michael@0: /* deflate.h -- internal compression state michael@0: * Copyright (C) 1995-2010 Jean-loup Gailly michael@0: * For conditions of distribution and use, see copyright notice in zlib.h michael@0: */ michael@0: michael@0: /* WARNING: this file should *not* be used by applications. It is michael@0: part of the implementation of the compression library and is michael@0: subject to change. Applications should only use zlib.h. michael@0: */ michael@0: michael@0: /* @(#) $Id$ */ michael@0: michael@0: #ifndef DEFLATE_H michael@0: #define DEFLATE_H michael@0: michael@0: #include "zutil.h" michael@0: michael@0: /* define NO_GZIP when compiling if you want to disable gzip header and michael@0: trailer creation by deflate(). NO_GZIP would be used to avoid linking in michael@0: the crc code when it is not needed. For shared libraries, gzip encoding michael@0: should be left enabled. */ michael@0: #ifndef NO_GZIP michael@0: # define GZIP michael@0: #endif michael@0: michael@0: /* =========================================================================== michael@0: * Internal compression state. michael@0: */ michael@0: michael@0: #define LENGTH_CODES 29 michael@0: /* number of length codes, not counting the special END_BLOCK code */ michael@0: michael@0: #define LITERALS 256 michael@0: /* number of literal bytes 0..255 */ michael@0: michael@0: #define L_CODES (LITERALS+1+LENGTH_CODES) michael@0: /* number of Literal or Length codes, including the END_BLOCK code */ michael@0: michael@0: #define D_CODES 30 michael@0: /* number of distance codes */ michael@0: michael@0: #define BL_CODES 19 michael@0: /* number of codes used to transfer the bit lengths */ michael@0: michael@0: #define HEAP_SIZE (2*L_CODES+1) michael@0: /* maximum heap size */ michael@0: michael@0: #define MAX_BITS 15 michael@0: /* All codes must not exceed MAX_BITS bits */ michael@0: michael@0: #define INIT_STATE 42 michael@0: #define EXTRA_STATE 69 michael@0: #define NAME_STATE 73 michael@0: #define COMMENT_STATE 91 michael@0: #define HCRC_STATE 103 michael@0: #define BUSY_STATE 113 michael@0: #define FINISH_STATE 666 michael@0: /* Stream status */ michael@0: michael@0: michael@0: /* Data structure describing a single value and its code string. */ michael@0: typedef struct ct_data_s { michael@0: union { michael@0: ush freq; /* frequency count */ michael@0: ush code; /* bit string */ michael@0: } fc; michael@0: union { michael@0: ush dad; /* father node in Huffman tree */ michael@0: ush len; /* length of bit string */ michael@0: } dl; michael@0: } FAR ct_data; michael@0: michael@0: #define Freq fc.freq michael@0: #define Code fc.code michael@0: #define Dad dl.dad michael@0: #define Len dl.len michael@0: michael@0: typedef struct static_tree_desc_s static_tree_desc; michael@0: michael@0: typedef struct tree_desc_s { michael@0: ct_data *dyn_tree; /* the dynamic tree */ michael@0: int max_code; /* largest code with non zero frequency */ michael@0: static_tree_desc *stat_desc; /* the corresponding static tree */ michael@0: } FAR tree_desc; michael@0: michael@0: typedef ush Pos; michael@0: typedef Pos FAR Posf; michael@0: typedef unsigned IPos; michael@0: michael@0: /* A Pos is an index in the character window. We use short instead of int to michael@0: * save space in the various tables. IPos is used only for parameter passing. michael@0: */ michael@0: michael@0: typedef struct internal_state { michael@0: z_streamp strm; /* pointer back to this zlib stream */ michael@0: int status; /* as the name implies */ michael@0: Bytef *pending_buf; /* output still pending */ michael@0: ulg pending_buf_size; /* size of pending_buf */ michael@0: Bytef *pending_out; /* next pending byte to output to the stream */ michael@0: uInt pending; /* nb of bytes in the pending buffer */ michael@0: int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ michael@0: gz_headerp gzhead; /* gzip header information to write */ michael@0: uInt gzindex; /* where in extra, name, or comment */ michael@0: Byte method; /* STORED (for zip only) or DEFLATED */ michael@0: int last_flush; /* value of flush param for previous deflate call */ michael@0: michael@0: /* used by deflate.c: */ michael@0: michael@0: uInt w_size; /* LZ77 window size (32K by default) */ michael@0: uInt w_bits; /* log2(w_size) (8..16) */ michael@0: uInt w_mask; /* w_size - 1 */ michael@0: michael@0: Bytef *window; michael@0: /* Sliding window. Input bytes are read into the second half of the window, michael@0: * and move to the first half later to keep a dictionary of at least wSize michael@0: * bytes. With this organization, matches are limited to a distance of michael@0: * wSize-MAX_MATCH bytes, but this ensures that IO is always michael@0: * performed with a length multiple of the block size. Also, it limits michael@0: * the window size to 64K, which is quite useful on MSDOS. michael@0: * To do: use the user input buffer as sliding window. michael@0: */ michael@0: michael@0: ulg window_size; michael@0: /* Actual size of window: 2*wSize, except when the user input buffer michael@0: * is directly used as sliding window. michael@0: */ michael@0: michael@0: Posf *prev; michael@0: /* Link to older string with same hash index. To limit the size of this michael@0: * array to 64K, this link is maintained only for the last 32K strings. michael@0: * An index in this array is thus a window index modulo 32K. michael@0: */ michael@0: michael@0: Posf *head; /* Heads of the hash chains or NIL. */ michael@0: michael@0: uInt ins_h; /* hash index of string to be inserted */ michael@0: uInt hash_size; /* number of elements in hash table */ michael@0: uInt hash_bits; /* log2(hash_size) */ michael@0: uInt hash_mask; /* hash_size-1 */ michael@0: michael@0: uInt hash_shift; michael@0: /* Number of bits by which ins_h must be shifted at each input michael@0: * step. It must be such that after MIN_MATCH steps, the oldest michael@0: * byte no longer takes part in the hash key, that is: michael@0: * hash_shift * MIN_MATCH >= hash_bits michael@0: */ michael@0: michael@0: long block_start; michael@0: /* Window position at the beginning of the current output block. Gets michael@0: * negative when the window is moved backwards. michael@0: */ michael@0: michael@0: uInt match_length; /* length of best match */ michael@0: IPos prev_match; /* previous match */ michael@0: int match_available; /* set if previous match exists */ michael@0: uInt strstart; /* start of string to insert */ michael@0: uInt match_start; /* start of matching string */ michael@0: uInt lookahead; /* number of valid bytes ahead in window */ michael@0: michael@0: uInt prev_length; michael@0: /* Length of the best match at previous step. Matches not greater than this michael@0: * are discarded. This is used in the lazy match evaluation. michael@0: */ michael@0: michael@0: uInt max_chain_length; michael@0: /* To speed up deflation, hash chains are never searched beyond this michael@0: * length. A higher limit improves compression ratio but degrades the michael@0: * speed. michael@0: */ michael@0: michael@0: uInt max_lazy_match; michael@0: /* Attempt to find a better match only when the current match is strictly michael@0: * smaller than this value. This mechanism is used only for compression michael@0: * levels >= 4. michael@0: */ michael@0: # define max_insert_length max_lazy_match michael@0: /* Insert new strings in the hash table only if the match length is not michael@0: * greater than this length. This saves time but degrades compression. michael@0: * max_insert_length is used only for compression levels <= 3. michael@0: */ michael@0: michael@0: int level; /* compression level (1..9) */ michael@0: int strategy; /* favor or force Huffman coding*/ michael@0: michael@0: uInt good_match; michael@0: /* Use a faster search when the previous match is longer than this */ michael@0: michael@0: int nice_match; /* Stop searching when current match exceeds this */ michael@0: michael@0: /* used by trees.c: */ michael@0: /* Didn't use ct_data typedef below to supress compiler warning */ michael@0: struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */ michael@0: struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */ michael@0: struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */ michael@0: michael@0: struct tree_desc_s l_desc; /* desc. for literal tree */ michael@0: struct tree_desc_s d_desc; /* desc. for distance tree */ michael@0: struct tree_desc_s bl_desc; /* desc. for bit length tree */ michael@0: michael@0: ush bl_count[MAX_BITS+1]; michael@0: /* number of codes at each bit length for an optimal tree */ michael@0: michael@0: int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */ michael@0: int heap_len; /* number of elements in the heap */ michael@0: int heap_max; /* element of largest frequency */ michael@0: /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. michael@0: * The same heap array is used to build all trees. michael@0: */ michael@0: michael@0: uch depth[2*L_CODES+1]; michael@0: /* Depth of each subtree used as tie breaker for trees of equal frequency michael@0: */ michael@0: michael@0: uchf *l_buf; /* buffer for literals or lengths */ michael@0: michael@0: uInt lit_bufsize; michael@0: /* Size of match buffer for literals/lengths. There are 4 reasons for michael@0: * limiting lit_bufsize to 64K: michael@0: * - frequencies can be kept in 16 bit counters michael@0: * - if compression is not successful for the first block, all input michael@0: * data is still in the window so we can still emit a stored block even michael@0: * when input comes from standard input. (This can also be done for michael@0: * all blocks if lit_bufsize is not greater than 32K.) michael@0: * - if compression is not successful for a file smaller than 64K, we can michael@0: * even emit a stored file instead of a stored block (saving 5 bytes). michael@0: * This is applicable only for zip (not gzip or zlib). michael@0: * - creating new Huffman trees less frequently may not provide fast michael@0: * adaptation to changes in the input data statistics. (Take for michael@0: * example a binary file with poorly compressible code followed by michael@0: * a highly compressible string table.) Smaller buffer sizes give michael@0: * fast adaptation but have of course the overhead of transmitting michael@0: * trees more frequently. michael@0: * - I can't count above 4 michael@0: */ michael@0: michael@0: uInt last_lit; /* running index in l_buf */ michael@0: michael@0: ushf *d_buf; michael@0: /* Buffer for distances. To simplify the code, d_buf and l_buf have michael@0: * the same number of elements. To use different lengths, an extra flag michael@0: * array would be necessary. michael@0: */ michael@0: michael@0: ulg opt_len; /* bit length of current block with optimal trees */ michael@0: ulg static_len; /* bit length of current block with static trees */ michael@0: uInt matches; /* number of string matches in current block */ michael@0: int last_eob_len; /* bit length of EOB code for last block */ michael@0: michael@0: #ifdef DEBUG michael@0: ulg compressed_len; /* total bit length of compressed file mod 2^32 */ michael@0: ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ michael@0: #endif michael@0: michael@0: ush bi_buf; michael@0: /* Output buffer. bits are inserted starting at the bottom (least michael@0: * significant bits). michael@0: */ michael@0: int bi_valid; michael@0: /* Number of valid bits in bi_buf. All bits above the last valid bit michael@0: * are always zero. michael@0: */ michael@0: michael@0: ulg high_water; michael@0: /* High water mark offset in window for initialized bytes -- bytes above michael@0: * this are set to zero in order to avoid memory check warnings when michael@0: * longest match routines access bytes past the input. This is then michael@0: * updated to the new high water mark. michael@0: */ michael@0: michael@0: } FAR deflate_state; michael@0: michael@0: /* Output a byte on the stream. michael@0: * IN assertion: there is enough room in pending_buf. michael@0: */ michael@0: #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);} michael@0: michael@0: michael@0: #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) michael@0: /* Minimum amount of lookahead, except at the end of the input file. michael@0: * See deflate.c for comments about the MIN_MATCH+1. michael@0: */ michael@0: michael@0: #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD) michael@0: /* In order to simplify the code, particularly on 16 bit machines, match michael@0: * distances are limited to MAX_DIST instead of WSIZE. michael@0: */ michael@0: michael@0: #define WIN_INIT MAX_MATCH michael@0: /* Number of bytes after end of data in window to initialize in order to avoid michael@0: memory checker errors from longest match routines */ michael@0: michael@0: /* in trees.c */ michael@0: void ZLIB_INTERNAL _tr_init OF((deflate_state *s)); michael@0: int ZLIB_INTERNAL _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc)); michael@0: void ZLIB_INTERNAL _tr_flush_block OF((deflate_state *s, charf *buf, michael@0: ulg stored_len, int last)); michael@0: void ZLIB_INTERNAL _tr_align OF((deflate_state *s)); michael@0: void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf, michael@0: ulg stored_len, int last)); michael@0: michael@0: #define d_code(dist) \ michael@0: ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) michael@0: /* Mapping from a distance to a distance code. dist is the distance - 1 and michael@0: * must not have side effects. _dist_code[256] and _dist_code[257] are never michael@0: * used. michael@0: */ michael@0: michael@0: #ifndef DEBUG michael@0: /* Inline versions of _tr_tally for speed: */ michael@0: michael@0: #if defined(GEN_TREES_H) || !defined(STDC) michael@0: extern uch ZLIB_INTERNAL _length_code[]; michael@0: extern uch ZLIB_INTERNAL _dist_code[]; michael@0: #else michael@0: extern const uch ZLIB_INTERNAL _length_code[]; michael@0: extern const uch ZLIB_INTERNAL _dist_code[]; michael@0: #endif michael@0: michael@0: # define _tr_tally_lit(s, c, flush) \ michael@0: { uch cc = (c); \ michael@0: s->d_buf[s->last_lit] = 0; \ michael@0: s->l_buf[s->last_lit++] = cc; \ michael@0: s->dyn_ltree[cc].Freq++; \ michael@0: flush = (s->last_lit == s->lit_bufsize-1); \ michael@0: } michael@0: # define _tr_tally_dist(s, distance, length, flush) \ michael@0: { uch len = (length); \ michael@0: ush dist = (distance); \ michael@0: s->d_buf[s->last_lit] = dist; \ michael@0: s->l_buf[s->last_lit++] = len; \ michael@0: dist--; \ michael@0: s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ michael@0: s->dyn_dtree[d_code(dist)].Freq++; \ michael@0: flush = (s->last_lit == s->lit_bufsize-1); \ michael@0: } michael@0: #else michael@0: # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c) michael@0: # define _tr_tally_dist(s, distance, length, flush) \ michael@0: flush = _tr_tally(s, distance, length) michael@0: #endif michael@0: michael@0: #endif /* DEFLATE_H */