1 /* deflate.h -- internal compression state
2 * Copyright (C) 1995 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: deflate.h,v 1.5 1995/05/03 17:27:09 jloup Exp $ */
15 /* ===========================================================================
16 * Internal compression state.
24 #define LENGTH_CODES 29
25 /* number of length codes, not counting the special END_BLOCK code */
28 /* number of literal bytes 0..255 */
30 #define L_CODES (LITERALS+1+LENGTH_CODES)
31 /* number of Literal or Length codes, including the END_BLOCK code */
34 /* number of distance codes */
37 /* number of codes used to transfer the bit lengths */
39 #define HEAP_SIZE (2*L_CODES+1)
40 /* maximum heap size */
43 /* All codes must not exceed MAX_BITS bits */
46 #define BUSY_STATE 113
47 #define FINISH_STATE 666
51 /* Data structure describing a single value and its code string. */
52 typedef struct ct_data_s {
54 ush freq; /* frequency count */
55 ush code; /* bit string */
58 ush dad; /* father node in Huffman tree */
59 ush len; /* length of bit string */
68 typedef struct static_tree_desc_s static_tree_desc;
70 typedef struct tree_desc_s {
71 ct_data *dyn_tree; /* the dynamic tree */
72 int max_code; /* largest code with non zero frequency */
73 static_tree_desc *stat_desc; /* the corresponding static tree */
78 typedef unsigned IPos;
80 /* A Pos is an index in the character window. We use short instead of int to
81 * save space in the various tables. IPos is used only for parameter passing.
84 typedef struct internal_state {
85 z_stream *strm; /* pointer back to this zlib stream */
86 int status; /* as the name implies */
87 Bytef *pending_buf; /* output still pending */
88 Bytef *pending_out; /* next pending byte to output to the stream */
89 int pending; /* nb of bytes in the pending buffer */
90 uLong adler; /* adler32 of uncompressed data */
91 int noheader; /* suppress zlib header and adler32 */
92 Byte data_type; /* UNKNOWN, BINARY or ASCII */
93 Byte method; /* STORED (for zip only) or DEFLATED */
95 /* used by deflate.c: */
97 uInt w_size; /* LZ77 window size (32K by default) */
98 uInt w_bits; /* log2(w_size) (8..16) */
99 uInt w_mask; /* w_size - 1 */
102 /* Sliding window. Input bytes are read into the second half of the window,
103 * and move to the first half later to keep a dictionary of at least wSize
104 * bytes. With this organization, matches are limited to a distance of
105 * wSize-MAX_MATCH bytes, but this ensures that IO is always
106 * performed with a length multiple of the block size. Also, it limits
107 * the window size to 64K, which is quite useful on MSDOS.
108 * To do: use the user input buffer as sliding window.
112 /* Actual size of window: 2*wSize, except when the user input buffer
113 * is directly used as sliding window.
117 /* Link to older string with same hash index. To limit the size of this
118 * array to 64K, this link is maintained only for the last 32K strings.
119 * An index in this array is thus a window index modulo 32K.
122 Posf *head; /* Heads of the hash chains or NIL. */
124 uInt ins_h; /* hash index of string to be inserted */
125 uInt hash_size; /* number of elements in hash table */
126 uInt hash_bits; /* log2(hash_size) */
127 uInt hash_mask; /* hash_size-1 */
130 /* Number of bits by which ins_h must be shifted at each input
131 * step. It must be such that after MIN_MATCH steps, the oldest
132 * byte no longer takes part in the hash key, that is:
133 * hash_shift * MIN_MATCH >= hash_bits
137 /* Window position at the beginning of the current output block. Gets
138 * negative when the window is moved backwards.
141 uInt match_length; /* length of best match */
142 IPos prev_match; /* previous match */
143 int match_available; /* set if previous match exists */
144 uInt strstart; /* start of string to insert */
145 uInt match_start; /* start of matching string */
146 uInt lookahead; /* number of valid bytes ahead in window */
149 /* Length of the best match at previous step. Matches not greater than this
150 * are discarded. This is used in the lazy match evaluation.
153 uInt max_chain_length;
154 /* To speed up deflation, hash chains are never searched beyond this
155 * length. A higher limit improves compression ratio but degrades the
160 /* Attempt to find a better match only when the current match is strictly
161 * smaller than this value. This mechanism is used only for compression
164 # define max_insert_length max_lazy_match
165 /* Insert new strings in the hash table only if the match length is not
166 * greater than this length. This saves time but degrades compression.
167 * max_insert_length is used only for compression levels <= 3.
170 int level; /* compression level (1..9) */
171 int strategy; /* favor or force Huffman coding*/
174 /* Use a faster search when the previous match is longer than this */
176 int nice_match; /* Stop searching when current match exceeds this */
178 /* used by trees.c: */
179 /* Didn't use ct_data typedef below to supress compiler warning */
180 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */
181 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
182 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */
184 struct tree_desc_s l_desc; /* desc. for literal tree */
185 struct tree_desc_s d_desc; /* desc. for distance tree */
186 struct tree_desc_s bl_desc; /* desc. for bit length tree */
188 ush bl_count[MAX_BITS+1];
189 /* number of codes at each bit length for an optimal tree */
191 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */
192 int heap_len; /* number of elements in the heap */
193 int heap_max; /* element of largest frequency */
194 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
195 * The same heap array is used to build all trees.
198 uch depth[2*L_CODES+1];
199 /* Depth of each subtree used as tie breaker for trees of equal frequency
202 uchf *l_buf; /* buffer for literals or lengths */
205 /* Size of match buffer for literals/lengths. There are 4 reasons for
206 * limiting lit_bufsize to 64K:
207 * - frequencies can be kept in 16 bit counters
208 * - if compression is not successful for the first block, all input
209 * data is still in the window so we can still emit a stored block even
210 * when input comes from standard input. (This can also be done for
211 * all blocks if lit_bufsize is not greater than 32K.)
212 * - if compression is not successful for a file smaller than 64K, we can
213 * even emit a stored file instead of a stored block (saving 5 bytes).
214 * This is applicable only for zip (not gzip or zlib).
215 * - creating new Huffman trees less frequently may not provide fast
216 * adaptation to changes in the input data statistics. (Take for
217 * example a binary file with poorly compressible code followed by
218 * a highly compressible string table.) Smaller buffer sizes give
219 * fast adaptation but have of course the overhead of transmitting
220 * trees more frequently.
221 * - I can't count above 4
224 uInt last_lit; /* running index in l_buf */
227 /* Buffer for distances. To simplify the code, d_buf and l_buf have
228 * the same number of elements. To use different lengths, an extra flag
229 * array would be necessary.
232 ulg opt_len; /* bit length of current block with optimal trees */
233 ulg static_len; /* bit length of current block with static trees */
234 ulg compressed_len; /* total bit length of compressed file */
235 uInt matches; /* number of string matches in current block */
238 ulg bits_sent; /* bit length of the compressed data */
242 /* Output buffer. bits are inserted starting at the bottom (least
246 /* Number of valid bits in bi_buf. All bits above the last valid bit
252 /* Output a byte on the stream.
253 * IN assertion: there is enough room in pending_buf.
255 #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
258 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
259 /* Minimum amount of lookahead, except at the end of the input file.
260 * See deflate.c for comments about the MIN_MATCH+1.
263 #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD)
264 /* In order to simplify the code, particularly on 16 bit machines, match
265 * distances are limited to MAX_DIST instead of WSIZE.
269 void ct_init OF((deflate_state *s));
270 int ct_tally OF((deflate_state *s, int dist, int lc));
271 ulg ct_flush_block OF((deflate_state *s, charf *buf, ulg stored_len, int eof));
272 void ct_align OF((deflate_state *s));
273 void ct_stored_block OF((deflate_state *s, charf *buf, ulg stored_len,