSH)
1823
break; /* flush the current block */
1827
/* Output a literal byte */
1828
s->match_length = 0;
1829
Tracevv((stderr,"%c", s->window[s->strstart]));
1830
_tr_tally_lit (s, s->window[s->strstart], bflush);
1833
if (bflush) FLUSH_BLOCK(s, 0);
1835
FLUSH_BLOCK(s, flush == Z_FINISH);
1836
return flush == Z_FINISH ? finish_done : block_done;
added
removed
tests/zlib/deflate.c
