1/* deflate.h -- internal compression state
2 * Copyright (C) 1995-2010 Jean-loup Gailly
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
5
6/* WARNING: this file should *not* be used by applications. It is
7   part of the implementation of the compression library and is
8   subject to change. Applications should only use zlib.h.
9 */
10
11/* @(#) $Id$ */
12
13#ifndef DEFLATE_H
14#define DEFLATE_H
15
16#include "zutil.h"
17
18/* define NO_GZIP when compiling if you want to disable gzip header and
19   trailer creation by deflate().  NO_GZIP would be used to avoid linking in
20   the crc code when it is not needed.  For shared libraries, gzip encoding
21   should be left enabled. */
22#ifndef NO_GZIP
23#  define GZIP
24#endif
25
26/* ===========================================================================
27 * Internal compression state.
28 */
29
30#define LENGTH_CODES 29
31/* number of length codes, not counting the special END_BLOCK code */
32
33#define LITERALS  256
34/* number of literal bytes 0..255 */
35
36#define L_CODES (LITERALS+1+LENGTH_CODES)
37/* number of Literal or Length codes, including the END_BLOCK code */
38
39#define D_CODES   30
40/* number of distance codes */
41
42#define BL_CODES  19
43/* number of codes used to transfer the bit lengths */
44
45#define HEAP_SIZE (2*L_CODES+1)
46/* maximum heap size */
47
48#define MAX_BITS 15
49/* All codes must not exceed MAX_BITS bits */
50
51#define INIT_STATE    42
52#define EXTRA_STATE   69
53#define NAME_STATE    73
54#define COMMENT_STATE 91
55#define HCRC_STATE   103
56#define BUSY_STATE   113
57#define FINISH_STATE 666
58/* Stream status */
59
60
61/* Data structure describing a single value and its code string. */
62typedef struct ct_data_s {
63    union {
64        ush  freq;       /* frequency count */
65        ush  code;       /* bit string */
66    } fc;
67    union {
68        ush  dad;        /* father node in Huffman tree */
69        ush  len;        /* length of bit string */
70    } dl;
71} FAR ct_data;
72
73#define Freq fc.freq
74#define Code fc.code
75#define Dad  dl.dad
76#define Len  dl.len
77
78typedef struct static_tree_desc_s  static_tree_desc;
79
80typedef struct tree_desc_s {
81    ct_data *dyn_tree;           /* the dynamic tree */
82    int     max_code;            /* largest code with non zero frequency */
83    static_tree_desc *stat_desc; /* the corresponding static tree */
84} FAR tree_desc;
85
86typedef ush Pos;
87typedef Pos FAR Posf;
88typedef unsigned IPos;
89
90/* A Pos is an index in the character window. We use short instead of int to
91 * save space in the various tables. IPos is used only for parameter passing.
92 */
93
94#define Z_COOKIE_HASH_SIZE 256
95#define Z_COOKIE_HASH_MASK (Z_COOKIE_HASH_SIZE-1)
96
97typedef struct internal_state {
98    z_streamp strm;      /* pointer back to this zlib stream */
99    int   status;        /* as the name implies */
100    Bytef *pending_buf;  /* output still pending */
101    ulg   pending_buf_size; /* size of pending_buf */
102    Bytef *pending_out;  /* next pending byte to output to the stream */
103    uInt   pending;      /* nb of bytes in the pending buffer */
104    int   wrap;          /* bit 0 true for zlib, bit 1 true for gzip */
105    gz_headerp  gzhead;  /* gzip header information to write */
106    uInt   gzindex;      /* where in extra, name, or comment */
107    Byte  method;        /* STORED (for zip only) or DEFLATED */
108    int   last_flush;    /* value of flush param for previous deflate call */
109
110                /* used by deflate.c: */
111
112    uInt  w_size;        /* LZ77 window size (32K by default) */
113    uInt  w_bits;        /* log2(w_size)  (8..16) */
114    uInt  w_mask;        /* w_size - 1 */
115
116    Bytef *window;
117    /* Sliding window. Input bytes are read into the second half of the window,
118     * and move to the first half later to keep a dictionary of at least wSize
119     * bytes. With this organization, matches are limited to a distance of
120     * wSize-MAX_MATCH bytes, but this ensures that IO is always
121     * performed with a length multiple of the block size. Also, it limits
122     * the window size to 64K, which is quite useful on MSDOS.
123     * To do: use the user input buffer as sliding window.
124     */
125
126    ulg window_size;
127    /* Actual size of window: 2*wSize, except when the user input buffer
128     * is directly used as sliding window.
129     */
130
131    Posf *prev;
132    /* Link to older string with same hash index. To limit the size of this
133     * array to 64K, this link is maintained only for the last 32K strings.
134     * An index in this array is thus a window index modulo 32K.
135     */
136
137    Posf *head; /* Heads of the hash chains or NIL. */
138
139    uInt  ins_h;          /* hash index of string to be inserted */
140    uInt  hash_size;      /* number of elements in hash table */
141    uInt  hash_bits;      /* log2(hash_size) */
142    uInt  hash_mask;      /* hash_size-1 */
143
144    uInt  hash_shift;
145    Bytef *class_bitmap;  /* bitmap of class for each byte in window */
146    IPos cookie_locations[Z_COOKIE_HASH_SIZE];
147    /* Number of bits by which ins_h must be shifted at each input
148     * step. It must be such that after MIN_MATCH steps, the oldest
149     * byte no longer takes part in the hash key, that is:
150     *   hash_shift * MIN_MATCH >= hash_bits
151     */
152
153    long block_start;
154    /* Window position at the beginning of the current output block. Gets
155     * negative when the window is moved backwards.
156     */
157
158    uInt match_length;           /* length of best match */
159    IPos prev_match;             /* previous match */
160    int match_available;         /* set if previous match exists */
161    uInt strstart;               /* start of string to insert */
162    uInt match_start;            /* start of matching string */
163    uInt lookahead;              /* number of valid bytes ahead in window */
164
165    uInt prev_length;
166    /* Length of the best match at previous step. Matches not greater than this
167     * are discarded. This is used in the lazy match evaluation.
168     */
169
170    uInt max_chain_length;
171    /* To speed up deflation, hash chains are never searched beyond this
172     * length.  A higher limit improves compression ratio but degrades the
173     * speed.
174     */
175
176    uInt max_lazy_match;
177    /* Attempt to find a better match only when the current match is strictly
178     * smaller than this value. This mechanism is used only for compression
179     * levels >= 4.
180     */
181#   define max_insert_length  max_lazy_match
182    /* Insert new strings in the hash table only if the match length is not
183     * greater than this length. This saves time but degrades compression.
184     * max_insert_length is used only for compression levels <= 3.
185     */
186
187    int level;    /* compression level (1..9) */
188    int strategy; /* favor or force Huffman coding*/
189
190    uInt good_match;
191    /* Use a faster search when the previous match is longer than this */
192
193    int nice_match; /* Stop searching when current match exceeds this */
194
195                /* used by trees.c: */
196    /* Didn't use ct_data typedef below to supress compiler warning */
197    struct ct_data_s dyn_ltree[HEAP_SIZE];   /* literal and length tree */
198    struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
199    struct ct_data_s bl_tree[2*BL_CODES+1];  /* Huffman tree for bit lengths */
200
201    struct tree_desc_s l_desc;               /* desc. for literal tree */
202    struct tree_desc_s d_desc;               /* desc. for distance tree */
203    struct tree_desc_s bl_desc;              /* desc. for bit length tree */
204
205    ush bl_count[MAX_BITS+1];
206    /* number of codes at each bit length for an optimal tree */
207
208    int heap[2*L_CODES+1];      /* heap used to build the Huffman trees */
209    int heap_len;               /* number of elements in the heap */
210    int heap_max;               /* element of largest frequency */
211    /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
212     * The same heap array is used to build all trees.
213     */
214
215    uch depth[2*L_CODES+1];
216    /* Depth of each subtree used as tie breaker for trees of equal frequency
217     */
218
219    uchf *l_buf;          /* buffer for literals or lengths */
220
221    uInt  lit_bufsize;
222    /* Size of match buffer for literals/lengths.  There are 4 reasons for
223     * limiting lit_bufsize to 64K:
224     *   - frequencies can be kept in 16 bit counters
225     *   - if compression is not successful for the first block, all input
226     *     data is still in the window so we can still emit a stored block even
227     *     when input comes from standard input.  (This can also be done for
228     *     all blocks if lit_bufsize is not greater than 32K.)
229     *   - if compression is not successful for a file smaller than 64K, we can
230     *     even emit a stored file instead of a stored block (saving 5 bytes).
231     *     This is applicable only for zip (not gzip or zlib).
232     *   - creating new Huffman trees less frequently may not provide fast
233     *     adaptation to changes in the input data statistics. (Take for
234     *     example a binary file with poorly compressible code followed by
235     *     a highly compressible string table.) Smaller buffer sizes give
236     *     fast adaptation but have of course the overhead of transmitting
237     *     trees more frequently.
238     *   - I can't count above 4
239     */
240
241    uInt last_lit;      /* running index in l_buf */
242
243    ushf *d_buf;
244    /* Buffer for distances. To simplify the code, d_buf and l_buf have
245     * the same number of elements. To use different lengths, an extra flag
246     * array would be necessary.
247     */
248
249    ulg opt_len;        /* bit length of current block with optimal trees */
250    ulg static_len;     /* bit length of current block with static trees */
251    uInt matches;       /* number of string matches in current block */
252    int last_eob_len;   /* bit length of EOB code for last block */
253
254#ifdef DEBUG
255    ulg compressed_len; /* total bit length of compressed file mod 2^32 */
256    ulg bits_sent;      /* bit length of compressed data sent mod 2^32 */
257#endif
258
259    ush bi_buf;
260    /* Output buffer. bits are inserted starting at the bottom (least
261     * significant bits).
262     */
263    int bi_valid;
264    /* Number of valid bits in bi_buf.  All bits above the last valid bit
265     * are always zero.
266     */
267
268    ulg high_water;
269    /* High water mark offset in window for initialized bytes -- bytes above
270     * this are set to zero in order to avoid memory check warnings when
271     * longest match routines access bytes past the input.  This is then
272     * updated to the new high water mark.
273     */
274
275} FAR deflate_state;
276
277/* Output a byte on the stream.
278 * IN assertion: there is enough room in pending_buf.
279 */
280#define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
281
282
283#define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
284/* Minimum amount of lookahead, except at the end of the input file.
285 * See deflate.c for comments about the MIN_MATCH+1.
286 */
287
288#define MAX_DIST(s)  ((s)->w_size-MIN_LOOKAHEAD)
289/* In order to simplify the code, particularly on 16 bit machines, match
290 * distances are limited to MAX_DIST instead of WSIZE.
291 */
292
293#define WIN_INIT MAX_MATCH
294/* Number of bytes after end of data in window to initialize in order to avoid
295   memory checker errors from longest match routines */
296
297        /* in trees.c */
298void ZLIB_INTERNAL _tr_init OF((deflate_state *s));
299int ZLIB_INTERNAL _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc));
300void ZLIB_INTERNAL _tr_flush_block OF((deflate_state *s, charf *buf,
301                        ulg stored_len, int last));
302void ZLIB_INTERNAL _tr_align OF((deflate_state *s));
303void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf,
304                        ulg stored_len, int last));
305
306#define d_code(dist) \
307   ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)])
308/* Mapping from a distance to a distance code. dist is the distance - 1 and
309 * must not have side effects. _dist_code[256] and _dist_code[257] are never
310 * used.
311 */
312
313#ifndef DEBUG
314/* Inline versions of _tr_tally for speed: */
315
316#if defined(GEN_TREES_H) || !defined(STDC)
317  extern uch ZLIB_INTERNAL _length_code[];
318  extern uch ZLIB_INTERNAL _dist_code[];
319#else
320  extern const uch ZLIB_INTERNAL _length_code[];
321  extern const uch ZLIB_INTERNAL _dist_code[];
322#endif
323
324# define _tr_tally_lit(s, c, flush) \
325  { uch cc = (c); \
326    s->d_buf[s->last_lit] = 0; \
327    s->l_buf[s->last_lit++] = cc; \
328    s->dyn_ltree[cc].Freq++; \
329    flush = (s->last_lit == s->lit_bufsize-1); \
330   }
331# define _tr_tally_dist(s, distance, length, flush) \
332  { uch len = (length); \
333    ush dist = (distance); \
334    s->d_buf[s->last_lit] = dist; \
335    s->l_buf[s->last_lit++] = len; \
336    dist--; \
337    s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \
338    s->dyn_dtree[d_code(dist)].Freq++; \
339    flush = (s->last_lit == s->lit_bufsize-1); \
340  }
341#else
342# define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c)
343# define _tr_tally_dist(s, distance, length, flush) \
344              flush = _tr_tally(s, distance, length)
345#endif
346
347#endif /* DEFLATE_H */
348