00001 /* deflate.h -- internal compression state 00002 * Copyright (C) 1995-2004 Jean-loup Gailly 00003 * For conditions of distribution and use, see copyright notice in zlib.h 00004 */ 00005 00006 /* WARNING: this file should *not* be used by applications. It is 00007 part of the implementation of the compression library and is 00008 subject to change. Applications should only use zlib.h. 00009 */ 00010 00011 /* @(#) $Id$ */ 00012 00013 #ifndef DEFLATE_H 00014 #define DEFLATE_H 00015 00016 #include "zutil.h" 00017 00018 /* define NO_GZIP when compiling if you want to disable gzip header and 00019 trailer creation by deflate(). NO_GZIP would be used to avoid linking in 00020 the crc code when it is not needed. For shared libraries, gzip encoding 00021 should be left enabled. */ 00022 #ifndef NO_GZIP 00023 # define GZIP 00024 #endif 00025 00026 /* =========================================================================== 00027 * Internal compression state. 00028 */ 00029 00030 #define LENGTH_CODES 29 00031 /* number of length codes, not counting the special END_BLOCK code */ 00032 00033 #define LITERALS 256 00034 /* number of literal bytes 0..255 */ 00035 00036 #define L_CODES (LITERALS+1+LENGTH_CODES) 00037 /* number of Literal or Length codes, including the END_BLOCK code */ 00038 00039 #define D_CODES 30 00040 /* number of distance codes */ 00041 00042 #define BL_CODES 19 00043 /* number of codes used to transfer the bit lengths */ 00044 00045 #define HEAP_SIZE (2*L_CODES+1) 00046 /* maximum heap size */ 00047 00048 #define MAX_BITS 15 00049 /* All codes must not exceed MAX_BITS bits */ 00050 00051 #define INIT_STATE 42 00052 #define EXTRA_STATE 69 00053 #define NAME_STATE 73 00054 #define COMMENT_STATE 91 00055 #define HCRC_STATE 103 00056 #define BUSY_STATE 113 00057 #define FINISH_STATE 666 00058 /* Stream status */ 00059 00060 00061 /* Data structure describing a single value and its code string. */ 00062 typedef struct ct_data_s { 00063 union { 00064 ush freq; /* frequency count */ 00065 ush code; /* bit string */ 00066 } fc; 00067 union { 00068 ush dad; /* father node in Huffman tree */ 00069 ush len; /* length of bit string */ 00070 } dl; 00071 } FAR ct_data; 00072 00073 #define Freq fc.freq 00074 #define Code fc.code 00075 #define Dad dl.dad 00076 #define Len dl.len 00077 00078 typedef struct static_tree_desc_s static_tree_desc; 00079 00080 typedef struct tree_desc_s { 00081 ct_data *dyn_tree; /* the dynamic tree */ 00082 int max_code; /* largest code with non zero frequency */ 00083 static_tree_desc *stat_desc; /* the corresponding static tree */ 00084 } FAR tree_desc; 00085 00086 typedef ush Pos; 00087 typedef Pos FAR Posf; 00088 typedef unsigned IPos; 00089 00090 /* A Pos is an index in the character window. We use short instead of int to 00091 * save space in the various tables. IPos is used only for parameter passing. 00092 */ 00093 00094 typedef struct internal_state { 00095 z_streamp strm; /* pointer back to this zlib stream */ 00096 int status; /* as the name implies */ 00097 Bytef *pending_buf; /* output still pending */ 00098 ulg pending_buf_size; /* size of pending_buf */ 00099 Bytef *pending_out; /* next pending byte to output to the stream */ 00100 uInt pending; /* nb of bytes in the pending buffer */ 00101 int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ 00102 gz_headerp gzhead; /* gzip header information to write */ 00103 uInt gzindex; /* where in extra, name, or comment */ 00104 Byte method; /* STORED (for zip only) or DEFLATED */ 00105 int last_flush; /* value of flush param for previous deflate call */ 00106 00107 /* used by deflate.c: */ 00108 00109 uInt w_size; /* LZ77 window size (32K by default) */ 00110 uInt w_bits; /* log2(w_size) (8..16) */ 00111 uInt w_mask; /* w_size - 1 */ 00112 00113 Bytef *window; 00114 /* Sliding window. Input bytes are read into the second half of the window, 00115 * and move to the first half later to keep a dictionary of at least wSize 00116 * bytes. With this organization, matches are limited to a distance of 00117 * wSize-MAX_MATCH bytes, but this ensures that IO is always 00118 * performed with a length multiple of the block size. Also, it limits 00119 * the window size to 64K, which is quite useful on MSDOS. 00120 * To do: use the user input buffer as sliding window. 00121 */ 00122 00123 ulg window_size; 00124 /* Actual size of window: 2*wSize, except when the user input buffer 00125 * is directly used as sliding window. 00126 */ 00127 00128 Posf *prev; 00129 /* Link to older string with same hash index. To limit the size of this 00130 * array to 64K, this link is maintained only for the last 32K strings. 00131 * An index in this array is thus a window index modulo 32K. 00132 */ 00133 00134 Posf *head; /* Heads of the hash chains or NIL. */ 00135 00136 uInt ins_h; /* hash index of string to be inserted */ 00137 uInt hash_size; /* number of elements in hash table */ 00138 uInt hash_bits; /* log2(hash_size) */ 00139 uInt hash_mask; /* hash_size-1 */ 00140 00141 uInt hash_shift; 00142 /* Number of bits by which ins_h must be shifted at each input 00143 * step. It must be such that after MIN_MATCH steps, the oldest 00144 * byte no longer takes part in the hash key, that is: 00145 * hash_shift * MIN_MATCH >= hash_bits 00146 */ 00147 00148 long block_start; 00149 /* Window position at the beginning of the current output block. Gets 00150 * negative when the window is moved backwards. 00151 */ 00152 00153 uInt match_length; /* length of best match */ 00154 IPos prev_match; /* previous match */ 00155 int match_available; /* set if previous match exists */ 00156 uInt strstart; /* start of string to insert */ 00157 uInt match_start; /* start of matching string */ 00158 uInt lookahead; /* number of valid bytes ahead in window */ 00159 00160 uInt prev_length; 00161 /* Length of the best match at previous step. Matches not greater than this 00162 * are discarded. This is used in the lazy match evaluation. 00163 */ 00164 00165 uInt max_chain_length; 00166 /* To speed up deflation, hash chains are never searched beyond this 00167 * length. A higher limit improves compression ratio but degrades the 00168 * speed. 00169 */ 00170 00171 uInt max_lazy_match; 00172 /* Attempt to find a better match only when the current match is strictly 00173 * smaller than this value. This mechanism is used only for compression 00174 * levels >= 4. 00175 */ 00176 # define max_insert_length max_lazy_match 00177 /* Insert new strings in the hash table only if the match length is not 00178 * greater than this length. This saves time but degrades compression. 00179 * max_insert_length is used only for compression levels <= 3. 00180 */ 00181 00182 int level; /* compression level (1..9) */ 00183 int strategy; /* favor or force Huffman coding*/ 00184 00185 uInt good_match; 00186 /* Use a faster search when the previous match is longer than this */ 00187 00188 int nice_match; /* Stop searching when current match exceeds this */ 00189 00190 /* used by trees.c: */ 00191 /* Didn't use ct_data typedef below to supress compiler warning */ 00192 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */ 00193 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */ 00194 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */ 00195 00196 struct tree_desc_s l_desc; /* desc. for literal tree */ 00197 struct tree_desc_s d_desc; /* desc. for distance tree */ 00198 struct tree_desc_s bl_desc; /* desc. for bit length tree */ 00199 00200 ush bl_count[MAX_BITS+1]; 00201 /* number of codes at each bit length for an optimal tree */ 00202 00203 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */ 00204 int heap_len; /* number of elements in the heap */ 00205 int heap_max; /* element of largest frequency */ 00206 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. 00207 * The same heap array is used to build all trees. 00208 */ 00209 00210 uch depth[2*L_CODES+1]; 00211 /* Depth of each subtree used as tie breaker for trees of equal frequency 00212 */ 00213 00214 uchf *l_buf; /* buffer for literals or lengths */ 00215 00216 uInt lit_bufsize; 00217 /* Size of match buffer for literals/lengths. There are 4 reasons for 00218 * limiting lit_bufsize to 64K: 00219 * - frequencies can be kept in 16 bit counters 00220 * - if compression is not successful for the first block, all input 00221 * data is still in the window so we can still emit a stored block even 00222 * when input comes from standard input. (This can also be done for 00223 * all blocks if lit_bufsize is not greater than 32K.) 00224 * - if compression is not successful for a file smaller than 64K, we can 00225 * even emit a stored file instead of a stored block (saving 5 bytes). 00226 * This is applicable only for zip (not gzip or zlib). 00227 * - creating new Huffman trees less frequently may not provide fast 00228 * adaptation to changes in the input data statistics. (Take for 00229 * example a binary file with poorly compressible code followed by 00230 * a highly compressible string table.) Smaller buffer sizes give 00231 * fast adaptation but have of course the overhead of transmitting 00232 * trees more frequently. 00233 * - I can't count above 4 00234 */ 00235 00236 uInt last_lit; /* running index in l_buf */ 00237 00238 ushf *d_buf; 00239 /* Buffer for distances. To simplify the code, d_buf and l_buf have 00240 * the same number of elements. To use different lengths, an extra flag 00241 * array would be necessary. 00242 */ 00243 00244 ulg opt_len; /* bit length of current block with optimal trees */ 00245 ulg static_len; /* bit length of current block with static trees */ 00246 uInt matches; /* number of string matches in current block */ 00247 int last_eob_len; /* bit length of EOB code for last block */ 00248 00249 #ifdef DEBUG 00250 ulg compressed_len; /* total bit length of compressed file mod 2^32 */ 00251 ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ 00252 #endif 00253 00254 ush bi_buf; 00255 /* Output buffer. bits are inserted starting at the bottom (least 00256 * significant bits). 00257 */ 00258 int bi_valid; 00259 /* Number of valid bits in bi_buf. All bits above the last valid bit 00260 * are always zero. 00261 */ 00262 00263 } FAR deflate_state; 00264 00265 /* Output a byte on the stream. 00266 * IN assertion: there is enough room in pending_buf. 00267 */ 00268 #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);} 00269 00270 00271 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) 00272 /* Minimum amount of lookahead, except at the end of the input file. 00273 * See deflate.c for comments about the MIN_MATCH+1. 00274 */ 00275 00276 #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD) 00277 /* In order to simplify the code, particularly on 16 bit machines, match 00278 * distances are limited to MAX_DIST instead of WSIZE. 00279 */ 00280 00281 /* in trees.c */ 00282 void _tr_init OF((deflate_state *s)); 00283 int _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc)); 00284 void _tr_flush_block OF((deflate_state *s, charf *buf, ulg stored_len, 00285 int eof)); 00286 void _tr_align OF((deflate_state *s)); 00287 void _tr_stored_block OF((deflate_state *s, charf *buf, ulg stored_len, 00288 int eof)); 00289 00290 #define d_code(dist) \ 00291 ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) 00292 /* Mapping from a distance to a distance code. dist is the distance - 1 and 00293 * must not have side effects. _dist_code[256] and _dist_code[257] are never 00294 * used. 00295 */ 00296 00297 #ifndef DEBUG 00298 /* Inline versions of _tr_tally for speed: */ 00299 00300 #if defined(GEN_TREES_H) || !defined(STDC) 00301 extern uch _length_code[]; 00302 extern uch _dist_code[]; 00303 #else 00304 extern const uch _length_code[]; 00305 extern const uch _dist_code[]; 00306 #endif 00307 00308 # define _tr_tally_lit(s, c, flush) \ 00309 { uch cc = (c); \ 00310 s->d_buf[s->last_lit] = 0; \ 00311 s->l_buf[s->last_lit++] = cc; \ 00312 s->dyn_ltree[cc].Freq++; \ 00313 flush = (s->last_lit == s->lit_bufsize-1); \ 00314 } 00315 # define _tr_tally_dist(s, distance, length, flush) \ 00316 { uch len = (length); \ 00317 ush dist = (distance); \ 00318 s->d_buf[s->last_lit] = dist; \ 00319 s->l_buf[s->last_lit++] = len; \ 00320 dist--; \ 00321 s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ 00322 s->dyn_dtree[d_code(dist)].Freq++; \ 00323 flush = (s->last_lit == s->lit_bufsize-1); \ 00324 } 00325 #else 00326 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c) 00327 # define _tr_tally_dist(s, distance, length, flush) \ 00328 flush = _tr_tally(s, distance, length) 00329 #endif 00330 00331 #endif /* DEFLATE_H */