blob: 9a20b7034483c32cdc528f2c14277796f1f1d09d [file] [log] [blame]
Lei Wen8a5f34e2012-09-28 04:26:42 +00001/* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2010 Jean-loup Gailly and Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
5
6/*
7 * ALGORITHM
8 *
9 * The "deflation" process depends on being able to identify portions
10 * of the input text which are identical to earlier input (within a
11 * sliding window trailing behind the input currently being processed).
12 *
13 * The most straightforward technique turns out to be the fastest for
14 * most input files: try all possible matches and select the longest.
15 * The key feature of this algorithm is that insertions into the string
16 * dictionary are very simple and thus fast, and deletions are avoided
17 * completely. Insertions are performed at each input character, whereas
18 * string matches are performed only when the previous match ends. So it
19 * is preferable to spend more time in matches to allow very fast string
20 * insertions and avoid deletions. The matching algorithm for small
21 * strings is inspired from that of Rabin & Karp. A brute force approach
22 * is used to find longer strings when a small match has been found.
23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
24 * (by Leonid Broukhis).
25 * A previous version of this file used a more sophisticated algorithm
26 * (by Fiala and Greene) which is guaranteed to run in linear amortized
27 * time, but has a larger average cost, uses more memory and is patented.
28 * However the F&G algorithm may be faster for some highly redundant
29 * files if the parameter max_chain_length (described below) is too large.
30 *
31 * ACKNOWLEDGEMENTS
32 *
33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
34 * I found it in 'freeze' written by Leonid Broukhis.
35 * Thanks to many people for bug reports and testing.
36 *
37 * REFERENCES
38 *
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40 * Available in http://www.ietf.org/rfc/rfc1951.txt
41 *
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
44 *
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
47 *
48 */
49
50/* @(#) $Id$ */
51
52#include "deflate.h"
53
54const char deflate_copyright[] =
55 " deflate 1.2.5 Copyright 1995-2010 Jean-loup Gailly and Mark Adler ";
56/*
57 If you use the zlib library in a product, an acknowledgment is welcome
58 in the documentation of your product. If for some reason you cannot
59 include such an acknowledgment, I would appreciate that you keep this
60 copyright string in the executable of your product.
61 */
62
63/* ===========================================================================
64 * Function prototypes.
65 */
66typedef 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
73typedef block_state (*compress_func) OF((deflate_state *s, int flush));
74/* Compression function. Returns the block state after the call. */
75
76local void fill_window OF((deflate_state *s));
77local block_state deflate_stored OF((deflate_state *s, int flush));
78local block_state deflate_fast OF((deflate_state *s, int flush));
79#ifndef FASTEST
80local block_state deflate_slow OF((deflate_state *s, int flush));
81#endif
82local block_state deflate_rle OF((deflate_state *s, int flush));
83local block_state deflate_huff OF((deflate_state *s, int flush));
84local void lm_init OF((deflate_state *s));
85local void putShortMSB OF((deflate_state *s, uInt b));
86local void flush_pending OF((z_streamp strm));
87local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
88#ifdef ASMV
89 void match_init OF((void)); /* asm code initialization */
90 uInt longest_match OF((deflate_state *s, IPos cur_match));
91#else
92local uInt longest_match OF((deflate_state *s, IPos cur_match));
93#endif
94
95#ifdef DEBUG
96local void check_match OF((deflate_state *s, IPos start, IPos match,
97 int length));
98#endif
99
100/* ===========================================================================
101 * Local data
102 */
103
104#define NIL 0
105/* Tail of hash chains */
106
107#ifndef TOO_FAR
108# define TOO_FAR 4096
109#endif
110/* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
111
112/* Values for max_lazy_match, good_match and max_chain_length, depending on
113 * the desired pack level (0..9). The values given below have been tuned to
114 * exclude worst case performance for pathological files. Better values may be
115 * found for specific files.
116 */
117typedef struct config_s {
118 ush good_length; /* reduce lazy search above this match length */
119 ush max_lazy; /* do not perform lazy search above this match length */
120 ush nice_length; /* quit search above this match length */
121 ush max_chain;
122 compress_func func;
123} config;
124
125#ifdef FASTEST
126local const config configuration_table[2] = {
127/* good lazy nice chain */
128/* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
129/* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
130#else
131local const config configuration_table[10] = {
132/* good lazy nice chain */
133/* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
134/* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
135/* 2 */ {4, 5, 16, 8, deflate_fast},
136/* 3 */ {4, 6, 32, 32, deflate_fast},
137
138/* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
139/* 5 */ {8, 16, 32, 32, deflate_slow},
140/* 6 */ {8, 16, 128, 128, deflate_slow},
141/* 7 */ {8, 32, 128, 256, deflate_slow},
142/* 8 */ {32, 128, 258, 1024, deflate_slow},
143/* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
144#endif
145
146/* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
147 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
148 * meaning.
149 */
150
151#define EQUAL 0
152/* result of memcmp for equal strings */
153
154#ifndef NO_DUMMY_DECL
155struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
156#endif
157
158/* ===========================================================================
159 * Update a hash value with the given input byte
160 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
161 * input characters, so that a running hash key can be computed from the
162 * previous key instead of complete recalculation each time.
163 */
164#define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
165
166
167/* ===========================================================================
168 * Insert string str in the dictionary and set match_head to the previous head
169 * of the hash chain (the most recent string with same hash key). Return
170 * the previous length of the hash chain.
171 * If this file is compiled with -DFASTEST, the compression level is forced
172 * to 1, and no hash chains are maintained.
173 * IN assertion: all calls to to INSERT_STRING are made with consecutive
174 * input characters and the first MIN_MATCH bytes of str are valid
175 * (except for the last MIN_MATCH-1 bytes of the input file).
176 */
177#ifdef FASTEST
178#define INSERT_STRING(s, str, match_head) \
179 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
180 match_head = s->head[s->ins_h], \
181 s->head[s->ins_h] = (Pos)(str))
182#else
183#define INSERT_STRING(s, str, match_head) \
184 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
185 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
186 s->head[s->ins_h] = (Pos)(str))
187#endif
188
189/* ===========================================================================
190 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
191 * prev[] will be initialized on the fly.
192 */
193#define CLEAR_HASH(s) \
194 s->head[s->hash_size-1] = NIL; \
195 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
196
197/* ========================================================================= */
198int ZEXPORT deflateInit_(strm, level, version, stream_size)
199 z_streamp strm;
200 int level;
201 const char *version;
202 int stream_size;
203{
204 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
205 Z_DEFAULT_STRATEGY, version, stream_size);
206 /* To do: ignore strm->next_in if we use it as window */
207}
208
209/* ========================================================================= */
210int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
211 version, stream_size)
212 z_streamp strm;
213 int level;
214 int method;
215 int windowBits;
216 int memLevel;
217 int strategy;
218 const char *version;
219 int stream_size;
220{
221 deflate_state *s;
222 int wrap = 1;
223 static const char my_version[] = ZLIB_VERSION;
224
225 ushf *overlay;
226 /* We overlay pending_buf and d_buf+l_buf. This works since the average
227 * output size for (length,distance) codes is <= 24 bits.
228 */
229
230 if (version == Z_NULL || version[0] != my_version[0] ||
231 stream_size != sizeof(z_stream)) {
232 return Z_VERSION_ERROR;
233 }
234 if (strm == Z_NULL) return Z_STREAM_ERROR;
235
236 strm->msg = Z_NULL;
237 if (strm->zalloc == (alloc_func)0) {
238 strm->zalloc = zcalloc;
239 strm->opaque = (voidpf)0;
240 }
241 if (strm->zfree == (free_func)0) strm->zfree = zcfree;
242
243#ifdef FASTEST
244 if (level != 0) level = 1;
245#else
246 if (level == Z_DEFAULT_COMPRESSION) level = 6;
247#endif
248
249 if (windowBits < 0) { /* suppress zlib wrapper */
250 wrap = 0;
251 windowBits = -windowBits;
252 }
253#ifdef GZIP
254 else if (windowBits > 15) {
255 wrap = 2; /* write gzip wrapper instead */
256 windowBits -= 16;
257 }
258#endif
259 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
260 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
261 strategy < 0 || strategy > Z_FIXED) {
262 return Z_STREAM_ERROR;
263 }
264 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
265 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
266 if (s == Z_NULL) return Z_MEM_ERROR;
267 strm->state = (struct internal_state FAR *)s;
268 s->strm = strm;
269
270 s->wrap = wrap;
271 s->gzhead = Z_NULL;
272 s->w_bits = windowBits;
273 s->w_size = 1 << s->w_bits;
274 s->w_mask = s->w_size - 1;
275
276 s->hash_bits = memLevel + 7;
277 s->hash_size = 1 << s->hash_bits;
278 s->hash_mask = s->hash_size - 1;
279 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
280
281 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
282 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
283 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
284
285 s->high_water = 0; /* nothing written to s->window yet */
286
287 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
288
289 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
290 s->pending_buf = (uchf *) overlay;
291 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
292
293 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
294 s->pending_buf == Z_NULL) {
295 s->status = FINISH_STATE;
296 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
297 deflateEnd (strm);
298 return Z_MEM_ERROR;
299 }
300 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
301 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
302
303 s->level = level;
304 s->strategy = strategy;
305 s->method = (Byte)method;
306
307 return deflateReset(strm);
308}
309
310/* ========================================================================= */
311int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
312 z_streamp strm;
313 const Bytef *dictionary;
314 uInt dictLength;
315{
316 deflate_state *s;
317 uInt length = dictLength;
318 uInt n;
319 IPos hash_head = 0;
320
321 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
322 strm->state->wrap == 2 ||
323 (strm->state->wrap == 1 && strm->state->status != INIT_STATE))
324 return Z_STREAM_ERROR;
325
326 s = strm->state;
327 if (s->wrap)
328 strm->adler = adler32(strm->adler, dictionary, dictLength);
329
330 if (length < MIN_MATCH) return Z_OK;
331 if (length > s->w_size) {
332 length = s->w_size;
333 dictionary += dictLength - length; /* use the tail of the dictionary */
334 }
335 zmemcpy(s->window, dictionary, length);
336 s->strstart = length;
337 s->block_start = (long)length;
338
339 /* Insert all strings in the hash table (except for the last two bytes).
340 * s->lookahead stays null, so s->ins_h will be recomputed at the next
341 * call of fill_window.
342 */
343 s->ins_h = s->window[0];
344 UPDATE_HASH(s, s->ins_h, s->window[1]);
345 for (n = 0; n <= length - MIN_MATCH; n++) {
346 INSERT_STRING(s, n, hash_head);
347 }
348 if (hash_head) hash_head = 0; /* to make compiler happy */
349 return Z_OK;
350}
351
352/* ========================================================================= */
353int ZEXPORT deflateReset (strm)
354 z_streamp strm;
355{
356 deflate_state *s;
357
358 if (strm == Z_NULL || strm->state == Z_NULL ||
359 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
360 return Z_STREAM_ERROR;
361 }
362
363 strm->total_in = strm->total_out = 0;
364 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
365 strm->data_type = Z_UNKNOWN;
366
367 s = (deflate_state *)strm->state;
368 s->pending = 0;
369 s->pending_out = s->pending_buf;
370
371 if (s->wrap < 0) {
372 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
373 }
374 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
375 strm->adler =
376#ifdef GZIP
377 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
378#endif
379 adler32(0L, Z_NULL, 0);
380 s->last_flush = Z_NO_FLUSH;
381
382 _tr_init(s);
383 lm_init(s);
384
385 return Z_OK;
386}
387
388/* ========================================================================= */
389int ZEXPORT deflateSetHeader (strm, head)
390 z_streamp strm;
391 gz_headerp head;
392{
393 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
394 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
395 strm->state->gzhead = head;
396 return Z_OK;
397}
398
399/* ========================================================================= */
400int ZEXPORT deflatePrime (strm, bits, value)
401 z_streamp strm;
402 int bits;
403 int value;
404{
405 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
406 strm->state->bi_valid = bits;
407 strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));
408 return Z_OK;
409}
410
411/* ========================================================================= */
412int ZEXPORT deflateParams(strm, level, strategy)
413 z_streamp strm;
414 int level;
415 int strategy;
416{
417 deflate_state *s;
418 compress_func func;
419 int err = Z_OK;
420
421 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
422 s = strm->state;
423
424#ifdef FASTEST
425 if (level != 0) level = 1;
426#else
427 if (level == Z_DEFAULT_COMPRESSION) level = 6;
428#endif
429 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
430 return Z_STREAM_ERROR;
431 }
432 func = configuration_table[s->level].func;
433
434 if ((strategy != s->strategy || func != configuration_table[level].func) &&
435 strm->total_in != 0) {
436 /* Flush the last buffer: */
437 err = deflate(strm, Z_BLOCK);
438 }
439 if (s->level != level) {
440 s->level = level;
441 s->max_lazy_match = configuration_table[level].max_lazy;
442 s->good_match = configuration_table[level].good_length;
443 s->nice_match = configuration_table[level].nice_length;
444 s->max_chain_length = configuration_table[level].max_chain;
445 }
446 s->strategy = strategy;
447 return err;
448}
449
450/* ========================================================================= */
451int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
452 z_streamp strm;
453 int good_length;
454 int max_lazy;
455 int nice_length;
456 int max_chain;
457{
458 deflate_state *s;
459
460 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
461 s = strm->state;
462 s->good_match = good_length;
463 s->max_lazy_match = max_lazy;
464 s->nice_match = nice_length;
465 s->max_chain_length = max_chain;
466 return Z_OK;
467}
468
469/* =========================================================================
470 * For the default windowBits of 15 and memLevel of 8, this function returns
471 * a close to exact, as well as small, upper bound on the compressed size.
472 * They are coded as constants here for a reason--if the #define's are
473 * changed, then this function needs to be changed as well. The return
474 * value for 15 and 8 only works for those exact settings.
475 *
476 * For any setting other than those defaults for windowBits and memLevel,
477 * the value returned is a conservative worst case for the maximum expansion
478 * resulting from using fixed blocks instead of stored blocks, which deflate
479 * can emit on compressed data for some combinations of the parameters.
480 *
481 * This function could be more sophisticated to provide closer upper bounds for
482 * every combination of windowBits and memLevel. But even the conservative
483 * upper bound of about 14% expansion does not seem onerous for output buffer
484 * allocation.
485 */
486uLong ZEXPORT deflateBound(strm, sourceLen)
487 z_streamp strm;
488 uLong sourceLen;
489{
490 deflate_state *s;
491 uLong complen, wraplen;
492 Bytef *str;
493
494 /* conservative upper bound for compressed data */
495 complen = sourceLen +
496 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
497
498 /* if can't get parameters, return conservative bound plus zlib wrapper */
499 if (strm == Z_NULL || strm->state == Z_NULL)
500 return complen + 6;
501
502 /* compute wrapper length */
503 s = strm->state;
504 switch (s->wrap) {
505 case 0: /* raw deflate */
506 wraplen = 0;
507 break;
508 case 1: /* zlib wrapper */
509 wraplen = 6 + (s->strstart ? 4 : 0);
510 break;
511 case 2: /* gzip wrapper */
512 wraplen = 18;
513 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
514 if (s->gzhead->extra != Z_NULL)
515 wraplen += 2 + s->gzhead->extra_len;
516 str = s->gzhead->name;
517 if (str != Z_NULL)
518 do {
519 wraplen++;
520 } while (*str++);
521 str = s->gzhead->comment;
522 if (str != Z_NULL)
523 do {
524 wraplen++;
525 } while (*str++);
526 if (s->gzhead->hcrc)
527 wraplen += 2;
528 }
529 break;
530 default: /* for compiler happiness */
531 wraplen = 6;
532 }
533
534 /* if not default parameters, return conservative bound */
535 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
536 return complen + wraplen;
537
538 /* default settings: return tight bound for that case */
539 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
540 (sourceLen >> 25) + 13 - 6 + wraplen;
541}
542
543/* =========================================================================
544 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
545 * IN assertion: the stream state is correct and there is enough room in
546 * pending_buf.
547 */
548local void putShortMSB (s, b)
549 deflate_state *s;
550 uInt b;
551{
552 put_byte(s, (Byte)(b >> 8));
553 put_byte(s, (Byte)(b & 0xff));
554}
555
556/* =========================================================================
557 * Flush as much pending output as possible. All deflate() output goes
558 * through this function so some applications may wish to modify it
559 * to avoid allocating a large strm->next_out buffer and copying into it.
560 * (See also read_buf()).
561 */
562local void flush_pending(strm)
563 z_streamp strm;
564{
565 unsigned len = strm->state->pending;
566
567 if (len > strm->avail_out) len = strm->avail_out;
568 if (len == 0) return;
569
570 zmemcpy(strm->next_out, strm->state->pending_out, len);
571 strm->next_out += len;
572 strm->state->pending_out += len;
573 strm->total_out += len;
574 strm->avail_out -= len;
575 strm->state->pending -= len;
576 if (strm->state->pending == 0) {
577 strm->state->pending_out = strm->state->pending_buf;
578 }
579}
580
581/* ========================================================================= */
582int ZEXPORT deflate (strm, flush)
583 z_streamp strm;
584 int flush;
585{
586 int old_flush; /* value of flush param for previous deflate call */
587 deflate_state *s;
588
589 if (strm == Z_NULL || strm->state == Z_NULL ||
590 flush > Z_BLOCK || flush < 0) {
591 return Z_STREAM_ERROR;
592 }
593 s = strm->state;
594
Lei Wen869c2ab2012-09-28 04:26:45 +0000595 if (s->status == FINISH_STATE && flush != Z_FINISH) {
Lei Wen8a5f34e2012-09-28 04:26:42 +0000596 ERR_RETURN(strm, Z_STREAM_ERROR);
597 }
598 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
599
600 s->strm = strm; /* just in case */
601 old_flush = s->last_flush;
602 s->last_flush = flush;
603
604 /* Write the header */
605 if (s->status == INIT_STATE) {
606#ifdef GZIP
607 if (s->wrap == 2) {
608 strm->adler = crc32(0L, Z_NULL, 0);
609 put_byte(s, 31);
610 put_byte(s, 139);
611 put_byte(s, 8);
612 if (s->gzhead == Z_NULL) {
613 put_byte(s, 0);
614 put_byte(s, 0);
615 put_byte(s, 0);
616 put_byte(s, 0);
617 put_byte(s, 0);
618 put_byte(s, s->level == 9 ? 2 :
619 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
620 4 : 0));
621 put_byte(s, OS_CODE);
622 s->status = BUSY_STATE;
623 }
624 else {
625 put_byte(s, (s->gzhead->text ? 1 : 0) +
626 (s->gzhead->hcrc ? 2 : 0) +
627 (s->gzhead->extra == Z_NULL ? 0 : 4) +
628 (s->gzhead->name == Z_NULL ? 0 : 8) +
629 (s->gzhead->comment == Z_NULL ? 0 : 16)
630 );
631 put_byte(s, (Byte)(s->gzhead->time & 0xff));
632 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
633 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
634 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
635 put_byte(s, s->level == 9 ? 2 :
636 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
637 4 : 0));
638 put_byte(s, s->gzhead->os & 0xff);
639 if (s->gzhead->extra != Z_NULL) {
640 put_byte(s, s->gzhead->extra_len & 0xff);
641 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
642 }
643 if (s->gzhead->hcrc)
644 strm->adler = crc32(strm->adler, s->pending_buf,
645 s->pending);
646 s->gzindex = 0;
647 s->status = EXTRA_STATE;
648 }
649 }
650 else
651#endif
652 {
653 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
654 uInt level_flags;
655
656 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
657 level_flags = 0;
658 else if (s->level < 6)
659 level_flags = 1;
660 else if (s->level == 6)
661 level_flags = 2;
662 else
663 level_flags = 3;
664 header |= (level_flags << 6);
665 if (s->strstart != 0) header |= PRESET_DICT;
666 header += 31 - (header % 31);
667
668 s->status = BUSY_STATE;
669 putShortMSB(s, header);
670
671 /* Save the adler32 of the preset dictionary: */
672 if (s->strstart != 0) {
673 putShortMSB(s, (uInt)(strm->adler >> 16));
674 putShortMSB(s, (uInt)(strm->adler & 0xffff));
675 }
676 strm->adler = adler32(0L, Z_NULL, 0);
677 }
678 }
679#ifdef GZIP
680 if (s->status == EXTRA_STATE) {
681 if (s->gzhead->extra != Z_NULL) {
682 uInt beg = s->pending; /* start of bytes to update crc */
683
684 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
685 if (s->pending == s->pending_buf_size) {
686 if (s->gzhead->hcrc && s->pending > beg)
687 strm->adler = crc32(strm->adler, s->pending_buf + beg,
688 s->pending - beg);
689 flush_pending(strm);
690 beg = s->pending;
691 if (s->pending == s->pending_buf_size)
692 break;
693 }
694 put_byte(s, s->gzhead->extra[s->gzindex]);
695 s->gzindex++;
696 }
697 if (s->gzhead->hcrc && s->pending > beg)
698 strm->adler = crc32(strm->adler, s->pending_buf + beg,
699 s->pending - beg);
700 if (s->gzindex == s->gzhead->extra_len) {
701 s->gzindex = 0;
702 s->status = NAME_STATE;
703 }
704 }
705 else
706 s->status = NAME_STATE;
707 }
708 if (s->status == NAME_STATE) {
709 if (s->gzhead->name != Z_NULL) {
710 uInt beg = s->pending; /* start of bytes to update crc */
711 int val;
712
713 do {
714 if (s->pending == s->pending_buf_size) {
715 if (s->gzhead->hcrc && s->pending > beg)
716 strm->adler = crc32(strm->adler, s->pending_buf + beg,
717 s->pending - beg);
718 flush_pending(strm);
719 beg = s->pending;
720 if (s->pending == s->pending_buf_size) {
721 val = 1;
722 break;
723 }
724 }
725 val = s->gzhead->name[s->gzindex++];
726 put_byte(s, val);
727 } while (val != 0);
728 if (s->gzhead->hcrc && s->pending > beg)
729 strm->adler = crc32(strm->adler, s->pending_buf + beg,
730 s->pending - beg);
731 if (val == 0) {
732 s->gzindex = 0;
733 s->status = COMMENT_STATE;
734 }
735 }
736 else
737 s->status = COMMENT_STATE;
738 }
739 if (s->status == COMMENT_STATE) {
740 if (s->gzhead->comment != Z_NULL) {
741 uInt beg = s->pending; /* start of bytes to update crc */
742 int val;
743
744 do {
745 if (s->pending == s->pending_buf_size) {
746 if (s->gzhead->hcrc && s->pending > beg)
747 strm->adler = crc32(strm->adler, s->pending_buf + beg,
748 s->pending - beg);
749 flush_pending(strm);
750 beg = s->pending;
751 if (s->pending == s->pending_buf_size) {
752 val = 1;
753 break;
754 }
755 }
756 val = s->gzhead->comment[s->gzindex++];
757 put_byte(s, val);
758 } while (val != 0);
759 if (s->gzhead->hcrc && s->pending > beg)
760 strm->adler = crc32(strm->adler, s->pending_buf + beg,
761 s->pending - beg);
762 if (val == 0)
763 s->status = HCRC_STATE;
764 }
765 else
766 s->status = HCRC_STATE;
767 }
768 if (s->status == HCRC_STATE) {
769 if (s->gzhead->hcrc) {
770 if (s->pending + 2 > s->pending_buf_size)
771 flush_pending(strm);
772 if (s->pending + 2 <= s->pending_buf_size) {
773 put_byte(s, (Byte)(strm->adler & 0xff));
774 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
775 strm->adler = crc32(0L, Z_NULL, 0);
776 s->status = BUSY_STATE;
777 }
778 }
779 else
780 s->status = BUSY_STATE;
781 }
782#endif
783
784 /* Flush as much pending output as possible */
785 if (s->pending != 0) {
786 flush_pending(strm);
787 if (strm->avail_out == 0) {
788 /* Since avail_out is 0, deflate will be called again with
789 * more output space, but possibly with both pending and
790 * avail_in equal to zero. There won't be anything to do,
791 * but this is not an error situation so make sure we
792 * return OK instead of BUF_ERROR at next call of deflate:
793 */
794 s->last_flush = -1;
795 return Z_OK;
796 }
797
798 /* Make sure there is something to do and avoid duplicate consecutive
799 * flushes. For repeated and useless calls with Z_FINISH, we keep
800 * returning Z_STREAM_END instead of Z_BUF_ERROR.
801 */
802 } else if (strm->avail_in == 0 && flush <= old_flush &&
803 flush != Z_FINISH) {
804 ERR_RETURN(strm, Z_BUF_ERROR);
805 }
806
807 /* User must not provide more input after the first FINISH: */
808 if (s->status == FINISH_STATE && strm->avail_in != 0) {
809 ERR_RETURN(strm, Z_BUF_ERROR);
810 }
811
812 /* Start a new block or continue the current one.
813 */
814 if (strm->avail_in != 0 || s->lookahead != 0 ||
815 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
816 block_state bstate;
817
818 bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
819 (s->strategy == Z_RLE ? deflate_rle(s, flush) :
820 (*(configuration_table[s->level].func))(s, flush));
821
822 if (bstate == finish_started || bstate == finish_done) {
823 s->status = FINISH_STATE;
824 }
825 if (bstate == need_more || bstate == finish_started) {
826 if (strm->avail_out == 0) {
827 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
828 }
829 return Z_OK;
830 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
831 * of deflate should use the same flush parameter to make sure
832 * that the flush is complete. So we don't have to output an
833 * empty block here, this will be done at next call. This also
834 * ensures that for a very small output buffer, we emit at most
835 * one empty block.
836 */
837 }
838 if (bstate == block_done) {
839 if (flush == Z_PARTIAL_FLUSH) {
840 _tr_align(s);
841 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
842 _tr_stored_block(s, (char*)0, 0L, 0);
843 /* For a full flush, this empty block will be recognized
844 * as a special marker by inflate_sync().
845 */
846 if (flush == Z_FULL_FLUSH) {
847 CLEAR_HASH(s); /* forget history */
848 if (s->lookahead == 0) {
849 s->strstart = 0;
850 s->block_start = 0L;
851 }
852 }
853 }
854 flush_pending(strm);
855 if (strm->avail_out == 0) {
856 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
857 return Z_OK;
858 }
859 }
860 }
861 Assert(strm->avail_out > 0, "bug2");
862
863 if (flush != Z_FINISH) return Z_OK;
864 if (s->wrap <= 0) return Z_STREAM_END;
865
866 /* Write the trailer */
867#ifdef GZIP
868 if (s->wrap == 2) {
869 put_byte(s, (Byte)(strm->adler & 0xff));
870 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
871 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
872 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
873 put_byte(s, (Byte)(strm->total_in & 0xff));
874 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
875 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
876 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
877 }
878 else
879#endif
880 {
881 putShortMSB(s, (uInt)(strm->adler >> 16));
882 putShortMSB(s, (uInt)(strm->adler & 0xffff));
883 }
884 flush_pending(strm);
885 /* If avail_out is zero, the application will call deflate again
886 * to flush the rest.
887 */
888 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
889 return s->pending != 0 ? Z_OK : Z_STREAM_END;
890}
891
892/* ========================================================================= */
893int ZEXPORT deflateEnd (strm)
894 z_streamp strm;
895{
896 int status;
897
898 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
899
900 status = strm->state->status;
901 if (status != INIT_STATE &&
902 status != EXTRA_STATE &&
903 status != NAME_STATE &&
904 status != COMMENT_STATE &&
905 status != HCRC_STATE &&
906 status != BUSY_STATE &&
907 status != FINISH_STATE) {
908 return Z_STREAM_ERROR;
909 }
910
911 /* Deallocate in reverse order of allocations: */
912 TRY_FREE(strm, strm->state->pending_buf);
913 TRY_FREE(strm, strm->state->head);
914 TRY_FREE(strm, strm->state->prev);
915 TRY_FREE(strm, strm->state->window);
916
917 ZFREE(strm, strm->state);
918 strm->state = Z_NULL;
919
920 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
921}
922
923/* =========================================================================
924 * Copy the source state to the destination state.
925 * To simplify the source, this is not supported for 16-bit MSDOS (which
926 * doesn't have enough memory anyway to duplicate compression states).
927 */
928int ZEXPORT deflateCopy (dest, source)
929 z_streamp dest;
930 z_streamp source;
931{
932#ifdef MAXSEG_64K
933 return Z_STREAM_ERROR;
934#else
935 deflate_state *ds;
936 deflate_state *ss;
937 ushf *overlay;
938
939
940 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
941 return Z_STREAM_ERROR;
942 }
943
944 ss = source->state;
945
946 zmemcpy(dest, source, sizeof(z_stream));
947
948 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
949 if (ds == Z_NULL) return Z_MEM_ERROR;
950 dest->state = (struct internal_state FAR *) ds;
951 zmemcpy(ds, ss, sizeof(deflate_state));
952 ds->strm = dest;
953
954 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
955 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
956 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
957 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
958 ds->pending_buf = (uchf *) overlay;
959
960 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
961 ds->pending_buf == Z_NULL) {
962 deflateEnd (dest);
963 return Z_MEM_ERROR;
964 }
965 /* following zmemcpy do not work for 16-bit MSDOS */
966 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
967 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
968 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
969 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
970
971 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
972 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
973 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
974
975 ds->l_desc.dyn_tree = ds->dyn_ltree;
976 ds->d_desc.dyn_tree = ds->dyn_dtree;
977 ds->bl_desc.dyn_tree = ds->bl_tree;
978
979 return Z_OK;
980#endif /* MAXSEG_64K */
981}
982
983/* ===========================================================================
984 * Read a new buffer from the current input stream, update the adler32
985 * and total number of bytes read. All deflate() input goes through
986 * this function so some applications may wish to modify it to avoid
987 * allocating a large strm->next_in buffer and copying from it.
988 * (See also flush_pending()).
989 */
990local int read_buf(strm, buf, size)
991 z_streamp strm;
992 Bytef *buf;
993 unsigned size;
994{
995 unsigned len = strm->avail_in;
996
997 if (len > size) len = size;
998 if (len == 0) return 0;
999
1000 strm->avail_in -= len;
1001
1002 if (strm->state->wrap == 1) {
1003 strm->adler = adler32(strm->adler, strm->next_in, len);
1004 }
1005#ifdef GZIP
1006 else if (strm->state->wrap == 2) {
1007 strm->adler = crc32(strm->adler, strm->next_in, len);
1008 }
1009#endif
1010 zmemcpy(buf, strm->next_in, len);
1011 strm->next_in += len;
1012 strm->total_in += len;
1013
1014 return (int)len;
1015}
1016
1017/* ===========================================================================
1018 * Initialize the "longest match" routines for a new zlib stream
1019 */
1020local void lm_init (s)
1021 deflate_state *s;
1022{
1023 s->window_size = (ulg)2L*s->w_size;
1024
1025 CLEAR_HASH(s);
1026
1027 /* Set the default configuration parameters:
1028 */
1029 s->max_lazy_match = configuration_table[s->level].max_lazy;
1030 s->good_match = configuration_table[s->level].good_length;
1031 s->nice_match = configuration_table[s->level].nice_length;
1032 s->max_chain_length = configuration_table[s->level].max_chain;
1033
1034 s->strstart = 0;
1035 s->block_start = 0L;
1036 s->lookahead = 0;
1037 s->match_length = s->prev_length = MIN_MATCH-1;
1038 s->match_available = 0;
1039 s->ins_h = 0;
1040#ifndef FASTEST
1041#ifdef ASMV
1042 match_init(); /* initialize the asm code */
1043#endif
1044#endif
1045}
1046
1047#ifndef FASTEST
1048/* ===========================================================================
1049 * Set match_start to the longest match starting at the given string and
1050 * return its length. Matches shorter or equal to prev_length are discarded,
1051 * in which case the result is equal to prev_length and match_start is
1052 * garbage.
1053 * IN assertions: cur_match is the head of the hash chain for the current
1054 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1055 * OUT assertion: the match length is not greater than s->lookahead.
1056 */
1057#ifndef ASMV
1058/* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1059 * match.S. The code will be functionally equivalent.
1060 */
1061local uInt longest_match(s, cur_match)
1062 deflate_state *s;
1063 IPos cur_match; /* current match */
1064{
1065 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1066 register Bytef *scan = s->window + s->strstart; /* current string */
1067 register Bytef *match; /* matched string */
1068 register int len; /* length of current match */
1069 int best_len = s->prev_length; /* best match length so far */
1070 int nice_match = s->nice_match; /* stop if match long enough */
1071 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1072 s->strstart - (IPos)MAX_DIST(s) : NIL;
1073 /* Stop when cur_match becomes <= limit. To simplify the code,
1074 * we prevent matches with the string of window index 0.
1075 */
1076 Posf *prev = s->prev;
1077 uInt wmask = s->w_mask;
1078
1079#ifdef UNALIGNED_OK
1080 /* Compare two bytes at a time. Note: this is not always beneficial.
1081 * Try with and without -DUNALIGNED_OK to check.
1082 */
1083 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1084 register ush scan_start = *(ushf*)scan;
1085 register ush scan_end = *(ushf*)(scan+best_len-1);
1086#else
1087 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1088 register Byte scan_end1 = scan[best_len-1];
1089 register Byte scan_end = scan[best_len];
1090#endif
1091
1092 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1093 * It is easy to get rid of this optimization if necessary.
1094 */
1095 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1096
1097 /* Do not waste too much time if we already have a good match: */
1098 if (s->prev_length >= s->good_match) {
1099 chain_length >>= 2;
1100 }
1101 /* Do not look for matches beyond the end of the input. This is necessary
1102 * to make deflate deterministic.
1103 */
1104 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1105
1106 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1107
1108 do {
1109 Assert(cur_match < s->strstart, "no future");
1110 match = s->window + cur_match;
1111
1112 /* Skip to next match if the match length cannot increase
1113 * or if the match length is less than 2. Note that the checks below
1114 * for insufficient lookahead only occur occasionally for performance
1115 * reasons. Therefore uninitialized memory will be accessed, and
1116 * conditional jumps will be made that depend on those values.
1117 * However the length of the match is limited to the lookahead, so
1118 * the output of deflate is not affected by the uninitialized values.
1119 */
1120#if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1121 /* This code assumes sizeof(unsigned short) == 2. Do not use
1122 * UNALIGNED_OK if your compiler uses a different size.
1123 */
1124 if (*(ushf*)(match+best_len-1) != scan_end ||
1125 *(ushf*)match != scan_start) continue;
1126
1127 /* It is not necessary to compare scan[2] and match[2] since they are
1128 * always equal when the other bytes match, given that the hash keys
1129 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1130 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1131 * lookahead only every 4th comparison; the 128th check will be made
1132 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1133 * necessary to put more guard bytes at the end of the window, or
1134 * to check more often for insufficient lookahead.
1135 */
1136 Assert(scan[2] == match[2], "scan[2]?");
1137 scan++, match++;
1138 do {
1139 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1140 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1141 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1142 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1143 scan < strend);
1144 /* The funny "do {}" generates better code on most compilers */
1145
1146 /* Here, scan <= window+strstart+257 */
1147 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1148 if (*scan == *match) scan++;
1149
1150 len = (MAX_MATCH - 1) - (int)(strend-scan);
1151 scan = strend - (MAX_MATCH-1);
1152
1153#else /* UNALIGNED_OK */
1154
1155 if (match[best_len] != scan_end ||
1156 match[best_len-1] != scan_end1 ||
1157 *match != *scan ||
1158 *++match != scan[1]) continue;
1159
1160 /* The check at best_len-1 can be removed because it will be made
1161 * again later. (This heuristic is not always a win.)
1162 * It is not necessary to compare scan[2] and match[2] since they
1163 * are always equal when the other bytes match, given that
1164 * the hash keys are equal and that HASH_BITS >= 8.
1165 */
1166 scan += 2, match++;
1167 Assert(*scan == *match, "match[2]?");
1168
1169 /* We check for insufficient lookahead only every 8th comparison;
1170 * the 256th check will be made at strstart+258.
1171 */
1172 do {
1173 } while (*++scan == *++match && *++scan == *++match &&
1174 *++scan == *++match && *++scan == *++match &&
1175 *++scan == *++match && *++scan == *++match &&
1176 *++scan == *++match && *++scan == *++match &&
1177 scan < strend);
1178
1179 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1180
1181 len = MAX_MATCH - (int)(strend - scan);
1182 scan = strend - MAX_MATCH;
1183
1184#endif /* UNALIGNED_OK */
1185
1186 if (len > best_len) {
1187 s->match_start = cur_match;
1188 best_len = len;
1189 if (len >= nice_match) break;
1190#ifdef UNALIGNED_OK
1191 scan_end = *(ushf*)(scan+best_len-1);
1192#else
1193 scan_end1 = scan[best_len-1];
1194 scan_end = scan[best_len];
1195#endif
1196 }
1197 } while ((cur_match = prev[cur_match & wmask]) > limit
1198 && --chain_length != 0);
1199
1200 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1201 return s->lookahead;
1202}
1203#endif /* ASMV */
1204
1205#else /* FASTEST */
1206
1207/* ---------------------------------------------------------------------------
1208 * Optimized version for FASTEST only
1209 */
1210local uInt longest_match(s, cur_match)
1211 deflate_state *s;
1212 IPos cur_match; /* current match */
1213{
1214 register Bytef *scan = s->window + s->strstart; /* current string */
1215 register Bytef *match; /* matched string */
1216 register int len; /* length of current match */
1217 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1218
1219 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1220 * It is easy to get rid of this optimization if necessary.
1221 */
1222 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1223
1224 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1225
1226 Assert(cur_match < s->strstart, "no future");
1227
1228 match = s->window + cur_match;
1229
1230 /* Return failure if the match length is less than 2:
1231 */
1232 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1233
1234 /* The check at best_len-1 can be removed because it will be made
1235 * again later. (This heuristic is not always a win.)
1236 * It is not necessary to compare scan[2] and match[2] since they
1237 * are always equal when the other bytes match, given that
1238 * the hash keys are equal and that HASH_BITS >= 8.
1239 */
1240 scan += 2, match += 2;
1241 Assert(*scan == *match, "match[2]?");
1242
1243 /* We check for insufficient lookahead only every 8th comparison;
1244 * the 256th check will be made at strstart+258.
1245 */
1246 do {
1247 } while (*++scan == *++match && *++scan == *++match &&
1248 *++scan == *++match && *++scan == *++match &&
1249 *++scan == *++match && *++scan == *++match &&
1250 *++scan == *++match && *++scan == *++match &&
1251 scan < strend);
1252
1253 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1254
1255 len = MAX_MATCH - (int)(strend - scan);
1256
1257 if (len < MIN_MATCH) return MIN_MATCH - 1;
1258
1259 s->match_start = cur_match;
1260 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1261}
1262
1263#endif /* FASTEST */
1264
1265#ifdef DEBUG
1266/* ===========================================================================
1267 * Check that the match at match_start is indeed a match.
1268 */
1269local void check_match(s, start, match, length)
1270 deflate_state *s;
1271 IPos start, match;
1272 int length;
1273{
1274 /* check that the match is indeed a match */
1275 if (zmemcmp(s->window + match,
1276 s->window + start, length) != EQUAL) {
1277 fprintf(stderr, " start %u, match %u, length %d\n",
1278 start, match, length);
1279 do {
1280 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1281 } while (--length != 0);
1282 z_error("invalid match");
1283 }
1284 if (z_verbose > 1) {
1285 fprintf(stderr,"\\[%d,%d]", start-match, length);
1286 do { putc(s->window[start++], stderr); } while (--length != 0);
1287 }
1288}
1289#else
1290# define check_match(s, start, match, length)
1291#endif /* DEBUG */
1292
1293/* ===========================================================================
1294 * Fill the window when the lookahead becomes insufficient.
1295 * Updates strstart and lookahead.
1296 *
1297 * IN assertion: lookahead < MIN_LOOKAHEAD
1298 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1299 * At least one byte has been read, or avail_in == 0; reads are
1300 * performed for at least two bytes (required for the zip translate_eol
1301 * option -- not supported here).
1302 */
1303local void fill_window(s)
1304 deflate_state *s;
1305{
1306 register unsigned n, m;
1307 register Posf *p;
1308 unsigned more; /* Amount of free space at the end of the window. */
1309 uInt wsize = s->w_size;
1310
1311 do {
1312 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1313
1314 /* Deal with !@#$% 64K limit: */
1315 if (sizeof(int) <= 2) {
1316 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1317 more = wsize;
1318
1319 } else if (more == (unsigned)(-1)) {
1320 /* Very unlikely, but possible on 16 bit machine if
1321 * strstart == 0 && lookahead == 1 (input done a byte at time)
1322 */
1323 more--;
1324 }
1325 }
1326
1327 /* If the window is almost full and there is insufficient lookahead,
1328 * move the upper half to the lower one to make room in the upper half.
1329 */
1330 if (s->strstart >= wsize+MAX_DIST(s)) {
1331
1332 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1333 s->match_start -= wsize;
1334 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1335 s->block_start -= (long) wsize;
1336
1337 /* Slide the hash table (could be avoided with 32 bit values
1338 at the expense of memory usage). We slide even when level == 0
1339 to keep the hash table consistent if we switch back to level > 0
1340 later. (Using level 0 permanently is not an optimal usage of
1341 zlib, so we don't care about this pathological case.)
1342 */
1343 n = s->hash_size;
1344 p = &s->head[n];
1345 do {
1346 m = *--p;
1347 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1348 } while (--n);
1349
1350 n = wsize;
1351#ifndef FASTEST
1352 p = &s->prev[n];
1353 do {
1354 m = *--p;
1355 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1356 /* If n is not on any hash chain, prev[n] is garbage but
1357 * its value will never be used.
1358 */
1359 } while (--n);
1360#endif
1361 more += wsize;
1362 }
1363 if (s->strm->avail_in == 0) return;
1364
1365 /* If there was no sliding:
1366 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1367 * more == window_size - lookahead - strstart
1368 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1369 * => more >= window_size - 2*WSIZE + 2
1370 * In the BIG_MEM or MMAP case (not yet supported),
1371 * window_size == input_size + MIN_LOOKAHEAD &&
1372 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1373 * Otherwise, window_size == 2*WSIZE so more >= 2.
1374 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1375 */
1376 Assert(more >= 2, "more < 2");
1377
1378 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1379 s->lookahead += n;
1380
1381 /* Initialize the hash value now that we have some input: */
1382 if (s->lookahead >= MIN_MATCH) {
1383 s->ins_h = s->window[s->strstart];
1384 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1385#if MIN_MATCH != 3
1386 Call UPDATE_HASH() MIN_MATCH-3 more times
1387#endif
1388 }
1389 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1390 * but this is not important since only literal bytes will be emitted.
1391 */
1392
1393 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1394
1395 /* If the WIN_INIT bytes after the end of the current data have never been
1396 * written, then zero those bytes in order to avoid memory check reports of
1397 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1398 * the longest match routines. Update the high water mark for the next
1399 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1400 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1401 */
1402 if (s->high_water < s->window_size) {
1403 ulg curr = s->strstart + (ulg)(s->lookahead);
1404 ulg init;
1405
1406 if (s->high_water < curr) {
1407 /* Previous high water mark below current data -- zero WIN_INIT
1408 * bytes or up to end of window, whichever is less.
1409 */
1410 init = s->window_size - curr;
1411 if (init > WIN_INIT)
1412 init = WIN_INIT;
1413 zmemzero(s->window + curr, (unsigned)init);
1414 s->high_water = curr + init;
1415 }
1416 else if (s->high_water < (ulg)curr + WIN_INIT) {
1417 /* High water mark at or above current data, but below current data
1418 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1419 * to end of window, whichever is less.
1420 */
1421 init = (ulg)curr + WIN_INIT - s->high_water;
1422 if (init > s->window_size - s->high_water)
1423 init = s->window_size - s->high_water;
1424 zmemzero(s->window + s->high_water, (unsigned)init);
1425 s->high_water += init;
1426 }
1427 }
1428}
1429
1430/* ===========================================================================
1431 * Flush the current block, with given end-of-file flag.
1432 * IN assertion: strstart is set to the end of the current match.
1433 */
1434#define FLUSH_BLOCK_ONLY(s, last) { \
1435 _tr_flush_block(s, (s->block_start >= 0L ? \
1436 (charf *)&s->window[(unsigned)s->block_start] : \
1437 (charf *)Z_NULL), \
1438 (ulg)((long)s->strstart - s->block_start), \
1439 (last)); \
1440 s->block_start = s->strstart; \
1441 flush_pending(s->strm); \
1442 Tracev((stderr,"[FLUSH]")); \
1443}
1444
1445/* Same but force premature exit if necessary. */
1446#define FLUSH_BLOCK(s, last) { \
1447 FLUSH_BLOCK_ONLY(s, last); \
1448 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1449}
1450
1451/* ===========================================================================
1452 * Copy without compression as much as possible from the input stream, return
1453 * the current block state.
1454 * This function does not insert new strings in the dictionary since
1455 * uncompressible data is probably not useful. This function is used
1456 * only for the level=0 compression option.
1457 * NOTE: this function should be optimized to avoid extra copying from
1458 * window to pending_buf.
1459 */
1460local block_state deflate_stored(s, flush)
1461 deflate_state *s;
1462 int flush;
1463{
1464 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1465 * to pending_buf_size, and each stored block has a 5 byte header:
1466 */
1467 ulg max_block_size = 0xffff;
1468 ulg max_start;
1469
1470 if (max_block_size > s->pending_buf_size - 5) {
1471 max_block_size = s->pending_buf_size - 5;
1472 }
1473
1474 /* Copy as much as possible from input to output: */
1475 for (;;) {
1476 /* Fill the window as much as possible: */
1477 if (s->lookahead <= 1) {
1478
1479 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1480 s->block_start >= (long)s->w_size, "slide too late");
1481
1482 fill_window(s);
1483 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1484
1485 if (s->lookahead == 0) break; /* flush the current block */
1486 }
1487 Assert(s->block_start >= 0L, "block gone");
1488
1489 s->strstart += s->lookahead;
1490 s->lookahead = 0;
1491
1492 /* Emit a stored block if pending_buf will be full: */
1493 max_start = s->block_start + max_block_size;
1494 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1495 /* strstart == 0 is possible when wraparound on 16-bit machine */
1496 s->lookahead = (uInt)(s->strstart - max_start);
1497 s->strstart = (uInt)max_start;
1498 FLUSH_BLOCK(s, 0);
1499 }
1500 /* Flush if we may have to slide, otherwise block_start may become
1501 * negative and the data will be gone:
1502 */
1503 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1504 FLUSH_BLOCK(s, 0);
1505 }
1506 }
1507 FLUSH_BLOCK(s, flush == Z_FINISH);
1508 return flush == Z_FINISH ? finish_done : block_done;
1509}
1510
1511/* ===========================================================================
1512 * Compress as much as possible from the input stream, return the current
1513 * block state.
1514 * This function does not perform lazy evaluation of matches and inserts
1515 * new strings in the dictionary only for unmatched strings or for short
1516 * matches. It is used only for the fast compression options.
1517 */
1518local block_state deflate_fast(s, flush)
1519 deflate_state *s;
1520 int flush;
1521{
1522 IPos hash_head; /* head of the hash chain */
1523 int bflush; /* set if current block must be flushed */
1524
1525 for (;;) {
1526 /* Make sure that we always have enough lookahead, except
1527 * at the end of the input file. We need MAX_MATCH bytes
1528 * for the next match, plus MIN_MATCH bytes to insert the
1529 * string following the next match.
1530 */
1531 if (s->lookahead < MIN_LOOKAHEAD) {
1532 fill_window(s);
1533 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1534 return need_more;
1535 }
1536 if (s->lookahead == 0) break; /* flush the current block */
1537 }
1538
1539 /* Insert the string window[strstart .. strstart+2] in the
1540 * dictionary, and set hash_head to the head of the hash chain:
1541 */
1542 hash_head = NIL;
1543 if (s->lookahead >= MIN_MATCH) {
1544 INSERT_STRING(s, s->strstart, hash_head);
1545 }
1546
1547 /* Find the longest match, discarding those <= prev_length.
1548 * At this point we have always match_length < MIN_MATCH
1549 */
1550 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1551 /* To simplify the code, we prevent matches with the string
1552 * of window index 0 (in particular we have to avoid a match
1553 * of the string with itself at the start of the input file).
1554 */
1555 s->match_length = longest_match (s, hash_head);
1556 /* longest_match() sets match_start */
1557 }
1558 if (s->match_length >= MIN_MATCH) {
1559 check_match(s, s->strstart, s->match_start, s->match_length);
1560
1561 _tr_tally_dist(s, s->strstart - s->match_start,
1562 s->match_length - MIN_MATCH, bflush);
1563
1564 s->lookahead -= s->match_length;
1565
1566 /* Insert new strings in the hash table only if the match length
1567 * is not too large. This saves time but degrades compression.
1568 */
1569#ifndef FASTEST
1570 if (s->match_length <= s->max_insert_length &&
1571 s->lookahead >= MIN_MATCH) {
1572 s->match_length--; /* string at strstart already in table */
1573 do {
1574 s->strstart++;
1575 INSERT_STRING(s, s->strstart, hash_head);
1576 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1577 * always MIN_MATCH bytes ahead.
1578 */
1579 } while (--s->match_length != 0);
1580 s->strstart++;
1581 } else
1582#endif
1583 {
1584 s->strstart += s->match_length;
1585 s->match_length = 0;
1586 s->ins_h = s->window[s->strstart];
1587 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1588#if MIN_MATCH != 3
1589 Call UPDATE_HASH() MIN_MATCH-3 more times
1590#endif
1591 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1592 * matter since it will be recomputed at next deflate call.
1593 */
1594 }
1595 } else {
1596 /* No match, output a literal byte */
1597 Tracevv((stderr,"%c", s->window[s->strstart]));
1598 _tr_tally_lit (s, s->window[s->strstart], bflush);
1599 s->lookahead--;
1600 s->strstart++;
1601 }
1602 if (bflush) FLUSH_BLOCK(s, 0);
1603 }
1604 FLUSH_BLOCK(s, flush == Z_FINISH);
1605 return flush == Z_FINISH ? finish_done : block_done;
1606}
1607
1608#ifndef FASTEST
1609/* ===========================================================================
1610 * Same as above, but achieves better compression. We use a lazy
1611 * evaluation for matches: a match is finally adopted only if there is
1612 * no better match at the next window position.
1613 */
1614local block_state deflate_slow(s, flush)
1615 deflate_state *s;
1616 int flush;
1617{
1618 IPos hash_head; /* head of hash chain */
1619 int bflush; /* set if current block must be flushed */
1620
1621 /* Process the input block. */
1622 for (;;) {
1623 /* Make sure that we always have enough lookahead, except
1624 * at the end of the input file. We need MAX_MATCH bytes
1625 * for the next match, plus MIN_MATCH bytes to insert the
1626 * string following the next match.
1627 */
1628 if (s->lookahead < MIN_LOOKAHEAD) {
1629 fill_window(s);
1630 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1631 return need_more;
1632 }
1633 if (s->lookahead == 0) break; /* flush the current block */
1634 }
1635
1636 /* Insert the string window[strstart .. strstart+2] in the
1637 * dictionary, and set hash_head to the head of the hash chain:
1638 */
1639 hash_head = NIL;
1640 if (s->lookahead >= MIN_MATCH) {
1641 INSERT_STRING(s, s->strstart, hash_head);
1642 }
1643
1644 /* Find the longest match, discarding those <= prev_length.
1645 */
1646 s->prev_length = s->match_length, s->prev_match = s->match_start;
1647 s->match_length = MIN_MATCH-1;
1648
1649 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1650 s->strstart - hash_head <= MAX_DIST(s)) {
1651 /* To simplify the code, we prevent matches with the string
1652 * of window index 0 (in particular we have to avoid a match
1653 * of the string with itself at the start of the input file).
1654 */
1655 s->match_length = longest_match (s, hash_head);
1656 /* longest_match() sets match_start */
1657
1658 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1659#if TOO_FAR <= 32767
1660 || (s->match_length == MIN_MATCH &&
1661 s->strstart - s->match_start > TOO_FAR)
1662#endif
1663 )) {
1664
1665 /* If prev_match is also MIN_MATCH, match_start is garbage
1666 * but we will ignore the current match anyway.
1667 */
1668 s->match_length = MIN_MATCH-1;
1669 }
1670 }
1671 /* If there was a match at the previous step and the current
1672 * match is not better, output the previous match:
1673 */
1674 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1675 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1676 /* Do not insert strings in hash table beyond this. */
1677
1678 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1679
1680 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1681 s->prev_length - MIN_MATCH, bflush);
1682
1683 /* Insert in hash table all strings up to the end of the match.
1684 * strstart-1 and strstart are already inserted. If there is not
1685 * enough lookahead, the last two strings are not inserted in
1686 * the hash table.
1687 */
1688 s->lookahead -= s->prev_length-1;
1689 s->prev_length -= 2;
1690 do {
1691 if (++s->strstart <= max_insert) {
1692 INSERT_STRING(s, s->strstart, hash_head);
1693 }
1694 } while (--s->prev_length != 0);
1695 s->match_available = 0;
1696 s->match_length = MIN_MATCH-1;
1697 s->strstart++;
1698
1699 if (bflush) FLUSH_BLOCK(s, 0);
1700
1701 } else if (s->match_available) {
1702 /* If there was no match at the previous position, output a
1703 * single literal. If there was a match but the current match
1704 * is longer, truncate the previous match to a single literal.
1705 */
1706 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1707 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1708 if (bflush) {
1709 FLUSH_BLOCK_ONLY(s, 0);
1710 }
1711 s->strstart++;
1712 s->lookahead--;
1713 if (s->strm->avail_out == 0) return need_more;
1714 } else {
1715 /* There is no previous match to compare with, wait for
1716 * the next step to decide.
1717 */
1718 s->match_available = 1;
1719 s->strstart++;
1720 s->lookahead--;
1721 }
1722 }
1723 Assert (flush != Z_NO_FLUSH, "no flush?");
1724 if (s->match_available) {
1725 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1726 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1727 s->match_available = 0;
1728 }
1729 FLUSH_BLOCK(s, flush == Z_FINISH);
1730 return flush == Z_FINISH ? finish_done : block_done;
1731}
1732#endif /* FASTEST */
1733
1734/* ===========================================================================
1735 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1736 * one. Do not maintain a hash table. (It will be regenerated if this run of
1737 * deflate switches away from Z_RLE.)
1738 */
1739local block_state deflate_rle(s, flush)
1740 deflate_state *s;
1741 int flush;
1742{
1743 int bflush; /* set if current block must be flushed */
1744 uInt prev; /* byte at distance one to match */
1745 Bytef *scan, *strend; /* scan goes up to strend for length of run */
1746
1747 for (;;) {
1748 /* Make sure that we always have enough lookahead, except
1749 * at the end of the input file. We need MAX_MATCH bytes
1750 * for the longest encodable run.
1751 */
1752 if (s->lookahead < MAX_MATCH) {
1753 fill_window(s);
1754 if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {
1755 return need_more;
1756 }
1757 if (s->lookahead == 0) break; /* flush the current block */
1758 }
1759
1760 /* See how many times the previous byte repeats */
1761 s->match_length = 0;
1762 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
1763 scan = s->window + s->strstart - 1;
1764 prev = *scan;
1765 if (prev == *++scan && prev == *++scan && prev == *++scan) {
1766 strend = s->window + s->strstart + MAX_MATCH;
1767 do {
1768 } while (prev == *++scan && prev == *++scan &&
1769 prev == *++scan && prev == *++scan &&
1770 prev == *++scan && prev == *++scan &&
1771 prev == *++scan && prev == *++scan &&
1772 scan < strend);
1773 s->match_length = MAX_MATCH - (int)(strend - scan);
1774 if (s->match_length > s->lookahead)
1775 s->match_length = s->lookahead;
1776 }
1777 }
1778
1779 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1780 if (s->match_length >= MIN_MATCH) {
1781 check_match(s, s->strstart, s->strstart - 1, s->match_length);
1782
1783 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
1784
1785 s->lookahead -= s->match_length;
1786 s->strstart += s->match_length;
1787 s->match_length = 0;
1788 } else {
1789 /* No match, output a literal byte */
1790 Tracevv((stderr,"%c", s->window[s->strstart]));
1791 _tr_tally_lit (s, s->window[s->strstart], bflush);
1792 s->lookahead--;
1793 s->strstart++;
1794 }
1795 if (bflush) FLUSH_BLOCK(s, 0);
1796 }
1797 FLUSH_BLOCK(s, flush == Z_FINISH);
1798 return flush == Z_FINISH ? finish_done : block_done;
1799}
1800
1801/* ===========================================================================
1802 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
1803 * (It will be regenerated if this run of deflate switches away from Huffman.)
1804 */
1805local block_state deflate_huff(s, flush)
1806 deflate_state *s;
1807 int flush;
1808{
1809 int bflush; /* set if current block must be flushed */
1810
1811 for (;;) {
1812 /* Make sure that we have a literal to write. */
1813 if (s->lookahead == 0) {
1814 fill_window(s);
1815 if (s->lookahead == 0) {
1816 if (flush == Z_NO_FLUSH)
1817 return need_more;
1818 break; /* flush the current block */
1819 }
1820 }
1821
1822 /* Output a literal byte */
1823 s->match_length = 0;
1824 Tracevv((stderr,"%c", s->window[s->strstart]));
1825 _tr_tally_lit (s, s->window[s->strstart], bflush);
1826 s->lookahead--;
1827 s->strstart++;
1828 if (bflush) FLUSH_BLOCK(s, 0);
1829 }
1830 FLUSH_BLOCK(s, flush == Z_FINISH);
1831 return flush == Z_FINISH ? finish_done : block_done;
1832}