Back to index

tetex-bin  3.0
deflate.h
Go to the documentation of this file.
00001 /* deflate.h -- internal compression state
00002  * Copyright (C) 1995-2002 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 BUSY_STATE   113
00053 #define FINISH_STATE 666
00054 /* Stream status */
00055 
00056 
00057 /* Data structure describing a single value and its code string. */
00058 typedef struct ct_data_s {
00059     union {
00060         ush  freq;       /* frequency count */
00061         ush  code;       /* bit string */
00062     } fc;
00063     union {
00064         ush  dad;        /* father node in Huffman tree */
00065         ush  len;        /* length of bit string */
00066     } dl;
00067 } FAR ct_data;
00068 
00069 #define Freq fc.freq
00070 #define Code fc.code
00071 #define Dad  dl.dad
00072 #define Len  dl.len
00073 
00074 typedef struct static_tree_desc_s  static_tree_desc;
00075 
00076 typedef struct tree_desc_s {
00077     ct_data *dyn_tree;           /* the dynamic tree */
00078     int     max_code;            /* largest code with non zero frequency */
00079     static_tree_desc *stat_desc; /* the corresponding static tree */
00080 } FAR tree_desc;
00081 
00082 typedef ush Pos;
00083 typedef Pos FAR Posf;
00084 typedef unsigned IPos;
00085 
00086 /* A Pos is an index in the character window. We use short instead of int to
00087  * save space in the various tables. IPos is used only for parameter passing.
00088  */
00089 
00090 typedef struct internal_state {
00091     z_streamp strm;      /* pointer back to this zlib stream */
00092     int   status;        /* as the name implies */
00093     Bytef *pending_buf;  /* output still pending */
00094     ulg   pending_buf_size; /* size of pending_buf */
00095     Bytef *pending_out;  /* next pending byte to output to the stream */
00096     int   pending;       /* nb of bytes in the pending buffer */
00097     int   wrap;          /* bit 0 true for zlib, bit 1 true for gzip */
00098     Byte  data_type;     /* UNKNOWN, BINARY or ASCII */
00099     Byte  method;        /* STORED (for zip only) or DEFLATED */
00100     int   last_flush;    /* value of flush param for previous deflate call */
00101 
00102                 /* used by deflate.c: */
00103 
00104     uInt  w_size;        /* LZ77 window size (32K by default) */
00105     uInt  w_bits;        /* log2(w_size)  (8..16) */
00106     uInt  w_mask;        /* w_size - 1 */
00107 
00108     Bytef *window;
00109     /* Sliding window. Input bytes are read into the second half of the window,
00110      * and move to the first half later to keep a dictionary of at least wSize
00111      * bytes. With this organization, matches are limited to a distance of
00112      * wSize-MAX_MATCH bytes, but this ensures that IO is always
00113      * performed with a length multiple of the block size. Also, it limits
00114      * the window size to 64K, which is quite useful on MSDOS.
00115      * To do: use the user input buffer as sliding window.
00116      */
00117 
00118     ulg window_size;
00119     /* Actual size of window: 2*wSize, except when the user input buffer
00120      * is directly used as sliding window.
00121      */
00122 
00123     Posf *prev;
00124     /* Link to older string with same hash index. To limit the size of this
00125      * array to 64K, this link is maintained only for the last 32K strings.
00126      * An index in this array is thus a window index modulo 32K.
00127      */
00128 
00129     Posf *head; /* Heads of the hash chains or NIL. */
00130 
00131     uInt  ins_h;          /* hash index of string to be inserted */
00132     uInt  hash_size;      /* number of elements in hash table */
00133     uInt  hash_bits;      /* log2(hash_size) */
00134     uInt  hash_mask;      /* hash_size-1 */
00135 
00136     uInt  hash_shift;
00137     /* Number of bits by which ins_h must be shifted at each input
00138      * step. It must be such that after MIN_MATCH steps, the oldest
00139      * byte no longer takes part in the hash key, that is:
00140      *   hash_shift * MIN_MATCH >= hash_bits
00141      */
00142 
00143     long block_start;
00144     /* Window position at the beginning of the current output block. Gets
00145      * negative when the window is moved backwards.
00146      */
00147 
00148     uInt match_length;           /* length of best match */
00149     IPos prev_match;             /* previous match */
00150     int match_available;         /* set if previous match exists */
00151     uInt strstart;               /* start of string to insert */
00152     uInt match_start;            /* start of matching string */
00153     uInt lookahead;              /* number of valid bytes ahead in window */
00154 
00155     uInt prev_length;
00156     /* Length of the best match at previous step. Matches not greater than this
00157      * are discarded. This is used in the lazy match evaluation.
00158      */
00159 
00160     uInt max_chain_length;
00161     /* To speed up deflation, hash chains are never searched beyond this
00162      * length.  A higher limit improves compression ratio but degrades the
00163      * speed.
00164      */
00165 
00166     uInt max_lazy_match;
00167     /* Attempt to find a better match only when the current match is strictly
00168      * smaller than this value. This mechanism is used only for compression
00169      * levels >= 4.
00170      */
00171 #   define max_insert_length  max_lazy_match
00172     /* Insert new strings in the hash table only if the match length is not
00173      * greater than this length. This saves time but degrades compression.
00174      * max_insert_length is used only for compression levels <= 3.
00175      */
00176 
00177     int level;    /* compression level (1..9) */
00178     int strategy; /* favor or force Huffman coding*/
00179 
00180     uInt good_match;
00181     /* Use a faster search when the previous match is longer than this */
00182 
00183     int nice_match; /* Stop searching when current match exceeds this */
00184 
00185                 /* used by trees.c: */
00186     /* Didn't use ct_data typedef below to supress compiler warning */
00187     struct ct_data_s dyn_ltree[HEAP_SIZE];   /* literal and length tree */
00188     struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
00189     struct ct_data_s bl_tree[2*BL_CODES+1];  /* Huffman tree for bit lengths */
00190 
00191     struct tree_desc_s l_desc;               /* desc. for literal tree */
00192     struct tree_desc_s d_desc;               /* desc. for distance tree */
00193     struct tree_desc_s bl_desc;              /* desc. for bit length tree */
00194 
00195     ush bl_count[MAX_BITS+1];
00196     /* number of codes at each bit length for an optimal tree */
00197 
00198     int heap[2*L_CODES+1];      /* heap used to build the Huffman trees */
00199     int heap_len;               /* number of elements in the heap */
00200     int heap_max;               /* element of largest frequency */
00201     /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
00202      * The same heap array is used to build all trees.
00203      */
00204 
00205     uch depth[2*L_CODES+1];
00206     /* Depth of each subtree used as tie breaker for trees of equal frequency
00207      */
00208 
00209     uchf *l_buf;          /* buffer for literals or lengths */
00210 
00211     uInt  lit_bufsize;
00212     /* Size of match buffer for literals/lengths.  There are 4 reasons for
00213      * limiting lit_bufsize to 64K:
00214      *   - frequencies can be kept in 16 bit counters
00215      *   - if compression is not successful for the first block, all input
00216      *     data is still in the window so we can still emit a stored block even
00217      *     when input comes from standard input.  (This can also be done for
00218      *     all blocks if lit_bufsize is not greater than 32K.)
00219      *   - if compression is not successful for a file smaller than 64K, we can
00220      *     even emit a stored file instead of a stored block (saving 5 bytes).
00221      *     This is applicable only for zip (not gzip or zlib).
00222      *   - creating new Huffman trees less frequently may not provide fast
00223      *     adaptation to changes in the input data statistics. (Take for
00224      *     example a binary file with poorly compressible code followed by
00225      *     a highly compressible string table.) Smaller buffer sizes give
00226      *     fast adaptation but have of course the overhead of transmitting
00227      *     trees more frequently.
00228      *   - I can't count above 4
00229      */
00230 
00231     uInt last_lit;      /* running index in l_buf */
00232 
00233     ushf *d_buf;
00234     /* Buffer for distances. To simplify the code, d_buf and l_buf have
00235      * the same number of elements. To use different lengths, an extra flag
00236      * array would be necessary.
00237      */
00238 
00239     ulg opt_len;        /* bit length of current block with optimal trees */
00240     ulg static_len;     /* bit length of current block with static trees */
00241     uInt matches;       /* number of string matches in current block */
00242     int last_eob_len;   /* bit length of EOB code for last block */
00243 
00244 #ifdef DEBUG
00245     ulg compressed_len; /* total bit length of compressed file mod 2^32 */
00246     ulg bits_sent;      /* bit length of compressed data sent mod 2^32 */
00247 #endif
00248 
00249     ush bi_buf;
00250     /* Output buffer. bits are inserted starting at the bottom (least
00251      * significant bits).
00252      */
00253     int bi_valid;
00254     /* Number of valid bits in bi_buf.  All bits above the last valid bit
00255      * are always zero.
00256      */
00257 
00258 } FAR deflate_state;
00259 
00260 /* Output a byte on the stream.
00261  * IN assertion: there is enough room in pending_buf.
00262  */
00263 #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
00264 
00265 
00266 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
00267 /* Minimum amount of lookahead, except at the end of the input file.
00268  * See deflate.c for comments about the MIN_MATCH+1.
00269  */
00270 
00271 #define MAX_DIST(s)  ((s)->w_size-MIN_LOOKAHEAD)
00272 /* In order to simplify the code, particularly on 16 bit machines, match
00273  * distances are limited to MAX_DIST instead of WSIZE.
00274  */
00275 
00276         /* in trees.c */
00277 void _tr_init         OF((deflate_state *s));
00278 int  _tr_tally        OF((deflate_state *s, unsigned dist, unsigned lc));
00279 void _tr_flush_block  OF((deflate_state *s, charf *buf, ulg stored_len,
00280                           int eof));
00281 void _tr_align        OF((deflate_state *s));
00282 void _tr_stored_block OF((deflate_state *s, charf *buf, ulg stored_len,
00283                           int eof));
00284 
00285 #define d_code(dist) \
00286    ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)])
00287 /* Mapping from a distance to a distance code. dist is the distance - 1 and
00288  * must not have side effects. _dist_code[256] and _dist_code[257] are never
00289  * used.
00290  */
00291 
00292 #ifndef DEBUG
00293 /* Inline versions of _tr_tally for speed: */
00294 
00295 #if defined(GEN_TREES_H) || !defined(STDC)
00296   extern uch _length_code[];
00297   extern uch _dist_code[];
00298 #else
00299   extern const uch _length_code[];
00300   extern const uch _dist_code[];
00301 #endif
00302 
00303 # define _tr_tally_lit(s, c, flush) \
00304   { uch cc = (c); \
00305     s->d_buf[s->last_lit] = 0; \
00306     s->l_buf[s->last_lit++] = cc; \
00307     s->dyn_ltree[cc].Freq++; \
00308     flush = (s->last_lit == s->lit_bufsize-1); \
00309    }
00310 # define _tr_tally_dist(s, distance, length, flush) \
00311   { uch len = (length); \
00312     ush dist = (distance); \
00313     s->d_buf[s->last_lit] = dist; \
00314     s->l_buf[s->last_lit++] = len; \
00315     dist--; \
00316     s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \
00317     s->dyn_dtree[d_code(dist)].Freq++; \
00318     flush = (s->last_lit == s->lit_bufsize-1); \
00319   }
00320 #else
00321 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c)
00322 # define _tr_tally_dist(s, distance, length, flush) \
00323               flush = _tr_tally(s, distance, length)
00324 #endif
00325 
00326 #endif /* DEFLATE_H */