1 /* deflate.h -- internal compression state 2 * Copyright (C) 1995-2024 Jean-loup Gailly 3 * For conditions of distribution and use, see copyright notice in zlib.h 4 */ 5 6 /* WARNING: this file should *not* be used by applications. It is 7 part of the implementation of the compression library and is 8 subject to change. Applications should only use zlib.h. 9 */ 10 11 #ifndef DEFLATE_H 12 #define DEFLATE_H 13 14 #include "zutil.h" 15 16 /* define NO_GZIP when compiling if you want to disable gzip header and 17 trailer creation by deflate(). NO_GZIP would be used to avoid linking in 18 the crc code when it is not needed. For shared libraries, gzip encoding 19 should be left enabled. */ 20 #ifndef NO_GZIP 21 # define GZIP 22 #endif 23 24 /* define LIT_MEM to slightly increase the speed of deflate (order 1% to 2%) at 25 the cost of a larger memory footprint */ 26 /* #define LIT_MEM */ 27 28 /* =========================================================================== 29 * Internal compression state. 30 */ 31 32 #define LENGTH_CODES 29 33 /* number of length codes, not counting the special END_BLOCK code */ 34 35 #define LITERALS 256 36 /* number of literal bytes 0..255 */ 37 38 #define L_CODES (LITERALS+1+LENGTH_CODES) 39 /* number of Literal or Length codes, including the END_BLOCK code */ 40 41 #define D_CODES 30 42 /* number of distance codes */ 43 44 #define BL_CODES 19 45 /* number of codes used to transfer the bit lengths */ 46 47 #define HEAP_SIZE (2*L_CODES+1) 48 /* maximum heap size */ 49 50 #define MAX_BITS 15 51 /* All codes must not exceed MAX_BITS bits */ 52 53 #define Buf_size 16 54 /* size of bit buffer in bi_buf */ 55 56 #define INIT_STATE 42 /* zlib header -> BUSY_STATE */ 57 #ifdef GZIP 58 # define GZIP_STATE 57 /* gzip header -> BUSY_STATE | EXTRA_STATE */ 59 #endif 60 #define EXTRA_STATE 69 /* gzip extra block -> NAME_STATE */ 61 #define NAME_STATE 73 /* gzip file name -> COMMENT_STATE */ 62 #define COMMENT_STATE 91 /* gzip comment -> HCRC_STATE */ 63 #define HCRC_STATE 103 /* gzip header CRC -> BUSY_STATE */ 64 #define BUSY_STATE 113 /* deflate -> FINISH_STATE */ 65 #define FINISH_STATE 666 /* stream complete */ 66 /* Stream status */ 67 68 69 /* Data structure describing a single value and its code string. */ 70 typedef struct ct_data_s { 71 union { 72 ush freq; /* frequency count */ 73 ush code; /* bit string */ 74 } fc; 75 union { 76 ush dad; /* father node in Huffman tree */ 77 ush len; /* length of bit string */ 78 } dl; 79 } FAR ct_data; 80 81 #define Freq fc.freq 82 #define Code fc.code 83 #define Dad dl.dad 84 #define Len dl.len 85 86 typedef struct static_tree_desc_s static_tree_desc; 87 88 typedef struct tree_desc_s { 89 ct_data *dyn_tree; /* the dynamic tree */ 90 int max_code; /* largest code with non zero frequency */ 91 const static_tree_desc *stat_desc; /* the corresponding static tree */ 92 } FAR tree_desc; 93 94 typedef ush Pos; 95 typedef Pos FAR Posf; 96 typedef unsigned IPos; 97 98 /* A Pos is an index in the character window. We use short instead of int to 99 * save space in the various tables. IPos is used only for parameter passing. 100 */ 101 102 typedef struct internal_state { 103 z_streamp strm; /* pointer back to this zlib stream */ 104 int status; /* as the name implies */ 105 Bytef *pending_buf; /* output still pending */ 106 ulg pending_buf_size; /* size of pending_buf */ 107 Bytef *pending_out; /* next pending byte to output to the stream */ 108 ulg pending; /* nb of bytes in the pending buffer */ 109 int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ 110 gz_headerp gzhead; /* gzip header information to write */ 111 ulg gzindex; /* where in extra, name, or comment */ 112 Byte method; /* can only be DEFLATED */ 113 int last_flush; /* value of flush param for previous deflate call */ 114 115 /* used by deflate.c: */ 116 117 uInt w_size; /* LZ77 window size (32K by default) */ 118 uInt w_bits; /* log2(w_size) (8..16) */ 119 uInt w_mask; /* w_size - 1 */ 120 121 Bytef *window; 122 /* Sliding window. Input bytes are read into the second half of the window, 123 * and move to the first half later to keep a dictionary of at least wSize 124 * bytes. With this organization, matches are limited to a distance of 125 * wSize-MAX_MATCH bytes, but this ensures that IO is always 126 * performed with a length multiple of the block size. Also, it limits 127 * the window size to 64K, which is quite useful on MSDOS. 128 * To do: use the user input buffer as sliding window. 129 */ 130 131 ulg window_size; 132 /* Actual size of window: 2*wSize, except when the user input buffer 133 * is directly used as sliding window. 134 */ 135 136 Posf *prev; 137 /* Link to older string with same hash index. To limit the size of this 138 * array to 64K, this link is maintained only for the last 32K strings. 139 * An index in this array is thus a window index modulo 32K. 140 */ 141 142 Posf *head; /* Heads of the hash chains or NIL. */ 143 144 uInt ins_h; /* hash index of string to be inserted */ 145 uInt hash_size; /* number of elements in hash table */ 146 uInt hash_bits; /* log2(hash_size) */ 147 uInt hash_mask; /* hash_size-1 */ 148 149 uInt hash_shift; 150 /* Number of bits by which ins_h must be shifted at each input 151 * step. It must be such that after MIN_MATCH steps, the oldest 152 * byte no longer takes part in the hash key, that is: 153 * hash_shift * MIN_MATCH >= hash_bits 154 */ 155 156 long block_start; 157 /* Window position at the beginning of the current output block. Gets 158 * negative when the window is moved backwards. 159 */ 160 161 uInt match_length; /* length of best match */ 162 IPos prev_match; /* previous match */ 163 int match_available; /* set if previous match exists */ 164 uInt strstart; /* start of string to insert */ 165 uInt match_start; /* start of matching string */ 166 uInt lookahead; /* number of valid bytes ahead in window */ 167 168 uInt prev_length; 169 /* Length of the best match at previous step. Matches not greater than this 170 * are discarded. This is used in the lazy match evaluation. 171 */ 172 173 uInt max_chain_length; 174 /* To speed up deflation, hash chains are never searched beyond this 175 * length. A higher limit improves compression ratio but degrades the 176 * speed. 177 */ 178 179 uInt max_lazy_match; 180 /* Attempt to find a better match only when the current match is strictly 181 * smaller than this value. This mechanism is used only for compression 182 * levels >= 4. 183 */ 184 # define max_insert_length max_lazy_match 185 /* Insert new strings in the hash table only if the match length is not 186 * greater than this length. This saves time but degrades compression. 187 * max_insert_length is used only for compression levels <= 3. 188 */ 189 190 int level; /* compression level (1..9) */ 191 int strategy; /* favor or force Huffman coding*/ 192 193 uInt good_match; 194 /* Use a faster search when the previous match is longer than this */ 195 196 int nice_match; /* Stop searching when current match exceeds this */ 197 198 /* used by trees.c: */ 199 /* Didn't use ct_data typedef below to suppress compiler warning */ 200 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */ 201 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */ 202 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */ 203 204 struct tree_desc_s l_desc; /* desc. for literal tree */ 205 struct tree_desc_s d_desc; /* desc. for distance tree */ 206 struct tree_desc_s bl_desc; /* desc. for bit length tree */ 207 208 ush bl_count[MAX_BITS+1]; 209 /* number of codes at each bit length for an optimal tree */ 210 211 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */ 212 int heap_len; /* number of elements in the heap */ 213 int heap_max; /* element of largest frequency */ 214 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. 215 * The same heap array is used to build all trees. 216 */ 217 218 uch depth[2*L_CODES+1]; 219 /* Depth of each subtree used as tie breaker for trees of equal frequency 220 */ 221 222 #ifdef LIT_MEM 223 # define LIT_BUFS 5 224 ushf *d_buf; /* buffer for distances */ 225 uchf *l_buf; /* buffer for literals/lengths */ 226 #else 227 # define LIT_BUFS 4 228 uchf *sym_buf; /* buffer for distances and literals/lengths */ 229 #endif 230 231 uInt lit_bufsize; 232 /* Size of match buffer for literals/lengths. There are 4 reasons for 233 * limiting lit_bufsize to 64K: 234 * - frequencies can be kept in 16 bit counters 235 * - if compression is not successful for the first block, all input 236 * data is still in the window so we can still emit a stored block even 237 * when input comes from standard input. (This can also be done for 238 * all blocks if lit_bufsize is not greater than 32K.) 239 * - if compression is not successful for a file smaller than 64K, we can 240 * even emit a stored file instead of a stored block (saving 5 bytes). 241 * This is applicable only for zip (not gzip or zlib). 242 * - creating new Huffman trees less frequently may not provide fast 243 * adaptation to changes in the input data statistics. (Take for 244 * example a binary file with poorly compressible code followed by 245 * a highly compressible string table.) Smaller buffer sizes give 246 * fast adaptation but have of course the overhead of transmitting 247 * trees more frequently. 248 * - I can't count above 4 249 */ 250 251 uInt sym_next; /* running index in symbol buffer */ 252 uInt sym_end; /* symbol table full when sym_next reaches this */ 253 254 ulg opt_len; /* bit length of current block with optimal trees */ 255 ulg static_len; /* bit length of current block with static trees */ 256 uInt matches; /* number of string matches in current block */ 257 uInt insert; /* bytes at end of window left to insert */ 258 259 #ifdef ZLIB_DEBUG 260 ulg compressed_len; /* total bit length of compressed file mod 2^32 */ 261 ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ 262 #endif 263 264 ush bi_buf; 265 /* Output buffer. bits are inserted starting at the bottom (least 266 * significant bits). 267 */ 268 int bi_valid; 269 /* Number of valid bits in bi_buf. All bits above the last valid bit 270 * are always zero. 271 */ 272 int bi_used; 273 /* Last number of used bits when going to a byte boundary. 274 */ 275 276 ulg high_water; 277 /* High water mark offset in window for initialized bytes -- bytes above 278 * this are set to zero in order to avoid memory check warnings when 279 * longest match routines access bytes past the input. This is then 280 * updated to the new high water mark. 281 */ 282 283 } FAR deflate_state; 284 285 /* Output a byte on the stream. 286 * IN assertion: there is enough room in pending_buf. 287 */ 288 #define put_byte(s, c) {s->pending_buf[s->pending++] = (Bytef)(c);} 289 290 291 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) 292 /* Minimum amount of lookahead, except at the end of the input file. 293 * See deflate.c for comments about the MIN_MATCH+1. 294 */ 295 296 #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD) 297 /* In order to simplify the code, particularly on 16 bit machines, match 298 * distances are limited to MAX_DIST instead of WSIZE. 299 */ 300 301 #define WIN_INIT MAX_MATCH 302 /* Number of bytes after end of data in window to initialize in order to avoid 303 memory checker errors from longest match routines */ 304 305 /* in trees.c */ 306 void ZLIB_INTERNAL _tr_init(deflate_state *s); 307 int ZLIB_INTERNAL _tr_tally(deflate_state *s, unsigned dist, unsigned lc); 308 void ZLIB_INTERNAL _tr_flush_block(deflate_state *s, charf *buf, 309 ulg stored_len, int last); 310 void ZLIB_INTERNAL _tr_flush_bits(deflate_state *s); 311 void ZLIB_INTERNAL _tr_align(deflate_state *s); 312 void ZLIB_INTERNAL _tr_stored_block(deflate_state *s, charf *buf, 313 ulg stored_len, int last); 314 315 #define d_code(dist) \ 316 ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) 317 /* Mapping from a distance to a distance code. dist is the distance - 1 and 318 * must not have side effects. _dist_code[256] and _dist_code[257] are never 319 * used. 320 */ 321 322 #ifndef ZLIB_DEBUG 323 /* Inline versions of _tr_tally for speed: */ 324 325 #if defined(GEN_TREES_H) || !defined(STDC) 326 extern uch ZLIB_INTERNAL _length_code[]; 327 extern uch ZLIB_INTERNAL _dist_code[]; 328 #else 329 extern const uch ZLIB_INTERNAL _length_code[]; 330 extern const uch ZLIB_INTERNAL _dist_code[]; 331 #endif 332 333 #ifdef LIT_MEM 334 # define _tr_tally_lit(s, c, flush) \ 335 { uch cc = (c); \ 336 s->d_buf[s->sym_next] = 0; \ 337 s->l_buf[s->sym_next++] = cc; \ 338 s->dyn_ltree[cc].Freq++; \ 339 flush = (s->sym_next == s->sym_end); \ 340 } 341 # define _tr_tally_dist(s, distance, length, flush) \ 342 { uch len = (uch)(length); \ 343 ush dist = (ush)(distance); \ 344 s->d_buf[s->sym_next] = dist; \ 345 s->l_buf[s->sym_next++] = len; \ 346 dist--; \ 347 s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ 348 s->dyn_dtree[d_code(dist)].Freq++; \ 349 flush = (s->sym_next == s->sym_end); \ 350 } 351 #else 352 # define _tr_tally_lit(s, c, flush) \ 353 { uch cc = (c); \ 354 s->sym_buf[s->sym_next++] = 0; \ 355 s->sym_buf[s->sym_next++] = 0; \ 356 s->sym_buf[s->sym_next++] = cc; \ 357 s->dyn_ltree[cc].Freq++; \ 358 flush = (s->sym_next == s->sym_end); \ 359 } 360 # define _tr_tally_dist(s, distance, length, flush) \ 361 { uch len = (uch)(length); \ 362 ush dist = (ush)(distance); \ 363 s->sym_buf[s->sym_next++] = (uch)dist; \ 364 s->sym_buf[s->sym_next++] = (uch)(dist >> 8); \ 365 s->sym_buf[s->sym_next++] = len; \ 366 dist--; \ 367 s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ 368 s->dyn_dtree[d_code(dist)].Freq++; \ 369 flush = (s->sym_next == s->sym_end); \ 370 } 371 #endif 372 #else 373 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c) 374 # define _tr_tally_dist(s, distance, length, flush) \ 375 flush = _tr_tally(s, distance, length) 376 #endif 377 378 #endif /* DEFLATE_H */ 379