Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* +++ deflate.c */ |
2 | /* deflate.c -- compress data using the deflation algorithm | |
3 | * Copyright (C) 1995-1996 Jean-loup Gailly. | |
4 | * For conditions of distribution and use, see copyright notice in zlib.h | |
5 | */ | |
6 | ||
7 | /* | |
8 | * ALGORITHM | |
9 | * | |
10 | * The "deflation" process depends on being able to identify portions | |
11 | * of the input text which are identical to earlier input (within a | |
12 | * sliding window trailing behind the input currently being processed). | |
13 | * | |
14 | * The most straightforward technique turns out to be the fastest for | |
15 | * most input files: try all possible matches and select the longest. | |
16 | * The key feature of this algorithm is that insertions into the string | |
17 | * dictionary are very simple and thus fast, and deletions are avoided | |
18 | * completely. Insertions are performed at each input character, whereas | |
19 | * string matches are performed only when the previous match ends. So it | |
20 | * is preferable to spend more time in matches to allow very fast string | |
21 | * insertions and avoid deletions. The matching algorithm for small | |
22 | * strings is inspired from that of Rabin & Karp. A brute force approach | |
23 | * is used to find longer strings when a small match has been found. | |
24 | * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze | |
25 | * (by Leonid Broukhis). | |
26 | * A previous version of this file used a more sophisticated algorithm | |
27 | * (by Fiala and Greene) which is guaranteed to run in linear amortized | |
28 | * time, but has a larger average cost, uses more memory and is patented. | |
29 | * However the F&G algorithm may be faster for some highly redundant | |
30 | * files if the parameter max_chain_length (described below) is too large. | |
31 | * | |
32 | * ACKNOWLEDGEMENTS | |
33 | * | |
34 | * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and | |
35 | * I found it in 'freeze' written by Leonid Broukhis. | |
36 | * Thanks to many people for bug reports and testing. | |
37 | * | |
38 | * REFERENCES | |
39 | * | |
40 | * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification". | |
41 | * Available in ftp://ds.internic.net/rfc/rfc1951.txt | |
42 | * | |
43 | * A description of the Rabin and Karp algorithm is given in the book | |
44 | * "Algorithms" by R. Sedgewick, Addison-Wesley, p252. | |
45 | * | |
46 | * Fiala,E.R., and Greene,D.H. | |
47 | * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595 | |
48 | * | |
49 | */ | |
50 | ||
51 | #include <linux/module.h> | |
52 | #include <linux/zutil.h> | |
53 | #include "defutil.h" | |
54 | ||
55 | ||
56 | /* =========================================================================== | |
57 | * Function prototypes. | |
58 | */ | |
59 | typedef enum { | |
60 | need_more, /* block not completed, need more input or more output */ | |
61 | block_done, /* block flush performed */ | |
62 | finish_started, /* finish started, need only more output at next deflate */ | |
63 | finish_done /* finish done, accept no more input or output */ | |
64 | } block_state; | |
65 | ||
66 | typedef block_state (*compress_func) (deflate_state *s, int flush); | |
67 | /* Compression function. Returns the block state after the call. */ | |
68 | ||
69 | static void fill_window (deflate_state *s); | |
70 | static block_state deflate_stored (deflate_state *s, int flush); | |
71 | static block_state deflate_fast (deflate_state *s, int flush); | |
72 | static block_state deflate_slow (deflate_state *s, int flush); | |
73 | static void lm_init (deflate_state *s); | |
74 | static void putShortMSB (deflate_state *s, uInt b); | |
75 | static void flush_pending (z_streamp strm); | |
76 | static int read_buf (z_streamp strm, Byte *buf, unsigned size); | |
77 | static uInt longest_match (deflate_state *s, IPos cur_match); | |
78 | ||
79 | #ifdef DEBUG_ZLIB | |
80 | static void check_match (deflate_state *s, IPos start, IPos match, | |
81 | int length); | |
82 | #endif | |
83 | ||
84 | /* =========================================================================== | |
85 | * Local data | |
86 | */ | |
87 | ||
88 | #define NIL 0 | |
89 | /* Tail of hash chains */ | |
90 | ||
91 | #ifndef TOO_FAR | |
92 | # define TOO_FAR 4096 | |
93 | #endif | |
94 | /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ | |
95 | ||
96 | #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) | |
97 | /* Minimum amount of lookahead, except at the end of the input file. | |
98 | * See deflate.c for comments about the MIN_MATCH+1. | |
99 | */ | |
100 | ||
101 | /* Values for max_lazy_match, good_match and max_chain_length, depending on | |
102 | * the desired pack level (0..9). The values given below have been tuned to | |
103 | * exclude worst case performance for pathological files. Better values may be | |
104 | * found for specific files. | |
105 | */ | |
106 | typedef struct config_s { | |
107 | ush good_length; /* reduce lazy search above this match length */ | |
108 | ush max_lazy; /* do not perform lazy search above this match length */ | |
109 | ush nice_length; /* quit search above this match length */ | |
110 | ush max_chain; | |
111 | compress_func func; | |
112 | } config; | |
113 | ||
114 | static const config configuration_table[10] = { | |
115 | /* good lazy nice chain */ | |
116 | /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ | |
117 | /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */ | |
118 | /* 2 */ {4, 5, 16, 8, deflate_fast}, | |
119 | /* 3 */ {4, 6, 32, 32, deflate_fast}, | |
120 | ||
121 | /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ | |
122 | /* 5 */ {8, 16, 32, 32, deflate_slow}, | |
123 | /* 6 */ {8, 16, 128, 128, deflate_slow}, | |
124 | /* 7 */ {8, 32, 128, 256, deflate_slow}, | |
125 | /* 8 */ {32, 128, 258, 1024, deflate_slow}, | |
126 | /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */ | |
127 | ||
128 | /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 | |
129 | * For deflate_fast() (levels <= 3) good is ignored and lazy has a different | |
130 | * meaning. | |
131 | */ | |
132 | ||
133 | #define EQUAL 0 | |
134 | /* result of memcmp for equal strings */ | |
135 | ||
136 | /* =========================================================================== | |
137 | * Update a hash value with the given input byte | |
138 | * IN assertion: all calls to to UPDATE_HASH are made with consecutive | |
139 | * input characters, so that a running hash key can be computed from the | |
140 | * previous key instead of complete recalculation each time. | |
141 | */ | |
142 | #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask) | |
143 | ||
144 | ||
145 | /* =========================================================================== | |
146 | * Insert string str in the dictionary and set match_head to the previous head | |
147 | * of the hash chain (the most recent string with same hash key). Return | |
148 | * the previous length of the hash chain. | |
149 | * IN assertion: all calls to to INSERT_STRING are made with consecutive | |
150 | * input characters and the first MIN_MATCH bytes of str are valid | |
151 | * (except for the last MIN_MATCH-1 bytes of the input file). | |
152 | */ | |
153 | #define INSERT_STRING(s, str, match_head) \ | |
154 | (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ | |
155 | s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \ | |
156 | s->head[s->ins_h] = (Pos)(str)) | |
157 | ||
158 | /* =========================================================================== | |
159 | * Initialize the hash table (avoiding 64K overflow for 16 bit systems). | |
160 | * prev[] will be initialized on the fly. | |
161 | */ | |
162 | #define CLEAR_HASH(s) \ | |
163 | s->head[s->hash_size-1] = NIL; \ | |
164 | memset((char *)s->head, 0, (unsigned)(s->hash_size-1)*sizeof(*s->head)); | |
165 | ||
166 | /* ========================================================================= */ | |
167 | int zlib_deflateInit_( | |
168 | z_streamp strm, | |
169 | int level, | |
170 | const char *version, | |
171 | int stream_size | |
172 | ) | |
173 | { | |
174 | return zlib_deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, | |
175 | DEF_MEM_LEVEL, | |
176 | Z_DEFAULT_STRATEGY, version, stream_size); | |
177 | /* To do: ignore strm->next_in if we use it as window */ | |
178 | } | |
179 | ||
180 | /* ========================================================================= */ | |
181 | int zlib_deflateInit2_( | |
182 | z_streamp strm, | |
183 | int level, | |
184 | int method, | |
185 | int windowBits, | |
186 | int memLevel, | |
187 | int strategy, | |
188 | const char *version, | |
189 | int stream_size | |
190 | ) | |
191 | { | |
192 | deflate_state *s; | |
193 | int noheader = 0; | |
194 | static char* my_version = ZLIB_VERSION; | |
195 | deflate_workspace *mem; | |
196 | ||
197 | ush *overlay; | |
198 | /* We overlay pending_buf and d_buf+l_buf. This works since the average | |
199 | * output size for (length,distance) codes is <= 24 bits. | |
200 | */ | |
201 | ||
202 | if (version == NULL || version[0] != my_version[0] || | |
203 | stream_size != sizeof(z_stream)) { | |
204 | return Z_VERSION_ERROR; | |
205 | } | |
206 | if (strm == NULL) return Z_STREAM_ERROR; | |
207 | ||
208 | strm->msg = NULL; | |
209 | ||
210 | if (level == Z_DEFAULT_COMPRESSION) level = 6; | |
211 | ||
212 | mem = (deflate_workspace *) strm->workspace; | |
213 | ||
214 | if (windowBits < 0) { /* undocumented feature: suppress zlib header */ | |
215 | noheader = 1; | |
216 | windowBits = -windowBits; | |
217 | } | |
218 | if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || | |
219 | windowBits < 9 || windowBits > 15 || level < 0 || level > 9 || | |
220 | strategy < 0 || strategy > Z_HUFFMAN_ONLY) { | |
221 | return Z_STREAM_ERROR; | |
222 | } | |
223 | s = (deflate_state *) &(mem->deflate_memory); | |
224 | strm->state = (struct internal_state *)s; | |
225 | s->strm = strm; | |
226 | ||
227 | s->noheader = noheader; | |
228 | s->w_bits = windowBits; | |
229 | s->w_size = 1 << s->w_bits; | |
230 | s->w_mask = s->w_size - 1; | |
231 | ||
232 | s->hash_bits = memLevel + 7; | |
233 | s->hash_size = 1 << s->hash_bits; | |
234 | s->hash_mask = s->hash_size - 1; | |
235 | s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); | |
236 | ||
237 | s->window = (Byte *) mem->window_memory; | |
238 | s->prev = (Pos *) mem->prev_memory; | |
239 | s->head = (Pos *) mem->head_memory; | |
240 | ||
241 | s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ | |
242 | ||
243 | overlay = (ush *) mem->overlay_memory; | |
244 | s->pending_buf = (uch *) overlay; | |
245 | s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); | |
246 | ||
247 | s->d_buf = overlay + s->lit_bufsize/sizeof(ush); | |
248 | s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; | |
249 | ||
250 | s->level = level; | |
251 | s->strategy = strategy; | |
252 | s->method = (Byte)method; | |
253 | ||
254 | return zlib_deflateReset(strm); | |
255 | } | |
256 | ||
257 | /* ========================================================================= */ | |
87c2ce3b | 258 | #if 0 |
1da177e4 LT |
259 | int zlib_deflateSetDictionary( |
260 | z_streamp strm, | |
261 | const Byte *dictionary, | |
262 | uInt dictLength | |
263 | ) | |
264 | { | |
265 | deflate_state *s; | |
266 | uInt length = dictLength; | |
267 | uInt n; | |
268 | IPos hash_head = 0; | |
269 | ||
270 | if (strm == NULL || strm->state == NULL || dictionary == NULL) | |
271 | return Z_STREAM_ERROR; | |
272 | ||
273 | s = (deflate_state *) strm->state; | |
274 | if (s->status != INIT_STATE) return Z_STREAM_ERROR; | |
275 | ||
276 | strm->adler = zlib_adler32(strm->adler, dictionary, dictLength); | |
277 | ||
278 | if (length < MIN_MATCH) return Z_OK; | |
279 | if (length > MAX_DIST(s)) { | |
280 | length = MAX_DIST(s); | |
281 | #ifndef USE_DICT_HEAD | |
282 | dictionary += dictLength - length; /* use the tail of the dictionary */ | |
283 | #endif | |
284 | } | |
285 | memcpy((char *)s->window, dictionary, length); | |
286 | s->strstart = length; | |
287 | s->block_start = (long)length; | |
288 | ||
289 | /* Insert all strings in the hash table (except for the last two bytes). | |
290 | * s->lookahead stays null, so s->ins_h will be recomputed at the next | |
291 | * call of fill_window. | |
292 | */ | |
293 | s->ins_h = s->window[0]; | |
294 | UPDATE_HASH(s, s->ins_h, s->window[1]); | |
295 | for (n = 0; n <= length - MIN_MATCH; n++) { | |
296 | INSERT_STRING(s, n, hash_head); | |
297 | } | |
298 | if (hash_head) hash_head = 0; /* to make compiler happy */ | |
299 | return Z_OK; | |
300 | } | |
87c2ce3b | 301 | #endif /* 0 */ |
1da177e4 LT |
302 | |
303 | /* ========================================================================= */ | |
304 | int zlib_deflateReset( | |
305 | z_streamp strm | |
306 | ) | |
307 | { | |
308 | deflate_state *s; | |
309 | ||
310 | if (strm == NULL || strm->state == NULL) | |
311 | return Z_STREAM_ERROR; | |
312 | ||
313 | strm->total_in = strm->total_out = 0; | |
314 | strm->msg = NULL; | |
315 | strm->data_type = Z_UNKNOWN; | |
316 | ||
317 | s = (deflate_state *)strm->state; | |
318 | s->pending = 0; | |
319 | s->pending_out = s->pending_buf; | |
320 | ||
321 | if (s->noheader < 0) { | |
322 | s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */ | |
323 | } | |
324 | s->status = s->noheader ? BUSY_STATE : INIT_STATE; | |
325 | strm->adler = 1; | |
326 | s->last_flush = Z_NO_FLUSH; | |
327 | ||
328 | zlib_tr_init(s); | |
329 | lm_init(s); | |
330 | ||
331 | return Z_OK; | |
332 | } | |
333 | ||
334 | /* ========================================================================= */ | |
87c2ce3b | 335 | #if 0 |
1da177e4 LT |
336 | int zlib_deflateParams( |
337 | z_streamp strm, | |
338 | int level, | |
339 | int strategy | |
340 | ) | |
341 | { | |
342 | deflate_state *s; | |
343 | compress_func func; | |
344 | int err = Z_OK; | |
345 | ||
346 | if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR; | |
347 | s = (deflate_state *) strm->state; | |
348 | ||
349 | if (level == Z_DEFAULT_COMPRESSION) { | |
350 | level = 6; | |
351 | } | |
352 | if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) { | |
353 | return Z_STREAM_ERROR; | |
354 | } | |
355 | func = configuration_table[s->level].func; | |
356 | ||
357 | if (func != configuration_table[level].func && strm->total_in != 0) { | |
358 | /* Flush the last buffer: */ | |
359 | err = zlib_deflate(strm, Z_PARTIAL_FLUSH); | |
360 | } | |
361 | if (s->level != level) { | |
362 | s->level = level; | |
363 | s->max_lazy_match = configuration_table[level].max_lazy; | |
364 | s->good_match = configuration_table[level].good_length; | |
365 | s->nice_match = configuration_table[level].nice_length; | |
366 | s->max_chain_length = configuration_table[level].max_chain; | |
367 | } | |
368 | s->strategy = strategy; | |
369 | return err; | |
370 | } | |
87c2ce3b | 371 | #endif /* 0 */ |
1da177e4 LT |
372 | |
373 | /* ========================================================================= | |
374 | * Put a short in the pending buffer. The 16-bit value is put in MSB order. | |
375 | * IN assertion: the stream state is correct and there is enough room in | |
376 | * pending_buf. | |
377 | */ | |
378 | static void putShortMSB( | |
379 | deflate_state *s, | |
380 | uInt b | |
381 | ) | |
382 | { | |
383 | put_byte(s, (Byte)(b >> 8)); | |
384 | put_byte(s, (Byte)(b & 0xff)); | |
385 | } | |
386 | ||
387 | /* ========================================================================= | |
388 | * Flush as much pending output as possible. All deflate() output goes | |
389 | * through this function so some applications may wish to modify it | |
390 | * to avoid allocating a large strm->next_out buffer and copying into it. | |
391 | * (See also read_buf()). | |
392 | */ | |
393 | static void flush_pending( | |
394 | z_streamp strm | |
395 | ) | |
396 | { | |
397 | deflate_state *s = (deflate_state *) strm->state; | |
398 | unsigned len = s->pending; | |
399 | ||
400 | if (len > strm->avail_out) len = strm->avail_out; | |
401 | if (len == 0) return; | |
402 | ||
403 | if (strm->next_out != NULL) { | |
404 | memcpy(strm->next_out, s->pending_out, len); | |
405 | strm->next_out += len; | |
406 | } | |
407 | s->pending_out += len; | |
408 | strm->total_out += len; | |
409 | strm->avail_out -= len; | |
410 | s->pending -= len; | |
411 | if (s->pending == 0) { | |
412 | s->pending_out = s->pending_buf; | |
413 | } | |
414 | } | |
415 | ||
416 | /* ========================================================================= */ | |
417 | int zlib_deflate( | |
418 | z_streamp strm, | |
419 | int flush | |
420 | ) | |
421 | { | |
422 | int old_flush; /* value of flush param for previous deflate call */ | |
423 | deflate_state *s; | |
424 | ||
425 | if (strm == NULL || strm->state == NULL || | |
426 | flush > Z_FINISH || flush < 0) { | |
427 | return Z_STREAM_ERROR; | |
428 | } | |
429 | s = (deflate_state *) strm->state; | |
430 | ||
431 | if ((strm->next_in == NULL && strm->avail_in != 0) || | |
432 | (s->status == FINISH_STATE && flush != Z_FINISH)) { | |
433 | return Z_STREAM_ERROR; | |
434 | } | |
435 | if (strm->avail_out == 0) return Z_BUF_ERROR; | |
436 | ||
437 | s->strm = strm; /* just in case */ | |
438 | old_flush = s->last_flush; | |
439 | s->last_flush = flush; | |
440 | ||
441 | /* Write the zlib header */ | |
442 | if (s->status == INIT_STATE) { | |
443 | ||
444 | uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; | |
445 | uInt level_flags = (s->level-1) >> 1; | |
446 | ||
447 | if (level_flags > 3) level_flags = 3; | |
448 | header |= (level_flags << 6); | |
449 | if (s->strstart != 0) header |= PRESET_DICT; | |
450 | header += 31 - (header % 31); | |
451 | ||
452 | s->status = BUSY_STATE; | |
453 | putShortMSB(s, header); | |
454 | ||
455 | /* Save the adler32 of the preset dictionary: */ | |
456 | if (s->strstart != 0) { | |
457 | putShortMSB(s, (uInt)(strm->adler >> 16)); | |
458 | putShortMSB(s, (uInt)(strm->adler & 0xffff)); | |
459 | } | |
460 | strm->adler = 1L; | |
461 | } | |
462 | ||
463 | /* Flush as much pending output as possible */ | |
464 | if (s->pending != 0) { | |
465 | flush_pending(strm); | |
466 | if (strm->avail_out == 0) { | |
467 | /* Since avail_out is 0, deflate will be called again with | |
468 | * more output space, but possibly with both pending and | |
469 | * avail_in equal to zero. There won't be anything to do, | |
470 | * but this is not an error situation so make sure we | |
471 | * return OK instead of BUF_ERROR at next call of deflate: | |
472 | */ | |
473 | s->last_flush = -1; | |
474 | return Z_OK; | |
475 | } | |
476 | ||
477 | /* Make sure there is something to do and avoid duplicate consecutive | |
478 | * flushes. For repeated and useless calls with Z_FINISH, we keep | |
479 | * returning Z_STREAM_END instead of Z_BUFF_ERROR. | |
480 | */ | |
481 | } else if (strm->avail_in == 0 && flush <= old_flush && | |
482 | flush != Z_FINISH) { | |
483 | return Z_BUF_ERROR; | |
484 | } | |
485 | ||
486 | /* User must not provide more input after the first FINISH: */ | |
487 | if (s->status == FINISH_STATE && strm->avail_in != 0) { | |
488 | return Z_BUF_ERROR; | |
489 | } | |
490 | ||
491 | /* Start a new block or continue the current one. | |
492 | */ | |
493 | if (strm->avail_in != 0 || s->lookahead != 0 || | |
494 | (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { | |
495 | block_state bstate; | |
496 | ||
497 | bstate = (*(configuration_table[s->level].func))(s, flush); | |
498 | ||
499 | if (bstate == finish_started || bstate == finish_done) { | |
500 | s->status = FINISH_STATE; | |
501 | } | |
502 | if (bstate == need_more || bstate == finish_started) { | |
503 | if (strm->avail_out == 0) { | |
504 | s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ | |
505 | } | |
506 | return Z_OK; | |
507 | /* If flush != Z_NO_FLUSH && avail_out == 0, the next call | |
508 | * of deflate should use the same flush parameter to make sure | |
509 | * that the flush is complete. So we don't have to output an | |
510 | * empty block here, this will be done at next call. This also | |
511 | * ensures that for a very small output buffer, we emit at most | |
512 | * one empty block. | |
513 | */ | |
514 | } | |
515 | if (bstate == block_done) { | |
516 | if (flush == Z_PARTIAL_FLUSH) { | |
517 | zlib_tr_align(s); | |
518 | } else if (flush == Z_PACKET_FLUSH) { | |
519 | /* Output just the 3-bit `stored' block type value, | |
520 | but not a zero length. */ | |
521 | zlib_tr_stored_type_only(s); | |
522 | } else { /* FULL_FLUSH or SYNC_FLUSH */ | |
523 | zlib_tr_stored_block(s, (char*)0, 0L, 0); | |
524 | /* For a full flush, this empty block will be recognized | |
525 | * as a special marker by inflate_sync(). | |
526 | */ | |
527 | if (flush == Z_FULL_FLUSH) { | |
528 | CLEAR_HASH(s); /* forget history */ | |
529 | } | |
530 | } | |
531 | flush_pending(strm); | |
532 | if (strm->avail_out == 0) { | |
533 | s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ | |
534 | return Z_OK; | |
535 | } | |
536 | } | |
537 | } | |
538 | Assert(strm->avail_out > 0, "bug2"); | |
539 | ||
540 | if (flush != Z_FINISH) return Z_OK; | |
541 | if (s->noheader) return Z_STREAM_END; | |
542 | ||
543 | /* Write the zlib trailer (adler32) */ | |
544 | putShortMSB(s, (uInt)(strm->adler >> 16)); | |
545 | putShortMSB(s, (uInt)(strm->adler & 0xffff)); | |
546 | flush_pending(strm); | |
547 | /* If avail_out is zero, the application will call deflate again | |
548 | * to flush the rest. | |
549 | */ | |
550 | s->noheader = -1; /* write the trailer only once! */ | |
551 | return s->pending != 0 ? Z_OK : Z_STREAM_END; | |
552 | } | |
553 | ||
554 | /* ========================================================================= */ | |
555 | int zlib_deflateEnd( | |
556 | z_streamp strm | |
557 | ) | |
558 | { | |
559 | int status; | |
560 | deflate_state *s; | |
561 | ||
562 | if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR; | |
563 | s = (deflate_state *) strm->state; | |
564 | ||
565 | status = s->status; | |
566 | if (status != INIT_STATE && status != BUSY_STATE && | |
567 | status != FINISH_STATE) { | |
568 | return Z_STREAM_ERROR; | |
569 | } | |
570 | ||
571 | strm->state = NULL; | |
572 | ||
573 | return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; | |
574 | } | |
575 | ||
576 | /* ========================================================================= | |
577 | * Copy the source state to the destination state. | |
578 | */ | |
87c2ce3b | 579 | #if 0 |
1da177e4 LT |
580 | int zlib_deflateCopy ( |
581 | z_streamp dest, | |
582 | z_streamp source | |
583 | ) | |
584 | { | |
585 | #ifdef MAXSEG_64K | |
586 | return Z_STREAM_ERROR; | |
587 | #else | |
588 | deflate_state *ds; | |
589 | deflate_state *ss; | |
590 | ush *overlay; | |
591 | deflate_workspace *mem; | |
592 | ||
593 | ||
594 | if (source == NULL || dest == NULL || source->state == NULL) { | |
595 | return Z_STREAM_ERROR; | |
596 | } | |
597 | ||
598 | ss = (deflate_state *) source->state; | |
599 | ||
600 | *dest = *source; | |
601 | ||
602 | mem = (deflate_workspace *) dest->workspace; | |
603 | ||
604 | ds = &(mem->deflate_memory); | |
605 | ||
606 | dest->state = (struct internal_state *) ds; | |
607 | *ds = *ss; | |
608 | ds->strm = dest; | |
609 | ||
610 | ds->window = (Byte *) mem->window_memory; | |
611 | ds->prev = (Pos *) mem->prev_memory; | |
612 | ds->head = (Pos *) mem->head_memory; | |
613 | overlay = (ush *) mem->overlay_memory; | |
614 | ds->pending_buf = (uch *) overlay; | |
615 | ||
616 | memcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); | |
617 | memcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos)); | |
618 | memcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos)); | |
619 | memcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); | |
620 | ||
621 | ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); | |
622 | ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); | |
623 | ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; | |
624 | ||
625 | ds->l_desc.dyn_tree = ds->dyn_ltree; | |
626 | ds->d_desc.dyn_tree = ds->dyn_dtree; | |
627 | ds->bl_desc.dyn_tree = ds->bl_tree; | |
628 | ||
629 | return Z_OK; | |
630 | #endif | |
631 | } | |
87c2ce3b | 632 | #endif /* 0 */ |
1da177e4 LT |
633 | |
634 | /* =========================================================================== | |
635 | * Read a new buffer from the current input stream, update the adler32 | |
636 | * and total number of bytes read. All deflate() input goes through | |
637 | * this function so some applications may wish to modify it to avoid | |
638 | * allocating a large strm->next_in buffer and copying from it. | |
639 | * (See also flush_pending()). | |
640 | */ | |
641 | static int read_buf( | |
642 | z_streamp strm, | |
643 | Byte *buf, | |
644 | unsigned size | |
645 | ) | |
646 | { | |
647 | unsigned len = strm->avail_in; | |
648 | ||
649 | if (len > size) len = size; | |
650 | if (len == 0) return 0; | |
651 | ||
652 | strm->avail_in -= len; | |
653 | ||
654 | if (!((deflate_state *)(strm->state))->noheader) { | |
655 | strm->adler = zlib_adler32(strm->adler, strm->next_in, len); | |
656 | } | |
657 | memcpy(buf, strm->next_in, len); | |
658 | strm->next_in += len; | |
659 | strm->total_in += len; | |
660 | ||
661 | return (int)len; | |
662 | } | |
663 | ||
664 | /* =========================================================================== | |
665 | * Initialize the "longest match" routines for a new zlib stream | |
666 | */ | |
667 | static void lm_init( | |
668 | deflate_state *s | |
669 | ) | |
670 | { | |
671 | s->window_size = (ulg)2L*s->w_size; | |
672 | ||
673 | CLEAR_HASH(s); | |
674 | ||
675 | /* Set the default configuration parameters: | |
676 | */ | |
677 | s->max_lazy_match = configuration_table[s->level].max_lazy; | |
678 | s->good_match = configuration_table[s->level].good_length; | |
679 | s->nice_match = configuration_table[s->level].nice_length; | |
680 | s->max_chain_length = configuration_table[s->level].max_chain; | |
681 | ||
682 | s->strstart = 0; | |
683 | s->block_start = 0L; | |
684 | s->lookahead = 0; | |
685 | s->match_length = s->prev_length = MIN_MATCH-1; | |
686 | s->match_available = 0; | |
687 | s->ins_h = 0; | |
688 | } | |
689 | ||
690 | /* =========================================================================== | |
691 | * Set match_start to the longest match starting at the given string and | |
692 | * return its length. Matches shorter or equal to prev_length are discarded, | |
693 | * in which case the result is equal to prev_length and match_start is | |
694 | * garbage. | |
695 | * IN assertions: cur_match is the head of the hash chain for the current | |
696 | * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 | |
697 | * OUT assertion: the match length is not greater than s->lookahead. | |
698 | */ | |
699 | /* For 80x86 and 680x0, an optimized version will be provided in match.asm or | |
700 | * match.S. The code will be functionally equivalent. | |
701 | */ | |
702 | static uInt longest_match( | |
703 | deflate_state *s, | |
704 | IPos cur_match /* current match */ | |
705 | ) | |
706 | { | |
707 | unsigned chain_length = s->max_chain_length;/* max hash chain length */ | |
708 | register Byte *scan = s->window + s->strstart; /* current string */ | |
709 | register Byte *match; /* matched string */ | |
710 | register int len; /* length of current match */ | |
711 | int best_len = s->prev_length; /* best match length so far */ | |
712 | int nice_match = s->nice_match; /* stop if match long enough */ | |
713 | IPos limit = s->strstart > (IPos)MAX_DIST(s) ? | |
714 | s->strstart - (IPos)MAX_DIST(s) : NIL; | |
715 | /* Stop when cur_match becomes <= limit. To simplify the code, | |
716 | * we prevent matches with the string of window index 0. | |
717 | */ | |
718 | Pos *prev = s->prev; | |
719 | uInt wmask = s->w_mask; | |
720 | ||
721 | #ifdef UNALIGNED_OK | |
722 | /* Compare two bytes at a time. Note: this is not always beneficial. | |
723 | * Try with and without -DUNALIGNED_OK to check. | |
724 | */ | |
725 | register Byte *strend = s->window + s->strstart + MAX_MATCH - 1; | |
726 | register ush scan_start = *(ush*)scan; | |
727 | register ush scan_end = *(ush*)(scan+best_len-1); | |
728 | #else | |
729 | register Byte *strend = s->window + s->strstart + MAX_MATCH; | |
730 | register Byte scan_end1 = scan[best_len-1]; | |
731 | register Byte scan_end = scan[best_len]; | |
732 | #endif | |
733 | ||
734 | /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. | |
735 | * It is easy to get rid of this optimization if necessary. | |
736 | */ | |
737 | Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); | |
738 | ||
739 | /* Do not waste too much time if we already have a good match: */ | |
740 | if (s->prev_length >= s->good_match) { | |
741 | chain_length >>= 2; | |
742 | } | |
743 | /* Do not look for matches beyond the end of the input. This is necessary | |
744 | * to make deflate deterministic. | |
745 | */ | |
746 | if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; | |
747 | ||
748 | Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); | |
749 | ||
750 | do { | |
751 | Assert(cur_match < s->strstart, "no future"); | |
752 | match = s->window + cur_match; | |
753 | ||
754 | /* Skip to next match if the match length cannot increase | |
755 | * or if the match length is less than 2: | |
756 | */ | |
757 | #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) | |
758 | /* This code assumes sizeof(unsigned short) == 2. Do not use | |
759 | * UNALIGNED_OK if your compiler uses a different size. | |
760 | */ | |
761 | if (*(ush*)(match+best_len-1) != scan_end || | |
762 | *(ush*)match != scan_start) continue; | |
763 | ||
764 | /* It is not necessary to compare scan[2] and match[2] since they are | |
765 | * always equal when the other bytes match, given that the hash keys | |
766 | * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at | |
767 | * strstart+3, +5, ... up to strstart+257. We check for insufficient | |
768 | * lookahead only every 4th comparison; the 128th check will be made | |
769 | * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is | |
770 | * necessary to put more guard bytes at the end of the window, or | |
771 | * to check more often for insufficient lookahead. | |
772 | */ | |
773 | Assert(scan[2] == match[2], "scan[2]?"); | |
774 | scan++, match++; | |
775 | do { | |
776 | } while (*(ush*)(scan+=2) == *(ush*)(match+=2) && | |
777 | *(ush*)(scan+=2) == *(ush*)(match+=2) && | |
778 | *(ush*)(scan+=2) == *(ush*)(match+=2) && | |
779 | *(ush*)(scan+=2) == *(ush*)(match+=2) && | |
780 | scan < strend); | |
781 | /* The funny "do {}" generates better code on most compilers */ | |
782 | ||
783 | /* Here, scan <= window+strstart+257 */ | |
784 | Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); | |
785 | if (*scan == *match) scan++; | |
786 | ||
787 | len = (MAX_MATCH - 1) - (int)(strend-scan); | |
788 | scan = strend - (MAX_MATCH-1); | |
789 | ||
790 | #else /* UNALIGNED_OK */ | |
791 | ||
792 | if (match[best_len] != scan_end || | |
793 | match[best_len-1] != scan_end1 || | |
794 | *match != *scan || | |
795 | *++match != scan[1]) continue; | |
796 | ||
797 | /* The check at best_len-1 can be removed because it will be made | |
798 | * again later. (This heuristic is not always a win.) | |
799 | * It is not necessary to compare scan[2] and match[2] since they | |
800 | * are always equal when the other bytes match, given that | |
801 | * the hash keys are equal and that HASH_BITS >= 8. | |
802 | */ | |
803 | scan += 2, match++; | |
804 | Assert(*scan == *match, "match[2]?"); | |
805 | ||
806 | /* We check for insufficient lookahead only every 8th comparison; | |
807 | * the 256th check will be made at strstart+258. | |
808 | */ | |
809 | do { | |
810 | } while (*++scan == *++match && *++scan == *++match && | |
811 | *++scan == *++match && *++scan == *++match && | |
812 | *++scan == *++match && *++scan == *++match && | |
813 | *++scan == *++match && *++scan == *++match && | |
814 | scan < strend); | |
815 | ||
816 | Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); | |
817 | ||
818 | len = MAX_MATCH - (int)(strend - scan); | |
819 | scan = strend - MAX_MATCH; | |
820 | ||
821 | #endif /* UNALIGNED_OK */ | |
822 | ||
823 | if (len > best_len) { | |
824 | s->match_start = cur_match; | |
825 | best_len = len; | |
826 | if (len >= nice_match) break; | |
827 | #ifdef UNALIGNED_OK | |
828 | scan_end = *(ush*)(scan+best_len-1); | |
829 | #else | |
830 | scan_end1 = scan[best_len-1]; | |
831 | scan_end = scan[best_len]; | |
832 | #endif | |
833 | } | |
834 | } while ((cur_match = prev[cur_match & wmask]) > limit | |
835 | && --chain_length != 0); | |
836 | ||
837 | if ((uInt)best_len <= s->lookahead) return best_len; | |
838 | return s->lookahead; | |
839 | } | |
840 | ||
841 | #ifdef DEBUG_ZLIB | |
842 | /* =========================================================================== | |
843 | * Check that the match at match_start is indeed a match. | |
844 | */ | |
845 | static void check_match( | |
846 | deflate_state *s, | |
847 | IPos start, | |
848 | IPos match, | |
849 | int length | |
850 | ) | |
851 | { | |
852 | /* check that the match is indeed a match */ | |
853 | if (memcmp((char *)s->window + match, | |
854 | (char *)s->window + start, length) != EQUAL) { | |
855 | fprintf(stderr, " start %u, match %u, length %d\n", | |
856 | start, match, length); | |
857 | do { | |
858 | fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); | |
859 | } while (--length != 0); | |
860 | z_error("invalid match"); | |
861 | } | |
862 | if (z_verbose > 1) { | |
863 | fprintf(stderr,"\\[%d,%d]", start-match, length); | |
864 | do { putc(s->window[start++], stderr); } while (--length != 0); | |
865 | } | |
866 | } | |
867 | #else | |
868 | # define check_match(s, start, match, length) | |
869 | #endif | |
870 | ||
871 | /* =========================================================================== | |
872 | * Fill the window when the lookahead becomes insufficient. | |
873 | * Updates strstart and lookahead. | |
874 | * | |
875 | * IN assertion: lookahead < MIN_LOOKAHEAD | |
876 | * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD | |
877 | * At least one byte has been read, or avail_in == 0; reads are | |
878 | * performed for at least two bytes (required for the zip translate_eol | |
879 | * option -- not supported here). | |
880 | */ | |
881 | static void fill_window( | |
882 | deflate_state *s | |
883 | ) | |
884 | { | |
885 | register unsigned n, m; | |
886 | register Pos *p; | |
887 | unsigned more; /* Amount of free space at the end of the window. */ | |
888 | uInt wsize = s->w_size; | |
889 | ||
890 | do { | |
891 | more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); | |
892 | ||
893 | /* Deal with !@#$% 64K limit: */ | |
894 | if (more == 0 && s->strstart == 0 && s->lookahead == 0) { | |
895 | more = wsize; | |
896 | ||
897 | } else if (more == (unsigned)(-1)) { | |
898 | /* Very unlikely, but possible on 16 bit machine if strstart == 0 | |
899 | * and lookahead == 1 (input done one byte at time) | |
900 | */ | |
901 | more--; | |
902 | ||
903 | /* If the window is almost full and there is insufficient lookahead, | |
904 | * move the upper half to the lower one to make room in the upper half. | |
905 | */ | |
906 | } else if (s->strstart >= wsize+MAX_DIST(s)) { | |
907 | ||
908 | memcpy((char *)s->window, (char *)s->window+wsize, | |
909 | (unsigned)wsize); | |
910 | s->match_start -= wsize; | |
911 | s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ | |
912 | s->block_start -= (long) wsize; | |
913 | ||
914 | /* Slide the hash table (could be avoided with 32 bit values | |
915 | at the expense of memory usage). We slide even when level == 0 | |
916 | to keep the hash table consistent if we switch back to level > 0 | |
917 | later. (Using level 0 permanently is not an optimal usage of | |
918 | zlib, so we don't care about this pathological case.) | |
919 | */ | |
920 | n = s->hash_size; | |
921 | p = &s->head[n]; | |
922 | do { | |
923 | m = *--p; | |
924 | *p = (Pos)(m >= wsize ? m-wsize : NIL); | |
925 | } while (--n); | |
926 | ||
927 | n = wsize; | |
928 | p = &s->prev[n]; | |
929 | do { | |
930 | m = *--p; | |
931 | *p = (Pos)(m >= wsize ? m-wsize : NIL); | |
932 | /* If n is not on any hash chain, prev[n] is garbage but | |
933 | * its value will never be used. | |
934 | */ | |
935 | } while (--n); | |
936 | more += wsize; | |
937 | } | |
938 | if (s->strm->avail_in == 0) return; | |
939 | ||
940 | /* If there was no sliding: | |
941 | * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && | |
942 | * more == window_size - lookahead - strstart | |
943 | * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) | |
944 | * => more >= window_size - 2*WSIZE + 2 | |
945 | * In the BIG_MEM or MMAP case (not yet supported), | |
946 | * window_size == input_size + MIN_LOOKAHEAD && | |
947 | * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. | |
948 | * Otherwise, window_size == 2*WSIZE so more >= 2. | |
949 | * If there was sliding, more >= WSIZE. So in all cases, more >= 2. | |
950 | */ | |
951 | Assert(more >= 2, "more < 2"); | |
952 | ||
953 | n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); | |
954 | s->lookahead += n; | |
955 | ||
956 | /* Initialize the hash value now that we have some input: */ | |
957 | if (s->lookahead >= MIN_MATCH) { | |
958 | s->ins_h = s->window[s->strstart]; | |
959 | UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); | |
960 | #if MIN_MATCH != 3 | |
961 | Call UPDATE_HASH() MIN_MATCH-3 more times | |
962 | #endif | |
963 | } | |
964 | /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, | |
965 | * but this is not important since only literal bytes will be emitted. | |
966 | */ | |
967 | ||
968 | } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); | |
969 | } | |
970 | ||
971 | /* =========================================================================== | |
972 | * Flush the current block, with given end-of-file flag. | |
973 | * IN assertion: strstart is set to the end of the current match. | |
974 | */ | |
975 | #define FLUSH_BLOCK_ONLY(s, eof) { \ | |
976 | zlib_tr_flush_block(s, (s->block_start >= 0L ? \ | |
977 | (char *)&s->window[(unsigned)s->block_start] : \ | |
978 | NULL), \ | |
979 | (ulg)((long)s->strstart - s->block_start), \ | |
980 | (eof)); \ | |
981 | s->block_start = s->strstart; \ | |
982 | flush_pending(s->strm); \ | |
983 | Tracev((stderr,"[FLUSH]")); \ | |
984 | } | |
985 | ||
986 | /* Same but force premature exit if necessary. */ | |
987 | #define FLUSH_BLOCK(s, eof) { \ | |
988 | FLUSH_BLOCK_ONLY(s, eof); \ | |
989 | if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \ | |
990 | } | |
991 | ||
992 | /* =========================================================================== | |
993 | * Copy without compression as much as possible from the input stream, return | |
994 | * the current block state. | |
995 | * This function does not insert new strings in the dictionary since | |
996 | * uncompressible data is probably not useful. This function is used | |
997 | * only for the level=0 compression option. | |
998 | * NOTE: this function should be optimized to avoid extra copying from | |
999 | * window to pending_buf. | |
1000 | */ | |
1001 | static block_state deflate_stored( | |
1002 | deflate_state *s, | |
1003 | int flush | |
1004 | ) | |
1005 | { | |
1006 | /* Stored blocks are limited to 0xffff bytes, pending_buf is limited | |
1007 | * to pending_buf_size, and each stored block has a 5 byte header: | |
1008 | */ | |
1009 | ulg max_block_size = 0xffff; | |
1010 | ulg max_start; | |
1011 | ||
1012 | if (max_block_size > s->pending_buf_size - 5) { | |
1013 | max_block_size = s->pending_buf_size - 5; | |
1014 | } | |
1015 | ||
1016 | /* Copy as much as possible from input to output: */ | |
1017 | for (;;) { | |
1018 | /* Fill the window as much as possible: */ | |
1019 | if (s->lookahead <= 1) { | |
1020 | ||
1021 | Assert(s->strstart < s->w_size+MAX_DIST(s) || | |
1022 | s->block_start >= (long)s->w_size, "slide too late"); | |
1023 | ||
1024 | fill_window(s); | |
1025 | if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more; | |
1026 | ||
1027 | if (s->lookahead == 0) break; /* flush the current block */ | |
1028 | } | |
1029 | Assert(s->block_start >= 0L, "block gone"); | |
1030 | ||
1031 | s->strstart += s->lookahead; | |
1032 | s->lookahead = 0; | |
1033 | ||
1034 | /* Emit a stored block if pending_buf will be full: */ | |
1035 | max_start = s->block_start + max_block_size; | |
1036 | if (s->strstart == 0 || (ulg)s->strstart >= max_start) { | |
1037 | /* strstart == 0 is possible when wraparound on 16-bit machine */ | |
1038 | s->lookahead = (uInt)(s->strstart - max_start); | |
1039 | s->strstart = (uInt)max_start; | |
1040 | FLUSH_BLOCK(s, 0); | |
1041 | } | |
1042 | /* Flush if we may have to slide, otherwise block_start may become | |
1043 | * negative and the data will be gone: | |
1044 | */ | |
1045 | if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) { | |
1046 | FLUSH_BLOCK(s, 0); | |
1047 | } | |
1048 | } | |
1049 | FLUSH_BLOCK(s, flush == Z_FINISH); | |
1050 | return flush == Z_FINISH ? finish_done : block_done; | |
1051 | } | |
1052 | ||
1053 | /* =========================================================================== | |
1054 | * Compress as much as possible from the input stream, return the current | |
1055 | * block state. | |
1056 | * This function does not perform lazy evaluation of matches and inserts | |
1057 | * new strings in the dictionary only for unmatched strings or for short | |
1058 | * matches. It is used only for the fast compression options. | |
1059 | */ | |
1060 | static block_state deflate_fast( | |
1061 | deflate_state *s, | |
1062 | int flush | |
1063 | ) | |
1064 | { | |
1065 | IPos hash_head = NIL; /* head of the hash chain */ | |
1066 | int bflush; /* set if current block must be flushed */ | |
1067 | ||
1068 | for (;;) { | |
1069 | /* Make sure that we always have enough lookahead, except | |
1070 | * at the end of the input file. We need MAX_MATCH bytes | |
1071 | * for the next match, plus MIN_MATCH bytes to insert the | |
1072 | * string following the next match. | |
1073 | */ | |
1074 | if (s->lookahead < MIN_LOOKAHEAD) { | |
1075 | fill_window(s); | |
1076 | if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { | |
1077 | return need_more; | |
1078 | } | |
1079 | if (s->lookahead == 0) break; /* flush the current block */ | |
1080 | } | |
1081 | ||
1082 | /* Insert the string window[strstart .. strstart+2] in the | |
1083 | * dictionary, and set hash_head to the head of the hash chain: | |
1084 | */ | |
1085 | if (s->lookahead >= MIN_MATCH) { | |
1086 | INSERT_STRING(s, s->strstart, hash_head); | |
1087 | } | |
1088 | ||
1089 | /* Find the longest match, discarding those <= prev_length. | |
1090 | * At this point we have always match_length < MIN_MATCH | |
1091 | */ | |
1092 | if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { | |
1093 | /* To simplify the code, we prevent matches with the string | |
1094 | * of window index 0 (in particular we have to avoid a match | |
1095 | * of the string with itself at the start of the input file). | |
1096 | */ | |
1097 | if (s->strategy != Z_HUFFMAN_ONLY) { | |
1098 | s->match_length = longest_match (s, hash_head); | |
1099 | } | |
1100 | /* longest_match() sets match_start */ | |
1101 | } | |
1102 | if (s->match_length >= MIN_MATCH) { | |
1103 | check_match(s, s->strstart, s->match_start, s->match_length); | |
1104 | ||
1105 | bflush = zlib_tr_tally(s, s->strstart - s->match_start, | |
1106 | s->match_length - MIN_MATCH); | |
1107 | ||
1108 | s->lookahead -= s->match_length; | |
1109 | ||
1110 | /* Insert new strings in the hash table only if the match length | |
1111 | * is not too large. This saves time but degrades compression. | |
1112 | */ | |
1113 | if (s->match_length <= s->max_insert_length && | |
1114 | s->lookahead >= MIN_MATCH) { | |
1115 | s->match_length--; /* string at strstart already in hash table */ | |
1116 | do { | |
1117 | s->strstart++; | |
1118 | INSERT_STRING(s, s->strstart, hash_head); | |
1119 | /* strstart never exceeds WSIZE-MAX_MATCH, so there are | |
1120 | * always MIN_MATCH bytes ahead. | |
1121 | */ | |
1122 | } while (--s->match_length != 0); | |
1123 | s->strstart++; | |
1124 | } else { | |
1125 | s->strstart += s->match_length; | |
1126 | s->match_length = 0; | |
1127 | s->ins_h = s->window[s->strstart]; | |
1128 | UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); | |
1129 | #if MIN_MATCH != 3 | |
1130 | Call UPDATE_HASH() MIN_MATCH-3 more times | |
1131 | #endif | |
1132 | /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not | |
1133 | * matter since it will be recomputed at next deflate call. | |
1134 | */ | |
1135 | } | |
1136 | } else { | |
1137 | /* No match, output a literal byte */ | |
1138 | Tracevv((stderr,"%c", s->window[s->strstart])); | |
1139 | bflush = zlib_tr_tally (s, 0, s->window[s->strstart]); | |
1140 | s->lookahead--; | |
1141 | s->strstart++; | |
1142 | } | |
1143 | if (bflush) FLUSH_BLOCK(s, 0); | |
1144 | } | |
1145 | FLUSH_BLOCK(s, flush == Z_FINISH); | |
1146 | return flush == Z_FINISH ? finish_done : block_done; | |
1147 | } | |
1148 | ||
1149 | /* =========================================================================== | |
1150 | * Same as above, but achieves better compression. We use a lazy | |
1151 | * evaluation for matches: a match is finally adopted only if there is | |
1152 | * no better match at the next window position. | |
1153 | */ | |
1154 | static block_state deflate_slow( | |
1155 | deflate_state *s, | |
1156 | int flush | |
1157 | ) | |
1158 | { | |
1159 | IPos hash_head = NIL; /* head of hash chain */ | |
1160 | int bflush; /* set if current block must be flushed */ | |
1161 | ||
1162 | /* Process the input block. */ | |
1163 | for (;;) { | |
1164 | /* Make sure that we always have enough lookahead, except | |
1165 | * at the end of the input file. We need MAX_MATCH bytes | |
1166 | * for the next match, plus MIN_MATCH bytes to insert the | |
1167 | * string following the next match. | |
1168 | */ | |
1169 | if (s->lookahead < MIN_LOOKAHEAD) { | |
1170 | fill_window(s); | |
1171 | if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { | |
1172 | return need_more; | |
1173 | } | |
1174 | if (s->lookahead == 0) break; /* flush the current block */ | |
1175 | } | |
1176 | ||
1177 | /* Insert the string window[strstart .. strstart+2] in the | |
1178 | * dictionary, and set hash_head to the head of the hash chain: | |
1179 | */ | |
1180 | if (s->lookahead >= MIN_MATCH) { | |
1181 | INSERT_STRING(s, s->strstart, hash_head); | |
1182 | } | |
1183 | ||
1184 | /* Find the longest match, discarding those <= prev_length. | |
1185 | */ | |
1186 | s->prev_length = s->match_length, s->prev_match = s->match_start; | |
1187 | s->match_length = MIN_MATCH-1; | |
1188 | ||
1189 | if (hash_head != NIL && s->prev_length < s->max_lazy_match && | |
1190 | s->strstart - hash_head <= MAX_DIST(s)) { | |
1191 | /* To simplify the code, we prevent matches with the string | |
1192 | * of window index 0 (in particular we have to avoid a match | |
1193 | * of the string with itself at the start of the input file). | |
1194 | */ | |
1195 | if (s->strategy != Z_HUFFMAN_ONLY) { | |
1196 | s->match_length = longest_match (s, hash_head); | |
1197 | } | |
1198 | /* longest_match() sets match_start */ | |
1199 | ||
1200 | if (s->match_length <= 5 && (s->strategy == Z_FILTERED || | |
1201 | (s->match_length == MIN_MATCH && | |
1202 | s->strstart - s->match_start > TOO_FAR))) { | |
1203 | ||
1204 | /* If prev_match is also MIN_MATCH, match_start is garbage | |
1205 | * but we will ignore the current match anyway. | |
1206 | */ | |
1207 | s->match_length = MIN_MATCH-1; | |
1208 | } | |
1209 | } | |
1210 | /* If there was a match at the previous step and the current | |
1211 | * match is not better, output the previous match: | |
1212 | */ | |
1213 | if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { | |
1214 | uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; | |
1215 | /* Do not insert strings in hash table beyond this. */ | |
1216 | ||
1217 | check_match(s, s->strstart-1, s->prev_match, s->prev_length); | |
1218 | ||
1219 | bflush = zlib_tr_tally(s, s->strstart -1 - s->prev_match, | |
1220 | s->prev_length - MIN_MATCH); | |
1221 | ||
1222 | /* Insert in hash table all strings up to the end of the match. | |
1223 | * strstart-1 and strstart are already inserted. If there is not | |
1224 | * enough lookahead, the last two strings are not inserted in | |
1225 | * the hash table. | |
1226 | */ | |
1227 | s->lookahead -= s->prev_length-1; | |
1228 | s->prev_length -= 2; | |
1229 | do { | |
1230 | if (++s->strstart <= max_insert) { | |
1231 | INSERT_STRING(s, s->strstart, hash_head); | |
1232 | } | |
1233 | } while (--s->prev_length != 0); | |
1234 | s->match_available = 0; | |
1235 | s->match_length = MIN_MATCH-1; | |
1236 | s->strstart++; | |
1237 | ||
1238 | if (bflush) FLUSH_BLOCK(s, 0); | |
1239 | ||
1240 | } else if (s->match_available) { | |
1241 | /* If there was no match at the previous position, output a | |
1242 | * single literal. If there was a match but the current match | |
1243 | * is longer, truncate the previous match to a single literal. | |
1244 | */ | |
1245 | Tracevv((stderr,"%c", s->window[s->strstart-1])); | |
1246 | if (zlib_tr_tally (s, 0, s->window[s->strstart-1])) { | |
1247 | FLUSH_BLOCK_ONLY(s, 0); | |
1248 | } | |
1249 | s->strstart++; | |
1250 | s->lookahead--; | |
1251 | if (s->strm->avail_out == 0) return need_more; | |
1252 | } else { | |
1253 | /* There is no previous match to compare with, wait for | |
1254 | * the next step to decide. | |
1255 | */ | |
1256 | s->match_available = 1; | |
1257 | s->strstart++; | |
1258 | s->lookahead--; | |
1259 | } | |
1260 | } | |
1261 | Assert (flush != Z_NO_FLUSH, "no flush?"); | |
1262 | if (s->match_available) { | |
1263 | Tracevv((stderr,"%c", s->window[s->strstart-1])); | |
1264 | zlib_tr_tally (s, 0, s->window[s->strstart-1]); | |
1265 | s->match_available = 0; | |
1266 | } | |
1267 | FLUSH_BLOCK(s, flush == Z_FINISH); | |
1268 | return flush == Z_FINISH ? finish_done : block_done; | |
1269 | } | |
1270 | ||
1271 | int zlib_deflate_workspacesize(void) | |
1272 | { | |
1273 | return sizeof(deflate_workspace); | |
1274 | } |