1 /* $NetBSD: deflate.h,v 1.6 2024/09/22 19:12:27 christos Exp $ */ 2 3 /* deflate.h -- internal compression state 4 * Copyright (C) 1995-2024 Jean-loup Gailly 5 * For conditions of distribution and use, see copyright notice in zlib.h 6 */ 7 8 /* WARNING: this file should *not* be used by applications. It is 9 part of the implementation of the compression library and is 10 subject to change. Applications should only use zlib.h. 11 */ 12 13 /* @(#) Id */ 14 15 #ifndef DEFLATE_H 16 #define DEFLATE_H 17 18 #include "zutil.h" 19 20 /* define NO_GZIP when compiling if you want to disable gzip header and 21 trailer creation by deflate(). NO_GZIP would be used to avoid linking in 22 the crc code when it is not needed. For shared libraries, gzip encoding 23 should be left enabled. */ 24 #ifndef NO_GZIP 25 # define GZIP 26 #endif 27 28 /* define LIT_MEM to slightly increase the speed of deflate (order 1% to 2%) at 29 the cost of a larger memory footprint */ 30 /* #define LIT_MEM */ 31 32 /* =========================================================================== 33 * Internal compression state. 34 */ 35 36 #define LENGTH_CODES 29 37 /* number of length codes, not counting the special END_BLOCK code */ 38 39 #define LITERALS 256 40 /* number of literal bytes 0..255 */ 41 42 #define L_CODES (LITERALS+1+LENGTH_CODES) 43 /* number of Literal or Length codes, including the END_BLOCK code */ 44 45 #define D_CODES 30 46 /* number of distance codes */ 47 48 #define BL_CODES 19 49 /* number of codes used to transfer the bit lengths */ 50 51 #define HEAP_SIZE (2*L_CODES+1) 52 /* maximum heap size */ 53 54 #define MAX_BITS 15 55 /* All codes must not exceed MAX_BITS bits */ 56 57 #define Buf_size 16 58 /* size of bit buffer in bi_buf */ 59 60 #define INIT_STATE 42 /* zlib header -> BUSY_STATE */ 61 #ifdef GZIP 62 # define GZIP_STATE 57 /* gzip header -> BUSY_STATE | EXTRA_STATE */ 63 #endif 64 #define EXTRA_STATE 69 /* gzip extra block -> NAME_STATE */ 65 #define NAME_STATE 73 /* gzip file name -> COMMENT_STATE */ 66 #define COMMENT_STATE 91 /* gzip comment -> HCRC_STATE */ 67 #define HCRC_STATE 103 /* gzip header CRC -> BUSY_STATE */ 68 #define BUSY_STATE 113 /* deflate -> FINISH_STATE */ 69 #define FINISH_STATE 666 /* stream complete */ 70 /* Stream status */ 71 72 73 /* Data structure describing a single value and its code string. */ 74 typedef struct ct_data_s { 75 union { 76 ush freq; /* frequency count */ 77 ush code; /* bit string */ 78 } fc; 79 union { 80 ush dad; /* father node in Huffman tree */ 81 ush len; /* length of bit string */ 82 } dl; 83 } FAR ct_data; 84 85 #define Freq fc.freq 86 #define Code fc.code 87 #define Dad dl.dad 88 #define Len dl.len 89 90 typedef struct static_tree_desc_s static_tree_desc; 91 92 typedef struct tree_desc_s { 93 ct_data *dyn_tree; /* the dynamic tree */ 94 int max_code; /* largest code with non zero frequency */ 95 const static_tree_desc *stat_desc; /* the corresponding static tree */ 96 } FAR tree_desc; 97 98 typedef ush Pos; 99 typedef Pos FAR Posf; 100 typedef unsigned IPos; 101 102 /* A Pos is an index in the character window. We use short instead of int to 103 * save space in the various tables. IPos is used only for parameter passing. 104 */ 105 106 typedef struct internal_state { 107 z_streamp strm; /* pointer back to this zlib stream */ 108 int status; /* as the name implies */ 109 Bytef *pending_buf; /* output still pending */ 110 ulg pending_buf_size; /* size of pending_buf */ 111 Bytef *pending_out; /* next pending byte to output to the stream */ 112 ulg pending; /* nb of bytes in the pending buffer */ 113 int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ 114 gz_headerp gzhead; /* gzip header information to write */ 115 ulg gzindex; /* where in extra, name, or comment */ 116 Byte method; /* can only be DEFLATED */ 117 int last_flush; /* value of flush param for previous deflate call */ 118 119 /* used by deflate.c: */ 120 121 uInt w_size; /* LZ77 window size (32K by default) */ 122 uInt w_bits; /* log2(w_size) (8..16) */ 123 uInt w_mask; /* w_size - 1 */ 124 125 Bytef *window; 126 /* Sliding window. Input bytes are read into the second half of the window, 127 * and move to the first half later to keep a dictionary of at least wSize 128 * bytes. With this organization, matches are limited to a distance of 129 * wSize-MAX_MATCH bytes, but this ensures that IO is always 130 * performed with a length multiple of the block size. Also, it limits 131 * the window size to 64K, which is quite useful on MSDOS. 132 * To do: use the user input buffer as sliding window. 133 */ 134 135 ulg window_size; 136 /* Actual size of window: 2*wSize, except when the user input buffer 137 * is directly used as sliding window. 138 */ 139 140 Posf *prev; 141 /* Link to older string with same hash index. To limit the size of this 142 * array to 64K, this link is maintained only for the last 32K strings. 143 * An index in this array is thus a window index modulo 32K. 144 */ 145 146 Posf *head; /* Heads of the hash chains or NIL. */ 147 148 uInt ins_h; /* hash index of string to be inserted */ 149 uInt hash_size; /* number of elements in hash table */ 150 uInt hash_bits; /* log2(hash_size) */ 151 uInt hash_mask; /* hash_size-1 */ 152 153 uInt hash_shift; 154 /* Number of bits by which ins_h must be shifted at each input 155 * step. It must be such that after MIN_MATCH steps, the oldest 156 * byte no longer takes part in the hash key, that is: 157 * hash_shift * MIN_MATCH >= hash_bits 158 */ 159 160 long block_start; 161 /* Window position at the beginning of the current output block. Gets 162 * negative when the window is moved backwards. 163 */ 164 165 uInt match_length; /* length of best match */ 166 IPos prev_match; /* previous match */ 167 int match_available; /* set if previous match exists */ 168 uInt strstart; /* start of string to insert */ 169 uInt match_start; /* start of matching string */ 170 uInt lookahead; /* number of valid bytes ahead in window */ 171 172 uInt prev_length; 173 /* Length of the best match at previous step. Matches not greater than this 174 * are discarded. This is used in the lazy match evaluation. 175 */ 176 177 uInt max_chain_length; 178 /* To speed up deflation, hash chains are never searched beyond this 179 * length. A higher limit improves compression ratio but degrades the 180 * speed. 181 */ 182 183 uInt max_lazy_match; 184 /* Attempt to find a better match only when the current match is strictly 185 * smaller than this value. This mechanism is used only for compression 186 * levels >= 4. 187 */ 188 # define max_insert_length max_lazy_match 189 /* Insert new strings in the hash table only if the match length is not 190 * greater than this length. This saves time but degrades compression. 191 * max_insert_length is used only for compression levels <= 3. 192 */ 193 194 int level; /* compression level (1..9) */ 195 int strategy; /* favor or force Huffman coding*/ 196 197 uInt good_match; 198 /* Use a faster search when the previous match is longer than this */ 199 200 int nice_match; /* Stop searching when current match exceeds this */ 201 202 /* used by trees.c: */ 203 /* Didn't use ct_data typedef below to suppress compiler warning */ 204 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */ 205 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */ 206 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */ 207 208 struct tree_desc_s l_desc; /* desc. for literal tree */ 209 struct tree_desc_s d_desc; /* desc. for distance tree */ 210 struct tree_desc_s bl_desc; /* desc. for bit length tree */ 211 212 ush bl_count[MAX_BITS+1]; 213 /* number of codes at each bit length for an optimal tree */ 214 215 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */ 216 int heap_len; /* number of elements in the heap */ 217 int heap_max; /* element of largest frequency */ 218 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. 219 * The same heap array is used to build all trees. 220 */ 221 222 uch depth[2*L_CODES+1]; 223 /* Depth of each subtree used as tie breaker for trees of equal frequency 224 */ 225 226 #ifdef LIT_MEM 227 # define LIT_BUFS 5 228 ushf *d_buf; /* buffer for distances */ 229 uchf *l_buf; /* buffer for literals/lengths */ 230 #else 231 # define LIT_BUFS 4 232 uchf *sym_buf; /* buffer for distances and literals/lengths */ 233 #endif 234 235 uInt lit_bufsize; 236 /* Size of match buffer for literals/lengths. There are 4 reasons for 237 * limiting lit_bufsize to 64K: 238 * - frequencies can be kept in 16 bit counters 239 * - if compression is not successful for the first block, all input 240 * data is still in the window so we can still emit a stored block even 241 * when input comes from standard input. (This can also be done for 242 * all blocks if lit_bufsize is not greater than 32K.) 243 * - if compression is not successful for a file smaller than 64K, we can 244 * even emit a stored file instead of a stored block (saving 5 bytes). 245 * This is applicable only for zip (not gzip or zlib). 246 * - creating new Huffman trees less frequently may not provide fast 247 * adaptation to changes in the input data statistics. (Take for 248 * example a binary file with poorly compressible code followed by 249 * a highly compressible string table.) Smaller buffer sizes give 250 * fast adaptation but have of course the overhead of transmitting 251 * trees more frequently. 252 * - I can't count above 4 253 */ 254 255 uInt sym_next; /* running index in symbol buffer */ 256 uInt sym_end; /* symbol table full when sym_next reaches this */ 257 258 ulg opt_len; /* bit length of current block with optimal trees */ 259 ulg static_len; /* bit length of current block with static trees */ 260 uInt matches; /* number of string matches in current block */ 261 uInt insert; /* bytes at end of window left to insert */ 262 263 #ifdef ZLIB_DEBUG 264 ulg compressed_len; /* total bit length of compressed file mod 2^32 */ 265 ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ 266 #endif 267 268 ush bi_buf; 269 /* Output buffer. bits are inserted starting at the bottom (least 270 * significant bits). 271 */ 272 int bi_valid; 273 /* Number of valid bits in bi_buf. All bits above the last valid bit 274 * are always zero. 275 */ 276 277 ulg high_water; 278 /* High water mark offset in window for initialized bytes -- bytes above 279 * this are set to zero in order to avoid memory check warnings when 280 * longest match routines access bytes past the input. This is then 281 * updated to the new high water mark. 282 */ 283 284 } FAR deflate_state; 285 286 /* Output a byte on the stream. 287 * IN assertion: there is enough room in pending_buf. 288 */ 289 #define put_byte(s, c) {s->pending_buf[s->pending++] = (Bytef)(c);} 290 291 292 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) 293 /* Minimum amount of lookahead, except at the end of the input file. 294 * See deflate.c for comments about the MIN_MATCH+1. 295 */ 296 297 #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD) 298 /* In order to simplify the code, particularly on 16 bit machines, match 299 * distances are limited to MAX_DIST instead of WSIZE. 300 */ 301 302 #define WIN_INIT MAX_MATCH 303 /* Number of bytes after end of data in window to initialize in order to avoid 304 memory checker errors from longest match routines */ 305 306 /* in trees.c */ 307 void ZLIB_INTERNAL _tr_init(deflate_state *s); 308 int ZLIB_INTERNAL _tr_tally(deflate_state *s, unsigned dist, unsigned lc); 309 void ZLIB_INTERNAL _tr_flush_block(deflate_state *s, charf *buf, 310 ulg stored_len, int last); 311 void ZLIB_INTERNAL _tr_flush_bits(deflate_state *s); 312 void ZLIB_INTERNAL _tr_align(deflate_state *s); 313 void ZLIB_INTERNAL _tr_stored_block(deflate_state *s, charf *buf, 314 ulg stored_len, int last); 315 316 #define d_code(dist) \ 317 ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) 318 /* Mapping from a distance to a distance code. dist is the distance - 1 and 319 * must not have side effects. _dist_code[256] and _dist_code[257] are never 320 * used. 321 */ 322 323 #ifndef ZLIB_DEBUG 324 /* Inline versions of _tr_tally for speed: */ 325 326 #if defined(GEN_TREES_H) || !defined(STDC) 327 extern uch ZLIB_INTERNAL _length_code[]; 328 extern uch ZLIB_INTERNAL _dist_code[]; 329 #else 330 extern const uch ZLIB_INTERNAL _length_code[]; 331 extern const uch ZLIB_INTERNAL _dist_code[]; 332 #endif 333 334 #ifdef LIT_MEM 335 # define _tr_tally_lit(s, c, flush) \ 336 { uch cc = (c); \ 337 s->d_buf[s->sym_next] = 0; \ 338 s->l_buf[s->sym_next++] = cc; \ 339 s->dyn_ltree[cc].Freq++; \ 340 flush = (s->sym_next == s->sym_end); \ 341 } 342 # define _tr_tally_dist(s, distance, length, flush) \ 343 { uch len = (uch)(length); \ 344 ush dist = (ush)(distance); \ 345 s->d_buf[s->sym_next] = dist; \ 346 s->l_buf[s->sym_next++] = len; \ 347 dist--; \ 348 s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ 349 s->dyn_dtree[d_code(dist)].Freq++; \ 350 flush = (s->sym_next == s->sym_end); \ 351 } 352 #else 353 # define _tr_tally_lit(s, c, flush) \ 354 { uch cc = (c); \ 355 s->sym_buf[s->sym_next++] = 0; \ 356 s->sym_buf[s->sym_next++] = 0; \ 357 s->sym_buf[s->sym_next++] = cc; \ 358 s->dyn_ltree[cc].Freq++; \ 359 flush = (s->sym_next == s->sym_end); \ 360 } 361 # define _tr_tally_dist(s, distance, length, flush) \ 362 { uch len = (uch)(length); \ 363 ush dist = (ush)(distance); \ 364 s->sym_buf[s->sym_next++] = (uch)dist; \ 365 s->sym_buf[s->sym_next++] = (uch)(dist >> 8); \ 366 s->sym_buf[s->sym_next++] = len; \ 367 dist--; \ 368 s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ 369 s->dyn_dtree[d_code(dist)].Freq++; \ 370 flush = (s->sym_next == s->sym_end); \ 371 } 372 #endif 373 #else 374 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c) 375 # define _tr_tally_dist(s, distance, length, flush) \ 376 flush = _tr_tally(s, distance, length) 377 #endif 378 379 #endif /* DEFLATE_H */ 380