deflate.h

00001 /* deflate.h -- internal compression state
00002  * Copyright (C) 1995-2012 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 Buf_size 16
00052 /* size of bit buffer in bi_buf */
00053 
00054 #define INIT_STATE    42
00055 #define EXTRA_STATE   69
00056 #define NAME_STATE    73
00057 #define COMMENT_STATE 91
00058 #define HCRC_STATE   103
00059 #define BUSY_STATE   113
00060 #define FINISH_STATE 666
00061 /* Stream status */
00062 
00063 
00064 /* Data structure describing a single value and its code string. */
00065 typedef struct ct_data_s {
00066     union {
00067         ush  freq;       /* frequency count */
00068         ush  code;       /* bit string */
00069     } fc;
00070     union {
00071         ush  dad;        /* father node in Huffman tree */
00072         ush  len;        /* length of bit string */
00073     } dl;
00074 } FAR ct_data;
00075 
00076 #define Freq fc.freq
00077 #define Code fc.code
00078 #define Dad  dl.dad
00079 #define Len  dl.len
00080 
00081 typedef struct static_tree_desc_s  static_tree_desc;
00082 
00083 typedef struct tree_desc_s {
00084     ct_data *dyn_tree;           /* the dynamic tree */
00085     int     max_code;            /* largest code with non zero frequency */
00086     static_tree_desc *stat_desc; /* the corresponding static tree */
00087 } FAR tree_desc;
00088 
00089 typedef ush Pos;
00090 typedef Pos FAR Posf;
00091 typedef unsigned IPos;
00092 
00093 /* A Pos is an index in the character window. We use short instead of int to
00094  * save space in the various tables. IPos is used only for parameter passing.
00095  */
00096 
00097 typedef struct internal_state {
00098     z_streamp strm;      /* pointer back to this zlib stream */
00099     int   status;        /* as the name implies */
00100     Bytef *pending_buf;  /* output still pending */
00101     ulg   pending_buf_size; /* size of pending_buf */
00102     Bytef *pending_out;  /* next pending byte to output to the stream */
00103     uInt   pending;      /* nb of bytes in the pending buffer */
00104     int   wrap;          /* bit 0 true for zlib, bit 1 true for gzip */
00105     gz_headerp  gzhead;  /* gzip header information to write */
00106     uInt   gzindex;      /* where in extra, name, or comment */
00107     Byte  method;        /* STORED (for zip only) or DEFLATED */
00108     int   last_flush;    /* value of flush param for previous deflate call */
00109 
00110                 /* used by deflate.c: */
00111 
00112     uInt  w_size;        /* LZ77 window size (32K by default) */
00113     uInt  w_bits;        /* log2(w_size)  (8..16) */
00114     uInt  w_mask;        /* w_size - 1 */
00115 
00116     Bytef *window;
00117     /* Sliding window. Input bytes are read into the second half of the window,
00118      * and move to the first half later to keep a dictionary of at least wSize
00119      * bytes. With this organization, matches are limited to a distance of
00120      * wSize-MAX_MATCH bytes, but this ensures that IO is always
00121      * performed with a length multiple of the block size. Also, it limits
00122      * the window size to 64K, which is quite useful on MSDOS.
00123      * To do: use the user input buffer as sliding window.
00124      */
00125 
00126     ulg window_size;
00127     /* Actual size of window: 2*wSize, except when the user input buffer
00128      * is directly used as sliding window.
00129      */
00130 
00131     Posf *prev;
00132     /* Link to older string with same hash index. To limit the size of this
00133      * array to 64K, this link is maintained only for the last 32K strings.
00134      * An index in this array is thus a window index modulo 32K.
00135      */
00136 
00137     Posf *head; /* Heads of the hash chains or NIL. */
00138 
00139     uInt  ins_h;          /* hash index of string to be inserted */
00140     uInt  hash_size;      /* number of elements in hash table */
00141     uInt  hash_bits;      /* log2(hash_size) */
00142     uInt  hash_mask;      /* hash_size-1 */
00143 
00144     uInt  hash_shift;
00145     /* Number of bits by which ins_h must be shifted at each input
00146      * step. It must be such that after MIN_MATCH steps, the oldest
00147      * byte no longer takes part in the hash key, that is:
00148      *   hash_shift * MIN_MATCH >= hash_bits
00149      */
00150 
00151     long block_start;
00152     /* Window position at the beginning of the current output block. Gets
00153      * negative when the window is moved backwards.
00154      */
00155 
00156     uInt match_length;           /* length of best match */
00157     IPos prev_match;             /* previous match */
00158     int match_available;         /* set if previous match exists */
00159     uInt strstart;               /* start of string to insert */
00160     uInt match_start;            /* start of matching string */
00161     uInt lookahead;              /* number of valid bytes ahead in window */
00162 
00163     uInt prev_length;
00164     /* Length of the best match at previous step. Matches not greater than this
00165      * are discarded. This is used in the lazy match evaluation.
00166      */
00167 
00168     uInt max_chain_length;
00169     /* To speed up deflation, hash chains are never searched beyond this
00170      * length.  A higher limit improves compression ratio but degrades the
00171      * speed.
00172      */
00173 
00174     uInt max_lazy_match;
00175     /* Attempt to find a better match only when the current match is strictly
00176      * smaller than this value. This mechanism is used only for compression
00177      * levels >= 4.
00178      */
00179 #   define max_insert_length  max_lazy_match
00180     /* Insert new strings in the hash table only if the match length is not
00181      * greater than this length. This saves time but degrades compression.
00182      * max_insert_length is used only for compression levels <= 3.
00183      */
00184 
00185     int level;    /* compression level (1..9) */
00186     int strategy; /* favor or force Huffman coding*/
00187 
00188     uInt good_match;
00189     /* Use a faster search when the previous match is longer than this */
00190 
00191     int nice_match; /* Stop searching when current match exceeds this */
00192 
00193                 /* used by trees.c: */
00194     /* Didn't use ct_data typedef below to suppress compiler warning */
00195     struct ct_data_s dyn_ltree[HEAP_SIZE];   /* literal and length tree */
00196     struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
00197     struct ct_data_s bl_tree[2*BL_CODES+1];  /* Huffman tree for bit lengths */
00198 
00199     struct tree_desc_s l_desc;               /* desc. for literal tree */
00200     struct tree_desc_s d_desc;               /* desc. for distance tree */
00201     struct tree_desc_s bl_desc;              /* desc. for bit length tree */
00202 
00203     ush bl_count[MAX_BITS+1];
00204     /* number of codes at each bit length for an optimal tree */
00205 
00206     int heap[2*L_CODES+1];      /* heap used to build the Huffman trees */
00207     int heap_len;               /* number of elements in the heap */
00208     int heap_max;               /* element of largest frequency */
00209     /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
00210      * The same heap array is used to build all trees.
00211      */
00212 
00213     uch depth[2*L_CODES+1];
00214     /* Depth of each subtree used as tie breaker for trees of equal frequency
00215      */
00216 
00217     uchf *l_buf;          /* buffer for literals or lengths */
00218 
00219     uInt  lit_bufsize;
00220     /* Size of match buffer for literals/lengths.  There are 4 reasons for
00221      * limiting lit_bufsize to 64K:
00222      *   - frequencies can be kept in 16 bit counters
00223      *   - if compression is not successful for the first block, all input
00224      *     data is still in the window so we can still emit a stored block even
00225      *     when input comes from standard input.  (This can also be done for
00226      *     all blocks if lit_bufsize is not greater than 32K.)
00227      *   - if compression is not successful for a file smaller than 64K, we can
00228      *     even emit a stored file instead of a stored block (saving 5 bytes).
00229      *     This is applicable only for zip (not gzip or zlib).
00230      *   - creating new Huffman trees less frequently may not provide fast
00231      *     adaptation to changes in the input data statistics. (Take for
00232      *     example a binary file with poorly compressible code followed by
00233      *     a highly compressible string table.) Smaller buffer sizes give
00234      *     fast adaptation but have of course the overhead of transmitting
00235      *     trees more frequently.
00236      *   - I can't count above 4
00237      */
00238 
00239     uInt last_lit;      /* running index in l_buf */
00240 
00241     ushf *d_buf;
00242     /* Buffer for distances. To simplify the code, d_buf and l_buf have
00243      * the same number of elements. To use different lengths, an extra flag
00244      * array would be necessary.
00245      */
00246 
00247     ulg opt_len;        /* bit length of current block with optimal trees */
00248     ulg static_len;     /* bit length of current block with static trees */
00249     uInt matches;       /* number of string matches in current block */
00250     uInt insert;        /* bytes at end of window left to insert */
00251 
00252 #ifdef DEBUG
00253     ulg compressed_len; /* total bit length of compressed file mod 2^32 */
00254     ulg bits_sent;      /* bit length of compressed data sent mod 2^32 */
00255 #endif
00256 
00257     ush bi_buf;
00258     /* Output buffer. bits are inserted starting at the bottom (least
00259      * significant bits).
00260      */
00261     int bi_valid;
00262     /* Number of valid bits in bi_buf.  All bits above the last valid bit
00263      * are always zero.
00264      */
00265 
00266     ulg high_water;
00267     /* High water mark offset in window for initialized bytes -- bytes above
00268      * this are set to zero in order to avoid memory check warnings when
00269      * longest match routines access bytes past the input.  This is then
00270      * updated to the new high water mark.
00271      */
00272 
00273 } FAR deflate_state;
00274 
00275 /* Output a byte on the stream.
00276  * IN assertion: there is enough room in pending_buf.
00277  */
00278 #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
00279 
00280 
00281 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
00282 /* Minimum amount of lookahead, except at the end of the input file.
00283  * See deflate.c for comments about the MIN_MATCH+1.
00284  */
00285 
00286 #define MAX_DIST(s)  ((s)->w_size-MIN_LOOKAHEAD)
00287 /* In order to simplify the code, particularly on 16 bit machines, match
00288  * distances are limited to MAX_DIST instead of WSIZE.
00289  */
00290 
00291 #define WIN_INIT MAX_MATCH
00292 /* Number of bytes after end of data in window to initialize in order to avoid
00293    memory checker errors from longest match routines */
00294 
00295         /* in trees.c */
00296 void ZLIB_INTERNAL _tr_init OF((deflate_state *s));
00297 int ZLIB_INTERNAL _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc));
00298 void ZLIB_INTERNAL _tr_flush_block OF((deflate_state *s, charf *buf,
00299                         ulg stored_len, int last));
00300 void ZLIB_INTERNAL _tr_flush_bits OF((deflate_state *s));
00301 void ZLIB_INTERNAL _tr_align OF((deflate_state *s));
00302 void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf,
00303                         ulg stored_len, int last));
00304 
00305 #define d_code(dist) \
00306    ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)])
00307 /* Mapping from a distance to a distance code. dist is the distance - 1 and
00308  * must not have side effects. _dist_code[256] and _dist_code[257] are never
00309  * used.
00310  */
00311 
00312 #ifndef DEBUG
00313 /* Inline versions of _tr_tally for speed: */
00314 
00315 #if defined(GEN_TREES_H) || !defined(STDC)
00316   extern uch ZLIB_INTERNAL _length_code[];
00317   extern uch ZLIB_INTERNAL _dist_code[];
00318 #else
00319   extern const uch ZLIB_INTERNAL _length_code[];
00320   extern const uch ZLIB_INTERNAL _dist_code[];
00321 #endif
00322 
00323 # define _tr_tally_lit(s, c, flush) \
00324   { uch cc = (c); \
00325     s->d_buf[s->last_lit] = 0; \
00326     s->l_buf[s->last_lit++] = cc; \
00327     s->dyn_ltree[cc].Freq++; \
00328     flush = (s->last_lit == s->lit_bufsize-1); \
00329    }
00330 # define _tr_tally_dist(s, distance, length, flush) \
00331   { uch len = (length); \
00332     ush dist = (distance); \
00333     s->d_buf[s->last_lit] = dist; \
00334     s->l_buf[s->last_lit++] = len; \
00335     dist--; \
00336     s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \
00337     s->dyn_dtree[d_code(dist)].Freq++; \
00338     flush = (s->last_lit == s->lit_bufsize-1); \
00339   }
00340 #else
00341 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c)
00342 # define _tr_tally_dist(s, distance, length, flush) \
00343               flush = _tr_tally(s, distance, length)
00344 #endif
00345 
00346 #endif /* DEFLATE_H */