1 | /* deflate.c -- compress data using the deflation algorithm␊ |
2 | * Copyright (C) 1995-2010 Jean-loup Gailly and Mark Adler␊ |
3 | * For conditions of distribution and use, see copyright notice in zlib.h␊ |
4 | */␊ |
5 | ␊ |
6 | /*␊ |
7 | * ALGORITHM␊ |
8 | *␊ |
9 | * The "deflation" process depends on being able to identify portions␊ |
10 | * of the input text which are identical to earlier input (within a␊ |
11 | * sliding window trailing behind the input currently being processed).␊ |
12 | *␊ |
13 | * The most straightforward technique turns out to be the fastest for␊ |
14 | * most input files: try all possible matches and select the longest.␊ |
15 | * The key feature of this algorithm is that insertions into the string␊ |
16 | * dictionary are very simple and thus fast, and deletions are avoided␊ |
17 | * completely. Insertions are performed at each input character, whereas␊ |
18 | * string matches are performed only when the previous match ends. So it␊ |
19 | * is preferable to spend more time in matches to allow very fast string␊ |
20 | * insertions and avoid deletions. The matching algorithm for small␊ |
21 | * strings is inspired from that of Rabin & Karp. A brute force approach␊ |
22 | * is used to find longer strings when a small match has been found.␊ |
23 | * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze␊ |
24 | * (by Leonid Broukhis).␊ |
25 | * A previous version of this file used a more sophisticated algorithm␊ |
26 | * (by Fiala and Greene) which is guaranteed to run in linear amortized␊ |
27 | * time, but has a larger average cost, uses more memory and is patented.␊ |
28 | * However the F&G algorithm may be faster for some highly redundant␊ |
29 | * files if the parameter max_chain_length (described below) is too large.␊ |
30 | *␊ |
31 | * ACKNOWLEDGEMENTS␊ |
32 | *␊ |
33 | * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and␊ |
34 | * I found it in 'freeze' written by Leonid Broukhis.␊ |
35 | * Thanks to many people for bug reports and testing.␊ |
36 | *␊ |
37 | * REFERENCES␊ |
38 | *␊ |
39 | * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".␊ |
40 | * Available in http://www.ietf.org/rfc/rfc1951.txt␊ |
41 | *␊ |
42 | * A description of the Rabin and Karp algorithm is given in the book␊ |
43 | * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.␊ |
44 | *␊ |
45 | * Fiala,E.R., and Greene,D.H.␊ |
46 | * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595␊ |
47 | *␊ |
48 | */␊ |
49 | ␊ |
50 | /* @(#) $Id$ */␊ |
51 | ␊ |
52 | #include "deflate.h"␊ |
53 | ␊ |
54 | const char deflate_copyright[] =␊ |
55 | " deflate 1.2.5 Copyright 1995-2010 Jean-loup Gailly and Mark Adler ";␊ |
56 | /*␊ |
57 | If you use the zlib library in a product, an acknowledgment is welcome␊ |
58 | in the documentation of your product. If for some reason you cannot␊ |
59 | include such an acknowledgment, I would appreciate that you keep this␊ |
60 | copyright string in the executable of your product.␊ |
61 | */␊ |
62 | ␊ |
63 | /* ===========================================================================␊ |
64 | * Function prototypes.␊ |
65 | */␊ |
66 | typedef enum {␊ |
67 | need_more, /* block not completed, need more input or more output */␊ |
68 | block_done, /* block flush performed */␊ |
69 | finish_started, /* finish started, need only more output at next deflate */␊ |
70 | finish_done /* finish done, accept no more input or output */␊ |
71 | } block_state;␊ |
72 | ␊ |
73 | typedef block_state (*compress_func) OF((deflate_state *s, int flush));␊ |
74 | /* Compression function. Returns the block state after the call. */␊ |
75 | ␊ |
76 | local void fill_window OF((deflate_state *s));␊ |
77 | local block_state deflate_stored OF((deflate_state *s, int flush));␊ |
78 | local block_state deflate_fast OF((deflate_state *s, int flush));␊ |
79 | #ifndef FASTEST␊ |
80 | local block_state deflate_slow OF((deflate_state *s, int flush));␊ |
81 | #endif␊ |
82 | local block_state deflate_rle OF((deflate_state *s, int flush));␊ |
83 | local block_state deflate_huff OF((deflate_state *s, int flush));␊ |
84 | local void lm_init OF((deflate_state *s));␊ |
85 | local void putShortMSB OF((deflate_state *s, uInt b));␊ |
86 | local void flush_pending OF((z_streamp strm));␊ |
87 | local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));␊ |
88 | #ifdef ASMV␊ |
89 | void match_init OF((void)); /* asm code initialization */␊ |
90 | uInt longest_match OF((deflate_state *s, IPos cur_match));␊ |
91 | #else␊ |
92 | local uInt longest_match OF((deflate_state *s, IPos cur_match));␊ |
93 | #endif␊ |
94 | ␊ |
95 | #ifdef DEBUG␊ |
96 | local void check_match OF((deflate_state *s, IPos start, IPos match,␊ |
97 | int length));␊ |
98 | #endif␊ |
99 | ␊ |
100 | /* ===========================================================================␊ |
101 | * Local data␊ |
102 | */␊ |
103 | ␊ |
104 | #define NIL 0␊ |
105 | /* Tail of hash chains */␊ |
106 | ␊ |
107 | #ifndef TOO_FAR␊ |
108 | # define TOO_FAR 4096␊ |
109 | #endif␊ |
110 | /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */␊ |
111 | ␊ |
112 | /* Values for max_lazy_match, good_match and max_chain_length, depending on␊ |
113 | * the desired pack level (0..9). The values given below have been tuned to␊ |
114 | * exclude worst case performance for pathological files. Better values may be␊ |
115 | * found for specific files.␊ |
116 | */␊ |
117 | typedef struct config_s {␊ |
118 | ush good_length; /* reduce lazy search above this match length */␊ |
119 | ush max_lazy; /* do not perform lazy search above this match length */␊ |
120 | ush nice_length; /* quit search above this match length */␊ |
121 | ush max_chain;␊ |
122 | compress_func func;␊ |
123 | } config;␊ |
124 | ␊ |
125 | #ifdef FASTEST␊ |
126 | local const config configuration_table[2] = {␊ |
127 | /* good lazy nice chain */␊ |
128 | /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */␊ |
129 | /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */␊ |
130 | #else␊ |
131 | local const config configuration_table[10] = {␊ |
132 | /* good lazy nice chain */␊ |
133 | /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */␊ |
134 | /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */␊ |
135 | /* 2 */ {4, 5, 16, 8, deflate_fast},␊ |
136 | /* 3 */ {4, 6, 32, 32, deflate_fast},␊ |
137 | ␊ |
138 | /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */␊ |
139 | /* 5 */ {8, 16, 32, 32, deflate_slow},␊ |
140 | /* 6 */ {8, 16, 128, 128, deflate_slow},␊ |
141 | /* 7 */ {8, 32, 128, 256, deflate_slow},␊ |
142 | /* 8 */ {32, 128, 258, 1024, deflate_slow},␊ |
143 | /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */␊ |
144 | #endif␊ |
145 | ␊ |
146 | /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4␊ |
147 | * For deflate_fast() (levels <= 3) good is ignored and lazy has a different␊ |
148 | * meaning.␊ |
149 | */␊ |
150 | ␊ |
151 | #define EQUAL 0␊ |
152 | /* result of memcmp for equal strings */␊ |
153 | ␊ |
154 | #ifndef NO_DUMMY_DECL␊ |
155 | struct static_tree_desc_s {int dummy;}; /* for buggy compilers */␊ |
156 | #endif␊ |
157 | ␊ |
158 | /* ===========================================================================␊ |
159 | * Update a hash value with the given input byte␊ |
160 | * IN assertion: all calls to to UPDATE_HASH are made with consecutive␊ |
161 | * input characters, so that a running hash key can be computed from the␊ |
162 | * previous key instead of complete recalculation each time.␊ |
163 | */␊ |
164 | #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)␊ |
165 | ␊ |
166 | ␊ |
167 | /* ===========================================================================␊ |
168 | * Insert string str in the dictionary and set match_head to the previous head␊ |
169 | * of the hash chain (the most recent string with same hash key). Return␊ |
170 | * the previous length of the hash chain.␊ |
171 | * If this file is compiled with -DFASTEST, the compression level is forced␊ |
172 | * to 1, and no hash chains are maintained.␊ |
173 | * IN assertion: all calls to to INSERT_STRING are made with consecutive␊ |
174 | * input characters and the first MIN_MATCH bytes of str are valid␊ |
175 | * (except for the last MIN_MATCH-1 bytes of the input file).␊ |
176 | */␊ |
177 | #ifdef FASTEST␊ |
178 | #define INSERT_STRING(s, str, match_head) \␊ |
179 | (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \␊ |
180 | match_head = s->head[s->ins_h], \␊ |
181 | s->head[s->ins_h] = (Pos)(str))␊ |
182 | #else␊ |
183 | #define INSERT_STRING(s, str, match_head) \␊ |
184 | (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \␊ |
185 | match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \␊ |
186 | s->head[s->ins_h] = (Pos)(str))␊ |
187 | #endif␊ |
188 | ␊ |
189 | /* ===========================================================================␊ |
190 | * Initialize the hash table (avoiding 64K overflow for 16 bit systems).␊ |
191 | * prev[] will be initialized on the fly.␊ |
192 | */␊ |
193 | #define CLEAR_HASH(s) \␊ |
194 | s->head[s->hash_size-1] = NIL; \␊ |
195 | zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));␊ |
196 | ␊ |
197 | /* ========================================================================= */␊ |
198 | int ZEXPORT deflateInit_(strm, level, version, stream_size)␊ |
199 | z_streamp strm;␊ |
200 | int level;␊ |
201 | const char *version;␊ |
202 | int stream_size;␊ |
203 | {␊ |
204 | return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,␊ |
205 | Z_DEFAULT_STRATEGY, version, stream_size);␊ |
206 | /* To do: ignore strm->next_in if we use it as window */␊ |
207 | }␊ |
208 | ␊ |
209 | /* ========================================================================= */␊ |
210 | int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,␊ |
211 | version, stream_size)␊ |
212 | z_streamp strm;␊ |
213 | int level;␊ |
214 | int method;␊ |
215 | int windowBits;␊ |
216 | int memLevel;␊ |
217 | int strategy;␊ |
218 | const char *version;␊ |
219 | int stream_size;␊ |
220 | {␊ |
221 | deflate_state *s;␊ |
222 | int wrap = 1;␊ |
223 | static const char my_version[] = ZLIB_VERSION;␊ |
224 | ␊ |
225 | ushf *overlay;␊ |
226 | /* We overlay pending_buf and d_buf+l_buf. This works since the average␊ |
227 | * output size for (length,distance) codes is <= 24 bits.␊ |
228 | */␊ |
229 | ␊ |
230 | if (version == Z_NULL || version[0] != my_version[0] ||␊ |
231 | stream_size != sizeof(z_stream)) {␊ |
232 | return Z_VERSION_ERROR;␊ |
233 | }␊ |
234 | if (strm == Z_NULL) return Z_STREAM_ERROR;␊ |
235 | ␊ |
236 | strm->msg = Z_NULL;␊ |
237 | if (strm->zalloc == (alloc_func)0) {␊ |
238 | strm->zalloc = zcalloc;␊ |
239 | strm->opaque = (voidpf)0;␊ |
240 | }␊ |
241 | if (strm->zfree == (free_func)0) strm->zfree = zcfree;␊ |
242 | ␊ |
243 | #ifdef FASTEST␊ |
244 | if (level != 0) level = 1;␊ |
245 | #else␊ |
246 | if (level == Z_DEFAULT_COMPRESSION) level = 6;␊ |
247 | #endif␊ |
248 | ␊ |
249 | if (windowBits < 0) { /* suppress zlib wrapper */␊ |
250 | wrap = 0;␊ |
251 | windowBits = -windowBits;␊ |
252 | }␊ |
253 | #ifdef GZIP␊ |
254 | else if (windowBits > 15) {␊ |
255 | wrap = 2; /* write gzip wrapper instead */␊ |
256 | windowBits -= 16;␊ |
257 | }␊ |
258 | #endif␊ |
259 | if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||␊ |
260 | windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||␊ |
261 | strategy < 0 || strategy > Z_FIXED) {␊ |
262 | return Z_STREAM_ERROR;␊ |
263 | }␊ |
264 | if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */␊ |
265 | s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));␊ |
266 | if (s == Z_NULL) return Z_MEM_ERROR;␊ |
267 | strm->state = (struct internal_state FAR *)s;␊ |
268 | s->strm = strm;␊ |
269 | ␊ |
270 | s->wrap = wrap;␊ |
271 | s->gzhead = Z_NULL;␊ |
272 | s->w_bits = windowBits;␊ |
273 | s->w_size = 1 << s->w_bits;␊ |
274 | s->w_mask = s->w_size - 1;␊ |
275 | ␊ |
276 | s->hash_bits = memLevel + 7;␊ |
277 | s->hash_size = 1 << s->hash_bits;␊ |
278 | s->hash_mask = s->hash_size - 1;␊ |
279 | s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);␊ |
280 | ␊ |
281 | s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));␊ |
282 | s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));␊ |
283 | s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));␊ |
284 | ␊ |
285 | s->high_water = 0; /* nothing written to s->window yet */␊ |
286 | ␊ |
287 | s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */␊ |
288 | ␊ |
289 | overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);␊ |
290 | s->pending_buf = (uchf *) overlay;␊ |
291 | s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);␊ |
292 | ␊ |
293 | if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||␊ |
294 | s->pending_buf == Z_NULL) {␊ |
295 | s->status = FINISH_STATE;␊ |
296 | strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);␊ |
297 | deflateEnd (strm);␊ |
298 | return Z_MEM_ERROR;␊ |
299 | }␊ |
300 | s->d_buf = overlay + s->lit_bufsize/sizeof(ush);␊ |
301 | s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;␊ |
302 | ␊ |
303 | s->level = level;␊ |
304 | s->strategy = strategy;␊ |
305 | s->method = (Byte)method;␊ |
306 | ␊ |
307 | return deflateReset(strm);␊ |
308 | }␊ |
309 | ␊ |
310 | /* ========================================================================= */␊ |
311 | int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)␊ |
312 | z_streamp strm;␊ |
313 | const Bytef *dictionary;␊ |
314 | uInt dictLength;␊ |
315 | {␊ |
316 | deflate_state *s;␊ |
317 | uInt length = dictLength;␊ |
318 | uInt n;␊ |
319 | IPos hash_head = 0;␊ |
320 | ␊ |
321 | if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||␊ |
322 | strm->state->wrap == 2 ||␊ |
323 | (strm->state->wrap == 1 && strm->state->status != INIT_STATE))␊ |
324 | return Z_STREAM_ERROR;␊ |
325 | ␊ |
326 | s = strm->state;␊ |
327 | if (s->wrap)␊ |
328 | strm->adler = adler32(strm->adler, dictionary, dictLength);␊ |
329 | ␊ |
330 | if (length < MIN_MATCH) return Z_OK;␊ |
331 | if (length > s->w_size) {␊ |
332 | length = s->w_size;␊ |
333 | dictionary += dictLength - length; /* use the tail of the dictionary */␊ |
334 | }␊ |
335 | zmemcpy(s->window, dictionary, length);␊ |
336 | s->strstart = length;␊ |
337 | s->block_start = (long)length;␊ |
338 | ␊ |
339 | /* Insert all strings in the hash table (except for the last two bytes).␊ |
340 | * s->lookahead stays null, so s->ins_h will be recomputed at the next␊ |
341 | * call of fill_window.␊ |
342 | */␊ |
343 | s->ins_h = s->window[0];␊ |
344 | UPDATE_HASH(s, s->ins_h, s->window[1]);␊ |
345 | for (n = 0; n <= length - MIN_MATCH; n++) {␊ |
346 | INSERT_STRING(s, n, hash_head);␊ |
347 | }␊ |
348 | #if 0␊ |
349 | #if UNUSED␊ |
350 | if (hash_head) hash_head = 0; /* to make compiler happy */␊ |
351 | #else␊ |
352 | ␉(void)hash_head;␊ |
353 | #endif␊ |
354 | #endif␊ |
355 | return Z_OK;␊ |
356 | }␊ |
357 | ␊ |
358 | /* ========================================================================= */␊ |
359 | int ZEXPORT deflateReset (strm)␊ |
360 | z_streamp strm;␊ |
361 | {␊ |
362 | deflate_state *s;␊ |
363 | ␊ |
364 | if (strm == Z_NULL || strm->state == Z_NULL ||␊ |
365 | strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {␊ |
366 | return Z_STREAM_ERROR;␊ |
367 | }␊ |
368 | ␊ |
369 | strm->total_in = strm->total_out = 0;␊ |
370 | strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */␊ |
371 | strm->data_type = Z_UNKNOWN;␊ |
372 | ␊ |
373 | s = (deflate_state *)strm->state;␊ |
374 | s->pending = 0;␊ |
375 | s->pending_out = s->pending_buf;␊ |
376 | ␊ |
377 | if (s->wrap < 0) {␊ |
378 | s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */␊ |
379 | }␊ |
380 | s->status = s->wrap ? INIT_STATE : BUSY_STATE;␊ |
381 | strm->adler =␊ |
382 | #ifdef GZIP␊ |
383 | s->wrap == 2 ? crc32(0L, Z_NULL, 0) :␊ |
384 | #endif␊ |
385 | adler32(0L, Z_NULL, 0);␊ |
386 | s->last_flush = Z_NO_FLUSH;␊ |
387 | ␊ |
388 | _tr_init(s);␊ |
389 | lm_init(s);␊ |
390 | ␊ |
391 | return Z_OK;␊ |
392 | }␊ |
393 | ␊ |
394 | /* ========================================================================= */␊ |
395 | int ZEXPORT deflateSetHeader (strm, head)␊ |
396 | z_streamp strm;␊ |
397 | gz_headerp head;␊ |
398 | {␊ |
399 | if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;␊ |
400 | if (strm->state->wrap != 2) return Z_STREAM_ERROR;␊ |
401 | strm->state->gzhead = head;␊ |
402 | return Z_OK;␊ |
403 | }␊ |
404 | ␊ |
405 | /* ========================================================================= */␊ |
406 | int ZEXPORT deflatePrime (strm, bits, value)␊ |
407 | z_streamp strm;␊ |
408 | int bits;␊ |
409 | int value;␊ |
410 | {␊ |
411 | if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;␊ |
412 | strm->state->bi_valid = bits;␊ |
413 | strm->state->bi_buf = (ush)(value & ((1 << bits) - 1));␊ |
414 | return Z_OK;␊ |
415 | }␊ |
416 | ␊ |
417 | /* ========================================================================= */␊ |
418 | int ZEXPORT deflateParams(strm, level, strategy)␊ |
419 | z_streamp strm;␊ |
420 | int level;␊ |
421 | int strategy;␊ |
422 | {␊ |
423 | deflate_state *s;␊ |
424 | compress_func func;␊ |
425 | int err = Z_OK;␊ |
426 | ␊ |
427 | if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;␊ |
428 | s = strm->state;␊ |
429 | ␊ |
430 | #ifdef FASTEST␊ |
431 | if (level != 0) level = 1;␊ |
432 | #else␊ |
433 | if (level == Z_DEFAULT_COMPRESSION) level = 6;␊ |
434 | #endif␊ |
435 | if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {␊ |
436 | return Z_STREAM_ERROR;␊ |
437 | }␊ |
438 | func = configuration_table[s->level].func;␊ |
439 | ␊ |
440 | if ((strategy != s->strategy || func != configuration_table[level].func) &&␊ |
441 | strm->total_in != 0) {␊ |
442 | /* Flush the last buffer: */␊ |
443 | err = deflate(strm, Z_BLOCK);␊ |
444 | }␊ |
445 | if (s->level != level) {␊ |
446 | s->level = level;␊ |
447 | s->max_lazy_match = configuration_table[level].max_lazy;␊ |
448 | s->good_match = configuration_table[level].good_length;␊ |
449 | s->nice_match = configuration_table[level].nice_length;␊ |
450 | s->max_chain_length = configuration_table[level].max_chain;␊ |
451 | }␊ |
452 | s->strategy = strategy;␊ |
453 | return err;␊ |
454 | }␊ |
455 | ␊ |
456 | /* ========================================================================= */␊ |
457 | int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)␊ |
458 | z_streamp strm;␊ |
459 | int good_length;␊ |
460 | int max_lazy;␊ |
461 | int nice_length;␊ |
462 | int max_chain;␊ |
463 | {␊ |
464 | deflate_state *s;␊ |
465 | ␊ |
466 | if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;␊ |
467 | s = strm->state;␊ |
468 | s->good_match = good_length;␊ |
469 | s->max_lazy_match = max_lazy;␊ |
470 | s->nice_match = nice_length;␊ |
471 | s->max_chain_length = max_chain;␊ |
472 | return Z_OK;␊ |
473 | }␊ |
474 | ␊ |
475 | /* =========================================================================␊ |
476 | * For the default windowBits of 15 and memLevel of 8, this function returns␊ |
477 | * a close to exact, as well as small, upper bound on the compressed size.␊ |
478 | * They are coded as constants here for a reason--if the #define's are␊ |
479 | * changed, then this function needs to be changed as well. The return␊ |
480 | * value for 15 and 8 only works for those exact settings.␊ |
481 | *␊ |
482 | * For any setting other than those defaults for windowBits and memLevel,␊ |
483 | * the value returned is a conservative worst case for the maximum expansion␊ |
484 | * resulting from using fixed blocks instead of stored blocks, which deflate␊ |
485 | * can emit on compressed data for some combinations of the parameters.␊ |
486 | *␊ |
487 | * This function could be more sophisticated to provide closer upper bounds for␊ |
488 | * every combination of windowBits and memLevel. But even the conservative␊ |
489 | * upper bound of about 14% expansion does not seem onerous for output buffer␊ |
490 | * allocation.␊ |
491 | */␊ |
492 | uLong ZEXPORT deflateBound(strm, sourceLen)␊ |
493 | z_streamp strm;␊ |
494 | uLong sourceLen;␊ |
495 | {␊ |
496 | deflate_state *s;␊ |
497 | uLong complen, wraplen;␊ |
498 | Bytef *str;␊ |
499 | ␊ |
500 | /* conservative upper bound for compressed data */␊ |
501 | complen = sourceLen +␊ |
502 | ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;␊ |
503 | ␊ |
504 | /* if can't get parameters, return conservative bound plus zlib wrapper */␊ |
505 | if (strm == Z_NULL || strm->state == Z_NULL)␊ |
506 | return complen + 6;␊ |
507 | ␊ |
508 | /* compute wrapper length */␊ |
509 | s = strm->state;␊ |
510 | switch (s->wrap) {␊ |
511 | case 0: /* raw deflate */␊ |
512 | wraplen = 0;␊ |
513 | break;␊ |
514 | case 1: /* zlib wrapper */␊ |
515 | wraplen = 6 + (s->strstart ? 4 : 0);␊ |
516 | break;␊ |
517 | case 2: /* gzip wrapper */␊ |
518 | wraplen = 18;␊ |
519 | if (s->gzhead != Z_NULL) { /* user-supplied gzip header */␊ |
520 | if (s->gzhead->extra != Z_NULL)␊ |
521 | wraplen += 2 + s->gzhead->extra_len;␊ |
522 | str = s->gzhead->name;␊ |
523 | if (str != Z_NULL)␊ |
524 | do {␊ |
525 | wraplen++;␊ |
526 | } while (*str++);␊ |
527 | str = s->gzhead->comment;␊ |
528 | if (str != Z_NULL)␊ |
529 | do {␊ |
530 | wraplen++;␊ |
531 | } while (*str++);␊ |
532 | if (s->gzhead->hcrc)␊ |
533 | wraplen += 2;␊ |
534 | }␊ |
535 | break;␊ |
536 | default: /* for compiler happiness */␊ |
537 | wraplen = 6;␊ |
538 | }␊ |
539 | ␊ |
540 | /* if not default parameters, return conservative bound */␊ |
541 | if (s->w_bits != 15 || s->hash_bits != 8 + 7)␊ |
542 | return complen + wraplen;␊ |
543 | ␊ |
544 | /* default settings: return tight bound for that case */␊ |
545 | return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +␊ |
546 | (sourceLen >> 25) + 13 - 6 + wraplen;␊ |
547 | }␊ |
548 | ␊ |
549 | /* =========================================================================␊ |
550 | * Put a short in the pending buffer. The 16-bit value is put in MSB order.␊ |
551 | * IN assertion: the stream state is correct and there is enough room in␊ |
552 | * pending_buf.␊ |
553 | */␊ |
554 | local void putShortMSB (s, b)␊ |
555 | deflate_state *s;␊ |
556 | uInt b;␊ |
557 | {␊ |
558 | put_byte(s, (Byte)(b >> 8));␊ |
559 | put_byte(s, (Byte)(b & 0xff));␊ |
560 | }␊ |
561 | ␊ |
562 | /* =========================================================================␊ |
563 | * Flush as much pending output as possible. All deflate() output goes␊ |
564 | * through this function so some applications may wish to modify it␊ |
565 | * to avoid allocating a large strm->next_out buffer and copying into it.␊ |
566 | * (See also read_buf()).␊ |
567 | */␊ |
568 | local void flush_pending(strm)␊ |
569 | z_streamp strm;␊ |
570 | {␊ |
571 | unsigned len = strm->state->pending;␊ |
572 | ␊ |
573 | if (len > strm->avail_out) len = strm->avail_out;␊ |
574 | if (len == 0) return;␊ |
575 | ␊ |
576 | zmemcpy(strm->next_out, strm->state->pending_out, len);␊ |
577 | strm->next_out += len;␊ |
578 | strm->state->pending_out += len;␊ |
579 | strm->total_out += len;␊ |
580 | strm->avail_out -= len;␊ |
581 | strm->state->pending -= len;␊ |
582 | if (strm->state->pending == 0) {␊ |
583 | strm->state->pending_out = strm->state->pending_buf;␊ |
584 | }␊ |
585 | }␊ |
586 | ␊ |
587 | /* ========================================================================= */␊ |
588 | int ZEXPORT deflate (strm, flush)␊ |
589 | z_streamp strm;␊ |
590 | int flush;␊ |
591 | {␊ |
592 | int old_flush; /* value of flush param for previous deflate call */␊ |
593 | deflate_state *s;␊ |
594 | ␊ |
595 | if (strm == Z_NULL || strm->state == Z_NULL ||␊ |
596 | flush > Z_BLOCK || flush < 0) {␊ |
597 | return Z_STREAM_ERROR;␊ |
598 | }␊ |
599 | s = strm->state;␊ |
600 | ␊ |
601 | if (strm->next_out == Z_NULL ||␊ |
602 | (strm->next_in == Z_NULL && strm->avail_in != 0) ||␊ |
603 | (s->status == FINISH_STATE && flush != Z_FINISH)) {␊ |
604 | ERR_RETURN(strm, Z_STREAM_ERROR);␊ |
605 | }␊ |
606 | if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);␊ |
607 | ␊ |
608 | s->strm = strm; /* just in case */␊ |
609 | old_flush = s->last_flush;␊ |
610 | s->last_flush = flush;␊ |
611 | ␊ |
612 | /* Write the header */␊ |
613 | if (s->status == INIT_STATE) {␊ |
614 | #ifdef GZIP␊ |
615 | if (s->wrap == 2) {␊ |
616 | strm->adler = crc32(0L, Z_NULL, 0);␊ |
617 | put_byte(s, 31);␊ |
618 | put_byte(s, 139);␊ |
619 | put_byte(s, 8);␊ |
620 | if (s->gzhead == Z_NULL) {␊ |
621 | put_byte(s, 0);␊ |
622 | put_byte(s, 0);␊ |
623 | put_byte(s, 0);␊ |
624 | put_byte(s, 0);␊ |
625 | put_byte(s, 0);␊ |
626 | put_byte(s, s->level == 9 ? 2 :␊ |
627 | (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?␊ |
628 | 4 : 0));␊ |
629 | put_byte(s, OS_CODE);␊ |
630 | s->status = BUSY_STATE;␊ |
631 | }␊ |
632 | else {␊ |
633 | put_byte(s, (s->gzhead->text ? 1 : 0) +␊ |
634 | (s->gzhead->hcrc ? 2 : 0) +␊ |
635 | (s->gzhead->extra == Z_NULL ? 0 : 4) +␊ |
636 | (s->gzhead->name == Z_NULL ? 0 : 8) +␊ |
637 | (s->gzhead->comment == Z_NULL ? 0 : 16)␊ |
638 | );␊ |
639 | put_byte(s, (Byte)(s->gzhead->time & 0xff));␊ |
640 | put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));␊ |
641 | put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));␊ |
642 | put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));␊ |
643 | put_byte(s, s->level == 9 ? 2 :␊ |
644 | (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?␊ |
645 | 4 : 0));␊ |
646 | put_byte(s, s->gzhead->os & 0xff);␊ |
647 | if (s->gzhead->extra != Z_NULL) {␊ |
648 | put_byte(s, s->gzhead->extra_len & 0xff);␊ |
649 | put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);␊ |
650 | }␊ |
651 | if (s->gzhead->hcrc)␊ |
652 | strm->adler = crc32(strm->adler, s->pending_buf,␊ |
653 | s->pending);␊ |
654 | s->gzindex = 0;␊ |
655 | s->status = EXTRA_STATE;␊ |
656 | }␊ |
657 | }␊ |
658 | else␊ |
659 | #endif␊ |
660 | {␊ |
661 | uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;␊ |
662 | uInt level_flags;␊ |
663 | ␊ |
664 | if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)␊ |
665 | level_flags = 0;␊ |
666 | else if (s->level < 6)␊ |
667 | level_flags = 1;␊ |
668 | else if (s->level == 6)␊ |
669 | level_flags = 2;␊ |
670 | else␊ |
671 | level_flags = 3;␊ |
672 | header |= (level_flags << 6);␊ |
673 | if (s->strstart != 0) header |= PRESET_DICT;␊ |
674 | header += 31 - (header % 31);␊ |
675 | ␊ |
676 | s->status = BUSY_STATE;␊ |
677 | putShortMSB(s, header);␊ |
678 | ␊ |
679 | /* Save the adler32 of the preset dictionary: */␊ |
680 | if (s->strstart != 0) {␊ |
681 | putShortMSB(s, (uInt)(strm->adler >> 16));␊ |
682 | putShortMSB(s, (uInt)(strm->adler & 0xffff));␊ |
683 | }␊ |
684 | strm->adler = adler32(0L, Z_NULL, 0);␊ |
685 | }␊ |
686 | }␊ |
687 | #ifdef GZIP␊ |
688 | if (s->status == EXTRA_STATE) {␊ |
689 | if (s->gzhead->extra != Z_NULL) {␊ |
690 | uInt beg = s->pending; /* start of bytes to update crc */␊ |
691 | ␊ |
692 | while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {␊ |
693 | if (s->pending == s->pending_buf_size) {␊ |
694 | if (s->gzhead->hcrc && s->pending > beg)␊ |
695 | strm->adler = crc32(strm->adler, s->pending_buf + beg,␊ |
696 | s->pending - beg);␊ |
697 | flush_pending(strm);␊ |
698 | beg = s->pending;␊ |
699 | if (s->pending == s->pending_buf_size)␊ |
700 | break;␊ |
701 | }␊ |
702 | put_byte(s, s->gzhead->extra[s->gzindex]);␊ |
703 | s->gzindex++;␊ |
704 | }␊ |
705 | if (s->gzhead->hcrc && s->pending > beg)␊ |
706 | strm->adler = crc32(strm->adler, s->pending_buf + beg,␊ |
707 | s->pending - beg);␊ |
708 | if (s->gzindex == s->gzhead->extra_len) {␊ |
709 | s->gzindex = 0;␊ |
710 | s->status = NAME_STATE;␊ |
711 | }␊ |
712 | }␊ |
713 | else␊ |
714 | s->status = NAME_STATE;␊ |
715 | }␊ |
716 | if (s->status == NAME_STATE) {␊ |
717 | if (s->gzhead->name != Z_NULL) {␊ |
718 | uInt beg = s->pending; /* start of bytes to update crc */␊ |
719 | int val;␊ |
720 | ␊ |
721 | do {␊ |
722 | if (s->pending == s->pending_buf_size) {␊ |
723 | if (s->gzhead->hcrc && s->pending > beg)␊ |
724 | strm->adler = crc32(strm->adler, s->pending_buf + beg,␊ |
725 | s->pending - beg);␊ |
726 | flush_pending(strm);␊ |
727 | beg = s->pending;␊ |
728 | if (s->pending == s->pending_buf_size) {␊ |
729 | val = 1;␊ |
730 | break;␊ |
731 | }␊ |
732 | }␊ |
733 | val = s->gzhead->name[s->gzindex++];␊ |
734 | put_byte(s, val);␊ |
735 | } while (val != 0);␊ |
736 | if (s->gzhead->hcrc && s->pending > beg)␊ |
737 | strm->adler = crc32(strm->adler, s->pending_buf + beg,␊ |
738 | s->pending - beg);␊ |
739 | if (val == 0) {␊ |
740 | s->gzindex = 0;␊ |
741 | s->status = COMMENT_STATE;␊ |
742 | }␊ |
743 | }␊ |
744 | else␊ |
745 | s->status = COMMENT_STATE;␊ |
746 | }␊ |
747 | if (s->status == COMMENT_STATE) {␊ |
748 | if (s->gzhead->comment != Z_NULL) {␊ |
749 | uInt beg = s->pending; /* start of bytes to update crc */␊ |
750 | int val;␊ |
751 | ␊ |
752 | do {␊ |
753 | if (s->pending == s->pending_buf_size) {␊ |
754 | if (s->gzhead->hcrc && s->pending > beg)␊ |
755 | strm->adler = crc32(strm->adler, s->pending_buf + beg,␊ |
756 | s->pending - beg);␊ |
757 | flush_pending(strm);␊ |
758 | beg = s->pending;␊ |
759 | if (s->pending == s->pending_buf_size) {␊ |
760 | val = 1;␊ |
761 | break;␊ |
762 | }␊ |
763 | }␊ |
764 | val = s->gzhead->comment[s->gzindex++];␊ |
765 | put_byte(s, val);␊ |
766 | } while (val != 0);␊ |
767 | if (s->gzhead->hcrc && s->pending > beg)␊ |
768 | strm->adler = crc32(strm->adler, s->pending_buf + beg,␊ |
769 | s->pending - beg);␊ |
770 | if (val == 0)␊ |
771 | s->status = HCRC_STATE;␊ |
772 | }␊ |
773 | else␊ |
774 | s->status = HCRC_STATE;␊ |
775 | }␊ |
776 | if (s->status == HCRC_STATE) {␊ |
777 | if (s->gzhead->hcrc) {␊ |
778 | if (s->pending + 2 > s->pending_buf_size)␊ |
779 | flush_pending(strm);␊ |
780 | if (s->pending + 2 <= s->pending_buf_size) {␊ |
781 | put_byte(s, (Byte)(strm->adler & 0xff));␊ |
782 | put_byte(s, (Byte)((strm->adler >> 8) & 0xff));␊ |
783 | strm->adler = crc32(0L, Z_NULL, 0);␊ |
784 | s->status = BUSY_STATE;␊ |
785 | }␊ |
786 | }␊ |
787 | else␊ |
788 | s->status = BUSY_STATE;␊ |
789 | }␊ |
790 | #endif␊ |
791 | ␊ |
792 | /* Flush as much pending output as possible */␊ |
793 | if (s->pending != 0) {␊ |
794 | flush_pending(strm);␊ |
795 | if (strm->avail_out == 0) {␊ |
796 | /* Since avail_out is 0, deflate will be called again with␊ |
797 | * more output space, but possibly with both pending and␊ |
798 | * avail_in equal to zero. There won't be anything to do,␊ |
799 | * but this is not an error situation so make sure we␊ |
800 | * return OK instead of BUF_ERROR at next call of deflate:␊ |
801 | */␊ |
802 | s->last_flush = -1;␊ |
803 | return Z_OK;␊ |
804 | }␊ |
805 | ␊ |
806 | /* Make sure there is something to do and avoid duplicate consecutive␊ |
807 | * flushes. For repeated and useless calls with Z_FINISH, we keep␊ |
808 | * returning Z_STREAM_END instead of Z_BUF_ERROR.␊ |
809 | */␊ |
810 | } else if (strm->avail_in == 0 && flush <= old_flush &&␊ |
811 | flush != Z_FINISH) {␊ |
812 | ERR_RETURN(strm, Z_BUF_ERROR);␊ |
813 | }␊ |
814 | ␊ |
815 | /* User must not provide more input after the first FINISH: */␊ |
816 | if (s->status == FINISH_STATE && strm->avail_in != 0) {␊ |
817 | ERR_RETURN(strm, Z_BUF_ERROR);␊ |
818 | }␊ |
819 | ␊ |
820 | /* Start a new block or continue the current one.␊ |
821 | */␊ |
822 | if (strm->avail_in != 0 || s->lookahead != 0 ||␊ |
823 | (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {␊ |
824 | block_state bstate;␊ |
825 | ␊ |
826 | bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :␊ |
827 | (s->strategy == Z_RLE ? deflate_rle(s, flush) :␊ |
828 | (*(configuration_table[s->level].func))(s, flush));␊ |
829 | ␊ |
830 | if (bstate == finish_started || bstate == finish_done) {␊ |
831 | s->status = FINISH_STATE;␊ |
832 | }␊ |
833 | if (bstate == need_more || bstate == finish_started) {␊ |
834 | if (strm->avail_out == 0) {␊ |
835 | s->last_flush = -1; /* avoid BUF_ERROR next call, see above */␊ |
836 | }␊ |
837 | return Z_OK;␊ |
838 | /* If flush != Z_NO_FLUSH && avail_out == 0, the next call␊ |
839 | * of deflate should use the same flush parameter to make sure␊ |
840 | * that the flush is complete. So we don't have to output an␊ |
841 | * empty block here, this will be done at next call. This also␊ |
842 | * ensures that for a very small output buffer, we emit at most␊ |
843 | * one empty block.␊ |
844 | */␊ |
845 | }␊ |
846 | if (bstate == block_done) {␊ |
847 | if (flush == Z_PARTIAL_FLUSH) {␊ |
848 | _tr_align(s);␊ |
849 | } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */␊ |
850 | _tr_stored_block(s, (char*)0, 0L, 0);␊ |
851 | /* For a full flush, this empty block will be recognized␊ |
852 | * as a special marker by inflate_sync().␊ |
853 | */␊ |
854 | if (flush == Z_FULL_FLUSH) {␊ |
855 | CLEAR_HASH(s); /* forget history */␊ |
856 | if (s->lookahead == 0) {␊ |
857 | s->strstart = 0;␊ |
858 | s->block_start = 0L;␊ |
859 | }␊ |
860 | }␊ |
861 | }␊ |
862 | flush_pending(strm);␊ |
863 | if (strm->avail_out == 0) {␊ |
864 | s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */␊ |
865 | return Z_OK;␊ |
866 | }␊ |
867 | }␊ |
868 | }␊ |
869 | Assert(strm->avail_out > 0, "bug2");␊ |
870 | ␊ |
871 | if (flush != Z_FINISH) return Z_OK;␊ |
872 | if (s->wrap <= 0) return Z_STREAM_END;␊ |
873 | ␊ |
874 | /* Write the trailer */␊ |
875 | #ifdef GZIP␊ |
876 | if (s->wrap == 2) {␊ |
877 | put_byte(s, (Byte)(strm->adler & 0xff));␊ |
878 | put_byte(s, (Byte)((strm->adler >> 8) & 0xff));␊ |
879 | put_byte(s, (Byte)((strm->adler >> 16) & 0xff));␊ |
880 | put_byte(s, (Byte)((strm->adler >> 24) & 0xff));␊ |
881 | put_byte(s, (Byte)(strm->total_in & 0xff));␊ |
882 | put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));␊ |
883 | put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));␊ |
884 | put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));␊ |
885 | }␊ |
886 | else␊ |
887 | #endif␊ |
888 | {␊ |
889 | putShortMSB(s, (uInt)(strm->adler >> 16));␊ |
890 | putShortMSB(s, (uInt)(strm->adler & 0xffff));␊ |
891 | }␊ |
892 | flush_pending(strm);␊ |
893 | /* If avail_out is zero, the application will call deflate again␊ |
894 | * to flush the rest.␊ |
895 | */␊ |
896 | if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */␊ |
897 | return s->pending != 0 ? Z_OK : Z_STREAM_END;␊ |
898 | }␊ |
899 | ␊ |
900 | /* ========================================================================= */␊ |
901 | int ZEXPORT deflateEnd (strm)␊ |
902 | z_streamp strm;␊ |
903 | {␊ |
904 | int status;␊ |
905 | ␊ |
906 | if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;␊ |
907 | ␊ |
908 | status = strm->state->status;␊ |
909 | if (status != INIT_STATE &&␊ |
910 | status != EXTRA_STATE &&␊ |
911 | status != NAME_STATE &&␊ |
912 | status != COMMENT_STATE &&␊ |
913 | status != HCRC_STATE &&␊ |
914 | status != BUSY_STATE &&␊ |
915 | status != FINISH_STATE) {␊ |
916 | return Z_STREAM_ERROR;␊ |
917 | }␊ |
918 | ␊ |
919 | /* Deallocate in reverse order of allocations: */␊ |
920 | TRY_FREE(strm, strm->state->pending_buf);␊ |
921 | TRY_FREE(strm, strm->state->head);␊ |
922 | TRY_FREE(strm, strm->state->prev);␊ |
923 | TRY_FREE(strm, strm->state->window);␊ |
924 | ␊ |
925 | ZFREE(strm, strm->state);␊ |
926 | strm->state = Z_NULL;␊ |
927 | ␊ |
928 | return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;␊ |
929 | }␊ |
930 | ␊ |
931 | /* =========================================================================␊ |
932 | * Copy the source state to the destination state.␊ |
933 | * To simplify the source, this is not supported for 16-bit MSDOS (which␊ |
934 | * doesn't have enough memory anyway to duplicate compression states).␊ |
935 | */␊ |
936 | int ZEXPORT deflateCopy (dest, source)␊ |
937 | z_streamp dest;␊ |
938 | z_streamp source;␊ |
939 | {␊ |
940 | #ifdef MAXSEG_64K␊ |
941 | return Z_STREAM_ERROR;␊ |
942 | #else␊ |
943 | deflate_state *ds;␊ |
944 | deflate_state *ss;␊ |
945 | ushf *overlay;␊ |
946 | ␊ |
947 | ␊ |
948 | if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {␊ |
949 | return Z_STREAM_ERROR;␊ |
950 | }␊ |
951 | ␊ |
952 | ss = source->state;␊ |
953 | ␊ |
954 | zmemcpy(dest, source, sizeof(z_stream));␊ |
955 | ␊ |
956 | ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));␊ |
957 | if (ds == Z_NULL) return Z_MEM_ERROR;␊ |
958 | dest->state = (struct internal_state FAR *) ds;␊ |
959 | zmemcpy(ds, ss, sizeof(deflate_state));␊ |
960 | ds->strm = dest;␊ |
961 | ␊ |
962 | ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));␊ |
963 | ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));␊ |
964 | ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));␊ |
965 | overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);␊ |
966 | ds->pending_buf = (uchf *) overlay;␊ |
967 | ␊ |
968 | if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||␊ |
969 | ds->pending_buf == Z_NULL) {␊ |
970 | deflateEnd (dest);␊ |
971 | return Z_MEM_ERROR;␊ |
972 | }␊ |
973 | /* following zmemcpy do not work for 16-bit MSDOS */␊ |
974 | zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));␊ |
975 | zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));␊ |
976 | zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));␊ |
977 | zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);␊ |
978 | ␊ |
979 | ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);␊ |
980 | ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);␊ |
981 | ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;␊ |
982 | ␊ |
983 | ds->l_desc.dyn_tree = ds->dyn_ltree;␊ |
984 | ds->d_desc.dyn_tree = ds->dyn_dtree;␊ |
985 | ds->bl_desc.dyn_tree = ds->bl_tree;␊ |
986 | ␊ |
987 | return Z_OK;␊ |
988 | #endif /* MAXSEG_64K */␊ |
989 | }␊ |
990 | ␊ |
991 | /* ===========================================================================␊ |
992 | * Read a new buffer from the current input stream, update the adler32␊ |
993 | * and total number of bytes read. All deflate() input goes through␊ |
994 | * this function so some applications may wish to modify it to avoid␊ |
995 | * allocating a large strm->next_in buffer and copying from it.␊ |
996 | * (See also flush_pending()).␊ |
997 | */␊ |
998 | local int read_buf(strm, buf, size)␊ |
999 | z_streamp strm;␊ |
1000 | Bytef *buf;␊ |
1001 | unsigned size;␊ |
1002 | {␊ |
1003 | unsigned len = strm->avail_in;␊ |
1004 | ␊ |
1005 | if (len > size) len = size;␊ |
1006 | if (len == 0) return 0;␊ |
1007 | ␊ |
1008 | strm->avail_in -= len;␊ |
1009 | ␊ |
1010 | if (strm->state->wrap == 1) {␊ |
1011 | strm->adler = adler32(strm->adler, strm->next_in, len);␊ |
1012 | }␊ |
1013 | #ifdef GZIP␊ |
1014 | else if (strm->state->wrap == 2) {␊ |
1015 | strm->adler = crc32(strm->adler, strm->next_in, len);␊ |
1016 | }␊ |
1017 | #endif␊ |
1018 | zmemcpy(buf, strm->next_in, len);␊ |
1019 | strm->next_in += len;␊ |
1020 | strm->total_in += len;␊ |
1021 | ␊ |
1022 | return (int)len;␊ |
1023 | }␊ |
1024 | ␊ |
1025 | /* ===========================================================================␊ |
1026 | * Initialize the "longest match" routines for a new zlib stream␊ |
1027 | */␊ |
1028 | local void lm_init (s)␊ |
1029 | deflate_state *s;␊ |
1030 | {␊ |
1031 | s->window_size = (ulg)2L*s->w_size;␊ |
1032 | ␊ |
1033 | CLEAR_HASH(s);␊ |
1034 | ␊ |
1035 | /* Set the default configuration parameters:␊ |
1036 | */␊ |
1037 | s->max_lazy_match = configuration_table[s->level].max_lazy;␊ |
1038 | s->good_match = configuration_table[s->level].good_length;␊ |
1039 | s->nice_match = configuration_table[s->level].nice_length;␊ |
1040 | s->max_chain_length = configuration_table[s->level].max_chain;␊ |
1041 | ␊ |
1042 | s->strstart = 0;␊ |
1043 | s->block_start = 0L;␊ |
1044 | s->lookahead = 0;␊ |
1045 | s->match_length = s->prev_length = MIN_MATCH-1;␊ |
1046 | s->match_available = 0;␊ |
1047 | s->ins_h = 0;␊ |
1048 | #ifndef FASTEST␊ |
1049 | #ifdef ASMV␊ |
1050 | match_init(); /* initialize the asm code */␊ |
1051 | #endif␊ |
1052 | #endif␊ |
1053 | }␊ |
1054 | ␊ |
1055 | #ifndef FASTEST␊ |
1056 | /* ===========================================================================␊ |
1057 | * Set match_start to the longest match starting at the given string and␊ |
1058 | * return its length. Matches shorter or equal to prev_length are discarded,␊ |
1059 | * in which case the result is equal to prev_length and match_start is␊ |
1060 | * garbage.␊ |
1061 | * IN assertions: cur_match is the head of the hash chain for the current␊ |
1062 | * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1␊ |
1063 | * OUT assertion: the match length is not greater than s->lookahead.␊ |
1064 | */␊ |
1065 | #ifndef ASMV␊ |
1066 | /* For 80x86 and 680x0, an optimized version will be provided in match.asm or␊ |
1067 | * match.S. The code will be functionally equivalent.␊ |
1068 | */␊ |
1069 | local uInt longest_match(s, cur_match)␊ |
1070 | deflate_state *s;␊ |
1071 | IPos cur_match; /* current match */␊ |
1072 | {␊ |
1073 | unsigned chain_length = s->max_chain_length;/* max hash chain length */␊ |
1074 | register Bytef *scan = s->window + s->strstart; /* current string */␊ |
1075 | register Bytef *match; /* matched string */␊ |
1076 | register int len; /* length of current match */␊ |
1077 | int best_len = s->prev_length; /* best match length so far */␊ |
1078 | int nice_match = s->nice_match; /* stop if match long enough */␊ |
1079 | IPos limit = s->strstart > (IPos)MAX_DIST(s) ?␊ |
1080 | s->strstart - (IPos)MAX_DIST(s) : NIL;␊ |
1081 | /* Stop when cur_match becomes <= limit. To simplify the code,␊ |
1082 | * we prevent matches with the string of window index 0.␊ |
1083 | */␊ |
1084 | Posf *prev = s->prev;␊ |
1085 | uInt wmask = s->w_mask;␊ |
1086 | ␊ |
1087 | #ifdef UNALIGNED_OK␊ |
1088 | /* Compare two bytes at a time. Note: this is not always beneficial.␊ |
1089 | * Try with and without -DUNALIGNED_OK to check.␊ |
1090 | */␊ |
1091 | register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;␊ |
1092 | register ush scan_start = *(ushf*)scan;␊ |
1093 | register ush scan_end = *(ushf*)(scan+best_len-1);␊ |
1094 | #else␊ |
1095 | register Bytef *strend = s->window + s->strstart + MAX_MATCH;␊ |
1096 | register Byte scan_end1 = scan[best_len-1];␊ |
1097 | register Byte scan_end = scan[best_len];␊ |
1098 | #endif␊ |
1099 | ␊ |
1100 | /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.␊ |
1101 | * It is easy to get rid of this optimization if necessary.␊ |
1102 | */␊ |
1103 | Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");␊ |
1104 | ␊ |
1105 | /* Do not waste too much time if we already have a good match: */␊ |
1106 | if (s->prev_length >= s->good_match) {␊ |
1107 | chain_length >>= 2;␊ |
1108 | }␊ |
1109 | /* Do not look for matches beyond the end of the input. This is necessary␊ |
1110 | * to make deflate deterministic.␊ |
1111 | */␊ |
1112 | if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;␊ |
1113 | ␊ |
1114 | Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");␊ |
1115 | ␊ |
1116 | do {␊ |
1117 | Assert(cur_match < s->strstart, "no future");␊ |
1118 | match = s->window + cur_match;␊ |
1119 | ␊ |
1120 | /* Skip to next match if the match length cannot increase␊ |
1121 | * or if the match length is less than 2. Note that the checks below␊ |
1122 | * for insufficient lookahead only occur occasionally for performance␊ |
1123 | * reasons. Therefore uninitialized memory will be accessed, and␊ |
1124 | * conditional jumps will be made that depend on those values.␊ |
1125 | * However the length of the match is limited to the lookahead, so␊ |
1126 | * the output of deflate is not affected by the uninitialized values.␊ |
1127 | */␊ |
1128 | #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)␊ |
1129 | /* This code assumes sizeof(unsigned short) == 2. Do not use␊ |
1130 | * UNALIGNED_OK if your compiler uses a different size.␊ |
1131 | */␊ |
1132 | if (*(ushf*)(match+best_len-1) != scan_end ||␊ |
1133 | *(ushf*)match != scan_start) continue;␊ |
1134 | ␊ |
1135 | /* It is not necessary to compare scan[2] and match[2] since they are␊ |
1136 | * always equal when the other bytes match, given that the hash keys␊ |
1137 | * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at␊ |
1138 | * strstart+3, +5, ... up to strstart+257. We check for insufficient␊ |
1139 | * lookahead only every 4th comparison; the 128th check will be made␊ |
1140 | * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is␊ |
1141 | * necessary to put more guard bytes at the end of the window, or␊ |
1142 | * to check more often for insufficient lookahead.␊ |
1143 | */␊ |
1144 | Assert(scan[2] == match[2], "scan[2]?");␊ |
1145 | scan++, match++;␊ |
1146 | do {␊ |
1147 | } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&␊ |
1148 | *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&␊ |
1149 | *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&␊ |
1150 | *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&␊ |
1151 | scan < strend);␊ |
1152 | /* The funny "do {}" generates better code on most compilers */␊ |
1153 | ␊ |
1154 | /* Here, scan <= window+strstart+257 */␊ |
1155 | Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");␊ |
1156 | if (*scan == *match) scan++;␊ |
1157 | ␊ |
1158 | len = (MAX_MATCH - 1) - (int)(strend-scan);␊ |
1159 | scan = strend - (MAX_MATCH-1);␊ |
1160 | ␊ |
1161 | #else /* UNALIGNED_OK */␊ |
1162 | ␊ |
1163 | if (match[best_len] != scan_end ||␊ |
1164 | match[best_len-1] != scan_end1 ||␊ |
1165 | *match != *scan ||␊ |
1166 | *++match != scan[1]) continue;␊ |
1167 | ␊ |
1168 | /* The check at best_len-1 can be removed because it will be made␊ |
1169 | * again later. (This heuristic is not always a win.)␊ |
1170 | * It is not necessary to compare scan[2] and match[2] since they␊ |
1171 | * are always equal when the other bytes match, given that␊ |
1172 | * the hash keys are equal and that HASH_BITS >= 8.␊ |
1173 | */␊ |
1174 | scan += 2, match++;␊ |
1175 | Assert(*scan == *match, "match[2]?");␊ |
1176 | ␊ |
1177 | /* We check for insufficient lookahead only every 8th comparison;␊ |
1178 | * the 256th check will be made at strstart+258.␊ |
1179 | */␊ |
1180 | do {␊ |
1181 | } while (*++scan == *++match && *++scan == *++match &&␊ |
1182 | *++scan == *++match && *++scan == *++match &&␊ |
1183 | *++scan == *++match && *++scan == *++match &&␊ |
1184 | *++scan == *++match && *++scan == *++match &&␊ |
1185 | scan < strend);␊ |
1186 | ␊ |
1187 | Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");␊ |
1188 | ␊ |
1189 | len = MAX_MATCH - (int)(strend - scan);␊ |
1190 | scan = strend - MAX_MATCH;␊ |
1191 | ␊ |
1192 | #endif /* UNALIGNED_OK */␊ |
1193 | ␊ |
1194 | if (len > best_len) {␊ |
1195 | s->match_start = cur_match;␊ |
1196 | best_len = len;␊ |
1197 | if (len >= nice_match) break;␊ |
1198 | #ifdef UNALIGNED_OK␊ |
1199 | scan_end = *(ushf*)(scan+best_len-1);␊ |
1200 | #else␊ |
1201 | scan_end1 = scan[best_len-1];␊ |
1202 | scan_end = scan[best_len];␊ |
1203 | #endif␊ |
1204 | }␊ |
1205 | } while ((cur_match = prev[cur_match & wmask]) > limit␊ |
1206 | && --chain_length != 0);␊ |
1207 | ␊ |
1208 | if ((uInt)best_len <= s->lookahead) return (uInt)best_len;␊ |
1209 | return s->lookahead;␊ |
1210 | }␊ |
1211 | #endif /* ASMV */␊ |
1212 | ␊ |
1213 | #else /* FASTEST */␊ |
1214 | ␊ |
1215 | /* ---------------------------------------------------------------------------␊ |
1216 | * Optimized version for FASTEST only␊ |
1217 | */␊ |
1218 | local uInt longest_match(s, cur_match)␊ |
1219 | deflate_state *s;␊ |
1220 | IPos cur_match; /* current match */␊ |
1221 | {␊ |
1222 | register Bytef *scan = s->window + s->strstart; /* current string */␊ |
1223 | register Bytef *match; /* matched string */␊ |
1224 | register int len; /* length of current match */␊ |
1225 | register Bytef *strend = s->window + s->strstart + MAX_MATCH;␊ |
1226 | ␊ |
1227 | /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.␊ |
1228 | * It is easy to get rid of this optimization if necessary.␊ |
1229 | */␊ |
1230 | Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");␊ |
1231 | ␊ |
1232 | Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");␊ |
1233 | ␊ |
1234 | Assert(cur_match < s->strstart, "no future");␊ |
1235 | ␊ |
1236 | match = s->window + cur_match;␊ |
1237 | ␊ |
1238 | /* Return failure if the match length is less than 2:␊ |
1239 | */␊ |
1240 | if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;␊ |
1241 | ␊ |
1242 | /* The check at best_len-1 can be removed because it will be made␊ |
1243 | * again later. (This heuristic is not always a win.)␊ |
1244 | * It is not necessary to compare scan[2] and match[2] since they␊ |
1245 | * are always equal when the other bytes match, given that␊ |
1246 | * the hash keys are equal and that HASH_BITS >= 8.␊ |
1247 | */␊ |
1248 | scan += 2, match += 2;␊ |
1249 | Assert(*scan == *match, "match[2]?");␊ |
1250 | ␊ |
1251 | /* We check for insufficient lookahead only every 8th comparison;␊ |
1252 | * the 256th check will be made at strstart+258.␊ |
1253 | */␊ |
1254 | do {␊ |
1255 | } while (*++scan == *++match && *++scan == *++match &&␊ |
1256 | *++scan == *++match && *++scan == *++match &&␊ |
1257 | *++scan == *++match && *++scan == *++match &&␊ |
1258 | *++scan == *++match && *++scan == *++match &&␊ |
1259 | scan < strend);␊ |
1260 | ␊ |
1261 | Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");␊ |
1262 | ␊ |
1263 | len = MAX_MATCH - (int)(strend - scan);␊ |
1264 | ␊ |
1265 | if (len < MIN_MATCH) return MIN_MATCH - 1;␊ |
1266 | ␊ |
1267 | s->match_start = cur_match;␊ |
1268 | return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;␊ |
1269 | }␊ |
1270 | ␊ |
1271 | #endif /* FASTEST */␊ |
1272 | ␊ |
1273 | #ifdef DEBUG␊ |
1274 | /* ===========================================================================␊ |
1275 | * Check that the match at match_start is indeed a match.␊ |
1276 | */␊ |
1277 | local void check_match(s, start, match, length)␊ |
1278 | deflate_state *s;␊ |
1279 | IPos start, match;␊ |
1280 | int length;␊ |
1281 | {␊ |
1282 | /* check that the match is indeed a match */␊ |
1283 | if (zmemcmp(s->window + match,␊ |
1284 | s->window + start, length) != EQUAL) {␊ |
1285 | fprintf(stderr, " start %u, match %u, length %d\n",␊ |
1286 | start, match, length);␊ |
1287 | do {␊ |
1288 | fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);␊ |
1289 | } while (--length != 0);␊ |
1290 | z_error("invalid match");␊ |
1291 | }␊ |
1292 | if (z_verbose > 1) {␊ |
1293 | fprintf(stderr,"\\[%d,%d]", start-match, length);␊ |
1294 | do { putc(s->window[start++], stderr); } while (--length != 0);␊ |
1295 | }␊ |
1296 | }␊ |
1297 | #else␊ |
1298 | # define check_match(s, start, match, length)␊ |
1299 | #endif /* DEBUG */␊ |
1300 | ␊ |
1301 | /* ===========================================================================␊ |
1302 | * Fill the window when the lookahead becomes insufficient.␊ |
1303 | * Updates strstart and lookahead.␊ |
1304 | *␊ |
1305 | * IN assertion: lookahead < MIN_LOOKAHEAD␊ |
1306 | * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD␊ |
1307 | * At least one byte has been read, or avail_in == 0; reads are␊ |
1308 | * performed for at least two bytes (required for the zip translate_eol␊ |
1309 | * option -- not supported here).␊ |
1310 | */␊ |
1311 | local void fill_window(s)␊ |
1312 | deflate_state *s;␊ |
1313 | {␊ |
1314 | register unsigned n, m;␊ |
1315 | register Posf *p;␊ |
1316 | unsigned more; /* Amount of free space at the end of the window. */␊ |
1317 | uInt wsize = s->w_size;␊ |
1318 | ␊ |
1319 | do {␊ |
1320 | more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);␊ |
1321 | ␊ |
1322 | /* Deal with !@#$% 64K limit: */␊ |
1323 | if (sizeof(int) <= 2) {␊ |
1324 | if (more == 0 && s->strstart == 0 && s->lookahead == 0) {␊ |
1325 | more = wsize;␊ |
1326 | ␊ |
1327 | } else if (more == (unsigned)(-1)) {␊ |
1328 | /* Very unlikely, but possible on 16 bit machine if␊ |
1329 | * strstart == 0 && lookahead == 1 (input done a byte at time)␊ |
1330 | */␊ |
1331 | more--;␊ |
1332 | }␊ |
1333 | }␊ |
1334 | ␊ |
1335 | /* If the window is almost full and there is insufficient lookahead,␊ |
1336 | * move the upper half to the lower one to make room in the upper half.␊ |
1337 | */␊ |
1338 | if (s->strstart >= wsize+MAX_DIST(s)) {␊ |
1339 | ␊ |
1340 | zmemcpy(s->window, s->window+wsize, (unsigned)wsize);␊ |
1341 | s->match_start -= wsize;␊ |
1342 | s->strstart -= wsize; /* we now have strstart >= MAX_DIST */␊ |
1343 | s->block_start -= (long) wsize;␊ |
1344 | ␊ |
1345 | /* Slide the hash table (could be avoided with 32 bit values␊ |
1346 | at the expense of memory usage). We slide even when level == 0␊ |
1347 | to keep the hash table consistent if we switch back to level > 0␊ |
1348 | later. (Using level 0 permanently is not an optimal usage of␊ |
1349 | zlib, so we don't care about this pathological case.)␊ |
1350 | */␊ |
1351 | n = s->hash_size;␊ |
1352 | p = &s->head[n];␊ |
1353 | do {␊ |
1354 | m = *--p;␊ |
1355 | *p = (Pos)(m >= wsize ? m-wsize : NIL);␊ |
1356 | } while (--n);␊ |
1357 | ␊ |
1358 | n = wsize;␊ |
1359 | #ifndef FASTEST␊ |
1360 | p = &s->prev[n];␊ |
1361 | do {␊ |
1362 | m = *--p;␊ |
1363 | *p = (Pos)(m >= wsize ? m-wsize : NIL);␊ |
1364 | /* If n is not on any hash chain, prev[n] is garbage but␊ |
1365 | * its value will never be used.␊ |
1366 | */␊ |
1367 | } while (--n);␊ |
1368 | #endif␊ |
1369 | more += wsize;␊ |
1370 | }␊ |
1371 | if (s->strm->avail_in == 0) return;␊ |
1372 | ␊ |
1373 | /* If there was no sliding:␊ |
1374 | * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&␊ |
1375 | * more == window_size - lookahead - strstart␊ |
1376 | * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)␊ |
1377 | * => more >= window_size - 2*WSIZE + 2␊ |
1378 | * In the BIG_MEM or MMAP case (not yet supported),␊ |
1379 | * window_size == input_size + MIN_LOOKAHEAD &&␊ |
1380 | * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.␊ |
1381 | * Otherwise, window_size == 2*WSIZE so more >= 2.␊ |
1382 | * If there was sliding, more >= WSIZE. So in all cases, more >= 2.␊ |
1383 | */␊ |
1384 | Assert(more >= 2, "more < 2");␊ |
1385 | ␊ |
1386 | n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);␊ |
1387 | s->lookahead += n;␊ |
1388 | ␊ |
1389 | /* Initialize the hash value now that we have some input: */␊ |
1390 | if (s->lookahead >= MIN_MATCH) {␊ |
1391 | s->ins_h = s->window[s->strstart];␊ |
1392 | UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);␊ |
1393 | #if MIN_MATCH != 3␊ |
1394 | Call UPDATE_HASH() MIN_MATCH-3 more times␊ |
1395 | #endif␊ |
1396 | }␊ |
1397 | /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,␊ |
1398 | * but this is not important since only literal bytes will be emitted.␊ |
1399 | */␊ |
1400 | ␊ |
1401 | } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);␊ |
1402 | ␊ |
1403 | /* If the WIN_INIT bytes after the end of the current data have never been␊ |
1404 | * written, then zero those bytes in order to avoid memory check reports of␊ |
1405 | * the use of uninitialized (or uninitialised as Julian writes) bytes by␊ |
1406 | * the longest match routines. Update the high water mark for the next␊ |
1407 | * time through here. WIN_INIT is set to MAX_MATCH since the longest match␊ |
1408 | * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.␊ |
1409 | */␊ |
1410 | if (s->high_water < s->window_size) {␊ |
1411 | ulg curr = s->strstart + (ulg)(s->lookahead);␊ |
1412 | ulg init;␊ |
1413 | ␊ |
1414 | if (s->high_water < curr) {␊ |
1415 | /* Previous high water mark below current data -- zero WIN_INIT␊ |
1416 | * bytes or up to end of window, whichever is less.␊ |
1417 | */␊ |
1418 | init = s->window_size - curr;␊ |
1419 | if (init > WIN_INIT)␊ |
1420 | init = WIN_INIT;␊ |
1421 | zmemzero(s->window + curr, (unsigned)init);␊ |
1422 | s->high_water = curr + init;␊ |
1423 | }␊ |
1424 | else if (s->high_water < (ulg)curr + WIN_INIT) {␊ |
1425 | /* High water mark at or above current data, but below current data␊ |
1426 | * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up␊ |
1427 | * to end of window, whichever is less.␊ |
1428 | */␊ |
1429 | init = (ulg)curr + WIN_INIT - s->high_water;␊ |
1430 | if (init > s->window_size - s->high_water)␊ |
1431 | init = s->window_size - s->high_water;␊ |
1432 | zmemzero(s->window + s->high_water, (unsigned)init);␊ |
1433 | s->high_water += init;␊ |
1434 | }␊ |
1435 | }␊ |
1436 | }␊ |
1437 | ␊ |
1438 | /* ===========================================================================␊ |
1439 | * Flush the current block, with given end-of-file flag.␊ |
1440 | * IN assertion: strstart is set to the end of the current match.␊ |
1441 | */␊ |
1442 | #define FLUSH_BLOCK_ONLY(s, last) { \␊ |
1443 | _tr_flush_block(s, (s->block_start >= 0L ? \␊ |
1444 | (charf *)&s->window[(unsigned)s->block_start] : \␊ |
1445 | (charf *)Z_NULL), \␊ |
1446 | (ulg)((long)s->strstart - s->block_start), \␊ |
1447 | (last)); \␊ |
1448 | s->block_start = s->strstart; \␊ |
1449 | flush_pending(s->strm); \␊ |
1450 | Tracev((stderr,"[FLUSH]")); \␊ |
1451 | }␊ |
1452 | ␊ |
1453 | /* Same but force premature exit if necessary. */␊ |
1454 | #define FLUSH_BLOCK(s, last) { \␊ |
1455 | FLUSH_BLOCK_ONLY(s, last); \␊ |
1456 | if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \␊ |
1457 | }␊ |
1458 | ␊ |
1459 | /* ===========================================================================␊ |
1460 | * Copy without compression as much as possible from the input stream, return␊ |
1461 | * the current block state.␊ |
1462 | * This function does not insert new strings in the dictionary since␊ |
1463 | * uncompressible data is probably not useful. This function is used␊ |
1464 | * only for the level=0 compression option.␊ |
1465 | * NOTE: this function should be optimized to avoid extra copying from␊ |
1466 | * window to pending_buf.␊ |
1467 | */␊ |
1468 | local block_state deflate_stored(s, flush)␊ |
1469 | deflate_state *s;␊ |
1470 | int flush;␊ |
1471 | {␊ |
1472 | /* Stored blocks are limited to 0xffff bytes, pending_buf is limited␊ |
1473 | * to pending_buf_size, and each stored block has a 5 byte header:␊ |
1474 | */␊ |
1475 | ulg max_block_size = 0xffff;␊ |
1476 | ulg max_start;␊ |
1477 | ␊ |
1478 | if (max_block_size > s->pending_buf_size - 5) {␊ |
1479 | max_block_size = s->pending_buf_size - 5;␊ |
1480 | }␊ |
1481 | ␊ |
1482 | /* Copy as much as possible from input to output: */␊ |
1483 | for (;;) {␊ |
1484 | /* Fill the window as much as possible: */␊ |
1485 | if (s->lookahead <= 1) {␊ |
1486 | ␊ |
1487 | Assert(s->strstart < s->w_size+MAX_DIST(s) ||␊ |
1488 | s->block_start >= (long)s->w_size, "slide too late");␊ |
1489 | ␊ |
1490 | fill_window(s);␊ |
1491 | if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;␊ |
1492 | ␊ |
1493 | if (s->lookahead == 0) break; /* flush the current block */␊ |
1494 | }␊ |
1495 | Assert(s->block_start >= 0L, "block gone");␊ |
1496 | ␊ |
1497 | s->strstart += s->lookahead;␊ |
1498 | s->lookahead = 0;␊ |
1499 | ␊ |
1500 | /* Emit a stored block if pending_buf will be full: */␊ |
1501 | max_start = s->block_start + max_block_size;␊ |
1502 | if (s->strstart == 0 || (ulg)s->strstart >= max_start) {␊ |
1503 | /* strstart == 0 is possible when wraparound on 16-bit machine */␊ |
1504 | s->lookahead = (uInt)(s->strstart - max_start);␊ |
1505 | s->strstart = (uInt)max_start;␊ |
1506 | FLUSH_BLOCK(s, 0);␊ |
1507 | }␊ |
1508 | /* Flush if we may have to slide, otherwise block_start may become␊ |
1509 | * negative and the data will be gone:␊ |
1510 | */␊ |
1511 | if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {␊ |
1512 | FLUSH_BLOCK(s, 0);␊ |
1513 | }␊ |
1514 | }␊ |
1515 | FLUSH_BLOCK(s, flush == Z_FINISH);␊ |
1516 | return flush == Z_FINISH ? finish_done : block_done;␊ |
1517 | }␊ |
1518 | ␊ |
1519 | /* ===========================================================================␊ |
1520 | * Compress as much as possible from the input stream, return the current␊ |
1521 | * block state.␊ |
1522 | * This function does not perform lazy evaluation of matches and inserts␊ |
1523 | * new strings in the dictionary only for unmatched strings or for short␊ |
1524 | * matches. It is used only for the fast compression options.␊ |
1525 | */␊ |
1526 | local block_state deflate_fast(s, flush)␊ |
1527 | deflate_state *s;␊ |
1528 | int flush;␊ |
1529 | {␊ |
1530 | IPos hash_head; /* head of the hash chain */␊ |
1531 | int bflush; /* set if current block must be flushed */␊ |
1532 | ␊ |
1533 | for (;;) {␊ |
1534 | /* Make sure that we always have enough lookahead, except␊ |
1535 | * at the end of the input file. We need MAX_MATCH bytes␊ |
1536 | * for the next match, plus MIN_MATCH bytes to insert the␊ |
1537 | * string following the next match.␊ |
1538 | */␊ |
1539 | if (s->lookahead < MIN_LOOKAHEAD) {␊ |
1540 | fill_window(s);␊ |
1541 | if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {␊ |
1542 | return need_more;␊ |
1543 | }␊ |
1544 | if (s->lookahead == 0) break; /* flush the current block */␊ |
1545 | }␊ |
1546 | ␊ |
1547 | /* Insert the string window[strstart .. strstart+2] in the␊ |
1548 | * dictionary, and set hash_head to the head of the hash chain:␊ |
1549 | */␊ |
1550 | hash_head = NIL;␊ |
1551 | if (s->lookahead >= MIN_MATCH) {␊ |
1552 | INSERT_STRING(s, s->strstart, hash_head);␊ |
1553 | }␊ |
1554 | ␊ |
1555 | /* Find the longest match, discarding those <= prev_length.␊ |
1556 | * At this point we have always match_length < MIN_MATCH␊ |
1557 | */␊ |
1558 | if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {␊ |
1559 | /* To simplify the code, we prevent matches with the string␊ |
1560 | * of window index 0 (in particular we have to avoid a match␊ |
1561 | * of the string with itself at the start of the input file).␊ |
1562 | */␊ |
1563 | s->match_length = longest_match (s, hash_head);␊ |
1564 | /* longest_match() sets match_start */␊ |
1565 | }␊ |
1566 | if (s->match_length >= MIN_MATCH) {␊ |
1567 | check_match(s, s->strstart, s->match_start, s->match_length);␊ |
1568 | ␊ |
1569 | _tr_tally_dist(s, s->strstart - s->match_start,␊ |
1570 | s->match_length - MIN_MATCH, bflush);␊ |
1571 | ␊ |
1572 | s->lookahead -= s->match_length;␊ |
1573 | ␊ |
1574 | /* Insert new strings in the hash table only if the match length␊ |
1575 | * is not too large. This saves time but degrades compression.␊ |
1576 | */␊ |
1577 | #ifndef FASTEST␊ |
1578 | if (s->match_length <= s->max_insert_length &&␊ |
1579 | s->lookahead >= MIN_MATCH) {␊ |
1580 | s->match_length--; /* string at strstart already in table */␊ |
1581 | do {␊ |
1582 | s->strstart++;␊ |
1583 | INSERT_STRING(s, s->strstart, hash_head);␊ |
1584 | /* strstart never exceeds WSIZE-MAX_MATCH, so there are␊ |
1585 | * always MIN_MATCH bytes ahead.␊ |
1586 | */␊ |
1587 | } while (--s->match_length != 0);␊ |
1588 | s->strstart++;␊ |
1589 | } else␊ |
1590 | #endif␊ |
1591 | {␊ |
1592 | s->strstart += s->match_length;␊ |
1593 | s->match_length = 0;␊ |
1594 | s->ins_h = s->window[s->strstart];␊ |
1595 | UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);␊ |
1596 | #if MIN_MATCH != 3␊ |
1597 | Call UPDATE_HASH() MIN_MATCH-3 more times␊ |
1598 | #endif␊ |
1599 | /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not␊ |
1600 | * matter since it will be recomputed at next deflate call.␊ |
1601 | */␊ |
1602 | }␊ |
1603 | } else {␊ |
1604 | /* No match, output a literal byte */␊ |
1605 | Tracevv((stderr,"%c", s->window[s->strstart]));␊ |
1606 | _tr_tally_lit (s, s->window[s->strstart], bflush);␊ |
1607 | s->lookahead--;␊ |
1608 | s->strstart++;␊ |
1609 | }␊ |
1610 | if (bflush) FLUSH_BLOCK(s, 0);␊ |
1611 | }␊ |
1612 | FLUSH_BLOCK(s, flush == Z_FINISH);␊ |
1613 | return flush == Z_FINISH ? finish_done : block_done;␊ |
1614 | }␊ |
1615 | ␊ |
1616 | #ifndef FASTEST␊ |
1617 | /* ===========================================================================␊ |
1618 | * Same as above, but achieves better compression. We use a lazy␊ |
1619 | * evaluation for matches: a match is finally adopted only if there is␊ |
1620 | * no better match at the next window position.␊ |
1621 | */␊ |
1622 | local block_state deflate_slow(s, flush)␊ |
1623 | deflate_state *s;␊ |
1624 | int flush;␊ |
1625 | {␊ |
1626 | IPos hash_head; /* head of hash chain */␊ |
1627 | int bflush; /* set if current block must be flushed */␊ |
1628 | ␊ |
1629 | /* Process the input block. */␊ |
1630 | for (;;) {␊ |
1631 | /* Make sure that we always have enough lookahead, except␊ |
1632 | * at the end of the input file. We need MAX_MATCH bytes␊ |
1633 | * for the next match, plus MIN_MATCH bytes to insert the␊ |
1634 | * string following the next match.␊ |
1635 | */␊ |
1636 | if (s->lookahead < MIN_LOOKAHEAD) {␊ |
1637 | fill_window(s);␊ |
1638 | if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {␊ |
1639 | return need_more;␊ |
1640 | }␊ |
1641 | if (s->lookahead == 0) break; /* flush the current block */␊ |
1642 | }␊ |
1643 | ␊ |
1644 | /* Insert the string window[strstart .. strstart+2] in the␊ |
1645 | * dictionary, and set hash_head to the head of the hash chain:␊ |
1646 | */␊ |
1647 | hash_head = NIL;␊ |
1648 | if (s->lookahead >= MIN_MATCH) {␊ |
1649 | INSERT_STRING(s, s->strstart, hash_head);␊ |
1650 | }␊ |
1651 | ␊ |
1652 | /* Find the longest match, discarding those <= prev_length.␊ |
1653 | */␊ |
1654 | s->prev_length = s->match_length, s->prev_match = s->match_start;␊ |
1655 | s->match_length = MIN_MATCH-1;␊ |
1656 | ␊ |
1657 | if (hash_head != NIL && s->prev_length < s->max_lazy_match &&␊ |
1658 | s->strstart - hash_head <= MAX_DIST(s)) {␊ |
1659 | /* To simplify the code, we prevent matches with the string␊ |
1660 | * of window index 0 (in particular we have to avoid a match␊ |
1661 | * of the string with itself at the start of the input file).␊ |
1662 | */␊ |
1663 | s->match_length = longest_match (s, hash_head);␊ |
1664 | /* longest_match() sets match_start */␊ |
1665 | ␊ |
1666 | if (s->match_length <= 5 && (s->strategy == Z_FILTERED␊ |
1667 | #if TOO_FAR <= 32767␊ |
1668 | || (s->match_length == MIN_MATCH &&␊ |
1669 | s->strstart - s->match_start > TOO_FAR)␊ |
1670 | #endif␊ |
1671 | )) {␊ |
1672 | ␊ |
1673 | /* If prev_match is also MIN_MATCH, match_start is garbage␊ |
1674 | * but we will ignore the current match anyway.␊ |
1675 | */␊ |
1676 | s->match_length = MIN_MATCH-1;␊ |
1677 | }␊ |
1678 | }␊ |
1679 | /* If there was a match at the previous step and the current␊ |
1680 | * match is not better, output the previous match:␊ |
1681 | */␊ |
1682 | if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {␊ |
1683 | uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;␊ |
1684 | /* Do not insert strings in hash table beyond this. */␊ |
1685 | ␊ |
1686 | check_match(s, s->strstart-1, s->prev_match, s->prev_length);␊ |
1687 | ␊ |
1688 | _tr_tally_dist(s, s->strstart -1 - s->prev_match,␊ |
1689 | s->prev_length - MIN_MATCH, bflush);␊ |
1690 | ␊ |
1691 | /* Insert in hash table all strings up to the end of the match.␊ |
1692 | * strstart-1 and strstart are already inserted. If there is not␊ |
1693 | * enough lookahead, the last two strings are not inserted in␊ |
1694 | * the hash table.␊ |
1695 | */␊ |
1696 | s->lookahead -= s->prev_length-1;␊ |
1697 | s->prev_length -= 2;␊ |
1698 | do {␊ |
1699 | if (++s->strstart <= max_insert) {␊ |
1700 | INSERT_STRING(s, s->strstart, hash_head);␊ |
1701 | }␊ |
1702 | } while (--s->prev_length != 0);␊ |
1703 | s->match_available = 0;␊ |
1704 | s->match_length = MIN_MATCH-1;␊ |
1705 | s->strstart++;␊ |
1706 | ␊ |
1707 | if (bflush) FLUSH_BLOCK(s, 0);␊ |
1708 | ␊ |
1709 | } else if (s->match_available) {␊ |
1710 | /* If there was no match at the previous position, output a␊ |
1711 | * single literal. If there was a match but the current match␊ |
1712 | * is longer, truncate the previous match to a single literal.␊ |
1713 | */␊ |
1714 | Tracevv((stderr,"%c", s->window[s->strstart-1]));␊ |
1715 | _tr_tally_lit(s, s->window[s->strstart-1], bflush);␊ |
1716 | if (bflush) {␊ |
1717 | FLUSH_BLOCK_ONLY(s, 0);␊ |
1718 | }␊ |
1719 | s->strstart++;␊ |
1720 | s->lookahead--;␊ |
1721 | if (s->strm->avail_out == 0) return need_more;␊ |
1722 | } else {␊ |
1723 | /* There is no previous match to compare with, wait for␊ |
1724 | * the next step to decide.␊ |
1725 | */␊ |
1726 | s->match_available = 1;␊ |
1727 | s->strstart++;␊ |
1728 | s->lookahead--;␊ |
1729 | }␊ |
1730 | }␊ |
1731 | Assert (flush != Z_NO_FLUSH, "no flush?");␊ |
1732 | if (s->match_available) {␊ |
1733 | Tracevv((stderr,"%c", s->window[s->strstart-1]));␊ |
1734 | _tr_tally_lit(s, s->window[s->strstart-1], bflush);␊ |
1735 | s->match_available = 0;␊ |
1736 | }␊ |
1737 | FLUSH_BLOCK(s, flush == Z_FINISH);␊ |
1738 | return flush == Z_FINISH ? finish_done : block_done;␊ |
1739 | }␊ |
1740 | #endif /* FASTEST */␊ |
1741 | ␊ |
1742 | /* ===========================================================================␊ |
1743 | * For Z_RLE, simply look for runs of bytes, generate matches only of distance␊ |
1744 | * one. Do not maintain a hash table. (It will be regenerated if this run of␊ |
1745 | * deflate switches away from Z_RLE.)␊ |
1746 | */␊ |
1747 | local block_state deflate_rle(s, flush)␊ |
1748 | deflate_state *s;␊ |
1749 | int flush;␊ |
1750 | {␊ |
1751 | int bflush; /* set if current block must be flushed */␊ |
1752 | uInt prev; /* byte at distance one to match */␊ |
1753 | Bytef *scan, *strend; /* scan goes up to strend for length of run */␊ |
1754 | ␊ |
1755 | for (;;) {␊ |
1756 | /* Make sure that we always have enough lookahead, except␊ |
1757 | * at the end of the input file. We need MAX_MATCH bytes␊ |
1758 | * for the longest encodable run.␊ |
1759 | */␊ |
1760 | if (s->lookahead < MAX_MATCH) {␊ |
1761 | fill_window(s);␊ |
1762 | if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) {␊ |
1763 | return need_more;␊ |
1764 | }␊ |
1765 | if (s->lookahead == 0) break; /* flush the current block */␊ |
1766 | }␊ |
1767 | ␊ |
1768 | /* See how many times the previous byte repeats */␊ |
1769 | s->match_length = 0;␊ |
1770 | if (s->lookahead >= MIN_MATCH && s->strstart > 0) {␊ |
1771 | scan = s->window + s->strstart - 1;␊ |
1772 | prev = *scan;␊ |
1773 | if (prev == *++scan && prev == *++scan && prev == *++scan) {␊ |
1774 | strend = s->window + s->strstart + MAX_MATCH;␊ |
1775 | do {␊ |
1776 | } while (prev == *++scan && prev == *++scan &&␊ |
1777 | prev == *++scan && prev == *++scan &&␊ |
1778 | prev == *++scan && prev == *++scan &&␊ |
1779 | prev == *++scan && prev == *++scan &&␊ |
1780 | scan < strend);␊ |
1781 | s->match_length = MAX_MATCH - (int)(strend - scan);␊ |
1782 | if (s->match_length > s->lookahead)␊ |
1783 | s->match_length = s->lookahead;␊ |
1784 | }␊ |
1785 | }␊ |
1786 | ␊ |
1787 | /* Emit match if have run of MIN_MATCH or longer, else emit literal */␊ |
1788 | if (s->match_length >= MIN_MATCH) {␊ |
1789 | check_match(s, s->strstart, s->strstart - 1, s->match_length);␊ |
1790 | ␊ |
1791 | _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);␊ |
1792 | ␊ |
1793 | s->lookahead -= s->match_length;␊ |
1794 | s->strstart += s->match_length;␊ |
1795 | s->match_length = 0;␊ |
1796 | } else {␊ |
1797 | /* No match, output a literal byte */␊ |
1798 | Tracevv((stderr,"%c", s->window[s->strstart]));␊ |
1799 | _tr_tally_lit (s, s->window[s->strstart], bflush);␊ |
1800 | s->lookahead--;␊ |
1801 | s->strstart++;␊ |
1802 | }␊ |
1803 | if (bflush) FLUSH_BLOCK(s, 0);␊ |
1804 | }␊ |
1805 | FLUSH_BLOCK(s, flush == Z_FINISH);␊ |
1806 | return flush == Z_FINISH ? finish_done : block_done;␊ |
1807 | }␊ |
1808 | ␊ |
1809 | /* ===========================================================================␊ |
1810 | * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.␊ |
1811 | * (It will be regenerated if this run of deflate switches away from Huffman.)␊ |
1812 | */␊ |
1813 | local block_state deflate_huff(s, flush)␊ |
1814 | deflate_state *s;␊ |
1815 | int flush;␊ |
1816 | {␊ |
1817 | int bflush; /* set if current block must be flushed */␊ |
1818 | ␊ |
1819 | for (;;) {␊ |
1820 | /* Make sure that we have a literal to write. */␊ |
1821 | if (s->lookahead == 0) {␊ |
1822 | fill_window(s);␊ |
1823 | if (s->lookahead == 0) {␊ |
1824 | if (flush == Z_NO_FLUSH)␊ |
1825 | return need_more;␊ |
1826 | break; /* flush the current block */␊ |
1827 | }␊ |
1828 | }␊ |
1829 | ␊ |
1830 | /* Output a literal byte */␊ |
1831 | s->match_length = 0;␊ |
1832 | Tracevv((stderr,"%c", s->window[s->strstart]));␊ |
1833 | _tr_tally_lit (s, s->window[s->strstart], bflush);␊ |
1834 | s->lookahead--;␊ |
1835 | s->strstart++;␊ |
1836 | if (bflush) FLUSH_BLOCK(s, 0);␊ |
1837 | }␊ |
1838 | FLUSH_BLOCK(s, flush == Z_FINISH);␊ |
1839 | return flush == Z_FINISH ? finish_done : block_done;␊ |
1840 | }␊ |
1841 | |