|
1 /* deflate.c -- compress data using the deflation algorithm |
|
2 * Copyright (C) 1995-2002 Jean-loup Gailly. |
|
3 * For conditions of distribution and use, see copyright notice in zlib.h |
|
4 */ |
|
5 |
|
6 /* |
|
7 * ALGORITHM |
|
8 * |
|
9 * The "deflation" process depends on being able to identify portions |
|
10 * of the input text which are identical to earlier input (within a |
|
11 * sliding window trailing behind the input currently being processed). |
|
12 * |
|
13 * The most straightforward technique turns out to be the fastest for |
|
14 * most input files: try all possible matches and select the longest. |
|
15 * The key feature of this algorithm is that insertions into the string |
|
16 * dictionary are very simple and thus fast, and deletions are avoided |
|
17 * completely. Insertions are performed at each input character, whereas |
|
18 * string matches are performed only when the previous match ends. So it |
|
19 * is preferable to spend more time in matches to allow very fast string |
|
20 * insertions and avoid deletions. The matching algorithm for small |
|
21 * strings is inspired from that of Rabin & Karp. A brute force approach |
|
22 * is used to find longer strings when a small match has been found. |
|
23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze |
|
24 * (by Leonid Broukhis). |
|
25 * A previous version of this file used a more sophisticated algorithm |
|
26 * (by Fiala and Greene) which is guaranteed to run in linear amortized |
|
27 * time, but has a larger average cost, uses more memory and is patented. |
|
28 * However the F&G algorithm may be faster for some highly redundant |
|
29 * files if the parameter max_chain_length (described below) is too large. |
|
30 * |
|
31 * ACKNOWLEDGEMENTS |
|
32 * |
|
33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and |
|
34 * I found it in 'freeze' written by Leonid Broukhis. |
|
35 * Thanks to many people for bug reports and testing. |
|
36 * |
|
37 * REFERENCES |
|
38 * |
|
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification". |
|
40 * Available in ftp://ds.internic.net/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 <e32def.h> |
|
53 |
|
54 #include "deflate.h" |
|
55 |
|
56 const char deflate_copyright[] = |
|
57 " deflate 1.1.4 Copyright 1995-2002 Jean-loup Gailly "; |
|
58 /* |
|
59 If you use the zlib library in a product, an acknowledgment is welcome |
|
60 in the documentation of your product. If for some reason you cannot |
|
61 include such an acknowledgment, I would appreciate that you keep this |
|
62 copyright string in the executable of your product. |
|
63 */ |
|
64 |
|
65 /* =========================================================================== |
|
66 * Function prototypes. |
|
67 */ |
|
68 typedef enum { |
|
69 need_more, /* block not completed, need more input or more output */ |
|
70 block_done, /* block flush performed */ |
|
71 finish_started, /* finish started, need only more output at next deflate */ |
|
72 finish_done /* finish done, accept no more input or output */ |
|
73 } block_state; |
|
74 |
|
75 typedef block_state (*compress_func) OF((deflate_state *s, int flush)); |
|
76 /* Compression function. Returns the block state after the call. */ |
|
77 |
|
78 local void fill_window OF((deflate_state *s)); |
|
79 local block_state deflate_stored OF((deflate_state *s, int flush)); |
|
80 local block_state deflate_fast OF((deflate_state *s, int flush)); |
|
81 local block_state deflate_slow OF((deflate_state *s, int flush)); |
|
82 local void lm_init OF((deflate_state *s)); |
|
83 local void putShortMSB OF((deflate_state *s, uInt b)); |
|
84 local void flush_pending OF((z_streamp strm)); |
|
85 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); |
|
86 #ifdef ASMV |
|
87 void match_init OF((void)); /* asm code initialization */ |
|
88 uInt longest_match OF((deflate_state *s, IPos cur_match)); |
|
89 #else |
|
90 local uInt longest_match OF((deflate_state *s, IPos cur_match)); |
|
91 #endif |
|
92 |
|
93 #ifdef DEBUG |
|
94 local void check_match OF((deflate_state *s, IPos start, IPos match, |
|
95 int length)); |
|
96 #endif |
|
97 |
|
98 /* =========================================================================== |
|
99 * Local data |
|
100 */ |
|
101 |
|
102 #define NIL 0 |
|
103 /* Tail of hash chains */ |
|
104 |
|
105 #ifndef TOO_FAR |
|
106 # define TOO_FAR 4096 |
|
107 #endif |
|
108 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ |
|
109 |
|
110 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) |
|
111 /* Minimum amount of lookahead, except at the end of the input file. |
|
112 * See deflate.c for comments about the MIN_MATCH+1. |
|
113 */ |
|
114 |
|
115 /* Values for max_lazy_match, good_match and max_chain_length, depending on |
|
116 * the desired pack level (0..9). The values given below have been tuned to |
|
117 * exclude worst case performance for pathological files. Better values may be |
|
118 * found for specific files. |
|
119 */ |
|
120 typedef struct config_s { |
|
121 ush good_length; /* reduce lazy search above this match length */ |
|
122 ush max_lazy; /* do not perform lazy search above this match length */ |
|
123 ush nice_length; /* quit search above this match length */ |
|
124 ush max_chain; |
|
125 compress_func func; |
|
126 } config; |
|
127 |
|
128 local const config configuration_table[10] = { |
|
129 /* good lazy nice chain */ |
|
130 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ |
|
131 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */ |
|
132 /* 2 */ {4, 5, 16, 8, deflate_fast}, |
|
133 /* 3 */ {4, 6, 32, 32, deflate_fast}, |
|
134 |
|
135 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ |
|
136 /* 5 */ {8, 16, 32, 32, deflate_slow}, |
|
137 /* 6 */ {8, 16, 128, 128, deflate_slow}, |
|
138 /* 7 */ {8, 32, 128, 256, deflate_slow}, |
|
139 /* 8 */ {32, 128, 258, 1024, deflate_slow}, |
|
140 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */ |
|
141 |
|
142 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 |
|
143 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different |
|
144 * meaning. |
|
145 */ |
|
146 |
|
147 #define EQUAL 0 |
|
148 /* result of memcmp for equal strings */ |
|
149 |
|
150 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */ |
|
151 |
|
152 /* =========================================================================== |
|
153 * Update a hash value with the given input byte |
|
154 * IN assertion: all calls to to UPDATE_HASH are made with consecutive |
|
155 * input characters, so that a running hash key can be computed from the |
|
156 * previous key instead of complete recalculation each time. |
|
157 */ |
|
158 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask) |
|
159 |
|
160 |
|
161 /* =========================================================================== |
|
162 * Insert string str in the dictionary and set match_head to the previous head |
|
163 * of the hash chain (the most recent string with same hash key). Return |
|
164 * the previous length of the hash chain. |
|
165 * If this file is compiled with -DFASTEST, the compression level is forced |
|
166 * to 1, and no hash chains are maintained. |
|
167 * IN assertion: all calls to to INSERT_STRING are made with consecutive |
|
168 * input characters and the first MIN_MATCH bytes of str are valid |
|
169 * (except for the last MIN_MATCH-1 bytes of the input file). |
|
170 */ |
|
171 #ifdef FASTEST |
|
172 #define INSERT_STRING(s, str, match_head) \ |
|
173 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ |
|
174 match_head = s->head[s->ins_h], \ |
|
175 s->head[s->ins_h] = STATIC_CAST(Pos,str)) |
|
176 #else |
|
177 #define INSERT_STRING(s, str, match_head) \ |
|
178 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ |
|
179 s->prev[(str) & s->w_mask] = STATIC_CAST(Posf,match_head = s->head[s->ins_h]), \ |
|
180 s->head[s->ins_h] = STATIC_CAST(Pos,str)) |
|
181 #endif |
|
182 |
|
183 /* =========================================================================== |
|
184 * Initialize the hash table (avoiding 64K overflow for 16 bit systems). |
|
185 * prev[] will be initialized on the fly. |
|
186 */ |
|
187 #define CLEAR_HASH(s) \ |
|
188 s->head[s->hash_size-1] = NIL; \ |
|
189 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head)); |
|
190 |
|
191 /* ========================================================================= */ |
|
192 EXPORT_C int ZEXPORT deflateInit_( |
|
193 z_streamp strm, |
|
194 int level, |
|
195 const char *version, |
|
196 int stream_size) |
|
197 { |
|
198 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, |
|
199 Z_DEFAULT_STRATEGY, version, stream_size); |
|
200 /* To do: ignore strm->next_in if we use it as window */ |
|
201 } |
|
202 |
|
203 /* ========================================================================= */ |
|
204 EXPORT_C int ZEXPORT deflateInit2_( |
|
205 z_streamp strm, |
|
206 int level, |
|
207 int method, |
|
208 int windowBits, |
|
209 int memLevel, |
|
210 int strategy, |
|
211 const char *version, |
|
212 int stream_size) |
|
213 { |
|
214 deflate_state *s; |
|
215 int noheader = 0; |
|
216 static const char* const my_version = ZLIB_VERSION; |
|
217 |
|
218 ushf *overlay; |
|
219 /* We overlay pending_buf and d_buf+l_buf. This works since the average |
|
220 * output size for (length,distance) codes is <= 24 bits. |
|
221 */ |
|
222 |
|
223 if (version == Z_NULL || version[0] != my_version[0] || |
|
224 stream_size != sizeof(z_stream)) { |
|
225 return Z_VERSION_ERROR; |
|
226 } |
|
227 if (strm == Z_NULL) return Z_STREAM_ERROR; |
|
228 |
|
229 strm->msg = Z_NULL; |
|
230 if (strm->zalloc == Z_NULL) { |
|
231 strm->zalloc = zcalloc; |
|
232 strm->opaque = (voidpf)0; |
|
233 } |
|
234 if (strm->zfree == Z_NULL) strm->zfree = zcfree; |
|
235 |
|
236 if (level == Z_DEFAULT_COMPRESSION) level = 6; |
|
237 #ifdef FASTEST |
|
238 level = 1; |
|
239 #endif |
|
240 |
|
241 if (windowBits < 0) { /* undocumented feature: suppress zlib header */ |
|
242 noheader = 1; |
|
243 windowBits = -windowBits; |
|
244 } |
|
245 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || |
|
246 windowBits < 9 || windowBits > 15 || level < 0 || level > 9 || |
|
247 strategy < 0 || strategy > Z_HUFFMAN_ONLY) { |
|
248 return Z_STREAM_ERROR; |
|
249 } |
|
250 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); |
|
251 if (s == Z_NULL) return Z_MEM_ERROR; |
|
252 strm->state = (struct internal_state FAR *)s; |
|
253 s->strm = strm; |
|
254 |
|
255 s->noheader = noheader; |
|
256 s->w_bits = windowBits; |
|
257 s->w_size = 1 << s->w_bits; |
|
258 s->w_mask = s->w_size - 1; |
|
259 |
|
260 s->hash_bits = memLevel + 7; |
|
261 s->hash_size = 1 << s->hash_bits; |
|
262 s->hash_mask = s->hash_size - 1; |
|
263 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); |
|
264 |
|
265 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); |
|
266 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); |
|
267 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); |
|
268 |
|
269 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ |
|
270 |
|
271 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2); |
|
272 s->pending_buf = (uchf *) overlay; |
|
273 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); |
|
274 |
|
275 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || |
|
276 s->pending_buf == Z_NULL) { |
|
277 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR); |
|
278 |
|
279 // The following line has been added by Markr. Up to this point s->status has not been |
|
280 // initialised. If this is the case when deflateEnd is called then it will fail to reclaim |
|
281 // the memory allocated for s->window etc when alloc fails. Setting the status to INIT_STATE |
|
282 // allows deflateEnd to reclaim memory. |
|
283 |
|
284 s->status = INIT_STATE; |
|
285 deflateEnd (strm); |
|
286 return Z_MEM_ERROR; |
|
287 } |
|
288 s->d_buf = overlay + s->lit_bufsize/sizeof(ush); |
|
289 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; |
|
290 |
|
291 s->level = level; |
|
292 s->strategy = strategy; |
|
293 s->method = (Byte)method; |
|
294 |
|
295 return deflateReset(strm); |
|
296 } |
|
297 |
|
298 /* ========================================================================= */ |
|
299 EXPORT_C int ZEXPORT deflateSetDictionary ( |
|
300 z_streamp strm, |
|
301 const Bytef *dictionary, |
|
302 uInt dictLength) |
|
303 { |
|
304 // Line to stop compiler warning about unused mandatory global variable |
|
305 char __z=deflate_copyright[0]; __z=__z; |
|
306 |
|
307 deflate_state *s; |
|
308 uInt length = dictLength; |
|
309 uInt n; |
|
310 IPos hash_head = 0; |
|
311 |
|
312 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL || |
|
313 strm->state->status != INIT_STATE) return Z_STREAM_ERROR; |
|
314 |
|
315 s = strm->state; |
|
316 strm->adler = adler32(strm->adler, dictionary, dictLength); |
|
317 |
|
318 if (length < MIN_MATCH) return Z_OK; |
|
319 if (length > MAX_DIST(s)) { |
|
320 length = MAX_DIST(s); |
|
321 #ifndef USE_DICT_HEAD |
|
322 dictionary += dictLength - length; /* use the tail of the dictionary */ |
|
323 #endif |
|
324 } |
|
325 zmemcpy(s->window, dictionary, length); |
|
326 s->strstart = length; |
|
327 s->block_start = (long)length; |
|
328 |
|
329 /* Insert all strings in the hash table (except for the last two bytes). |
|
330 * s->lookahead stays null, so s->ins_h will be recomputed at the next |
|
331 * call of fill_window. |
|
332 */ |
|
333 s->ins_h = s->window[0]; |
|
334 UPDATE_HASH(s, s->ins_h, s->window[1]); |
|
335 for (n = 0; n <= length - MIN_MATCH; n++) { |
|
336 INSERT_STRING(s, n, hash_head); |
|
337 } |
|
338 if (hash_head) hash_head = 0; /* to make compiler happy */ |
|
339 return Z_OK; |
|
340 } |
|
341 |
|
342 /* ========================================================================= */ |
|
343 EXPORT_C int ZEXPORT deflateReset ( |
|
344 z_streamp strm) |
|
345 { |
|
346 deflate_state *s; |
|
347 |
|
348 if (strm == Z_NULL || strm->state == Z_NULL || |
|
349 strm->zalloc == Z_NULL || strm->zfree == Z_NULL) return Z_STREAM_ERROR; |
|
350 |
|
351 strm->total_in = strm->total_out = 0; |
|
352 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ |
|
353 strm->data_type = Z_UNKNOWN; |
|
354 |
|
355 s = (deflate_state *)strm->state; |
|
356 s->pending = 0; |
|
357 s->pending_out = s->pending_buf; |
|
358 |
|
359 if (s->noheader < 0) { |
|
360 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */ |
|
361 } |
|
362 s->status = s->noheader ? BUSY_STATE : INIT_STATE; |
|
363 strm->adler = 1; |
|
364 s->last_flush = Z_NO_FLUSH; |
|
365 |
|
366 _tr_init(s); |
|
367 lm_init(s); |
|
368 |
|
369 return Z_OK; |
|
370 } |
|
371 |
|
372 /* ========================================================================= */ |
|
373 EXPORT_C int ZEXPORT deflateParams( |
|
374 z_streamp strm, |
|
375 int level, |
|
376 int strategy) |
|
377 { |
|
378 deflate_state *s; |
|
379 compress_func func; |
|
380 int err = Z_OK; |
|
381 |
|
382 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; |
|
383 s = strm->state; |
|
384 |
|
385 if (level == Z_DEFAULT_COMPRESSION) { |
|
386 level = 6; |
|
387 } |
|
388 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) { |
|
389 return Z_STREAM_ERROR; |
|
390 } |
|
391 func = configuration_table[s->level].func; |
|
392 |
|
393 if (func != configuration_table[level].func && strm->total_in != 0) { |
|
394 /* Flush the last buffer: */ |
|
395 err = deflate(strm, Z_PARTIAL_FLUSH); |
|
396 } |
|
397 if (s->level != level) { |
|
398 s->level = level; |
|
399 s->max_lazy_match = configuration_table[level].max_lazy; |
|
400 s->good_match = configuration_table[level].good_length; |
|
401 s->nice_match = configuration_table[level].nice_length; |
|
402 s->max_chain_length = configuration_table[level].max_chain; |
|
403 } |
|
404 s->strategy = strategy; |
|
405 return err; |
|
406 } |
|
407 |
|
408 /* ========================================================================= |
|
409 * Put a short in the pending buffer. The 16-bit value is put in MSB order. |
|
410 * IN assertion: the stream state is correct and there is enough room in |
|
411 * pending_buf. |
|
412 */ |
|
413 local void putShortMSB ( |
|
414 deflate_state *s, |
|
415 uInt b) |
|
416 { |
|
417 put_byte(s, (Byte)(b >> 8)); |
|
418 put_byte(s, (Byte)(b & 0xff)); |
|
419 } |
|
420 |
|
421 /* ========================================================================= |
|
422 * Flush as much pending output as possible. All deflate() output goes |
|
423 * through this function so some applications may wish to modify it |
|
424 * to avoid allocating a large strm->next_out buffer and copying into it. |
|
425 * (See also read_buf()). |
|
426 */ |
|
427 local void flush_pending( |
|
428 z_streamp strm) |
|
429 { |
|
430 unsigned len = strm->state->pending; |
|
431 |
|
432 if (len > strm->avail_out) len = strm->avail_out; |
|
433 if (len == 0) return; |
|
434 |
|
435 zmemcpy(strm->next_out, strm->state->pending_out, len); |
|
436 strm->next_out += len; |
|
437 strm->state->pending_out += len; |
|
438 strm->total_out += len; |
|
439 strm->avail_out -= len; |
|
440 strm->state->pending -= len; |
|
441 if (strm->state->pending == 0) { |
|
442 strm->state->pending_out = strm->state->pending_buf; |
|
443 } |
|
444 } |
|
445 |
|
446 /* ========================================================================= */ |
|
447 EXPORT_C int ZEXPORT deflate ( |
|
448 z_streamp strm, |
|
449 int flush) |
|
450 { |
|
451 int old_flush; /* value of flush param for previous deflate call */ |
|
452 deflate_state *s; |
|
453 |
|
454 if (strm == Z_NULL || strm->state == Z_NULL || |
|
455 flush > Z_FINISH || flush < 0) { |
|
456 return Z_STREAM_ERROR; |
|
457 } |
|
458 s = strm->state; |
|
459 |
|
460 if (strm->next_out == Z_NULL || |
|
461 (strm->next_in == Z_NULL && strm->avail_in != 0) || |
|
462 (s->status == FINISH_STATE && flush != Z_FINISH)) { |
|
463 ERR_RETURN(strm, Z_STREAM_ERROR); |
|
464 } |
|
465 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); |
|
466 |
|
467 s->strm = strm; /* just in case */ |
|
468 old_flush = s->last_flush; |
|
469 s->last_flush = flush; |
|
470 |
|
471 /* Write the zlib header */ |
|
472 if (s->status == INIT_STATE) { |
|
473 |
|
474 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; |
|
475 uInt level_flags = (s->level-1) >> 1; |
|
476 |
|
477 if (level_flags > 3) level_flags = 3; |
|
478 header |= (level_flags << 6); |
|
479 if (s->strstart != 0) header |= PRESET_DICT; |
|
480 header += 31 - (header % 31); |
|
481 |
|
482 s->status = BUSY_STATE; |
|
483 putShortMSB(s, header); |
|
484 |
|
485 /* Save the adler32 of the preset dictionary: */ |
|
486 if (s->strstart != 0) { |
|
487 putShortMSB(s, (uInt)(strm->adler >> 16)); |
|
488 putShortMSB(s, (uInt)(strm->adler & 0xffff)); |
|
489 } |
|
490 strm->adler = 1L; |
|
491 } |
|
492 |
|
493 /* Flush as much pending output as possible */ |
|
494 if (s->pending != 0) { |
|
495 flush_pending(strm); |
|
496 if (strm->avail_out == 0) { |
|
497 /* Since avail_out is 0, deflate will be called again with |
|
498 * more output space, but possibly with both pending and |
|
499 * avail_in equal to zero. There won't be anything to do, |
|
500 * but this is not an error situation so make sure we |
|
501 * return OK instead of BUF_ERROR at next call of deflate: |
|
502 */ |
|
503 s->last_flush = -1; |
|
504 return Z_OK; |
|
505 } |
|
506 |
|
507 /* Make sure there is something to do and avoid duplicate consecutive |
|
508 * flushes. For repeated and useless calls with Z_FINISH, we keep |
|
509 * returning Z_STREAM_END instead of Z_BUFF_ERROR. |
|
510 */ |
|
511 } else if (strm->avail_in == 0 && flush <= old_flush && |
|
512 flush != Z_FINISH) { |
|
513 ERR_RETURN(strm, Z_BUF_ERROR); |
|
514 } |
|
515 |
|
516 /* User must not provide more input after the first FINISH: */ |
|
517 if (s->status == FINISH_STATE && strm->avail_in != 0) { |
|
518 ERR_RETURN(strm, Z_BUF_ERROR); |
|
519 } |
|
520 |
|
521 /* Start a new block or continue the current one. |
|
522 */ |
|
523 if (strm->avail_in != 0 || s->lookahead != 0 || |
|
524 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { |
|
525 block_state bstate; |
|
526 |
|
527 bstate = (*(configuration_table[s->level].func))(s, flush); |
|
528 |
|
529 if (bstate == finish_started || bstate == finish_done) { |
|
530 s->status = FINISH_STATE; |
|
531 } |
|
532 if (bstate == need_more || bstate == finish_started) { |
|
533 if (strm->avail_out == 0) { |
|
534 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ |
|
535 } |
|
536 return Z_OK; |
|
537 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call |
|
538 * of deflate should use the same flush parameter to make sure |
|
539 * that the flush is complete. So we don't have to output an |
|
540 * empty block here, this will be done at next call. This also |
|
541 * ensures that for a very small output buffer, we emit at most |
|
542 * one empty block. |
|
543 */ |
|
544 } |
|
545 if (bstate == block_done) { |
|
546 if (flush == Z_PARTIAL_FLUSH) { |
|
547 _tr_align(s); |
|
548 } else { /* FULL_FLUSH or SYNC_FLUSH */ |
|
549 _tr_stored_block(s, (char*)0, 0L, 0); |
|
550 /* For a full flush, this empty block will be recognized |
|
551 * as a special marker by inflate_sync(). |
|
552 */ |
|
553 if (flush == Z_FULL_FLUSH) { |
|
554 CLEAR_HASH(s); /* forget history */ |
|
555 } |
|
556 } |
|
557 flush_pending(strm); |
|
558 if (strm->avail_out == 0) { |
|
559 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ |
|
560 return Z_OK; |
|
561 } |
|
562 } |
|
563 } |
|
564 Assert(strm->avail_out > 0, "bug2"); |
|
565 |
|
566 if (flush != Z_FINISH) return Z_OK; |
|
567 if (s->noheader) return Z_STREAM_END; |
|
568 |
|
569 /* Write the zlib trailer (adler32) */ |
|
570 putShortMSB(s, (uInt)(strm->adler >> 16)); |
|
571 putShortMSB(s, (uInt)(strm->adler & 0xffff)); |
|
572 flush_pending(strm); |
|
573 /* If avail_out is zero, the application will call deflate again |
|
574 * to flush the rest. |
|
575 */ |
|
576 s->noheader = -1; /* write the trailer only once! */ |
|
577 return s->pending != 0 ? Z_OK : Z_STREAM_END; |
|
578 } |
|
579 |
|
580 /* ========================================================================= */ |
|
581 EXPORT_C int ZEXPORT deflateEnd ( |
|
582 z_streamp strm) |
|
583 { |
|
584 int status; |
|
585 |
|
586 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; |
|
587 |
|
588 status = strm->state->status; |
|
589 if (status != INIT_STATE && status != BUSY_STATE && |
|
590 status != FINISH_STATE) { |
|
591 return Z_STREAM_ERROR; |
|
592 } |
|
593 |
|
594 /* Deallocate in reverse order of allocations: */ |
|
595 TRY_FREE(strm, strm->state->pending_buf); |
|
596 TRY_FREE(strm, strm->state->head); |
|
597 TRY_FREE(strm, strm->state->prev); |
|
598 TRY_FREE(strm, strm->state->window); |
|
599 |
|
600 ZFREE(strm, strm->state); |
|
601 strm->state = Z_NULL; |
|
602 |
|
603 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; |
|
604 } |
|
605 |
|
606 /* ========================================================================= |
|
607 * Copy the source state to the destination state. |
|
608 * To simplify the source, this is not supported for 16-bit MSDOS (which |
|
609 * doesn't have enough memory anyway to duplicate compression states). |
|
610 */ |
|
611 EXPORT_C int ZEXPORT deflateCopy ( |
|
612 z_streamp dest, |
|
613 z_streamp source) |
|
614 { |
|
615 #ifdef MAXSEG_64K |
|
616 return Z_STREAM_ERROR; |
|
617 #else |
|
618 deflate_state *ds; |
|
619 deflate_state *ss; |
|
620 ushf *overlay; |
|
621 |
|
622 |
|
623 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) { |
|
624 return Z_STREAM_ERROR; |
|
625 } |
|
626 |
|
627 ss = source->state; |
|
628 |
|
629 *dest = *source; |
|
630 |
|
631 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); |
|
632 if (ds == Z_NULL) return Z_MEM_ERROR; |
|
633 dest->state = (struct internal_state FAR *) ds; |
|
634 *ds = *ss; |
|
635 ds->strm = dest; |
|
636 |
|
637 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); |
|
638 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); |
|
639 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); |
|
640 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2); |
|
641 ds->pending_buf = (uchf *) overlay; |
|
642 |
|
643 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || |
|
644 ds->pending_buf == Z_NULL) { |
|
645 deflateEnd (dest); |
|
646 return Z_MEM_ERROR; |
|
647 } |
|
648 /* following zmemcpy do not work for 16-bit MSDOS */ |
|
649 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); |
|
650 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos)); |
|
651 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos)); |
|
652 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); |
|
653 |
|
654 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); |
|
655 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); |
|
656 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; |
|
657 |
|
658 ds->l_desc.dyn_tree = ds->dyn_ltree; |
|
659 ds->d_desc.dyn_tree = ds->dyn_dtree; |
|
660 ds->bl_desc.dyn_tree = ds->bl_tree; |
|
661 |
|
662 return Z_OK; |
|
663 #endif |
|
664 } |
|
665 |
|
666 /* =========================================================================== |
|
667 * Read a new buffer from the current input stream, update the adler32 |
|
668 * and total number of bytes read. All deflate() input goes through |
|
669 * this function so some applications may wish to modify it to avoid |
|
670 * allocating a large strm->next_in buffer and copying from it. |
|
671 * (See also flush_pending()). |
|
672 */ |
|
673 local int read_buf( |
|
674 z_streamp strm, |
|
675 Bytef *buf, |
|
676 unsigned size) |
|
677 { |
|
678 unsigned len = strm->avail_in; |
|
679 |
|
680 if (len > size) len = size; |
|
681 if (len == 0) return 0; |
|
682 |
|
683 strm->avail_in -= len; |
|
684 |
|
685 if (!strm->state->noheader) { |
|
686 strm->adler = adler32(strm->adler, strm->next_in, len); |
|
687 } |
|
688 zmemcpy(buf, strm->next_in, len); |
|
689 strm->next_in += len; |
|
690 strm->total_in += len; |
|
691 |
|
692 return (int)len; |
|
693 } |
|
694 |
|
695 /* =========================================================================== |
|
696 * Initialize the "longest match" routines for a new zlib stream |
|
697 */ |
|
698 local void lm_init ( |
|
699 deflate_state *s) |
|
700 { |
|
701 s->window_size = (ulg)2L*s->w_size; |
|
702 |
|
703 CLEAR_HASH(s); |
|
704 |
|
705 /* Set the default configuration parameters: |
|
706 */ |
|
707 s->max_lazy_match = configuration_table[s->level].max_lazy; |
|
708 s->good_match = configuration_table[s->level].good_length; |
|
709 s->nice_match = configuration_table[s->level].nice_length; |
|
710 s->max_chain_length = configuration_table[s->level].max_chain; |
|
711 |
|
712 s->strstart = 0; |
|
713 s->block_start = 0L; |
|
714 s->lookahead = 0; |
|
715 s->match_length = s->prev_length = MIN_MATCH-1; |
|
716 s->match_available = 0; |
|
717 s->ins_h = 0; |
|
718 #ifdef ASMV |
|
719 match_init(); /* initialize the asm code */ |
|
720 #endif |
|
721 } |
|
722 |
|
723 /* =========================================================================== |
|
724 * Set match_start to the longest match starting at the given string and |
|
725 * return its length. Matches shorter or equal to prev_length are discarded, |
|
726 * in which case the result is equal to prev_length and match_start is |
|
727 * garbage. |
|
728 * IN assertions: cur_match is the head of the hash chain for the current |
|
729 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 |
|
730 * OUT assertion: the match length is not greater than s->lookahead. |
|
731 */ |
|
732 #ifndef ASMV |
|
733 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or |
|
734 * match.S. The code will be functionally equivalent. |
|
735 */ |
|
736 #ifndef FASTEST |
|
737 local uInt longest_match( |
|
738 deflate_state *s, |
|
739 IPos cur_match) /* current match */ |
|
740 { |
|
741 unsigned chain_length = s->max_chain_length;/* max hash chain length */ |
|
742 register Bytef *scan = s->window + s->strstart; /* current string */ |
|
743 register Bytef *match; /* matched string */ |
|
744 register int len; /* length of current match */ |
|
745 int best_len = s->prev_length; /* best match length so far */ |
|
746 int nice_match = s->nice_match; /* stop if match long enough */ |
|
747 IPos limit = s->strstart > (IPos)MAX_DIST(s) ? |
|
748 s->strstart - (IPos)MAX_DIST(s) : NIL; |
|
749 /* Stop when cur_match becomes <= limit. To simplify the code, |
|
750 * we prevent matches with the string of window index 0. |
|
751 */ |
|
752 Posf *prev = s->prev; |
|
753 uInt wmask = s->w_mask; |
|
754 |
|
755 #ifdef UNALIGNED_OK |
|
756 /* Compare two bytes at a time. Note: this is not always beneficial. |
|
757 * Try with and without -DUNALIGNED_OK to check. |
|
758 */ |
|
759 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; |
|
760 register ush scan_start = *(ushf*)scan; |
|
761 register ush scan_end = *(ushf*)(scan+best_len-1); |
|
762 #else |
|
763 register Bytef *strend = s->window + s->strstart + MAX_MATCH; |
|
764 register Byte scan_end1 = scan[best_len-1]; |
|
765 register Byte scan_end = scan[best_len]; |
|
766 #endif |
|
767 |
|
768 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. |
|
769 * It is easy to get rid of this optimization if necessary. |
|
770 */ |
|
771 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); |
|
772 |
|
773 /* Do not waste too much time if we already have a good match: */ |
|
774 if (s->prev_length >= s->good_match) { |
|
775 chain_length >>= 2; |
|
776 } |
|
777 /* Do not look for matches beyond the end of the input. This is necessary |
|
778 * to make deflate deterministic. |
|
779 */ |
|
780 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; |
|
781 |
|
782 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); |
|
783 |
|
784 do { |
|
785 Assert(cur_match < s->strstart, "no future"); |
|
786 match = s->window + cur_match; |
|
787 |
|
788 /* Skip to next match if the match length cannot increase |
|
789 * or if the match length is less than 2: |
|
790 */ |
|
791 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) |
|
792 /* This code assumes sizeof(unsigned short) == 2. Do not use |
|
793 * UNALIGNED_OK if your compiler uses a different size. |
|
794 */ |
|
795 if (*(ushf*)(match+best_len-1) != scan_end || |
|
796 *(ushf*)match != scan_start) continue; |
|
797 |
|
798 /* It is not necessary to compare scan[2] and match[2] since they are |
|
799 * always equal when the other bytes match, given that the hash keys |
|
800 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at |
|
801 * strstart+3, +5, ... up to strstart+257. We check for insufficient |
|
802 * lookahead only every 4th comparison; the 128th check will be made |
|
803 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is |
|
804 * necessary to put more guard bytes at the end of the window, or |
|
805 * to check more often for insufficient lookahead. |
|
806 */ |
|
807 Assert(scan[2] == match[2], "scan[2]?"); |
|
808 scan++, match++; |
|
809 do { |
|
810 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) && |
|
811 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && |
|
812 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && |
|
813 *(ushf*)(scan+=2) == *(ushf*)(match+=2) && |
|
814 scan < strend); |
|
815 /* The funny "do {}" generates better code on most compilers */ |
|
816 |
|
817 /* Here, scan <= window+strstart+257 */ |
|
818 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); |
|
819 if (*scan == *match) scan++; |
|
820 |
|
821 len = (MAX_MATCH - 1) - (int)(strend-scan); |
|
822 scan = strend - (MAX_MATCH-1); |
|
823 |
|
824 #else /* UNALIGNED_OK */ |
|
825 |
|
826 if (match[best_len] != scan_end || |
|
827 match[best_len-1] != scan_end1 || |
|
828 *match != *scan || |
|
829 *++match != scan[1]) continue; |
|
830 |
|
831 /* The check at best_len-1 can be removed because it will be made |
|
832 * again later. (This heuristic is not always a win.) |
|
833 * It is not necessary to compare scan[2] and match[2] since they |
|
834 * are always equal when the other bytes match, given that |
|
835 * the hash keys are equal and that HASH_BITS >= 8. |
|
836 */ |
|
837 scan += 2, match++; |
|
838 Assert(*scan == *match, "match[2]?"); |
|
839 |
|
840 /* We check for insufficient lookahead only every 8th comparison; |
|
841 * the 256th check will be made at strstart+258. |
|
842 */ |
|
843 do { |
|
844 } while (*++scan == *++match && *++scan == *++match && |
|
845 *++scan == *++match && *++scan == *++match && |
|
846 *++scan == *++match && *++scan == *++match && |
|
847 *++scan == *++match && *++scan == *++match && |
|
848 scan < strend); |
|
849 |
|
850 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); |
|
851 |
|
852 len = MAX_MATCH - (int)(strend - scan); |
|
853 scan = strend - MAX_MATCH; |
|
854 |
|
855 #endif /* UNALIGNED_OK */ |
|
856 |
|
857 if (len > best_len) { |
|
858 s->match_start = cur_match; |
|
859 best_len = len; |
|
860 if (len >= nice_match) break; |
|
861 #ifdef UNALIGNED_OK |
|
862 scan_end = *(ushf*)(scan+best_len-1); |
|
863 #else |
|
864 scan_end1 = scan[best_len-1]; |
|
865 scan_end = scan[best_len]; |
|
866 #endif |
|
867 } |
|
868 } while ((cur_match = prev[cur_match & wmask]) > limit |
|
869 && --chain_length != 0); |
|
870 |
|
871 if ((uInt)best_len <= s->lookahead) return (uInt)best_len; |
|
872 return s->lookahead; |
|
873 } |
|
874 |
|
875 #else /* FASTEST */ |
|
876 /* --------------------------------------------------------------------------- |
|
877 * Optimized version for level == 1 only |
|
878 */ |
|
879 local uInt longest_match( |
|
880 deflate_state *s, |
|
881 IPos cur_match) /* current match */ |
|
882 { |
|
883 register Bytef *scan = s->window + s->strstart; /* current string */ |
|
884 register Bytef *match; /* matched string */ |
|
885 register int len; /* length of current match */ |
|
886 register Bytef *strend = s->window + s->strstart + MAX_MATCH; |
|
887 |
|
888 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. |
|
889 * It is easy to get rid of this optimization if necessary. |
|
890 */ |
|
891 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); |
|
892 |
|
893 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); |
|
894 |
|
895 Assert(cur_match < s->strstart, "no future"); |
|
896 |
|
897 match = s->window + cur_match; |
|
898 |
|
899 /* Return failure if the match length is less than 2: |
|
900 */ |
|
901 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; |
|
902 |
|
903 /* The check at best_len-1 can be removed because it will be made |
|
904 * again later. (This heuristic is not always a win.) |
|
905 * It is not necessary to compare scan[2] and match[2] since they |
|
906 * are always equal when the other bytes match, given that |
|
907 * the hash keys are equal and that HASH_BITS >= 8. |
|
908 */ |
|
909 scan += 2, match += 2; |
|
910 Assert(*scan == *match, "match[2]?"); |
|
911 |
|
912 /* We check for insufficient lookahead only every 8th comparison; |
|
913 * the 256th check will be made at strstart+258. |
|
914 */ |
|
915 do { |
|
916 } while (*++scan == *++match && *++scan == *++match && |
|
917 *++scan == *++match && *++scan == *++match && |
|
918 *++scan == *++match && *++scan == *++match && |
|
919 *++scan == *++match && *++scan == *++match && |
|
920 scan < strend); |
|
921 |
|
922 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); |
|
923 |
|
924 len = MAX_MATCH - (int)(strend - scan); |
|
925 |
|
926 if (len < MIN_MATCH) return MIN_MATCH - 1; |
|
927 |
|
928 s->match_start = cur_match; |
|
929 return len <= s->lookahead ? len : s->lookahead; |
|
930 } |
|
931 #endif /* FASTEST */ |
|
932 #endif /* ASMV */ |
|
933 |
|
934 #ifdef DEBUG |
|
935 /* =========================================================================== |
|
936 * Check that the match at match_start is indeed a match. |
|
937 */ |
|
938 local void check_match( |
|
939 deflate_state *s, |
|
940 IPos start, match, |
|
941 int length) |
|
942 { |
|
943 /* check that the match is indeed a match */ |
|
944 if (zmemcmp(s->window + match, |
|
945 s->window + start, length) != EQUAL) { |
|
946 fprintf(stderr, " start %u, match %u, length %d\n", |
|
947 start, match, length); |
|
948 do { |
|
949 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); |
|
950 } while (--length != 0); |
|
951 z_error("invalid match"); |
|
952 } |
|
953 if (z_verbose > 1) { |
|
954 fprintf(stderr,"\\[%d,%d]", start-match, length); |
|
955 do { putc(s->window[start++], stderr); } while (--length != 0); |
|
956 } |
|
957 } |
|
958 #else |
|
959 # define check_match(s, start, match, length) |
|
960 #endif |
|
961 |
|
962 /* =========================================================================== |
|
963 * Fill the window when the lookahead becomes insufficient. |
|
964 * Updates strstart and lookahead. |
|
965 * |
|
966 * IN assertion: lookahead < MIN_LOOKAHEAD |
|
967 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD |
|
968 * At least one byte has been read, or avail_in == 0; reads are |
|
969 * performed for at least two bytes (required for the zip translate_eol |
|
970 * option -- not supported here). |
|
971 */ |
|
972 local void fill_window( |
|
973 deflate_state *s) |
|
974 { |
|
975 register unsigned n, m; |
|
976 register Posf *p; |
|
977 unsigned more; /* Amount of free space at the end of the window. */ |
|
978 uInt wsize = s->w_size; |
|
979 |
|
980 do { |
|
981 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); |
|
982 |
|
983 /* Deal with !@#$% 64K limit: */ |
|
984 if (more == 0 && s->strstart == 0 && s->lookahead == 0) { |
|
985 more = wsize; |
|
986 |
|
987 } else if (more == (unsigned)(-1)) { |
|
988 /* Very unlikely, but possible on 16 bit machine if strstart == 0 |
|
989 * and lookahead == 1 (input done one byte at time) |
|
990 */ |
|
991 more--; |
|
992 |
|
993 /* If the window is almost full and there is insufficient lookahead, |
|
994 * move the upper half to the lower one to make room in the upper half. |
|
995 */ |
|
996 } else if (s->strstart >= wsize+MAX_DIST(s)) { |
|
997 |
|
998 zmemcpy(s->window, s->window+wsize, (unsigned)wsize); |
|
999 s->match_start -= wsize; |
|
1000 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ |
|
1001 s->block_start -= (long) wsize; |
|
1002 |
|
1003 /* Slide the hash table (could be avoided with 32 bit values |
|
1004 at the expense of memory usage). We slide even when level == 0 |
|
1005 to keep the hash table consistent if we switch back to level > 0 |
|
1006 later. (Using level 0 permanently is not an optimal usage of |
|
1007 zlib, so we don't care about this pathological case.) |
|
1008 */ |
|
1009 n = s->hash_size; |
|
1010 p = &s->head[n]; |
|
1011 do { |
|
1012 m = *--p; |
|
1013 *p = (Pos)(m >= wsize ? m-wsize : NIL); |
|
1014 } while (--n); |
|
1015 |
|
1016 n = wsize; |
|
1017 #ifndef FASTEST |
|
1018 p = &s->prev[n]; |
|
1019 do { |
|
1020 m = *--p; |
|
1021 *p = (Pos)(m >= wsize ? m-wsize : NIL); |
|
1022 /* If n is not on any hash chain, prev[n] is garbage but |
|
1023 * its value will never be used. |
|
1024 */ |
|
1025 } while (--n); |
|
1026 #endif |
|
1027 more += wsize; |
|
1028 } |
|
1029 if (s->strm->avail_in == 0) return; |
|
1030 |
|
1031 /* If there was no sliding: |
|
1032 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && |
|
1033 * more == window_size - lookahead - strstart |
|
1034 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) |
|
1035 * => more >= window_size - 2*WSIZE + 2 |
|
1036 * In the BIG_MEM or MMAP case (not yet supported), |
|
1037 * window_size == input_size + MIN_LOOKAHEAD && |
|
1038 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. |
|
1039 * Otherwise, window_size == 2*WSIZE so more >= 2. |
|
1040 * If there was sliding, more >= WSIZE. So in all cases, more >= 2. |
|
1041 */ |
|
1042 Assert(more >= 2, "more < 2"); |
|
1043 |
|
1044 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); |
|
1045 s->lookahead += n; |
|
1046 |
|
1047 /* Initialize the hash value now that we have some input: */ |
|
1048 if (s->lookahead >= MIN_MATCH) { |
|
1049 s->ins_h = s->window[s->strstart]; |
|
1050 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); |
|
1051 #if MIN_MATCH != 3 |
|
1052 // Call UPDATE_HASH() MIN_MATCH-3 more times |
|
1053 #endif |
|
1054 } |
|
1055 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, |
|
1056 * but this is not important since only literal bytes will be emitted. |
|
1057 */ |
|
1058 |
|
1059 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); |
|
1060 } |
|
1061 |
|
1062 /* =========================================================================== |
|
1063 * Flush the current block, with given end-of-file flag. |
|
1064 * IN assertion: strstart is set to the end of the current match. |
|
1065 */ |
|
1066 #define FLUSH_BLOCK_ONLY(s, eof) { \ |
|
1067 _tr_flush_block(s, (s->block_start >= 0L ? \ |
|
1068 (charf *)&s->window[(unsigned)s->block_start] : \ |
|
1069 (charf *)Z_NULL), \ |
|
1070 (ulg)((long)s->strstart - s->block_start), \ |
|
1071 (eof)); \ |
|
1072 s->block_start = s->strstart; \ |
|
1073 flush_pending(s->strm); \ |
|
1074 Tracev((stderr,"[FLUSH]")); \ |
|
1075 } |
|
1076 |
|
1077 /* Same but force premature exit if necessary. */ |
|
1078 #define FLUSH_BLOCK(s, eof) { \ |
|
1079 FLUSH_BLOCK_ONLY(s, eof); \ |
|
1080 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \ |
|
1081 } |
|
1082 |
|
1083 /* =========================================================================== |
|
1084 * Copy without compression as much as possible from the input stream, return |
|
1085 * the current block state. |
|
1086 * This function does not insert new strings in the dictionary since |
|
1087 * uncompressible data is probably not useful. This function is used |
|
1088 * only for the level=0 compression option. |
|
1089 * NOTE: this function should be optimized to avoid extra copying from |
|
1090 * window to pending_buf. |
|
1091 */ |
|
1092 local block_state deflate_stored( |
|
1093 deflate_state *s, |
|
1094 int flush) |
|
1095 { |
|
1096 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited |
|
1097 * to pending_buf_size, and each stored block has a 5 byte header: |
|
1098 */ |
|
1099 ulg max_block_size = 0xffff; |
|
1100 ulg max_start; |
|
1101 |
|
1102 if (max_block_size > s->pending_buf_size - 5) { |
|
1103 max_block_size = s->pending_buf_size - 5; |
|
1104 } |
|
1105 |
|
1106 /* Copy as much as possible from input to output: */ |
|
1107 for (;;) { |
|
1108 /* Fill the window as much as possible: */ |
|
1109 if (s->lookahead <= 1) { |
|
1110 |
|
1111 Assert(s->strstart < s->w_size+MAX_DIST(s) || |
|
1112 s->block_start >= (long)s->w_size, "slide too late"); |
|
1113 |
|
1114 fill_window(s); |
|
1115 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more; |
|
1116 |
|
1117 if (s->lookahead == 0) break; /* flush the current block */ |
|
1118 } |
|
1119 Assert(s->block_start >= 0L, "block gone"); |
|
1120 |
|
1121 s->strstart += s->lookahead; |
|
1122 s->lookahead = 0; |
|
1123 |
|
1124 /* Emit a stored block if pending_buf will be full: */ |
|
1125 max_start = s->block_start + max_block_size; |
|
1126 if (s->strstart == 0 || (ulg)s->strstart >= max_start) { |
|
1127 /* strstart == 0 is possible when wraparound on 16-bit machine */ |
|
1128 s->lookahead = (uInt)(s->strstart - max_start); |
|
1129 s->strstart = (uInt)max_start; |
|
1130 FLUSH_BLOCK(s, 0); |
|
1131 } |
|
1132 /* Flush if we may have to slide, otherwise block_start may become |
|
1133 * negative and the data will be gone: |
|
1134 */ |
|
1135 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) { |
|
1136 FLUSH_BLOCK(s, 0); |
|
1137 } |
|
1138 } |
|
1139 FLUSH_BLOCK(s, flush == Z_FINISH); |
|
1140 return flush == Z_FINISH ? finish_done : block_done; |
|
1141 } |
|
1142 |
|
1143 /* =========================================================================== |
|
1144 * Compress as much as possible from the input stream, return the current |
|
1145 * block state. |
|
1146 * This function does not perform lazy evaluation of matches and inserts |
|
1147 * new strings in the dictionary only for unmatched strings or for short |
|
1148 * matches. It is used only for the fast compression options. |
|
1149 */ |
|
1150 local block_state deflate_fast( |
|
1151 deflate_state *s, |
|
1152 int flush) |
|
1153 { |
|
1154 IPos hash_head = NIL; /* head of the hash chain */ |
|
1155 int bflush; /* set if current block must be flushed */ |
|
1156 |
|
1157 for (;;) { |
|
1158 /* Make sure that we always have enough lookahead, except |
|
1159 * at the end of the input file. We need MAX_MATCH bytes |
|
1160 * for the next match, plus MIN_MATCH bytes to insert the |
|
1161 * string following the next match. |
|
1162 */ |
|
1163 if (s->lookahead < MIN_LOOKAHEAD) { |
|
1164 fill_window(s); |
|
1165 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { |
|
1166 return need_more; |
|
1167 } |
|
1168 if (s->lookahead == 0) break; /* flush the current block */ |
|
1169 } |
|
1170 |
|
1171 /* Insert the string window[strstart .. strstart+2] in the |
|
1172 * dictionary, and set hash_head to the head of the hash chain: |
|
1173 */ |
|
1174 if (s->lookahead >= MIN_MATCH) { |
|
1175 INSERT_STRING(s, s->strstart, hash_head); |
|
1176 } |
|
1177 |
|
1178 /* Find the longest match, discarding those <= prev_length. |
|
1179 * At this point we have always match_length < MIN_MATCH |
|
1180 */ |
|
1181 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { |
|
1182 /* To simplify the code, we prevent matches with the string |
|
1183 * of window index 0 (in particular we have to avoid a match |
|
1184 * of the string with itself at the start of the input file). |
|
1185 */ |
|
1186 if (s->strategy != Z_HUFFMAN_ONLY) { |
|
1187 s->match_length = longest_match (s, hash_head); |
|
1188 } |
|
1189 /* longest_match() sets match_start */ |
|
1190 } |
|
1191 if (s->match_length >= MIN_MATCH) { |
|
1192 check_match(s, s->strstart, s->match_start, s->match_length); |
|
1193 |
|
1194 _tr_tally_dist(s, s->strstart - s->match_start, |
|
1195 s->match_length - MIN_MATCH, bflush); |
|
1196 |
|
1197 s->lookahead -= s->match_length; |
|
1198 |
|
1199 /* Insert new strings in the hash table only if the match length |
|
1200 * is not too large. This saves time but degrades compression. |
|
1201 */ |
|
1202 #ifndef FASTEST |
|
1203 if (s->match_length <= s->max_insert_length && |
|
1204 s->lookahead >= MIN_MATCH) { |
|
1205 s->match_length--; /* string at strstart already in hash table */ |
|
1206 do { |
|
1207 s->strstart++; |
|
1208 INSERT_STRING(s, s->strstart, hash_head); |
|
1209 /* strstart never exceeds WSIZE-MAX_MATCH, so there are |
|
1210 * always MIN_MATCH bytes ahead. |
|
1211 */ |
|
1212 } while (--s->match_length != 0); |
|
1213 s->strstart++; |
|
1214 } else |
|
1215 #endif |
|
1216 { |
|
1217 s->strstart += s->match_length; |
|
1218 s->match_length = 0; |
|
1219 s->ins_h = s->window[s->strstart]; |
|
1220 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); |
|
1221 #if MIN_MATCH != 3 |
|
1222 // Call UPDATE_HASH() MIN_MATCH-3 more times |
|
1223 #endif |
|
1224 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not |
|
1225 * matter since it will be recomputed at next deflate call. |
|
1226 */ |
|
1227 } |
|
1228 } else { |
|
1229 /* No match, output a literal byte */ |
|
1230 Tracevv((stderr,"%c", s->window[s->strstart])); |
|
1231 _tr_tally_lit (s, s->window[s->strstart], bflush); |
|
1232 s->lookahead--; |
|
1233 s->strstart++; |
|
1234 } |
|
1235 if (bflush) FLUSH_BLOCK(s, 0); |
|
1236 } |
|
1237 FLUSH_BLOCK(s, flush == Z_FINISH); |
|
1238 return flush == Z_FINISH ? finish_done : block_done; |
|
1239 } |
|
1240 |
|
1241 /* =========================================================================== |
|
1242 * Same as above, but achieves better compression. We use a lazy |
|
1243 * evaluation for matches: a match is finally adopted only if there is |
|
1244 * no better match at the next window position. |
|
1245 */ |
|
1246 local block_state deflate_slow( |
|
1247 deflate_state *s, |
|
1248 int flush) |
|
1249 { |
|
1250 IPos hash_head = NIL; /* head of hash chain */ |
|
1251 int bflush; /* set if current block must be flushed */ |
|
1252 |
|
1253 /* Process the input block. */ |
|
1254 for (;;) { |
|
1255 /* Make sure that we always have enough lookahead, except |
|
1256 * at the end of the input file. We need MAX_MATCH bytes |
|
1257 * for the next match, plus MIN_MATCH bytes to insert the |
|
1258 * string following the next match. |
|
1259 */ |
|
1260 if (s->lookahead < MIN_LOOKAHEAD) { |
|
1261 fill_window(s); |
|
1262 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { |
|
1263 return need_more; |
|
1264 } |
|
1265 if (s->lookahead == 0) break; /* flush the current block */ |
|
1266 } |
|
1267 |
|
1268 /* Insert the string window[strstart .. strstart+2] in the |
|
1269 * dictionary, and set hash_head to the head of the hash chain: |
|
1270 */ |
|
1271 if (s->lookahead >= MIN_MATCH) { |
|
1272 INSERT_STRING(s, s->strstart, hash_head); |
|
1273 } |
|
1274 |
|
1275 /* Find the longest match, discarding those <= prev_length. |
|
1276 */ |
|
1277 s->prev_length = s->match_length, s->prev_match = s->match_start; |
|
1278 s->match_length = MIN_MATCH-1; |
|
1279 |
|
1280 if (hash_head != NIL && s->prev_length < s->max_lazy_match && |
|
1281 s->strstart - hash_head <= MAX_DIST(s)) { |
|
1282 /* To simplify the code, we prevent matches with the string |
|
1283 * of window index 0 (in particular we have to avoid a match |
|
1284 * of the string with itself at the start of the input file). |
|
1285 */ |
|
1286 if (s->strategy != Z_HUFFMAN_ONLY) { |
|
1287 s->match_length = longest_match (s, hash_head); |
|
1288 } |
|
1289 /* longest_match() sets match_start */ |
|
1290 |
|
1291 if (s->match_length <= 5 && (s->strategy == Z_FILTERED || |
|
1292 (s->match_length == MIN_MATCH && |
|
1293 s->strstart - s->match_start > TOO_FAR))) { |
|
1294 |
|
1295 /* If prev_match is also MIN_MATCH, match_start is garbage |
|
1296 * but we will ignore the current match anyway. |
|
1297 */ |
|
1298 s->match_length = MIN_MATCH-1; |
|
1299 } |
|
1300 } |
|
1301 /* If there was a match at the previous step and the current |
|
1302 * match is not better, output the previous match: |
|
1303 */ |
|
1304 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { |
|
1305 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; |
|
1306 /* Do not insert strings in hash table beyond this. */ |
|
1307 |
|
1308 check_match(s, s->strstart-1, s->prev_match, s->prev_length); |
|
1309 |
|
1310 _tr_tally_dist(s, s->strstart -1 - s->prev_match, |
|
1311 s->prev_length - MIN_MATCH, bflush); |
|
1312 |
|
1313 /* Insert in hash table all strings up to the end of the match. |
|
1314 * strstart-1 and strstart are already inserted. If there is not |
|
1315 * enough lookahead, the last two strings are not inserted in |
|
1316 * the hash table. |
|
1317 */ |
|
1318 s->lookahead -= s->prev_length-1; |
|
1319 s->prev_length -= 2; |
|
1320 do { |
|
1321 if (++s->strstart <= max_insert) { |
|
1322 INSERT_STRING(s, s->strstart, hash_head); |
|
1323 } |
|
1324 } while (--s->prev_length != 0); |
|
1325 s->match_available = 0; |
|
1326 s->match_length = MIN_MATCH-1; |
|
1327 s->strstart++; |
|
1328 |
|
1329 if (bflush) FLUSH_BLOCK(s, 0); |
|
1330 |
|
1331 } else if (s->match_available) { |
|
1332 /* If there was no match at the previous position, output a |
|
1333 * single literal. If there was a match but the current match |
|
1334 * is longer, truncate the previous match to a single literal. |
|
1335 */ |
|
1336 Tracevv((stderr,"%c", s->window[s->strstart-1])); |
|
1337 _tr_tally_lit(s, s->window[s->strstart-1], bflush); |
|
1338 if (bflush) { |
|
1339 FLUSH_BLOCK_ONLY(s, 0); |
|
1340 } |
|
1341 s->strstart++; |
|
1342 s->lookahead--; |
|
1343 if (s->strm->avail_out == 0) return need_more; |
|
1344 } else { |
|
1345 /* There is no previous match to compare with, wait for |
|
1346 * the next step to decide. |
|
1347 */ |
|
1348 s->match_available = 1; |
|
1349 s->strstart++; |
|
1350 s->lookahead--; |
|
1351 } |
|
1352 } |
|
1353 Assert (flush != Z_NO_FLUSH, "no flush?"); |
|
1354 if (s->match_available) { |
|
1355 Tracevv((stderr,"%c", s->window[s->strstart-1])); |
|
1356 _tr_tally_lit(s, s->window[s->strstart-1], bflush); |
|
1357 s->match_available = 0; |
|
1358 } |
|
1359 FLUSH_BLOCK(s, flush == Z_FINISH); |
|
1360 return flush == Z_FINISH ? finish_done : block_done; |
|
1361 } |