Thu, 22 Jan 2015 13:21:57 +0100
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
michael@0 | 1 | /* deflate.h -- internal compression state |
michael@0 | 2 | * Copyright (C) 1995-2010 Jean-loup Gailly |
michael@0 | 3 | * For conditions of distribution and use, see copyright notice in zlib.h |
michael@0 | 4 | */ |
michael@0 | 5 | |
michael@0 | 6 | /* WARNING: this file should *not* be used by applications. It is |
michael@0 | 7 | part of the implementation of the compression library and is |
michael@0 | 8 | subject to change. Applications should only use zlib.h. |
michael@0 | 9 | */ |
michael@0 | 10 | |
michael@0 | 11 | /* @(#) $Id$ */ |
michael@0 | 12 | |
michael@0 | 13 | #ifndef DEFLATE_H |
michael@0 | 14 | #define DEFLATE_H |
michael@0 | 15 | |
michael@0 | 16 | #include "zutil.h" |
michael@0 | 17 | |
michael@0 | 18 | /* define NO_GZIP when compiling if you want to disable gzip header and |
michael@0 | 19 | trailer creation by deflate(). NO_GZIP would be used to avoid linking in |
michael@0 | 20 | the crc code when it is not needed. For shared libraries, gzip encoding |
michael@0 | 21 | should be left enabled. */ |
michael@0 | 22 | #ifndef NO_GZIP |
michael@0 | 23 | # define GZIP |
michael@0 | 24 | #endif |
michael@0 | 25 | |
michael@0 | 26 | /* =========================================================================== |
michael@0 | 27 | * Internal compression state. |
michael@0 | 28 | */ |
michael@0 | 29 | |
michael@0 | 30 | #define LENGTH_CODES 29 |
michael@0 | 31 | /* number of length codes, not counting the special END_BLOCK code */ |
michael@0 | 32 | |
michael@0 | 33 | #define LITERALS 256 |
michael@0 | 34 | /* number of literal bytes 0..255 */ |
michael@0 | 35 | |
michael@0 | 36 | #define L_CODES (LITERALS+1+LENGTH_CODES) |
michael@0 | 37 | /* number of Literal or Length codes, including the END_BLOCK code */ |
michael@0 | 38 | |
michael@0 | 39 | #define D_CODES 30 |
michael@0 | 40 | /* number of distance codes */ |
michael@0 | 41 | |
michael@0 | 42 | #define BL_CODES 19 |
michael@0 | 43 | /* number of codes used to transfer the bit lengths */ |
michael@0 | 44 | |
michael@0 | 45 | #define HEAP_SIZE (2*L_CODES+1) |
michael@0 | 46 | /* maximum heap size */ |
michael@0 | 47 | |
michael@0 | 48 | #define MAX_BITS 15 |
michael@0 | 49 | /* All codes must not exceed MAX_BITS bits */ |
michael@0 | 50 | |
michael@0 | 51 | #define INIT_STATE 42 |
michael@0 | 52 | #define EXTRA_STATE 69 |
michael@0 | 53 | #define NAME_STATE 73 |
michael@0 | 54 | #define COMMENT_STATE 91 |
michael@0 | 55 | #define HCRC_STATE 103 |
michael@0 | 56 | #define BUSY_STATE 113 |
michael@0 | 57 | #define FINISH_STATE 666 |
michael@0 | 58 | /* Stream status */ |
michael@0 | 59 | |
michael@0 | 60 | |
michael@0 | 61 | /* Data structure describing a single value and its code string. */ |
michael@0 | 62 | typedef struct ct_data_s { |
michael@0 | 63 | union { |
michael@0 | 64 | ush freq; /* frequency count */ |
michael@0 | 65 | ush code; /* bit string */ |
michael@0 | 66 | } fc; |
michael@0 | 67 | union { |
michael@0 | 68 | ush dad; /* father node in Huffman tree */ |
michael@0 | 69 | ush len; /* length of bit string */ |
michael@0 | 70 | } dl; |
michael@0 | 71 | } FAR ct_data; |
michael@0 | 72 | |
michael@0 | 73 | #define Freq fc.freq |
michael@0 | 74 | #define Code fc.code |
michael@0 | 75 | #define Dad dl.dad |
michael@0 | 76 | #define Len dl.len |
michael@0 | 77 | |
michael@0 | 78 | typedef struct static_tree_desc_s static_tree_desc; |
michael@0 | 79 | |
michael@0 | 80 | typedef struct tree_desc_s { |
michael@0 | 81 | ct_data *dyn_tree; /* the dynamic tree */ |
michael@0 | 82 | int max_code; /* largest code with non zero frequency */ |
michael@0 | 83 | static_tree_desc *stat_desc; /* the corresponding static tree */ |
michael@0 | 84 | } FAR tree_desc; |
michael@0 | 85 | |
michael@0 | 86 | typedef ush Pos; |
michael@0 | 87 | typedef Pos FAR Posf; |
michael@0 | 88 | typedef unsigned IPos; |
michael@0 | 89 | |
michael@0 | 90 | /* A Pos is an index in the character window. We use short instead of int to |
michael@0 | 91 | * save space in the various tables. IPos is used only for parameter passing. |
michael@0 | 92 | */ |
michael@0 | 93 | |
michael@0 | 94 | typedef struct internal_state { |
michael@0 | 95 | z_streamp strm; /* pointer back to this zlib stream */ |
michael@0 | 96 | int status; /* as the name implies */ |
michael@0 | 97 | Bytef *pending_buf; /* output still pending */ |
michael@0 | 98 | ulg pending_buf_size; /* size of pending_buf */ |
michael@0 | 99 | Bytef *pending_out; /* next pending byte to output to the stream */ |
michael@0 | 100 | uInt pending; /* nb of bytes in the pending buffer */ |
michael@0 | 101 | int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ |
michael@0 | 102 | gz_headerp gzhead; /* gzip header information to write */ |
michael@0 | 103 | uInt gzindex; /* where in extra, name, or comment */ |
michael@0 | 104 | Byte method; /* STORED (for zip only) or DEFLATED */ |
michael@0 | 105 | int last_flush; /* value of flush param for previous deflate call */ |
michael@0 | 106 | |
michael@0 | 107 | /* used by deflate.c: */ |
michael@0 | 108 | |
michael@0 | 109 | uInt w_size; /* LZ77 window size (32K by default) */ |
michael@0 | 110 | uInt w_bits; /* log2(w_size) (8..16) */ |
michael@0 | 111 | uInt w_mask; /* w_size - 1 */ |
michael@0 | 112 | |
michael@0 | 113 | Bytef *window; |
michael@0 | 114 | /* Sliding window. Input bytes are read into the second half of the window, |
michael@0 | 115 | * and move to the first half later to keep a dictionary of at least wSize |
michael@0 | 116 | * bytes. With this organization, matches are limited to a distance of |
michael@0 | 117 | * wSize-MAX_MATCH bytes, but this ensures that IO is always |
michael@0 | 118 | * performed with a length multiple of the block size. Also, it limits |
michael@0 | 119 | * the window size to 64K, which is quite useful on MSDOS. |
michael@0 | 120 | * To do: use the user input buffer as sliding window. |
michael@0 | 121 | */ |
michael@0 | 122 | |
michael@0 | 123 | ulg window_size; |
michael@0 | 124 | /* Actual size of window: 2*wSize, except when the user input buffer |
michael@0 | 125 | * is directly used as sliding window. |
michael@0 | 126 | */ |
michael@0 | 127 | |
michael@0 | 128 | Posf *prev; |
michael@0 | 129 | /* Link to older string with same hash index. To limit the size of this |
michael@0 | 130 | * array to 64K, this link is maintained only for the last 32K strings. |
michael@0 | 131 | * An index in this array is thus a window index modulo 32K. |
michael@0 | 132 | */ |
michael@0 | 133 | |
michael@0 | 134 | Posf *head; /* Heads of the hash chains or NIL. */ |
michael@0 | 135 | |
michael@0 | 136 | uInt ins_h; /* hash index of string to be inserted */ |
michael@0 | 137 | uInt hash_size; /* number of elements in hash table */ |
michael@0 | 138 | uInt hash_bits; /* log2(hash_size) */ |
michael@0 | 139 | uInt hash_mask; /* hash_size-1 */ |
michael@0 | 140 | |
michael@0 | 141 | uInt hash_shift; |
michael@0 | 142 | /* Number of bits by which ins_h must be shifted at each input |
michael@0 | 143 | * step. It must be such that after MIN_MATCH steps, the oldest |
michael@0 | 144 | * byte no longer takes part in the hash key, that is: |
michael@0 | 145 | * hash_shift * MIN_MATCH >= hash_bits |
michael@0 | 146 | */ |
michael@0 | 147 | |
michael@0 | 148 | long block_start; |
michael@0 | 149 | /* Window position at the beginning of the current output block. Gets |
michael@0 | 150 | * negative when the window is moved backwards. |
michael@0 | 151 | */ |
michael@0 | 152 | |
michael@0 | 153 | uInt match_length; /* length of best match */ |
michael@0 | 154 | IPos prev_match; /* previous match */ |
michael@0 | 155 | int match_available; /* set if previous match exists */ |
michael@0 | 156 | uInt strstart; /* start of string to insert */ |
michael@0 | 157 | uInt match_start; /* start of matching string */ |
michael@0 | 158 | uInt lookahead; /* number of valid bytes ahead in window */ |
michael@0 | 159 | |
michael@0 | 160 | uInt prev_length; |
michael@0 | 161 | /* Length of the best match at previous step. Matches not greater than this |
michael@0 | 162 | * are discarded. This is used in the lazy match evaluation. |
michael@0 | 163 | */ |
michael@0 | 164 | |
michael@0 | 165 | uInt max_chain_length; |
michael@0 | 166 | /* To speed up deflation, hash chains are never searched beyond this |
michael@0 | 167 | * length. A higher limit improves compression ratio but degrades the |
michael@0 | 168 | * speed. |
michael@0 | 169 | */ |
michael@0 | 170 | |
michael@0 | 171 | uInt max_lazy_match; |
michael@0 | 172 | /* Attempt to find a better match only when the current match is strictly |
michael@0 | 173 | * smaller than this value. This mechanism is used only for compression |
michael@0 | 174 | * levels >= 4. |
michael@0 | 175 | */ |
michael@0 | 176 | # define max_insert_length max_lazy_match |
michael@0 | 177 | /* Insert new strings in the hash table only if the match length is not |
michael@0 | 178 | * greater than this length. This saves time but degrades compression. |
michael@0 | 179 | * max_insert_length is used only for compression levels <= 3. |
michael@0 | 180 | */ |
michael@0 | 181 | |
michael@0 | 182 | int level; /* compression level (1..9) */ |
michael@0 | 183 | int strategy; /* favor or force Huffman coding*/ |
michael@0 | 184 | |
michael@0 | 185 | uInt good_match; |
michael@0 | 186 | /* Use a faster search when the previous match is longer than this */ |
michael@0 | 187 | |
michael@0 | 188 | int nice_match; /* Stop searching when current match exceeds this */ |
michael@0 | 189 | |
michael@0 | 190 | /* used by trees.c: */ |
michael@0 | 191 | /* Didn't use ct_data typedef below to supress compiler warning */ |
michael@0 | 192 | struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */ |
michael@0 | 193 | struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */ |
michael@0 | 194 | struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */ |
michael@0 | 195 | |
michael@0 | 196 | struct tree_desc_s l_desc; /* desc. for literal tree */ |
michael@0 | 197 | struct tree_desc_s d_desc; /* desc. for distance tree */ |
michael@0 | 198 | struct tree_desc_s bl_desc; /* desc. for bit length tree */ |
michael@0 | 199 | |
michael@0 | 200 | ush bl_count[MAX_BITS+1]; |
michael@0 | 201 | /* number of codes at each bit length for an optimal tree */ |
michael@0 | 202 | |
michael@0 | 203 | int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */ |
michael@0 | 204 | int heap_len; /* number of elements in the heap */ |
michael@0 | 205 | int heap_max; /* element of largest frequency */ |
michael@0 | 206 | /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. |
michael@0 | 207 | * The same heap array is used to build all trees. |
michael@0 | 208 | */ |
michael@0 | 209 | |
michael@0 | 210 | uch depth[2*L_CODES+1]; |
michael@0 | 211 | /* Depth of each subtree used as tie breaker for trees of equal frequency |
michael@0 | 212 | */ |
michael@0 | 213 | |
michael@0 | 214 | uchf *l_buf; /* buffer for literals or lengths */ |
michael@0 | 215 | |
michael@0 | 216 | uInt lit_bufsize; |
michael@0 | 217 | /* Size of match buffer for literals/lengths. There are 4 reasons for |
michael@0 | 218 | * limiting lit_bufsize to 64K: |
michael@0 | 219 | * - frequencies can be kept in 16 bit counters |
michael@0 | 220 | * - if compression is not successful for the first block, all input |
michael@0 | 221 | * data is still in the window so we can still emit a stored block even |
michael@0 | 222 | * when input comes from standard input. (This can also be done for |
michael@0 | 223 | * all blocks if lit_bufsize is not greater than 32K.) |
michael@0 | 224 | * - if compression is not successful for a file smaller than 64K, we can |
michael@0 | 225 | * even emit a stored file instead of a stored block (saving 5 bytes). |
michael@0 | 226 | * This is applicable only for zip (not gzip or zlib). |
michael@0 | 227 | * - creating new Huffman trees less frequently may not provide fast |
michael@0 | 228 | * adaptation to changes in the input data statistics. (Take for |
michael@0 | 229 | * example a binary file with poorly compressible code followed by |
michael@0 | 230 | * a highly compressible string table.) Smaller buffer sizes give |
michael@0 | 231 | * fast adaptation but have of course the overhead of transmitting |
michael@0 | 232 | * trees more frequently. |
michael@0 | 233 | * - I can't count above 4 |
michael@0 | 234 | */ |
michael@0 | 235 | |
michael@0 | 236 | uInt last_lit; /* running index in l_buf */ |
michael@0 | 237 | |
michael@0 | 238 | ushf *d_buf; |
michael@0 | 239 | /* Buffer for distances. To simplify the code, d_buf and l_buf have |
michael@0 | 240 | * the same number of elements. To use different lengths, an extra flag |
michael@0 | 241 | * array would be necessary. |
michael@0 | 242 | */ |
michael@0 | 243 | |
michael@0 | 244 | ulg opt_len; /* bit length of current block with optimal trees */ |
michael@0 | 245 | ulg static_len; /* bit length of current block with static trees */ |
michael@0 | 246 | uInt matches; /* number of string matches in current block */ |
michael@0 | 247 | int last_eob_len; /* bit length of EOB code for last block */ |
michael@0 | 248 | |
michael@0 | 249 | #ifdef DEBUG |
michael@0 | 250 | ulg compressed_len; /* total bit length of compressed file mod 2^32 */ |
michael@0 | 251 | ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ |
michael@0 | 252 | #endif |
michael@0 | 253 | |
michael@0 | 254 | ush bi_buf; |
michael@0 | 255 | /* Output buffer. bits are inserted starting at the bottom (least |
michael@0 | 256 | * significant bits). |
michael@0 | 257 | */ |
michael@0 | 258 | int bi_valid; |
michael@0 | 259 | /* Number of valid bits in bi_buf. All bits above the last valid bit |
michael@0 | 260 | * are always zero. |
michael@0 | 261 | */ |
michael@0 | 262 | |
michael@0 | 263 | ulg high_water; |
michael@0 | 264 | /* High water mark offset in window for initialized bytes -- bytes above |
michael@0 | 265 | * this are set to zero in order to avoid memory check warnings when |
michael@0 | 266 | * longest match routines access bytes past the input. This is then |
michael@0 | 267 | * updated to the new high water mark. |
michael@0 | 268 | */ |
michael@0 | 269 | |
michael@0 | 270 | } FAR deflate_state; |
michael@0 | 271 | |
michael@0 | 272 | /* Output a byte on the stream. |
michael@0 | 273 | * IN assertion: there is enough room in pending_buf. |
michael@0 | 274 | */ |
michael@0 | 275 | #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);} |
michael@0 | 276 | |
michael@0 | 277 | |
michael@0 | 278 | #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) |
michael@0 | 279 | /* Minimum amount of lookahead, except at the end of the input file. |
michael@0 | 280 | * See deflate.c for comments about the MIN_MATCH+1. |
michael@0 | 281 | */ |
michael@0 | 282 | |
michael@0 | 283 | #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD) |
michael@0 | 284 | /* In order to simplify the code, particularly on 16 bit machines, match |
michael@0 | 285 | * distances are limited to MAX_DIST instead of WSIZE. |
michael@0 | 286 | */ |
michael@0 | 287 | |
michael@0 | 288 | #define WIN_INIT MAX_MATCH |
michael@0 | 289 | /* Number of bytes after end of data in window to initialize in order to avoid |
michael@0 | 290 | memory checker errors from longest match routines */ |
michael@0 | 291 | |
michael@0 | 292 | /* in trees.c */ |
michael@0 | 293 | void ZLIB_INTERNAL _tr_init OF((deflate_state *s)); |
michael@0 | 294 | int ZLIB_INTERNAL _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc)); |
michael@0 | 295 | void ZLIB_INTERNAL _tr_flush_block OF((deflate_state *s, charf *buf, |
michael@0 | 296 | ulg stored_len, int last)); |
michael@0 | 297 | void ZLIB_INTERNAL _tr_align OF((deflate_state *s)); |
michael@0 | 298 | void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf, |
michael@0 | 299 | ulg stored_len, int last)); |
michael@0 | 300 | |
michael@0 | 301 | #define d_code(dist) \ |
michael@0 | 302 | ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) |
michael@0 | 303 | /* Mapping from a distance to a distance code. dist is the distance - 1 and |
michael@0 | 304 | * must not have side effects. _dist_code[256] and _dist_code[257] are never |
michael@0 | 305 | * used. |
michael@0 | 306 | */ |
michael@0 | 307 | |
michael@0 | 308 | #ifndef DEBUG |
michael@0 | 309 | /* Inline versions of _tr_tally for speed: */ |
michael@0 | 310 | |
michael@0 | 311 | #if defined(GEN_TREES_H) || !defined(STDC) |
michael@0 | 312 | extern uch ZLIB_INTERNAL _length_code[]; |
michael@0 | 313 | extern uch ZLIB_INTERNAL _dist_code[]; |
michael@0 | 314 | #else |
michael@0 | 315 | extern const uch ZLIB_INTERNAL _length_code[]; |
michael@0 | 316 | extern const uch ZLIB_INTERNAL _dist_code[]; |
michael@0 | 317 | #endif |
michael@0 | 318 | |
michael@0 | 319 | # define _tr_tally_lit(s, c, flush) \ |
michael@0 | 320 | { uch cc = (c); \ |
michael@0 | 321 | s->d_buf[s->last_lit] = 0; \ |
michael@0 | 322 | s->l_buf[s->last_lit++] = cc; \ |
michael@0 | 323 | s->dyn_ltree[cc].Freq++; \ |
michael@0 | 324 | flush = (s->last_lit == s->lit_bufsize-1); \ |
michael@0 | 325 | } |
michael@0 | 326 | # define _tr_tally_dist(s, distance, length, flush) \ |
michael@0 | 327 | { uch len = (length); \ |
michael@0 | 328 | ush dist = (distance); \ |
michael@0 | 329 | s->d_buf[s->last_lit] = dist; \ |
michael@0 | 330 | s->l_buf[s->last_lit++] = len; \ |
michael@0 | 331 | dist--; \ |
michael@0 | 332 | s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ |
michael@0 | 333 | s->dyn_dtree[d_code(dist)].Freq++; \ |
michael@0 | 334 | flush = (s->last_lit == s->lit_bufsize-1); \ |
michael@0 | 335 | } |
michael@0 | 336 | #else |
michael@0 | 337 | # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c) |
michael@0 | 338 | # define _tr_tally_dist(s, distance, length, flush) \ |
michael@0 | 339 | flush = _tr_tally(s, distance, length) |
michael@0 | 340 | #endif |
michael@0 | 341 | |
michael@0 | 342 | #endif /* DEFLATE_H */ |