1 /* deflate.h -- internal compression state
2 * Copyright (C) 1995-2004 Jean-loup Gailly
3 * For conditions of distribution and use, see copyright notice in zlib.h
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.
11 /* @(#) $Id: ./src/external/zlib-1.2.3/deflate.h, 2011/09/08 dcid Exp $
19 /* define NO_GZIP when compiling if you want to disable gzip header and
20 trailer creation by deflate(). NO_GZIP would be used to avoid linking in
21 the crc code when it is not needed. For shared libraries, gzip encoding
22 should be left enabled. */
27 /* ===========================================================================
28 * Internal compression state.
31 #define LENGTH_CODES 29
32 /* number of length codes, not counting the special END_BLOCK code */
35 /* number of literal bytes 0..255 */
37 #define L_CODES (LITERALS+1+LENGTH_CODES)
38 /* number of Literal or Length codes, including the END_BLOCK code */
41 /* number of distance codes */
44 /* number of codes used to transfer the bit lengths */
46 #define HEAP_SIZE (2*L_CODES+1)
47 /* maximum heap size */
50 /* All codes must not exceed MAX_BITS bits */
53 #define EXTRA_STATE 69
55 #define COMMENT_STATE 91
56 #define HCRC_STATE 103
57 #define BUSY_STATE 113
58 #define FINISH_STATE 666
62 /* Data structure describing a single value and its code string. */
63 typedef struct ct_data_s {
65 ush freq; /* frequency count */
66 ush code; /* bit string */
69 ush dad; /* father node in Huffman tree */
70 ush len; /* length of bit string */
79 typedef struct static_tree_desc_s static_tree_desc;
81 typedef struct tree_desc_s {
82 ct_data *dyn_tree; /* the dynamic tree */
83 int max_code; /* largest code with non zero frequency */
84 static_tree_desc *stat_desc; /* the corresponding static tree */
89 typedef unsigned IPos;
91 /* A Pos is an index in the character window. We use short instead of int to
92 * save space in the various tables. IPos is used only for parameter passing.
95 typedef struct internal_state {
96 z_streamp strm; /* pointer back to this zlib stream */
97 int status; /* as the name implies */
98 Bytef *pending_buf; /* output still pending */
99 ulg pending_buf_size; /* size of pending_buf */
100 Bytef *pending_out; /* next pending byte to output to the stream */
101 uInt pending; /* nb of bytes in the pending buffer */
102 int wrap; /* bit 0 true for zlib, bit 1 true for gzip */
103 gz_headerp gzhead; /* gzip header information to write */
104 uInt gzindex; /* where in extra, name, or comment */
105 Byte method; /* STORED (for zip only) or DEFLATED */
106 int last_flush; /* value of flush param for previous deflate call */
108 /* used by deflate.c: */
110 uInt w_size; /* LZ77 window size (32K by default) */
111 uInt w_bits; /* log2(w_size) (8..16) */
112 uInt w_mask; /* w_size - 1 */
115 /* Sliding window. Input bytes are read into the second half of the window,
116 * and move to the first half later to keep a dictionary of at least wSize
117 * bytes. With this organization, matches are limited to a distance of
118 * wSize-MAX_MATCH bytes, but this ensures that IO is always
119 * performed with a length multiple of the block size. Also, it limits
120 * the window size to 64K, which is quite useful on MSDOS.
121 * To do: use the user input buffer as sliding window.
125 /* Actual size of window: 2*wSize, except when the user input buffer
126 * is directly used as sliding window.
130 /* Link to older string with same hash index. To limit the size of this
131 * array to 64K, this link is maintained only for the last 32K strings.
132 * An index in this array is thus a window index modulo 32K.
135 Posf *head; /* Heads of the hash chains or NIL. */
137 uInt ins_h; /* hash index of string to be inserted */
138 uInt hash_size; /* number of elements in hash table */
139 uInt hash_bits; /* log2(hash_size) */
140 uInt hash_mask; /* hash_size-1 */
143 /* Number of bits by which ins_h must be shifted at each input
144 * step. It must be such that after MIN_MATCH steps, the oldest
145 * byte no longer takes part in the hash key, that is:
146 * hash_shift * MIN_MATCH >= hash_bits
150 /* Window position at the beginning of the current output block. Gets
151 * negative when the window is moved backwards.
154 uInt match_length; /* length of best match */
155 IPos prev_match; /* previous match */
156 int match_available; /* set if previous match exists */
157 uInt strstart; /* start of string to insert */
158 uInt match_start; /* start of matching string */
159 uInt lookahead; /* number of valid bytes ahead in window */
162 /* Length of the best match at previous step. Matches not greater than this
163 * are discarded. This is used in the lazy match evaluation.
166 uInt max_chain_length;
167 /* To speed up deflation, hash chains are never searched beyond this
168 * length. A higher limit improves compression ratio but degrades the
173 /* Attempt to find a better match only when the current match is strictly
174 * smaller than this value. This mechanism is used only for compression
177 # define max_insert_length max_lazy_match
178 /* Insert new strings in the hash table only if the match length is not
179 * greater than this length. This saves time but degrades compression.
180 * max_insert_length is used only for compression levels <= 3.
183 int level; /* compression level (1..9) */
184 int strategy; /* favor or force Huffman coding*/
187 /* Use a faster search when the previous match is longer than this */
189 int nice_match; /* Stop searching when current match exceeds this */
191 /* used by trees.c: */
192 /* Didn't use ct_data typedef below to supress compiler warning */
193 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */
194 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
195 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */
197 struct tree_desc_s l_desc; /* desc. for literal tree */
198 struct tree_desc_s d_desc; /* desc. for distance tree */
199 struct tree_desc_s bl_desc; /* desc. for bit length tree */
201 ush bl_count[MAX_BITS+1];
202 /* number of codes at each bit length for an optimal tree */
204 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */
205 int heap_len; /* number of elements in the heap */
206 int heap_max; /* element of largest frequency */
207 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
208 * The same heap array is used to build all trees.
211 uch depth[2*L_CODES+1];
212 /* Depth of each subtree used as tie breaker for trees of equal frequency
215 uchf *l_buf; /* buffer for literals or lengths */
218 /* Size of match buffer for literals/lengths. There are 4 reasons for
219 * limiting lit_bufsize to 64K:
220 * - frequencies can be kept in 16 bit counters
221 * - if compression is not successful for the first block, all input
222 * data is still in the window so we can still emit a stored block even
223 * when input comes from standard input. (This can also be done for
224 * all blocks if lit_bufsize is not greater than 32K.)
225 * - if compression is not successful for a file smaller than 64K, we can
226 * even emit a stored file instead of a stored block (saving 5 bytes).
227 * This is applicable only for zip (not gzip or zlib).
228 * - creating new Huffman trees less frequently may not provide fast
229 * adaptation to changes in the input data statistics. (Take for
230 * example a binary file with poorly compressible code followed by
231 * a highly compressible string table.) Smaller buffer sizes give
232 * fast adaptation but have of course the overhead of transmitting
233 * trees more frequently.
234 * - I can't count above 4
237 uInt last_lit; /* running index in l_buf */
240 /* Buffer for distances. To simplify the code, d_buf and l_buf have
241 * the same number of elements. To use different lengths, an extra flag
242 * array would be necessary.
245 ulg opt_len; /* bit length of current block with optimal trees */
246 ulg static_len; /* bit length of current block with static trees */
247 uInt matches; /* number of string matches in current block */
248 int last_eob_len; /* bit length of EOB code for last block */
251 ulg compressed_len; /* total bit length of compressed file mod 2^32 */
252 ulg bits_sent; /* bit length of compressed data sent mod 2^32 */
256 /* Output buffer. bits are inserted starting at the bottom (least
260 /* Number of valid bits in bi_buf. All bits above the last valid bit
266 /* Output a byte on the stream.
267 * IN assertion: there is enough room in pending_buf.
269 #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
272 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
273 /* Minimum amount of lookahead, except at the end of the input file.
274 * See deflate.c for comments about the MIN_MATCH+1.
277 #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD)
278 /* In order to simplify the code, particularly on 16 bit machines, match
279 * distances are limited to MAX_DIST instead of WSIZE.
283 void _tr_init OF((deflate_state *s));
284 int _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc));
285 void _tr_flush_block OF((deflate_state *s, charf *buf, ulg stored_len,
287 void _tr_align OF((deflate_state *s));
288 void _tr_stored_block OF((deflate_state *s, charf *buf, ulg stored_len,
291 #define d_code(dist) \
292 ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)])
293 /* Mapping from a distance to a distance code. dist is the distance - 1 and
294 * must not have side effects. _dist_code[256] and _dist_code[257] are never
299 /* Inline versions of _tr_tally for speed: */
301 #if defined(GEN_TREES_H) || !defined(STDC)
302 extern uch _length_code[];
303 extern uch _dist_code[];
305 extern const uch _length_code[];
306 extern const uch _dist_code[];
309 # define _tr_tally_lit(s, c, flush) \
311 s->d_buf[s->last_lit] = 0; \
312 s->l_buf[s->last_lit++] = cc; \
313 s->dyn_ltree[cc].Freq++; \
314 flush = (s->last_lit == s->lit_bufsize-1); \
316 # define _tr_tally_dist(s, distance, length, flush) \
317 { uch len = (length); \
318 ush dist = (distance); \
319 s->d_buf[s->last_lit] = dist; \
320 s->l_buf[s->last_lit++] = len; \
322 s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \
323 s->dyn_dtree[d_code(dist)].Freq++; \
324 flush = (s->last_lit == s->lit_bufsize-1); \
327 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c)
328 # define _tr_tally_dist(s, distance, length, flush) \
329 flush = _tr_tally(s, distance, length)
332 #endif /* DEFLATE_H */