core/zlib/deflate.h
author David Anderson <dvander@alliedmods.net>
Fri Feb 16 03:47:10 2007 +0000 (2007-02-16)
changeset 501 e209a7296936
parent 159 43c471413f70
permissions -rw-r--r--
removed this potential feature, not useful enough
dvander@159
     1
/* deflate.h -- internal compression state
dvander@159
     2
 * Copyright (C) 1995-2004 Jean-loup Gailly
dvander@159
     3
 * For conditions of distribution and use, see copyright notice in zlib.h
dvander@159
     4
 */
dvander@159
     5
dvander@159
     6
/* WARNING: this file should *not* be used by applications. It is
dvander@159
     7
   part of the implementation of the compression library and is
dvander@159
     8
   subject to change. Applications should only use zlib.h.
dvander@159
     9
 */
dvander@159
    10
dvander@159
    11
/* @(#) $Id$ */
dvander@159
    12
dvander@159
    13
#ifndef DEFLATE_H
dvander@159
    14
#define DEFLATE_H
dvander@159
    15
dvander@159
    16
#include "zutil.h"
dvander@159
    17
dvander@159
    18
/* define NO_GZIP when compiling if you want to disable gzip header and
dvander@159
    19
   trailer creation by deflate().  NO_GZIP would be used to avoid linking in
dvander@159
    20
   the crc code when it is not needed.  For shared libraries, gzip encoding
dvander@159
    21
   should be left enabled. */
dvander@159
    22
#ifndef NO_GZIP
dvander@159
    23
#  define GZIP
dvander@159
    24
#endif
dvander@159
    25
dvander@159
    26
/* ===========================================================================
dvander@159
    27
 * Internal compression state.
dvander@159
    28
 */
dvander@159
    29
dvander@159
    30
#define LENGTH_CODES 29
dvander@159
    31
/* number of length codes, not counting the special END_BLOCK code */
dvander@159
    32
dvander@159
    33
#define LITERALS  256
dvander@159
    34
/* number of literal bytes 0..255 */
dvander@159
    35
dvander@159
    36
#define L_CODES (LITERALS+1+LENGTH_CODES)
dvander@159
    37
/* number of Literal or Length codes, including the END_BLOCK code */
dvander@159
    38
dvander@159
    39
#define D_CODES   30
dvander@159
    40
/* number of distance codes */
dvander@159
    41
dvander@159
    42
#define BL_CODES  19
dvander@159
    43
/* number of codes used to transfer the bit lengths */
dvander@159
    44
dvander@159
    45
#define HEAP_SIZE (2*L_CODES+1)
dvander@159
    46
/* maximum heap size */
dvander@159
    47
dvander@159
    48
#define MAX_BITS 15
dvander@159
    49
/* All codes must not exceed MAX_BITS bits */
dvander@159
    50
dvander@159
    51
#define INIT_STATE    42
dvander@159
    52
#define EXTRA_STATE   69
dvander@159
    53
#define NAME_STATE    73
dvander@159
    54
#define COMMENT_STATE 91
dvander@159
    55
#define HCRC_STATE   103
dvander@159
    56
#define BUSY_STATE   113
dvander@159
    57
#define FINISH_STATE 666
dvander@159
    58
/* Stream status */
dvander@159
    59
dvander@159
    60
dvander@159
    61
/* Data structure describing a single value and its code string. */
dvander@159
    62
typedef struct ct_data_s {
dvander@159
    63
    union {
dvander@159
    64
        ush  freq;       /* frequency count */
dvander@159
    65
        ush  code;       /* bit string */
dvander@159
    66
    } fc;
dvander@159
    67
    union {
dvander@159
    68
        ush  dad;        /* father node in Huffman tree */
dvander@159
    69
        ush  len;        /* length of bit string */
dvander@159
    70
    } dl;
dvander@159
    71
} FAR ct_data;
dvander@159
    72
dvander@159
    73
#define Freq fc.freq
dvander@159
    74
#define Code fc.code
dvander@159
    75
#define Dad  dl.dad
dvander@159
    76
#define Len  dl.len
dvander@159
    77
dvander@159
    78
typedef struct static_tree_desc_s  static_tree_desc;
dvander@159
    79
dvander@159
    80
typedef struct tree_desc_s {
dvander@159
    81
    ct_data *dyn_tree;           /* the dynamic tree */
dvander@159
    82
    int     max_code;            /* largest code with non zero frequency */
dvander@159
    83
    static_tree_desc *stat_desc; /* the corresponding static tree */
dvander@159
    84
} FAR tree_desc;
dvander@159
    85
dvander@159
    86
typedef ush Pos;
dvander@159
    87
typedef Pos FAR Posf;
dvander@159
    88
typedef unsigned IPos;
dvander@159
    89
dvander@159
    90
/* A Pos is an index in the character window. We use short instead of int to
dvander@159
    91
 * save space in the various tables. IPos is used only for parameter passing.
dvander@159
    92
 */
dvander@159
    93
dvander@159
    94
typedef struct internal_state {
dvander@159
    95
    z_streamp strm;      /* pointer back to this zlib stream */
dvander@159
    96
    int   status;        /* as the name implies */
dvander@159
    97
    Bytef *pending_buf;  /* output still pending */
dvander@159
    98
    ulg   pending_buf_size; /* size of pending_buf */
dvander@159
    99
    Bytef *pending_out;  /* next pending byte to output to the stream */
dvander@159
   100
    uInt   pending;      /* nb of bytes in the pending buffer */
dvander@159
   101
    int   wrap;          /* bit 0 true for zlib, bit 1 true for gzip */
dvander@159
   102
    gz_headerp  gzhead;  /* gzip header information to write */
dvander@159
   103
    uInt   gzindex;      /* where in extra, name, or comment */
dvander@159
   104
    Byte  method;        /* STORED (for zip only) or DEFLATED */
dvander@159
   105
    int   last_flush;    /* value of flush param for previous deflate call */
dvander@159
   106
dvander@159
   107
                /* used by deflate.c: */
dvander@159
   108
dvander@159
   109
    uInt  w_size;        /* LZ77 window size (32K by default) */
dvander@159
   110
    uInt  w_bits;        /* log2(w_size)  (8..16) */
dvander@159
   111
    uInt  w_mask;        /* w_size - 1 */
dvander@159
   112
dvander@159
   113
    Bytef *window;
dvander@159
   114
    /* Sliding window. Input bytes are read into the second half of the window,
dvander@159
   115
     * and move to the first half later to keep a dictionary of at least wSize
dvander@159
   116
     * bytes. With this organization, matches are limited to a distance of
dvander@159
   117
     * wSize-MAX_MATCH bytes, but this ensures that IO is always
dvander@159
   118
     * performed with a length multiple of the block size. Also, it limits
dvander@159
   119
     * the window size to 64K, which is quite useful on MSDOS.
dvander@159
   120
     * To do: use the user input buffer as sliding window.
dvander@159
   121
     */
dvander@159
   122
dvander@159
   123
    ulg window_size;
dvander@159
   124
    /* Actual size of window: 2*wSize, except when the user input buffer
dvander@159
   125
     * is directly used as sliding window.
dvander@159
   126
     */
dvander@159
   127
dvander@159
   128
    Posf *prev;
dvander@159
   129
    /* Link to older string with same hash index. To limit the size of this
dvander@159
   130
     * array to 64K, this link is maintained only for the last 32K strings.
dvander@159
   131
     * An index in this array is thus a window index modulo 32K.
dvander@159
   132
     */
dvander@159
   133
dvander@159
   134
    Posf *head; /* Heads of the hash chains or NIL. */
dvander@159
   135
dvander@159
   136
    uInt  ins_h;          /* hash index of string to be inserted */
dvander@159
   137
    uInt  hash_size;      /* number of elements in hash table */
dvander@159
   138
    uInt  hash_bits;      /* log2(hash_size) */
dvander@159
   139
    uInt  hash_mask;      /* hash_size-1 */
dvander@159
   140
dvander@159
   141
    uInt  hash_shift;
dvander@159
   142
    /* Number of bits by which ins_h must be shifted at each input
dvander@159
   143
     * step. It must be such that after MIN_MATCH steps, the oldest
dvander@159
   144
     * byte no longer takes part in the hash key, that is:
dvander@159
   145
     *   hash_shift * MIN_MATCH >= hash_bits
dvander@159
   146
     */
dvander@159
   147
dvander@159
   148
    long block_start;
dvander@159
   149
    /* Window position at the beginning of the current output block. Gets
dvander@159
   150
     * negative when the window is moved backwards.
dvander@159
   151
     */
dvander@159
   152
dvander@159
   153
    uInt match_length;           /* length of best match */
dvander@159
   154
    IPos prev_match;             /* previous match */
dvander@159
   155
    int match_available;         /* set if previous match exists */
dvander@159
   156
    uInt strstart;               /* start of string to insert */
dvander@159
   157
    uInt match_start;            /* start of matching string */
dvander@159
   158
    uInt lookahead;              /* number of valid bytes ahead in window */
dvander@159
   159
dvander@159
   160
    uInt prev_length;
dvander@159
   161
    /* Length of the best match at previous step. Matches not greater than this
dvander@159
   162
     * are discarded. This is used in the lazy match evaluation.
dvander@159
   163
     */
dvander@159
   164
dvander@159
   165
    uInt max_chain_length;
dvander@159
   166
    /* To speed up deflation, hash chains are never searched beyond this
dvander@159
   167
     * length.  A higher limit improves compression ratio but degrades the
dvander@159
   168
     * speed.
dvander@159
   169
     */
dvander@159
   170
dvander@159
   171
    uInt max_lazy_match;
dvander@159
   172
    /* Attempt to find a better match only when the current match is strictly
dvander@159
   173
     * smaller than this value. This mechanism is used only for compression
dvander@159
   174
     * levels >= 4.
dvander@159
   175
     */
dvander@159
   176
#   define max_insert_length  max_lazy_match
dvander@159
   177
    /* Insert new strings in the hash table only if the match length is not
dvander@159
   178
     * greater than this length. This saves time but degrades compression.
dvander@159
   179
     * max_insert_length is used only for compression levels <= 3.
dvander@159
   180
     */
dvander@159
   181
dvander@159
   182
    int level;    /* compression level (1..9) */
dvander@159
   183
    int strategy; /* favor or force Huffman coding*/
dvander@159
   184
dvander@159
   185
    uInt good_match;
dvander@159
   186
    /* Use a faster search when the previous match is longer than this */
dvander@159
   187
dvander@159
   188
    int nice_match; /* Stop searching when current match exceeds this */
dvander@159
   189
dvander@159
   190
                /* used by trees.c: */
dvander@159
   191
    /* Didn't use ct_data typedef below to supress compiler warning */
dvander@159
   192
    struct ct_data_s dyn_ltree[HEAP_SIZE];   /* literal and length tree */
dvander@159
   193
    struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
dvander@159
   194
    struct ct_data_s bl_tree[2*BL_CODES+1];  /* Huffman tree for bit lengths */
dvander@159
   195
dvander@159
   196
    struct tree_desc_s l_desc;               /* desc. for literal tree */
dvander@159
   197
    struct tree_desc_s d_desc;               /* desc. for distance tree */
dvander@159
   198
    struct tree_desc_s bl_desc;              /* desc. for bit length tree */
dvander@159
   199
dvander@159
   200
    ush bl_count[MAX_BITS+1];
dvander@159
   201
    /* number of codes at each bit length for an optimal tree */
dvander@159
   202
dvander@159
   203
    int heap[2*L_CODES+1];      /* heap used to build the Huffman trees */
dvander@159
   204
    int heap_len;               /* number of elements in the heap */
dvander@159
   205
    int heap_max;               /* element of largest frequency */
dvander@159
   206
    /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
dvander@159
   207
     * The same heap array is used to build all trees.
dvander@159
   208
     */
dvander@159
   209
dvander@159
   210
    uch depth[2*L_CODES+1];
dvander@159
   211
    /* Depth of each subtree used as tie breaker for trees of equal frequency
dvander@159
   212
     */
dvander@159
   213
dvander@159
   214
    uchf *l_buf;          /* buffer for literals or lengths */
dvander@159
   215
dvander@159
   216
    uInt  lit_bufsize;
dvander@159
   217
    /* Size of match buffer for literals/lengths.  There are 4 reasons for
dvander@159
   218
     * limiting lit_bufsize to 64K:
dvander@159
   219
     *   - frequencies can be kept in 16 bit counters
dvander@159
   220
     *   - if compression is not successful for the first block, all input
dvander@159
   221
     *     data is still in the window so we can still emit a stored block even
dvander@159
   222
     *     when input comes from standard input.  (This can also be done for
dvander@159
   223
     *     all blocks if lit_bufsize is not greater than 32K.)
dvander@159
   224
     *   - if compression is not successful for a file smaller than 64K, we can
dvander@159
   225
     *     even emit a stored file instead of a stored block (saving 5 bytes).
dvander@159
   226
     *     This is applicable only for zip (not gzip or zlib).
dvander@159
   227
     *   - creating new Huffman trees less frequently may not provide fast
dvander@159
   228
     *     adaptation to changes in the input data statistics. (Take for
dvander@159
   229
     *     example a binary file with poorly compressible code followed by
dvander@159
   230
     *     a highly compressible string table.) Smaller buffer sizes give
dvander@159
   231
     *     fast adaptation but have of course the overhead of transmitting
dvander@159
   232
     *     trees more frequently.
dvander@159
   233
     *   - I can't count above 4
dvander@159
   234
     */
dvander@159
   235
dvander@159
   236
    uInt last_lit;      /* running index in l_buf */
dvander@159
   237
dvander@159
   238
    ushf *d_buf;
dvander@159
   239
    /* Buffer for distances. To simplify the code, d_buf and l_buf have
dvander@159
   240
     * the same number of elements. To use different lengths, an extra flag
dvander@159
   241
     * array would be necessary.
dvander@159
   242
     */
dvander@159
   243
dvander@159
   244
    ulg opt_len;        /* bit length of current block with optimal trees */
dvander@159
   245
    ulg static_len;     /* bit length of current block with static trees */
dvander@159
   246
    uInt matches;       /* number of string matches in current block */
dvander@159
   247
    int last_eob_len;   /* bit length of EOB code for last block */
dvander@159
   248
dvander@159
   249
#ifdef DEBUG
dvander@159
   250
    ulg compressed_len; /* total bit length of compressed file mod 2^32 */
dvander@159
   251
    ulg bits_sent;      /* bit length of compressed data sent mod 2^32 */
dvander@159
   252
#endif
dvander@159
   253
dvander@159
   254
    ush bi_buf;
dvander@159
   255
    /* Output buffer. bits are inserted starting at the bottom (least
dvander@159
   256
     * significant bits).
dvander@159
   257
     */
dvander@159
   258
    int bi_valid;
dvander@159
   259
    /* Number of valid bits in bi_buf.  All bits above the last valid bit
dvander@159
   260
     * are always zero.
dvander@159
   261
     */
dvander@159
   262
dvander@159
   263
} FAR deflate_state;
dvander@159
   264
dvander@159
   265
/* Output a byte on the stream.
dvander@159
   266
 * IN assertion: there is enough room in pending_buf.
dvander@159
   267
 */
dvander@159
   268
#define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
dvander@159
   269
dvander@159
   270
dvander@159
   271
#define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
dvander@159
   272
/* Minimum amount of lookahead, except at the end of the input file.
dvander@159
   273
 * See deflate.c for comments about the MIN_MATCH+1.
dvander@159
   274
 */
dvander@159
   275
dvander@159
   276
#define MAX_DIST(s)  ((s)->w_size-MIN_LOOKAHEAD)
dvander@159
   277
/* In order to simplify the code, particularly on 16 bit machines, match
dvander@159
   278
 * distances are limited to MAX_DIST instead of WSIZE.
dvander@159
   279
 */
dvander@159
   280
dvander@159
   281
        /* in trees.c */
dvander@159
   282
void _tr_init         OF((deflate_state *s));
dvander@159
   283
int  _tr_tally        OF((deflate_state *s, unsigned dist, unsigned lc));
dvander@159
   284
void _tr_flush_block  OF((deflate_state *s, charf *buf, ulg stored_len,
dvander@159
   285
                          int eof));
dvander@159
   286
void _tr_align        OF((deflate_state *s));
dvander@159
   287
void _tr_stored_block OF((deflate_state *s, charf *buf, ulg stored_len,
dvander@159
   288
                          int eof));
dvander@159
   289
dvander@159
   290
#define d_code(dist) \
dvander@159
   291
   ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)])
dvander@159
   292
/* Mapping from a distance to a distance code. dist is the distance - 1 and
dvander@159
   293
 * must not have side effects. _dist_code[256] and _dist_code[257] are never
dvander@159
   294
 * used.
dvander@159
   295
 */
dvander@159
   296
dvander@159
   297
#ifndef DEBUG
dvander@159
   298
/* Inline versions of _tr_tally for speed: */
dvander@159
   299
dvander@159
   300
#if defined(GEN_TREES_H) || !defined(STDC)
dvander@159
   301
  extern uch _length_code[];
dvander@159
   302
  extern uch _dist_code[];
dvander@159
   303
#else
dvander@159
   304
  extern const uch _length_code[];
dvander@159
   305
  extern const uch _dist_code[];
dvander@159
   306
#endif
dvander@159
   307
dvander@159
   308
# define _tr_tally_lit(s, c, flush) \
dvander@159
   309
  { uch cc = (c); \
dvander@159
   310
    s->d_buf[s->last_lit] = 0; \
dvander@159
   311
    s->l_buf[s->last_lit++] = cc; \
dvander@159
   312
    s->dyn_ltree[cc].Freq++; \
dvander@159
   313
    flush = (s->last_lit == s->lit_bufsize-1); \
dvander@159
   314
   }
dvander@159
   315
# define _tr_tally_dist(s, distance, length, flush) \
dvander@159
   316
  { uch len = (length); \
dvander@159
   317
    ush dist = (distance); \
dvander@159
   318
    s->d_buf[s->last_lit] = dist; \
dvander@159
   319
    s->l_buf[s->last_lit++] = len; \
dvander@159
   320
    dist--; \
dvander@159
   321
    s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \
dvander@159
   322
    s->dyn_dtree[d_code(dist)].Freq++; \
dvander@159
   323
    flush = (s->last_lit == s->lit_bufsize-1); \
dvander@159
   324
  }
dvander@159
   325
#else
dvander@159
   326
# define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c)
dvander@159
   327
# define _tr_tally_dist(s, distance, length, flush) \
dvander@159
   328
              flush = _tr_tally(s, distance, length)
dvander@159
   329
#endif
dvander@159
   330
dvander@159
   331
#endif /* DEFLATE_H */