cannam@89: /* deflate.c -- compress data using the deflation algorithm cannam@89: * Copyright (C) 1995-2012 Jean-loup Gailly and Mark Adler cannam@89: * For conditions of distribution and use, see copyright notice in zlib.h cannam@89: */ cannam@89: cannam@89: /* cannam@89: * ALGORITHM cannam@89: * cannam@89: * The "deflation" process depends on being able to identify portions cannam@89: * of the input text which are identical to earlier input (within a cannam@89: * sliding window trailing behind the input currently being processed). cannam@89: * cannam@89: * The most straightforward technique turns out to be the fastest for cannam@89: * most input files: try all possible matches and select the longest. cannam@89: * The key feature of this algorithm is that insertions into the string cannam@89: * dictionary are very simple and thus fast, and deletions are avoided cannam@89: * completely. Insertions are performed at each input character, whereas cannam@89: * string matches are performed only when the previous match ends. So it cannam@89: * is preferable to spend more time in matches to allow very fast string cannam@89: * insertions and avoid deletions. The matching algorithm for small cannam@89: * strings is inspired from that of Rabin & Karp. A brute force approach cannam@89: * is used to find longer strings when a small match has been found. cannam@89: * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze cannam@89: * (by Leonid Broukhis). cannam@89: * A previous version of this file used a more sophisticated algorithm cannam@89: * (by Fiala and Greene) which is guaranteed to run in linear amortized cannam@89: * time, but has a larger average cost, uses more memory and is patented. cannam@89: * However the F&G algorithm may be faster for some highly redundant cannam@89: * files if the parameter max_chain_length (described below) is too large. cannam@89: * cannam@89: * ACKNOWLEDGEMENTS cannam@89: * cannam@89: * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and cannam@89: * I found it in 'freeze' written by Leonid Broukhis. cannam@89: * Thanks to many people for bug reports and testing. cannam@89: * cannam@89: * REFERENCES cannam@89: * cannam@89: * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification". cannam@89: * Available in http://tools.ietf.org/html/rfc1951 cannam@89: * cannam@89: * A description of the Rabin and Karp algorithm is given in the book cannam@89: * "Algorithms" by R. Sedgewick, Addison-Wesley, p252. cannam@89: * cannam@89: * Fiala,E.R., and Greene,D.H. cannam@89: * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595 cannam@89: * cannam@89: */ cannam@89: cannam@89: /* @(#) $Id$ */ cannam@89: cannam@89: #include "deflate.h" cannam@89: cannam@89: const char deflate_copyright[] = cannam@89: " deflate 1.2.7 Copyright 1995-2012 Jean-loup Gailly and Mark Adler "; cannam@89: /* cannam@89: If you use the zlib library in a product, an acknowledgment is welcome cannam@89: in the documentation of your product. If for some reason you cannot cannam@89: include such an acknowledgment, I would appreciate that you keep this cannam@89: copyright string in the executable of your product. cannam@89: */ cannam@89: cannam@89: /* =========================================================================== cannam@89: * Function prototypes. cannam@89: */ cannam@89: typedef enum { cannam@89: need_more, /* block not completed, need more input or more output */ cannam@89: block_done, /* block flush performed */ cannam@89: finish_started, /* finish started, need only more output at next deflate */ cannam@89: finish_done /* finish done, accept no more input or output */ cannam@89: } block_state; cannam@89: cannam@89: typedef block_state (*compress_func) OF((deflate_state *s, int flush)); cannam@89: /* Compression function. Returns the block state after the call. */ cannam@89: cannam@89: local void fill_window OF((deflate_state *s)); cannam@89: local block_state deflate_stored OF((deflate_state *s, int flush)); cannam@89: local block_state deflate_fast OF((deflate_state *s, int flush)); cannam@89: #ifndef FASTEST cannam@89: local block_state deflate_slow OF((deflate_state *s, int flush)); cannam@89: #endif cannam@89: local block_state deflate_rle OF((deflate_state *s, int flush)); cannam@89: local block_state deflate_huff OF((deflate_state *s, int flush)); cannam@89: local void lm_init OF((deflate_state *s)); cannam@89: local void putShortMSB OF((deflate_state *s, uInt b)); cannam@89: local void flush_pending OF((z_streamp strm)); cannam@89: local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); cannam@89: #ifdef ASMV cannam@89: void match_init OF((void)); /* asm code initialization */ cannam@89: uInt longest_match OF((deflate_state *s, IPos cur_match)); cannam@89: #else cannam@89: local uInt longest_match OF((deflate_state *s, IPos cur_match)); cannam@89: #endif cannam@89: cannam@89: #ifdef DEBUG cannam@89: local void check_match OF((deflate_state *s, IPos start, IPos match, cannam@89: int length)); cannam@89: #endif cannam@89: cannam@89: /* =========================================================================== cannam@89: * Local data cannam@89: */ cannam@89: cannam@89: #define NIL 0 cannam@89: /* Tail of hash chains */ cannam@89: cannam@89: #ifndef TOO_FAR cannam@89: # define TOO_FAR 4096 cannam@89: #endif cannam@89: /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ cannam@89: cannam@89: /* Values for max_lazy_match, good_match and max_chain_length, depending on cannam@89: * the desired pack level (0..9). The values given below have been tuned to cannam@89: * exclude worst case performance for pathological files. Better values may be cannam@89: * found for specific files. cannam@89: */ cannam@89: typedef struct config_s { cannam@89: ush good_length; /* reduce lazy search above this match length */ cannam@89: ush max_lazy; /* do not perform lazy search above this match length */ cannam@89: ush nice_length; /* quit search above this match length */ cannam@89: ush max_chain; cannam@89: compress_func func; cannam@89: } config; cannam@89: cannam@89: #ifdef FASTEST cannam@89: local const config configuration_table[2] = { cannam@89: /* good lazy nice chain */ cannam@89: /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ cannam@89: /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */ cannam@89: #else cannam@89: local const config configuration_table[10] = { cannam@89: /* good lazy nice chain */ cannam@89: /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ cannam@89: /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */ cannam@89: /* 2 */ {4, 5, 16, 8, deflate_fast}, cannam@89: /* 3 */ {4, 6, 32, 32, deflate_fast}, cannam@89: cannam@89: /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ cannam@89: /* 5 */ {8, 16, 32, 32, deflate_slow}, cannam@89: /* 6 */ {8, 16, 128, 128, deflate_slow}, cannam@89: /* 7 */ {8, 32, 128, 256, deflate_slow}, cannam@89: /* 8 */ {32, 128, 258, 1024, deflate_slow}, cannam@89: /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */ cannam@89: #endif cannam@89: cannam@89: /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 cannam@89: * For deflate_fast() (levels <= 3) good is ignored and lazy has a different cannam@89: * meaning. cannam@89: */ cannam@89: cannam@89: #define EQUAL 0 cannam@89: /* result of memcmp for equal strings */ cannam@89: cannam@89: #ifndef NO_DUMMY_DECL cannam@89: struct static_tree_desc_s {int dummy;}; /* for buggy compilers */ cannam@89: #endif cannam@89: cannam@89: /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */ cannam@89: #define RANK(f) (((f) << 1) - ((f) > 4 ? 9 : 0)) cannam@89: cannam@89: /* =========================================================================== cannam@89: * Update a hash value with the given input byte cannam@89: * IN assertion: all calls to to UPDATE_HASH are made with consecutive cannam@89: * input characters, so that a running hash key can be computed from the cannam@89: * previous key instead of complete recalculation each time. cannam@89: */ cannam@89: #define UPDATE_HASH(s,h,c) (h = (((h)<hash_shift) ^ (c)) & s->hash_mask) cannam@89: cannam@89: cannam@89: /* =========================================================================== cannam@89: * Insert string str in the dictionary and set match_head to the previous head cannam@89: * of the hash chain (the most recent string with same hash key). Return cannam@89: * the previous length of the hash chain. cannam@89: * If this file is compiled with -DFASTEST, the compression level is forced cannam@89: * to 1, and no hash chains are maintained. cannam@89: * IN assertion: all calls to to INSERT_STRING are made with consecutive cannam@89: * input characters and the first MIN_MATCH bytes of str are valid cannam@89: * (except for the last MIN_MATCH-1 bytes of the input file). cannam@89: */ cannam@89: #ifdef FASTEST cannam@89: #define INSERT_STRING(s, str, match_head) \ cannam@89: (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ cannam@89: match_head = s->head[s->ins_h], \ cannam@89: s->head[s->ins_h] = (Pos)(str)) cannam@89: #else cannam@89: #define INSERT_STRING(s, str, match_head) \ cannam@89: (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ cannam@89: match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \ cannam@89: s->head[s->ins_h] = (Pos)(str)) cannam@89: #endif cannam@89: cannam@89: /* =========================================================================== cannam@89: * Initialize the hash table (avoiding 64K overflow for 16 bit systems). cannam@89: * prev[] will be initialized on the fly. cannam@89: */ cannam@89: #define CLEAR_HASH(s) \ cannam@89: s->head[s->hash_size-1] = NIL; \ cannam@89: zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head)); cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateInit_(strm, level, version, stream_size) cannam@89: z_streamp strm; cannam@89: int level; cannam@89: const char *version; cannam@89: int stream_size; cannam@89: { cannam@89: return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, cannam@89: Z_DEFAULT_STRATEGY, version, stream_size); cannam@89: /* To do: ignore strm->next_in if we use it as window */ cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, cannam@89: version, stream_size) cannam@89: z_streamp strm; cannam@89: int level; cannam@89: int method; cannam@89: int windowBits; cannam@89: int memLevel; cannam@89: int strategy; cannam@89: const char *version; cannam@89: int stream_size; cannam@89: { cannam@89: deflate_state *s; cannam@89: int wrap = 1; cannam@89: static const char my_version[] = ZLIB_VERSION; cannam@89: cannam@89: ushf *overlay; cannam@89: /* We overlay pending_buf and d_buf+l_buf. This works since the average cannam@89: * output size for (length,distance) codes is <= 24 bits. cannam@89: */ cannam@89: cannam@89: if (version == Z_NULL || version[0] != my_version[0] || cannam@89: stream_size != sizeof(z_stream)) { cannam@89: return Z_VERSION_ERROR; cannam@89: } cannam@89: if (strm == Z_NULL) return Z_STREAM_ERROR; cannam@89: cannam@89: strm->msg = Z_NULL; cannam@89: if (strm->zalloc == (alloc_func)0) { cannam@89: #ifdef Z_SOLO cannam@89: return Z_STREAM_ERROR; cannam@89: #else cannam@89: strm->zalloc = zcalloc; cannam@89: strm->opaque = (voidpf)0; cannam@89: #endif cannam@89: } cannam@89: if (strm->zfree == (free_func)0) cannam@89: #ifdef Z_SOLO cannam@89: return Z_STREAM_ERROR; cannam@89: #else cannam@89: strm->zfree = zcfree; cannam@89: #endif cannam@89: cannam@89: #ifdef FASTEST cannam@89: if (level != 0) level = 1; cannam@89: #else cannam@89: if (level == Z_DEFAULT_COMPRESSION) level = 6; cannam@89: #endif cannam@89: cannam@89: if (windowBits < 0) { /* suppress zlib wrapper */ cannam@89: wrap = 0; cannam@89: windowBits = -windowBits; cannam@89: } cannam@89: #ifdef GZIP cannam@89: else if (windowBits > 15) { cannam@89: wrap = 2; /* write gzip wrapper instead */ cannam@89: windowBits -= 16; cannam@89: } cannam@89: #endif cannam@89: if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || cannam@89: windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || cannam@89: strategy < 0 || strategy > Z_FIXED) { cannam@89: return Z_STREAM_ERROR; cannam@89: } cannam@89: if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ cannam@89: s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); cannam@89: if (s == Z_NULL) return Z_MEM_ERROR; cannam@89: strm->state = (struct internal_state FAR *)s; cannam@89: s->strm = strm; cannam@89: cannam@89: s->wrap = wrap; cannam@89: s->gzhead = Z_NULL; cannam@89: s->w_bits = windowBits; cannam@89: s->w_size = 1 << s->w_bits; cannam@89: s->w_mask = s->w_size - 1; cannam@89: cannam@89: s->hash_bits = memLevel + 7; cannam@89: s->hash_size = 1 << s->hash_bits; cannam@89: s->hash_mask = s->hash_size - 1; cannam@89: s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); cannam@89: cannam@89: s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); cannam@89: s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); cannam@89: s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); cannam@89: cannam@89: s->high_water = 0; /* nothing written to s->window yet */ cannam@89: cannam@89: s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ cannam@89: cannam@89: overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2); cannam@89: s->pending_buf = (uchf *) overlay; cannam@89: s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); cannam@89: cannam@89: if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || cannam@89: s->pending_buf == Z_NULL) { cannam@89: s->status = FINISH_STATE; cannam@89: strm->msg = (char*)ERR_MSG(Z_MEM_ERROR); cannam@89: deflateEnd (strm); cannam@89: return Z_MEM_ERROR; cannam@89: } cannam@89: s->d_buf = overlay + s->lit_bufsize/sizeof(ush); cannam@89: s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; cannam@89: cannam@89: s->level = level; cannam@89: s->strategy = strategy; cannam@89: s->method = (Byte)method; cannam@89: cannam@89: return deflateReset(strm); cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength) cannam@89: z_streamp strm; cannam@89: const Bytef *dictionary; cannam@89: uInt dictLength; cannam@89: { cannam@89: deflate_state *s; cannam@89: uInt str, n; cannam@89: int wrap; cannam@89: unsigned avail; cannam@89: unsigned char *next; cannam@89: cannam@89: if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL) cannam@89: return Z_STREAM_ERROR; cannam@89: s = strm->state; cannam@89: wrap = s->wrap; cannam@89: if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead) cannam@89: return Z_STREAM_ERROR; cannam@89: cannam@89: /* when using zlib wrappers, compute Adler-32 for provided dictionary */ cannam@89: if (wrap == 1) cannam@89: strm->adler = adler32(strm->adler, dictionary, dictLength); cannam@89: s->wrap = 0; /* avoid computing Adler-32 in read_buf */ cannam@89: cannam@89: /* if dictionary would fill window, just replace the history */ cannam@89: if (dictLength >= s->w_size) { cannam@89: if (wrap == 0) { /* already empty otherwise */ cannam@89: CLEAR_HASH(s); cannam@89: s->strstart = 0; cannam@89: s->block_start = 0L; cannam@89: s->insert = 0; cannam@89: } cannam@89: dictionary += dictLength - s->w_size; /* use the tail */ cannam@89: dictLength = s->w_size; cannam@89: } cannam@89: cannam@89: /* insert dictionary into window and hash */ cannam@89: avail = strm->avail_in; cannam@89: next = strm->next_in; cannam@89: strm->avail_in = dictLength; cannam@89: strm->next_in = (Bytef *)dictionary; cannam@89: fill_window(s); cannam@89: while (s->lookahead >= MIN_MATCH) { cannam@89: str = s->strstart; cannam@89: n = s->lookahead - (MIN_MATCH-1); cannam@89: do { cannam@89: UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); cannam@89: #ifndef FASTEST cannam@89: s->prev[str & s->w_mask] = s->head[s->ins_h]; cannam@89: #endif cannam@89: s->head[s->ins_h] = (Pos)str; cannam@89: str++; cannam@89: } while (--n); cannam@89: s->strstart = str; cannam@89: s->lookahead = MIN_MATCH-1; cannam@89: fill_window(s); cannam@89: } cannam@89: s->strstart += s->lookahead; cannam@89: s->block_start = (long)s->strstart; cannam@89: s->insert = s->lookahead; cannam@89: s->lookahead = 0; cannam@89: s->match_length = s->prev_length = MIN_MATCH-1; cannam@89: s->match_available = 0; cannam@89: strm->next_in = next; cannam@89: strm->avail_in = avail; cannam@89: s->wrap = wrap; cannam@89: return Z_OK; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateResetKeep (strm) cannam@89: z_streamp strm; cannam@89: { cannam@89: deflate_state *s; cannam@89: cannam@89: if (strm == Z_NULL || strm->state == Z_NULL || cannam@89: strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) { cannam@89: return Z_STREAM_ERROR; cannam@89: } cannam@89: cannam@89: strm->total_in = strm->total_out = 0; cannam@89: strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ cannam@89: strm->data_type = Z_UNKNOWN; cannam@89: cannam@89: s = (deflate_state *)strm->state; cannam@89: s->pending = 0; cannam@89: s->pending_out = s->pending_buf; cannam@89: cannam@89: if (s->wrap < 0) { cannam@89: s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ cannam@89: } cannam@89: s->status = s->wrap ? INIT_STATE : BUSY_STATE; cannam@89: strm->adler = cannam@89: #ifdef GZIP cannam@89: s->wrap == 2 ? crc32(0L, Z_NULL, 0) : cannam@89: #endif cannam@89: adler32(0L, Z_NULL, 0); cannam@89: s->last_flush = Z_NO_FLUSH; cannam@89: cannam@89: _tr_init(s); cannam@89: cannam@89: return Z_OK; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateReset (strm) cannam@89: z_streamp strm; cannam@89: { cannam@89: int ret; cannam@89: cannam@89: ret = deflateResetKeep(strm); cannam@89: if (ret == Z_OK) cannam@89: lm_init(strm->state); cannam@89: return ret; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateSetHeader (strm, head) cannam@89: z_streamp strm; cannam@89: gz_headerp head; cannam@89: { cannam@89: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; cannam@89: if (strm->state->wrap != 2) return Z_STREAM_ERROR; cannam@89: strm->state->gzhead = head; cannam@89: return Z_OK; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflatePending (strm, pending, bits) cannam@89: unsigned *pending; cannam@89: int *bits; cannam@89: z_streamp strm; cannam@89: { cannam@89: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; cannam@89: if (pending != Z_NULL) cannam@89: *pending = strm->state->pending; cannam@89: if (bits != Z_NULL) cannam@89: *bits = strm->state->bi_valid; cannam@89: return Z_OK; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflatePrime (strm, bits, value) cannam@89: z_streamp strm; cannam@89: int bits; cannam@89: int value; cannam@89: { cannam@89: deflate_state *s; cannam@89: int put; cannam@89: cannam@89: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; cannam@89: s = strm->state; cannam@89: if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3)) cannam@89: return Z_BUF_ERROR; cannam@89: do { cannam@89: put = Buf_size - s->bi_valid; cannam@89: if (put > bits) cannam@89: put = bits; cannam@89: s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid); cannam@89: s->bi_valid += put; cannam@89: _tr_flush_bits(s); cannam@89: value >>= put; cannam@89: bits -= put; cannam@89: } while (bits); cannam@89: return Z_OK; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateParams(strm, level, strategy) cannam@89: z_streamp strm; cannam@89: int level; cannam@89: int strategy; cannam@89: { cannam@89: deflate_state *s; cannam@89: compress_func func; cannam@89: int err = Z_OK; cannam@89: cannam@89: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; cannam@89: s = strm->state; cannam@89: cannam@89: #ifdef FASTEST cannam@89: if (level != 0) level = 1; cannam@89: #else cannam@89: if (level == Z_DEFAULT_COMPRESSION) level = 6; cannam@89: #endif cannam@89: if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { cannam@89: return Z_STREAM_ERROR; cannam@89: } cannam@89: func = configuration_table[s->level].func; cannam@89: cannam@89: if ((strategy != s->strategy || func != configuration_table[level].func) && cannam@89: strm->total_in != 0) { cannam@89: /* Flush the last buffer: */ cannam@89: err = deflate(strm, Z_BLOCK); cannam@89: } cannam@89: if (s->level != level) { cannam@89: s->level = level; cannam@89: s->max_lazy_match = configuration_table[level].max_lazy; cannam@89: s->good_match = configuration_table[level].good_length; cannam@89: s->nice_match = configuration_table[level].nice_length; cannam@89: s->max_chain_length = configuration_table[level].max_chain; cannam@89: } cannam@89: s->strategy = strategy; cannam@89: return err; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain) cannam@89: z_streamp strm; cannam@89: int good_length; cannam@89: int max_lazy; cannam@89: int nice_length; cannam@89: int max_chain; cannam@89: { cannam@89: deflate_state *s; cannam@89: cannam@89: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; cannam@89: s = strm->state; cannam@89: s->good_match = good_length; cannam@89: s->max_lazy_match = max_lazy; cannam@89: s->nice_match = nice_length; cannam@89: s->max_chain_length = max_chain; cannam@89: return Z_OK; cannam@89: } cannam@89: cannam@89: /* ========================================================================= cannam@89: * For the default windowBits of 15 and memLevel of 8, this function returns cannam@89: * a close to exact, as well as small, upper bound on the compressed size. cannam@89: * They are coded as constants here for a reason--if the #define's are cannam@89: * changed, then this function needs to be changed as well. The return cannam@89: * value for 15 and 8 only works for those exact settings. cannam@89: * cannam@89: * For any setting other than those defaults for windowBits and memLevel, cannam@89: * the value returned is a conservative worst case for the maximum expansion cannam@89: * resulting from using fixed blocks instead of stored blocks, which deflate cannam@89: * can emit on compressed data for some combinations of the parameters. cannam@89: * cannam@89: * This function could be more sophisticated to provide closer upper bounds for cannam@89: * every combination of windowBits and memLevel. But even the conservative cannam@89: * upper bound of about 14% expansion does not seem onerous for output buffer cannam@89: * allocation. cannam@89: */ cannam@89: uLong ZEXPORT deflateBound(strm, sourceLen) cannam@89: z_streamp strm; cannam@89: uLong sourceLen; cannam@89: { cannam@89: deflate_state *s; cannam@89: uLong complen, wraplen; cannam@89: Bytef *str; cannam@89: cannam@89: /* conservative upper bound for compressed data */ cannam@89: complen = sourceLen + cannam@89: ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5; cannam@89: cannam@89: /* if can't get parameters, return conservative bound plus zlib wrapper */ cannam@89: if (strm == Z_NULL || strm->state == Z_NULL) cannam@89: return complen + 6; cannam@89: cannam@89: /* compute wrapper length */ cannam@89: s = strm->state; cannam@89: switch (s->wrap) { cannam@89: case 0: /* raw deflate */ cannam@89: wraplen = 0; cannam@89: break; cannam@89: case 1: /* zlib wrapper */ cannam@89: wraplen = 6 + (s->strstart ? 4 : 0); cannam@89: break; cannam@89: case 2: /* gzip wrapper */ cannam@89: wraplen = 18; cannam@89: if (s->gzhead != Z_NULL) { /* user-supplied gzip header */ cannam@89: if (s->gzhead->extra != Z_NULL) cannam@89: wraplen += 2 + s->gzhead->extra_len; cannam@89: str = s->gzhead->name; cannam@89: if (str != Z_NULL) cannam@89: do { cannam@89: wraplen++; cannam@89: } while (*str++); cannam@89: str = s->gzhead->comment; cannam@89: if (str != Z_NULL) cannam@89: do { cannam@89: wraplen++; cannam@89: } while (*str++); cannam@89: if (s->gzhead->hcrc) cannam@89: wraplen += 2; cannam@89: } cannam@89: break; cannam@89: default: /* for compiler happiness */ cannam@89: wraplen = 6; cannam@89: } cannam@89: cannam@89: /* if not default parameters, return conservative bound */ cannam@89: if (s->w_bits != 15 || s->hash_bits != 8 + 7) cannam@89: return complen + wraplen; cannam@89: cannam@89: /* default settings: return tight bound for that case */ cannam@89: return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) + cannam@89: (sourceLen >> 25) + 13 - 6 + wraplen; cannam@89: } cannam@89: cannam@89: /* ========================================================================= cannam@89: * Put a short in the pending buffer. The 16-bit value is put in MSB order. cannam@89: * IN assertion: the stream state is correct and there is enough room in cannam@89: * pending_buf. cannam@89: */ cannam@89: local void putShortMSB (s, b) cannam@89: deflate_state *s; cannam@89: uInt b; cannam@89: { cannam@89: put_byte(s, (Byte)(b >> 8)); cannam@89: put_byte(s, (Byte)(b & 0xff)); cannam@89: } cannam@89: cannam@89: /* ========================================================================= cannam@89: * Flush as much pending output as possible. All deflate() output goes cannam@89: * through this function so some applications may wish to modify it cannam@89: * to avoid allocating a large strm->next_out buffer and copying into it. cannam@89: * (See also read_buf()). cannam@89: */ cannam@89: local void flush_pending(strm) cannam@89: z_streamp strm; cannam@89: { cannam@89: unsigned len; cannam@89: deflate_state *s = strm->state; cannam@89: cannam@89: _tr_flush_bits(s); cannam@89: len = s->pending; cannam@89: if (len > strm->avail_out) len = strm->avail_out; cannam@89: if (len == 0) return; cannam@89: cannam@89: zmemcpy(strm->next_out, s->pending_out, len); cannam@89: strm->next_out += len; cannam@89: s->pending_out += len; cannam@89: strm->total_out += len; cannam@89: strm->avail_out -= len; cannam@89: s->pending -= len; cannam@89: if (s->pending == 0) { cannam@89: s->pending_out = s->pending_buf; cannam@89: } cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflate (strm, flush) cannam@89: z_streamp strm; cannam@89: int flush; cannam@89: { cannam@89: int old_flush; /* value of flush param for previous deflate call */ cannam@89: deflate_state *s; cannam@89: cannam@89: if (strm == Z_NULL || strm->state == Z_NULL || cannam@89: flush > Z_BLOCK || flush < 0) { cannam@89: return Z_STREAM_ERROR; cannam@89: } cannam@89: s = strm->state; cannam@89: cannam@89: if (strm->next_out == Z_NULL || cannam@89: (strm->next_in == Z_NULL && strm->avail_in != 0) || cannam@89: (s->status == FINISH_STATE && flush != Z_FINISH)) { cannam@89: ERR_RETURN(strm, Z_STREAM_ERROR); cannam@89: } cannam@89: if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); cannam@89: cannam@89: s->strm = strm; /* just in case */ cannam@89: old_flush = s->last_flush; cannam@89: s->last_flush = flush; cannam@89: cannam@89: /* Write the header */ cannam@89: if (s->status == INIT_STATE) { cannam@89: #ifdef GZIP cannam@89: if (s->wrap == 2) { cannam@89: strm->adler = crc32(0L, Z_NULL, 0); cannam@89: put_byte(s, 31); cannam@89: put_byte(s, 139); cannam@89: put_byte(s, 8); cannam@89: if (s->gzhead == Z_NULL) { cannam@89: put_byte(s, 0); cannam@89: put_byte(s, 0); cannam@89: put_byte(s, 0); cannam@89: put_byte(s, 0); cannam@89: put_byte(s, 0); cannam@89: put_byte(s, s->level == 9 ? 2 : cannam@89: (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? cannam@89: 4 : 0)); cannam@89: put_byte(s, OS_CODE); cannam@89: s->status = BUSY_STATE; cannam@89: } cannam@89: else { cannam@89: put_byte(s, (s->gzhead->text ? 1 : 0) + cannam@89: (s->gzhead->hcrc ? 2 : 0) + cannam@89: (s->gzhead->extra == Z_NULL ? 0 : 4) + cannam@89: (s->gzhead->name == Z_NULL ? 0 : 8) + cannam@89: (s->gzhead->comment == Z_NULL ? 0 : 16) cannam@89: ); cannam@89: put_byte(s, (Byte)(s->gzhead->time & 0xff)); cannam@89: put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); cannam@89: put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); cannam@89: put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); cannam@89: put_byte(s, s->level == 9 ? 2 : cannam@89: (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? cannam@89: 4 : 0)); cannam@89: put_byte(s, s->gzhead->os & 0xff); cannam@89: if (s->gzhead->extra != Z_NULL) { cannam@89: put_byte(s, s->gzhead->extra_len & 0xff); cannam@89: put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); cannam@89: } cannam@89: if (s->gzhead->hcrc) cannam@89: strm->adler = crc32(strm->adler, s->pending_buf, cannam@89: s->pending); cannam@89: s->gzindex = 0; cannam@89: s->status = EXTRA_STATE; cannam@89: } cannam@89: } cannam@89: else cannam@89: #endif cannam@89: { cannam@89: uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; cannam@89: uInt level_flags; cannam@89: cannam@89: if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) cannam@89: level_flags = 0; cannam@89: else if (s->level < 6) cannam@89: level_flags = 1; cannam@89: else if (s->level == 6) cannam@89: level_flags = 2; cannam@89: else cannam@89: level_flags = 3; cannam@89: header |= (level_flags << 6); cannam@89: if (s->strstart != 0) header |= PRESET_DICT; cannam@89: header += 31 - (header % 31); cannam@89: cannam@89: s->status = BUSY_STATE; cannam@89: putShortMSB(s, header); cannam@89: cannam@89: /* Save the adler32 of the preset dictionary: */ cannam@89: if (s->strstart != 0) { cannam@89: putShortMSB(s, (uInt)(strm->adler >> 16)); cannam@89: putShortMSB(s, (uInt)(strm->adler & 0xffff)); cannam@89: } cannam@89: strm->adler = adler32(0L, Z_NULL, 0); cannam@89: } cannam@89: } cannam@89: #ifdef GZIP cannam@89: if (s->status == EXTRA_STATE) { cannam@89: if (s->gzhead->extra != Z_NULL) { cannam@89: uInt beg = s->pending; /* start of bytes to update crc */ cannam@89: cannam@89: while (s->gzindex < (s->gzhead->extra_len & 0xffff)) { cannam@89: if (s->pending == s->pending_buf_size) { cannam@89: if (s->gzhead->hcrc && s->pending > beg) cannam@89: strm->adler = crc32(strm->adler, s->pending_buf + beg, cannam@89: s->pending - beg); cannam@89: flush_pending(strm); cannam@89: beg = s->pending; cannam@89: if (s->pending == s->pending_buf_size) cannam@89: break; cannam@89: } cannam@89: put_byte(s, s->gzhead->extra[s->gzindex]); cannam@89: s->gzindex++; cannam@89: } cannam@89: if (s->gzhead->hcrc && s->pending > beg) cannam@89: strm->adler = crc32(strm->adler, s->pending_buf + beg, cannam@89: s->pending - beg); cannam@89: if (s->gzindex == s->gzhead->extra_len) { cannam@89: s->gzindex = 0; cannam@89: s->status = NAME_STATE; cannam@89: } cannam@89: } cannam@89: else cannam@89: s->status = NAME_STATE; cannam@89: } cannam@89: if (s->status == NAME_STATE) { cannam@89: if (s->gzhead->name != Z_NULL) { cannam@89: uInt beg = s->pending; /* start of bytes to update crc */ cannam@89: int val; cannam@89: cannam@89: do { cannam@89: if (s->pending == s->pending_buf_size) { cannam@89: if (s->gzhead->hcrc && s->pending > beg) cannam@89: strm->adler = crc32(strm->adler, s->pending_buf + beg, cannam@89: s->pending - beg); cannam@89: flush_pending(strm); cannam@89: beg = s->pending; cannam@89: if (s->pending == s->pending_buf_size) { cannam@89: val = 1; cannam@89: break; cannam@89: } cannam@89: } cannam@89: val = s->gzhead->name[s->gzindex++]; cannam@89: put_byte(s, val); cannam@89: } while (val != 0); cannam@89: if (s->gzhead->hcrc && s->pending > beg) cannam@89: strm->adler = crc32(strm->adler, s->pending_buf + beg, cannam@89: s->pending - beg); cannam@89: if (val == 0) { cannam@89: s->gzindex = 0; cannam@89: s->status = COMMENT_STATE; cannam@89: } cannam@89: } cannam@89: else cannam@89: s->status = COMMENT_STATE; cannam@89: } cannam@89: if (s->status == COMMENT_STATE) { cannam@89: if (s->gzhead->comment != Z_NULL) { cannam@89: uInt beg = s->pending; /* start of bytes to update crc */ cannam@89: int val; cannam@89: cannam@89: do { cannam@89: if (s->pending == s->pending_buf_size) { cannam@89: if (s->gzhead->hcrc && s->pending > beg) cannam@89: strm->adler = crc32(strm->adler, s->pending_buf + beg, cannam@89: s->pending - beg); cannam@89: flush_pending(strm); cannam@89: beg = s->pending; cannam@89: if (s->pending == s->pending_buf_size) { cannam@89: val = 1; cannam@89: break; cannam@89: } cannam@89: } cannam@89: val = s->gzhead->comment[s->gzindex++]; cannam@89: put_byte(s, val); cannam@89: } while (val != 0); cannam@89: if (s->gzhead->hcrc && s->pending > beg) cannam@89: strm->adler = crc32(strm->adler, s->pending_buf + beg, cannam@89: s->pending - beg); cannam@89: if (val == 0) cannam@89: s->status = HCRC_STATE; cannam@89: } cannam@89: else cannam@89: s->status = HCRC_STATE; cannam@89: } cannam@89: if (s->status == HCRC_STATE) { cannam@89: if (s->gzhead->hcrc) { cannam@89: if (s->pending + 2 > s->pending_buf_size) cannam@89: flush_pending(strm); cannam@89: if (s->pending + 2 <= s->pending_buf_size) { cannam@89: put_byte(s, (Byte)(strm->adler & 0xff)); cannam@89: put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); cannam@89: strm->adler = crc32(0L, Z_NULL, 0); cannam@89: s->status = BUSY_STATE; cannam@89: } cannam@89: } cannam@89: else cannam@89: s->status = BUSY_STATE; cannam@89: } cannam@89: #endif cannam@89: cannam@89: /* Flush as much pending output as possible */ cannam@89: if (s->pending != 0) { cannam@89: flush_pending(strm); cannam@89: if (strm->avail_out == 0) { cannam@89: /* Since avail_out is 0, deflate will be called again with cannam@89: * more output space, but possibly with both pending and cannam@89: * avail_in equal to zero. There won't be anything to do, cannam@89: * but this is not an error situation so make sure we cannam@89: * return OK instead of BUF_ERROR at next call of deflate: cannam@89: */ cannam@89: s->last_flush = -1; cannam@89: return Z_OK; cannam@89: } cannam@89: cannam@89: /* Make sure there is something to do and avoid duplicate consecutive cannam@89: * flushes. For repeated and useless calls with Z_FINISH, we keep cannam@89: * returning Z_STREAM_END instead of Z_BUF_ERROR. cannam@89: */ cannam@89: } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) && cannam@89: flush != Z_FINISH) { cannam@89: ERR_RETURN(strm, Z_BUF_ERROR); cannam@89: } cannam@89: cannam@89: /* User must not provide more input after the first FINISH: */ cannam@89: if (s->status == FINISH_STATE && strm->avail_in != 0) { cannam@89: ERR_RETURN(strm, Z_BUF_ERROR); cannam@89: } cannam@89: cannam@89: /* Start a new block or continue the current one. cannam@89: */ cannam@89: if (strm->avail_in != 0 || s->lookahead != 0 || cannam@89: (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { cannam@89: block_state bstate; cannam@89: cannam@89: bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : cannam@89: (s->strategy == Z_RLE ? deflate_rle(s, flush) : cannam@89: (*(configuration_table[s->level].func))(s, flush)); cannam@89: cannam@89: if (bstate == finish_started || bstate == finish_done) { cannam@89: s->status = FINISH_STATE; cannam@89: } cannam@89: if (bstate == need_more || bstate == finish_started) { cannam@89: if (strm->avail_out == 0) { cannam@89: s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ cannam@89: } cannam@89: return Z_OK; cannam@89: /* If flush != Z_NO_FLUSH && avail_out == 0, the next call cannam@89: * of deflate should use the same flush parameter to make sure cannam@89: * that the flush is complete. So we don't have to output an cannam@89: * empty block here, this will be done at next call. This also cannam@89: * ensures that for a very small output buffer, we emit at most cannam@89: * one empty block. cannam@89: */ cannam@89: } cannam@89: if (bstate == block_done) { cannam@89: if (flush == Z_PARTIAL_FLUSH) { cannam@89: _tr_align(s); cannam@89: } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */ cannam@89: _tr_stored_block(s, (char*)0, 0L, 0); cannam@89: /* For a full flush, this empty block will be recognized cannam@89: * as a special marker by inflate_sync(). cannam@89: */ cannam@89: if (flush == Z_FULL_FLUSH) { cannam@89: CLEAR_HASH(s); /* forget history */ cannam@89: if (s->lookahead == 0) { cannam@89: s->strstart = 0; cannam@89: s->block_start = 0L; cannam@89: s->insert = 0; cannam@89: } cannam@89: } cannam@89: } cannam@89: flush_pending(strm); cannam@89: if (strm->avail_out == 0) { cannam@89: s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ cannam@89: return Z_OK; cannam@89: } cannam@89: } cannam@89: } cannam@89: Assert(strm->avail_out > 0, "bug2"); cannam@89: cannam@89: if (flush != Z_FINISH) return Z_OK; cannam@89: if (s->wrap <= 0) return Z_STREAM_END; cannam@89: cannam@89: /* Write the trailer */ cannam@89: #ifdef GZIP cannam@89: if (s->wrap == 2) { cannam@89: put_byte(s, (Byte)(strm->adler & 0xff)); cannam@89: put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); cannam@89: put_byte(s, (Byte)((strm->adler >> 16) & 0xff)); cannam@89: put_byte(s, (Byte)((strm->adler >> 24) & 0xff)); cannam@89: put_byte(s, (Byte)(strm->total_in & 0xff)); cannam@89: put_byte(s, (Byte)((strm->total_in >> 8) & 0xff)); cannam@89: put_byte(s, (Byte)((strm->total_in >> 16) & 0xff)); cannam@89: put_byte(s, (Byte)((strm->total_in >> 24) & 0xff)); cannam@89: } cannam@89: else cannam@89: #endif cannam@89: { cannam@89: putShortMSB(s, (uInt)(strm->adler >> 16)); cannam@89: putShortMSB(s, (uInt)(strm->adler & 0xffff)); cannam@89: } cannam@89: flush_pending(strm); cannam@89: /* If avail_out is zero, the application will call deflate again cannam@89: * to flush the rest. cannam@89: */ cannam@89: if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ cannam@89: return s->pending != 0 ? Z_OK : Z_STREAM_END; cannam@89: } cannam@89: cannam@89: /* ========================================================================= */ cannam@89: int ZEXPORT deflateEnd (strm) cannam@89: z_streamp strm; cannam@89: { cannam@89: int status; cannam@89: cannam@89: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; cannam@89: cannam@89: status = strm->state->status; cannam@89: if (status != INIT_STATE && cannam@89: status != EXTRA_STATE && cannam@89: status != NAME_STATE && cannam@89: status != COMMENT_STATE && cannam@89: status != HCRC_STATE && cannam@89: status != BUSY_STATE && cannam@89: status != FINISH_STATE) { cannam@89: return Z_STREAM_ERROR; cannam@89: } cannam@89: cannam@89: /* Deallocate in reverse order of allocations: */ cannam@89: TRY_FREE(strm, strm->state->pending_buf); cannam@89: TRY_FREE(strm, strm->state->head); cannam@89: TRY_FREE(strm, strm->state->prev); cannam@89: TRY_FREE(strm, strm->state->window); cannam@89: cannam@89: ZFREE(strm, strm->state); cannam@89: strm->state = Z_NULL; cannam@89: cannam@89: return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; cannam@89: } cannam@89: cannam@89: /* ========================================================================= cannam@89: * Copy the source state to the destination state. cannam@89: * To simplify the source, this is not supported for 16-bit MSDOS (which cannam@89: * doesn't have enough memory anyway to duplicate compression states). cannam@89: */ cannam@89: int ZEXPORT deflateCopy (dest, source) cannam@89: z_streamp dest; cannam@89: z_streamp source; cannam@89: { cannam@89: #ifdef MAXSEG_64K cannam@89: return Z_STREAM_ERROR; cannam@89: #else cannam@89: deflate_state *ds; cannam@89: deflate_state *ss; cannam@89: ushf *overlay; cannam@89: cannam@89: cannam@89: if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) { cannam@89: return Z_STREAM_ERROR; cannam@89: } cannam@89: cannam@89: ss = source->state; cannam@89: cannam@89: zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream)); cannam@89: cannam@89: ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); cannam@89: if (ds == Z_NULL) return Z_MEM_ERROR; cannam@89: dest->state = (struct internal_state FAR *) ds; cannam@89: zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state)); cannam@89: ds->strm = dest; cannam@89: cannam@89: ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); cannam@89: ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); cannam@89: ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); cannam@89: overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2); cannam@89: ds->pending_buf = (uchf *) overlay; cannam@89: cannam@89: if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || cannam@89: ds->pending_buf == Z_NULL) { cannam@89: deflateEnd (dest); cannam@89: return Z_MEM_ERROR; cannam@89: } cannam@89: /* following zmemcpy do not work for 16-bit MSDOS */ cannam@89: zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); cannam@89: zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos)); cannam@89: zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos)); cannam@89: zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); cannam@89: cannam@89: ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); cannam@89: ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); cannam@89: ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; cannam@89: cannam@89: ds->l_desc.dyn_tree = ds->dyn_ltree; cannam@89: ds->d_desc.dyn_tree = ds->dyn_dtree; cannam@89: ds->bl_desc.dyn_tree = ds->bl_tree; cannam@89: cannam@89: return Z_OK; cannam@89: #endif /* MAXSEG_64K */ cannam@89: } cannam@89: cannam@89: /* =========================================================================== cannam@89: * Read a new buffer from the current input stream, update the adler32 cannam@89: * and total number of bytes read. All deflate() input goes through cannam@89: * this function so some applications may wish to modify it to avoid cannam@89: * allocating a large strm->next_in buffer and copying from it. cannam@89: * (See also flush_pending()). cannam@89: */ cannam@89: local int read_buf(strm, buf, size) cannam@89: z_streamp strm; cannam@89: Bytef *buf; cannam@89: unsigned size; cannam@89: { cannam@89: unsigned len = strm->avail_in; cannam@89: cannam@89: if (len > size) len = size; cannam@89: if (len == 0) return 0; cannam@89: cannam@89: strm->avail_in -= len; cannam@89: cannam@89: zmemcpy(buf, strm->next_in, len); cannam@89: if (strm->state->wrap == 1) { cannam@89: strm->adler = adler32(strm->adler, buf, len); cannam@89: } cannam@89: #ifdef GZIP cannam@89: else if (strm->state->wrap == 2) { cannam@89: strm->adler = crc32(strm->adler, buf, len); cannam@89: } cannam@89: #endif cannam@89: strm->next_in += len; cannam@89: strm->total_in += len; cannam@89: cannam@89: return (int)len; cannam@89: } cannam@89: cannam@89: /* =========================================================================== cannam@89: * Initialize the "longest match" routines for a new zlib stream cannam@89: */ cannam@89: local void lm_init (s) cannam@89: deflate_state *s; cannam@89: { cannam@89: s->window_size = (ulg)2L*s->w_size; cannam@89: cannam@89: CLEAR_HASH(s); cannam@89: cannam@89: /* Set the default configuration parameters: cannam@89: */ cannam@89: s->max_lazy_match = configuration_table[s->level].max_lazy; cannam@89: s->good_match = configuration_table[s->level].good_length; cannam@89: s->nice_match = configuration_table[s->level].nice_length; cannam@89: s->max_chain_length = configuration_table[s->level].max_chain; cannam@89: cannam@89: s->strstart = 0; cannam@89: s->block_start = 0L; cannam@89: s->lookahead = 0; cannam@89: s->insert = 0; cannam@89: s->match_length = s->prev_length = MIN_MATCH-1; cannam@89: s->match_available = 0; cannam@89: s->ins_h = 0; cannam@89: #ifndef FASTEST cannam@89: #ifdef ASMV cannam@89: match_init(); /* initialize the asm code */ cannam@89: #endif cannam@89: #endif cannam@89: } cannam@89: cannam@89: #ifndef FASTEST cannam@89: /* =========================================================================== cannam@89: * Set match_start to the longest match starting at the given string and cannam@89: * return its length. Matches shorter or equal to prev_length are discarded, cannam@89: * in which case the result is equal to prev_length and match_start is cannam@89: * garbage. cannam@89: * IN assertions: cur_match is the head of the hash chain for the current cannam@89: * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 cannam@89: * OUT assertion: the match length is not greater than s->lookahead. cannam@89: */ cannam@89: #ifndef ASMV cannam@89: /* For 80x86 and 680x0, an optimized version will be provided in match.asm or cannam@89: * match.S. The code will be functionally equivalent. cannam@89: */ cannam@89: local uInt longest_match(s, cur_match) cannam@89: deflate_state *s; cannam@89: IPos cur_match; /* current match */ cannam@89: { cannam@89: unsigned chain_length = s->max_chain_length;/* max hash chain length */ cannam@89: register Bytef *scan = s->window + s->strstart; /* current string */ cannam@89: register Bytef *match; /* matched string */ cannam@89: register int len; /* length of current match */ cannam@89: int best_len = s->prev_length; /* best match length so far */ cannam@89: int nice_match = s->nice_match; /* stop if match long enough */ cannam@89: IPos limit = s->strstart > (IPos)MAX_DIST(s) ? cannam@89: s->strstart - (IPos)MAX_DIST(s) : NIL; cannam@89: /* Stop when cur_match becomes <= limit. To simplify the code, cannam@89: * we prevent matches with the string of window index 0. cannam@89: */ cannam@89: Posf *prev = s->prev; cannam@89: uInt wmask = s->w_mask; cannam@89: cannam@89: #ifdef UNALIGNED_OK cannam@89: /* Compare two bytes at a time. Note: this is not always beneficial. cannam@89: * Try with and without -DUNALIGNED_OK to check. cannam@89: */ cannam@89: register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; cannam@89: register ush scan_start = *(ushf*)scan; cannam@89: register ush scan_end = *(ushf*)(scan+best_len-1); cannam@89: #else cannam@89: register Bytef *strend = s->window + s->strstart + MAX_MATCH; cannam@89: register Byte scan_end1 = scan[best_len-1]; cannam@89: register Byte scan_end = scan[best_len]; cannam@89: #endif cannam@89: cannam@89: /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. cannam@89: * It is easy to get rid of this optimization if necessary. cannam@89: */ cannam@89: Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); cannam@89: cannam@89: /* Do not waste too much time if we already have a good match: */ cannam@89: if (s->prev_length >= s->good_match) { cannam@89: chain_length >>= 2; cannam@89: } cannam@89: /* Do not look for matches beyond the end of the input. This is necessary cannam@89: * to make deflate deterministic. cannam@89: */ cannam@89: if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; cannam@89: cannam@89: Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); cannam@89: cannam@89: do { cannam@89: Assert(cur_match < s->strstart, "no future"); cannam@89: match = s->window + cur_match; cannam@89: cannam@89: /* Skip to next match if the match length cannot increase cannam@89: * or if the match length is less than 2. Note that the checks below cannam@89: * for insufficient lookahead only occur occasionally for performance cannam@89: * reasons. Therefore uninitialized memory will be accessed, and cannam@89: * conditional jumps will be made that depend on those values. cannam@89: * However the length of the match is limited to the lookahead, so cannam@89: * the output of deflate is not affected by the uninitialized values. cannam@89: */ cannam@89: #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) cannam@89: /* This code assumes sizeof(unsigned short) == 2. Do not use cannam@89: * UNALIGNED_OK if your compiler uses a different size. cannam@89: */ cannam@89: if (*(ushf*)(match+best_len-1) != scan_end || cannam@89: *(ushf*)match != scan_start) continue; cannam@89: cannam@89: /* It is not necessary to compare scan[2] and match[2] since they are cannam@89: * always equal when the other bytes match, given that the hash keys cannam@89: * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at cannam@89: * strstart+3, +5, ... up to strstart+257. We check for insufficient cannam@89: * lookahead only every 4th comparison; the 128th check will be made cannam@89: * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is cannam@89: * necessary to put more guard bytes at the end of the window, or cannam@89: * to check more often for insufficient lookahead. cannam@89: */ cannam@89: Assert(scan[2] == match[2], "scan[2]?"); cannam@89: scan++, match++; cannam@89: do { cannam@89: } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) && cannam@89: *(ushf*)(scan+=2) == *(ushf*)(match+=2) && cannam@89: *(ushf*)(scan+=2) == *(ushf*)(match+=2) && cannam@89: *(ushf*)(scan+=2) == *(ushf*)(match+=2) && cannam@89: scan < strend); cannam@89: /* The funny "do {}" generates better code on most compilers */ cannam@89: cannam@89: /* Here, scan <= window+strstart+257 */ cannam@89: Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); cannam@89: if (*scan == *match) scan++; cannam@89: cannam@89: len = (MAX_MATCH - 1) - (int)(strend-scan); cannam@89: scan = strend - (MAX_MATCH-1); cannam@89: cannam@89: #else /* UNALIGNED_OK */ cannam@89: cannam@89: if (match[best_len] != scan_end || cannam@89: match[best_len-1] != scan_end1 || cannam@89: *match != *scan || cannam@89: *++match != scan[1]) continue; cannam@89: cannam@89: /* The check at best_len-1 can be removed because it will be made cannam@89: * again later. (This heuristic is not always a win.) cannam@89: * It is not necessary to compare scan[2] and match[2] since they cannam@89: * are always equal when the other bytes match, given that cannam@89: * the hash keys are equal and that HASH_BITS >= 8. cannam@89: */ cannam@89: scan += 2, match++; cannam@89: Assert(*scan == *match, "match[2]?"); cannam@89: cannam@89: /* We check for insufficient lookahead only every 8th comparison; cannam@89: * the 256th check will be made at strstart+258. cannam@89: */ cannam@89: do { cannam@89: } while (*++scan == *++match && *++scan == *++match && cannam@89: *++scan == *++match && *++scan == *++match && cannam@89: *++scan == *++match && *++scan == *++match && cannam@89: *++scan == *++match && *++scan == *++match && cannam@89: scan < strend); cannam@89: cannam@89: Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); cannam@89: cannam@89: len = MAX_MATCH - (int)(strend - scan); cannam@89: scan = strend - MAX_MATCH; cannam@89: cannam@89: #endif /* UNALIGNED_OK */ cannam@89: cannam@89: if (len > best_len) { cannam@89: s->match_start = cur_match; cannam@89: best_len = len; cannam@89: if (len >= nice_match) break; cannam@89: #ifdef UNALIGNED_OK cannam@89: scan_end = *(ushf*)(scan+best_len-1); cannam@89: #else cannam@89: scan_end1 = scan[best_len-1]; cannam@89: scan_end = scan[best_len]; cannam@89: #endif cannam@89: } cannam@89: } while ((cur_match = prev[cur_match & wmask]) > limit cannam@89: && --chain_length != 0); cannam@89: cannam@89: if ((uInt)best_len <= s->lookahead) return (uInt)best_len; cannam@89: return s->lookahead; cannam@89: } cannam@89: #endif /* ASMV */ cannam@89: cannam@89: #else /* FASTEST */ cannam@89: cannam@89: /* --------------------------------------------------------------------------- cannam@89: * Optimized version for FASTEST only cannam@89: */ cannam@89: local uInt longest_match(s, cur_match) cannam@89: deflate_state *s; cannam@89: IPos cur_match; /* current match */ cannam@89: { cannam@89: register Bytef *scan = s->window + s->strstart; /* current string */ cannam@89: register Bytef *match; /* matched string */ cannam@89: register int len; /* length of current match */ cannam@89: register Bytef *strend = s->window + s->strstart + MAX_MATCH; cannam@89: cannam@89: /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. cannam@89: * It is easy to get rid of this optimization if necessary. cannam@89: */ cannam@89: Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); cannam@89: cannam@89: Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); cannam@89: cannam@89: Assert(cur_match < s->strstart, "no future"); cannam@89: cannam@89: match = s->window + cur_match; cannam@89: cannam@89: /* Return failure if the match length is less than 2: cannam@89: */ cannam@89: if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; cannam@89: cannam@89: /* The check at best_len-1 can be removed because it will be made cannam@89: * again later. (This heuristic is not always a win.) cannam@89: * It is not necessary to compare scan[2] and match[2] since they cannam@89: * are always equal when the other bytes match, given that cannam@89: * the hash keys are equal and that HASH_BITS >= 8. cannam@89: */ cannam@89: scan += 2, match += 2; cannam@89: Assert(*scan == *match, "match[2]?"); cannam@89: cannam@89: /* We check for insufficient lookahead only every 8th comparison; cannam@89: * the 256th check will be made at strstart+258. cannam@89: */ cannam@89: do { cannam@89: } while (*++scan == *++match && *++scan == *++match && cannam@89: *++scan == *++match && *++scan == *++match && cannam@89: *++scan == *++match && *++scan == *++match && cannam@89: *++scan == *++match && *++scan == *++match && cannam@89: scan < strend); cannam@89: cannam@89: Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); cannam@89: cannam@89: len = MAX_MATCH - (int)(strend - scan); cannam@89: cannam@89: if (len < MIN_MATCH) return MIN_MATCH - 1; cannam@89: cannam@89: s->match_start = cur_match; cannam@89: return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead; cannam@89: } cannam@89: cannam@89: #endif /* FASTEST */ cannam@89: cannam@89: #ifdef DEBUG cannam@89: /* =========================================================================== cannam@89: * Check that the match at match_start is indeed a match. cannam@89: */ cannam@89: local void check_match(s, start, match, length) cannam@89: deflate_state *s; cannam@89: IPos start, match; cannam@89: int length; cannam@89: { cannam@89: /* check that the match is indeed a match */ cannam@89: if (zmemcmp(s->window + match, cannam@89: s->window + start, length) != EQUAL) { cannam@89: fprintf(stderr, " start %u, match %u, length %d\n", cannam@89: start, match, length); cannam@89: do { cannam@89: fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); cannam@89: } while (--length != 0); cannam@89: z_error("invalid match"); cannam@89: } cannam@89: if (z_verbose > 1) { cannam@89: fprintf(stderr,"\\[%d,%d]", start-match, length); cannam@89: do { putc(s->window[start++], stderr); } while (--length != 0); cannam@89: } cannam@89: } cannam@89: #else cannam@89: # define check_match(s, start, match, length) cannam@89: #endif /* DEBUG */ cannam@89: cannam@89: /* =========================================================================== cannam@89: * Fill the window when the lookahead becomes insufficient. cannam@89: * Updates strstart and lookahead. cannam@89: * cannam@89: * IN assertion: lookahead < MIN_LOOKAHEAD cannam@89: * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD cannam@89: * At least one byte has been read, or avail_in == 0; reads are cannam@89: * performed for at least two bytes (required for the zip translate_eol cannam@89: * option -- not supported here). cannam@89: */ cannam@89: local void fill_window(s) cannam@89: deflate_state *s; cannam@89: { cannam@89: register unsigned n, m; cannam@89: register Posf *p; cannam@89: unsigned more; /* Amount of free space at the end of the window. */ cannam@89: uInt wsize = s->w_size; cannam@89: cannam@89: Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead"); cannam@89: cannam@89: do { cannam@89: more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); cannam@89: cannam@89: /* Deal with !@#$% 64K limit: */ cannam@89: if (sizeof(int) <= 2) { cannam@89: if (more == 0 && s->strstart == 0 && s->lookahead == 0) { cannam@89: more = wsize; cannam@89: cannam@89: } else if (more == (unsigned)(-1)) { cannam@89: /* Very unlikely, but possible on 16 bit machine if cannam@89: * strstart == 0 && lookahead == 1 (input done a byte at time) cannam@89: */ cannam@89: more--; cannam@89: } cannam@89: } cannam@89: cannam@89: /* If the window is almost full and there is insufficient lookahead, cannam@89: * move the upper half to the lower one to make room in the upper half. cannam@89: */ cannam@89: if (s->strstart >= wsize+MAX_DIST(s)) { cannam@89: cannam@89: zmemcpy(s->window, s->window+wsize, (unsigned)wsize); cannam@89: s->match_start -= wsize; cannam@89: s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ cannam@89: s->block_start -= (long) wsize; cannam@89: cannam@89: /* Slide the hash table (could be avoided with 32 bit values cannam@89: at the expense of memory usage). We slide even when level == 0 cannam@89: to keep the hash table consistent if we switch back to level > 0 cannam@89: later. (Using level 0 permanently is not an optimal usage of cannam@89: zlib, so we don't care about this pathological case.) cannam@89: */ cannam@89: n = s->hash_size; cannam@89: p = &s->head[n]; cannam@89: do { cannam@89: m = *--p; cannam@89: *p = (Pos)(m >= wsize ? m-wsize : NIL); cannam@89: } while (--n); cannam@89: cannam@89: n = wsize; cannam@89: #ifndef FASTEST cannam@89: p = &s->prev[n]; cannam@89: do { cannam@89: m = *--p; cannam@89: *p = (Pos)(m >= wsize ? m-wsize : NIL); cannam@89: /* If n is not on any hash chain, prev[n] is garbage but cannam@89: * its value will never be used. cannam@89: */ cannam@89: } while (--n); cannam@89: #endif cannam@89: more += wsize; cannam@89: } cannam@89: if (s->strm->avail_in == 0) break; cannam@89: cannam@89: /* If there was no sliding: cannam@89: * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && cannam@89: * more == window_size - lookahead - strstart cannam@89: * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) cannam@89: * => more >= window_size - 2*WSIZE + 2 cannam@89: * In the BIG_MEM or MMAP case (not yet supported), cannam@89: * window_size == input_size + MIN_LOOKAHEAD && cannam@89: * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. cannam@89: * Otherwise, window_size == 2*WSIZE so more >= 2. cannam@89: * If there was sliding, more >= WSIZE. So in all cases, more >= 2. cannam@89: */ cannam@89: Assert(more >= 2, "more < 2"); cannam@89: cannam@89: n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); cannam@89: s->lookahead += n; cannam@89: cannam@89: /* Initialize the hash value now that we have some input: */ cannam@89: if (s->lookahead + s->insert >= MIN_MATCH) { cannam@89: uInt str = s->strstart - s->insert; cannam@89: s->ins_h = s->window[str]; cannam@89: UPDATE_HASH(s, s->ins_h, s->window[str + 1]); cannam@89: #if MIN_MATCH != 3 cannam@89: Call UPDATE_HASH() MIN_MATCH-3 more times cannam@89: #endif cannam@89: while (s->insert) { cannam@89: UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]); cannam@89: #ifndef FASTEST cannam@89: s->prev[str & s->w_mask] = s->head[s->ins_h]; cannam@89: #endif cannam@89: s->head[s->ins_h] = (Pos)str; cannam@89: str++; cannam@89: s->insert--; cannam@89: if (s->lookahead + s->insert < MIN_MATCH) cannam@89: break; cannam@89: } cannam@89: } cannam@89: /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, cannam@89: * but this is not important since only literal bytes will be emitted. cannam@89: */ cannam@89: cannam@89: } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); cannam@89: cannam@89: /* If the WIN_INIT bytes after the end of the current data have never been cannam@89: * written, then zero those bytes in order to avoid memory check reports of cannam@89: * the use of uninitialized (or uninitialised as Julian writes) bytes by cannam@89: * the longest match routines. Update the high water mark for the next cannam@89: * time through here. WIN_INIT is set to MAX_MATCH since the longest match cannam@89: * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead. cannam@89: */ cannam@89: if (s->high_water < s->window_size) { cannam@89: ulg curr = s->strstart + (ulg)(s->lookahead); cannam@89: ulg init; cannam@89: cannam@89: if (s->high_water < curr) { cannam@89: /* Previous high water mark below current data -- zero WIN_INIT cannam@89: * bytes or up to end of window, whichever is less. cannam@89: */ cannam@89: init = s->window_size - curr; cannam@89: if (init > WIN_INIT) cannam@89: init = WIN_INIT; cannam@89: zmemzero(s->window + curr, (unsigned)init); cannam@89: s->high_water = curr + init; cannam@89: } cannam@89: else if (s->high_water < (ulg)curr + WIN_INIT) { cannam@89: /* High water mark at or above current data, but below current data cannam@89: * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up cannam@89: * to end of window, whichever is less. cannam@89: */ cannam@89: init = (ulg)curr + WIN_INIT - s->high_water; cannam@89: if (init > s->window_size - s->high_water) cannam@89: init = s->window_size - s->high_water; cannam@89: zmemzero(s->window + s->high_water, (unsigned)init); cannam@89: s->high_water += init; cannam@89: } cannam@89: } cannam@89: cannam@89: Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD, cannam@89: "not enough room for search"); cannam@89: } cannam@89: cannam@89: /* =========================================================================== cannam@89: * Flush the current block, with given end-of-file flag. cannam@89: * IN assertion: strstart is set to the end of the current match. cannam@89: */ cannam@89: #define FLUSH_BLOCK_ONLY(s, last) { \ cannam@89: _tr_flush_block(s, (s->block_start >= 0L ? \ cannam@89: (charf *)&s->window[(unsigned)s->block_start] : \ cannam@89: (charf *)Z_NULL), \ cannam@89: (ulg)((long)s->strstart - s->block_start), \ cannam@89: (last)); \ cannam@89: s->block_start = s->strstart; \ cannam@89: flush_pending(s->strm); \ cannam@89: Tracev((stderr,"[FLUSH]")); \ cannam@89: } cannam@89: cannam@89: /* Same but force premature exit if necessary. */ cannam@89: #define FLUSH_BLOCK(s, last) { \ cannam@89: FLUSH_BLOCK_ONLY(s, last); \ cannam@89: if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \ cannam@89: } cannam@89: cannam@89: /* =========================================================================== cannam@89: * Copy without compression as much as possible from the input stream, return cannam@89: * the current block state. cannam@89: * This function does not insert new strings in the dictionary since cannam@89: * uncompressible data is probably not useful. This function is used cannam@89: * only for the level=0 compression option. cannam@89: * NOTE: this function should be optimized to avoid extra copying from cannam@89: * window to pending_buf. cannam@89: */ cannam@89: local block_state deflate_stored(s, flush) cannam@89: deflate_state *s; cannam@89: int flush; cannam@89: { cannam@89: /* Stored blocks are limited to 0xffff bytes, pending_buf is limited cannam@89: * to pending_buf_size, and each stored block has a 5 byte header: cannam@89: */ cannam@89: ulg max_block_size = 0xffff; cannam@89: ulg max_start; cannam@89: cannam@89: if (max_block_size > s->pending_buf_size - 5) { cannam@89: max_block_size = s->pending_buf_size - 5; cannam@89: } cannam@89: cannam@89: /* Copy as much as possible from input to output: */ cannam@89: for (;;) { cannam@89: /* Fill the window as much as possible: */ cannam@89: if (s->lookahead <= 1) { cannam@89: cannam@89: Assert(s->strstart < s->w_size+MAX_DIST(s) || cannam@89: s->block_start >= (long)s->w_size, "slide too late"); cannam@89: cannam@89: fill_window(s); cannam@89: if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more; cannam@89: cannam@89: if (s->lookahead == 0) break; /* flush the current block */ cannam@89: } cannam@89: Assert(s->block_start >= 0L, "block gone"); cannam@89: cannam@89: s->strstart += s->lookahead; cannam@89: s->lookahead = 0; cannam@89: cannam@89: /* Emit a stored block if pending_buf will be full: */ cannam@89: max_start = s->block_start + max_block_size; cannam@89: if (s->strstart == 0 || (ulg)s->strstart >= max_start) { cannam@89: /* strstart == 0 is possible when wraparound on 16-bit machine */ cannam@89: s->lookahead = (uInt)(s->strstart - max_start); cannam@89: s->strstart = (uInt)max_start; cannam@89: FLUSH_BLOCK(s, 0); cannam@89: } cannam@89: /* Flush if we may have to slide, otherwise block_start may become cannam@89: * negative and the data will be gone: cannam@89: */ cannam@89: if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) { cannam@89: FLUSH_BLOCK(s, 0); cannam@89: } cannam@89: } cannam@89: s->insert = 0; cannam@89: if (flush == Z_FINISH) { cannam@89: FLUSH_BLOCK(s, 1); cannam@89: return finish_done; cannam@89: } cannam@89: if ((long)s->strstart > s->block_start) cannam@89: FLUSH_BLOCK(s, 0); cannam@89: return block_done; cannam@89: } cannam@89: cannam@89: /* =========================================================================== cannam@89: * Compress as much as possible from the input stream, return the current cannam@89: * block state. cannam@89: * This function does not perform lazy evaluation of matches and inserts cannam@89: * new strings in the dictionary only for unmatched strings or for short cannam@89: * matches. It is used only for the fast compression options. cannam@89: */ cannam@89: local block_state deflate_fast(s, flush) cannam@89: deflate_state *s; cannam@89: int flush; cannam@89: { cannam@89: IPos hash_head; /* head of the hash chain */ cannam@89: int bflush; /* set if current block must be flushed */ cannam@89: cannam@89: for (;;) { cannam@89: /* Make sure that we always have enough lookahead, except cannam@89: * at the end of the input file. We need MAX_MATCH bytes cannam@89: * for the next match, plus MIN_MATCH bytes to insert the cannam@89: * string following the next match. cannam@89: */ cannam@89: if (s->lookahead < MIN_LOOKAHEAD) { cannam@89: fill_window(s); cannam@89: if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { cannam@89: return need_more; cannam@89: } cannam@89: if (s->lookahead == 0) break; /* flush the current block */ cannam@89: } cannam@89: cannam@89: /* Insert the string window[strstart .. strstart+2] in the cannam@89: * dictionary, and set hash_head to the head of the hash chain: cannam@89: */ cannam@89: hash_head = NIL; cannam@89: if (s->lookahead >= MIN_MATCH) { cannam@89: INSERT_STRING(s, s->strstart, hash_head); cannam@89: } cannam@89: cannam@89: /* Find the longest match, discarding those <= prev_length. cannam@89: * At this point we have always match_length < MIN_MATCH cannam@89: */ cannam@89: if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { cannam@89: /* To simplify the code, we prevent matches with the string cannam@89: * of window index 0 (in particular we have to avoid a match cannam@89: * of the string with itself at the start of the input file). cannam@89: */ cannam@89: s->match_length = longest_match (s, hash_head); cannam@89: /* longest_match() sets match_start */ cannam@89: } cannam@89: if (s->match_length >= MIN_MATCH) { cannam@89: check_match(s, s->strstart, s->match_start, s->match_length); cannam@89: cannam@89: _tr_tally_dist(s, s->strstart - s->match_start, cannam@89: s->match_length - MIN_MATCH, bflush); cannam@89: cannam@89: s->lookahead -= s->match_length; cannam@89: cannam@89: /* Insert new strings in the hash table only if the match length cannam@89: * is not too large. This saves time but degrades compression. cannam@89: */ cannam@89: #ifndef FASTEST cannam@89: if (s->match_length <= s->max_insert_length && cannam@89: s->lookahead >= MIN_MATCH) { cannam@89: s->match_length--; /* string at strstart already in table */ cannam@89: do { cannam@89: s->strstart++; cannam@89: INSERT_STRING(s, s->strstart, hash_head); cannam@89: /* strstart never exceeds WSIZE-MAX_MATCH, so there are cannam@89: * always MIN_MATCH bytes ahead. cannam@89: */ cannam@89: } while (--s->match_length != 0); cannam@89: s->strstart++; cannam@89: } else cannam@89: #endif cannam@89: { cannam@89: s->strstart += s->match_length; cannam@89: s->match_length = 0; cannam@89: s->ins_h = s->window[s->strstart]; cannam@89: UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); cannam@89: #if MIN_MATCH != 3 cannam@89: Call UPDATE_HASH() MIN_MATCH-3 more times cannam@89: #endif cannam@89: /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not cannam@89: * matter since it will be recomputed at next deflate call. cannam@89: */ cannam@89: } cannam@89: } else { cannam@89: /* No match, output a literal byte */ cannam@89: Tracevv((stderr,"%c", s->window[s->strstart])); cannam@89: _tr_tally_lit (s, s->window[s->strstart], bflush); cannam@89: s->lookahead--; cannam@89: s->strstart++; cannam@89: } cannam@89: if (bflush) FLUSH_BLOCK(s, 0); cannam@89: } cannam@89: s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; cannam@89: if (flush == Z_FINISH) { cannam@89: FLUSH_BLOCK(s, 1); cannam@89: return finish_done; cannam@89: } cannam@89: if (s->last_lit) cannam@89: FLUSH_BLOCK(s, 0); cannam@89: return block_done; cannam@89: } cannam@89: cannam@89: #ifndef FASTEST cannam@89: /* =========================================================================== cannam@89: * Same as above, but achieves better compression. We use a lazy cannam@89: * evaluation for matches: a match is finally adopted only if there is cannam@89: * no better match at the next window position. cannam@89: */ cannam@89: local block_state deflate_slow(s, flush) cannam@89: deflate_state *s; cannam@89: int flush; cannam@89: { cannam@89: IPos hash_head; /* head of hash chain */ cannam@89: int bflush; /* set if current block must be flushed */ cannam@89: cannam@89: /* Process the input block. */ cannam@89: for (;;) { cannam@89: /* Make sure that we always have enough lookahead, except cannam@89: * at the end of the input file. We need MAX_MATCH bytes cannam@89: * for the next match, plus MIN_MATCH bytes to insert the cannam@89: * string following the next match. cannam@89: */ cannam@89: if (s->lookahead < MIN_LOOKAHEAD) { cannam@89: fill_window(s); cannam@89: if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { cannam@89: return need_more; cannam@89: } cannam@89: if (s->lookahead == 0) break; /* flush the current block */ cannam@89: } cannam@89: cannam@89: /* Insert the string window[strstart .. strstart+2] in the cannam@89: * dictionary, and set hash_head to the head of the hash chain: cannam@89: */ cannam@89: hash_head = NIL; cannam@89: if (s->lookahead >= MIN_MATCH) { cannam@89: INSERT_STRING(s, s->strstart, hash_head); cannam@89: } cannam@89: cannam@89: /* Find the longest match, discarding those <= prev_length. cannam@89: */ cannam@89: s->prev_length = s->match_length, s->prev_match = s->match_start; cannam@89: s->match_length = MIN_MATCH-1; cannam@89: cannam@89: if (hash_head != NIL && s->prev_length < s->max_lazy_match && cannam@89: s->strstart - hash_head <= MAX_DIST(s)) { cannam@89: /* To simplify the code, we prevent matches with the string cannam@89: * of window index 0 (in particular we have to avoid a match cannam@89: * of the string with itself at the start of the input file). cannam@89: */ cannam@89: s->match_length = longest_match (s, hash_head); cannam@89: /* longest_match() sets match_start */ cannam@89: cannam@89: if (s->match_length <= 5 && (s->strategy == Z_FILTERED cannam@89: #if TOO_FAR <= 32767 cannam@89: || (s->match_length == MIN_MATCH && cannam@89: s->strstart - s->match_start > TOO_FAR) cannam@89: #endif cannam@89: )) { cannam@89: cannam@89: /* If prev_match is also MIN_MATCH, match_start is garbage cannam@89: * but we will ignore the current match anyway. cannam@89: */ cannam@89: s->match_length = MIN_MATCH-1; cannam@89: } cannam@89: } cannam@89: /* If there was a match at the previous step and the current cannam@89: * match is not better, output the previous match: cannam@89: */ cannam@89: if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { cannam@89: uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; cannam@89: /* Do not insert strings in hash table beyond this. */ cannam@89: cannam@89: check_match(s, s->strstart-1, s->prev_match, s->prev_length); cannam@89: cannam@89: _tr_tally_dist(s, s->strstart -1 - s->prev_match, cannam@89: s->prev_length - MIN_MATCH, bflush); cannam@89: cannam@89: /* Insert in hash table all strings up to the end of the match. cannam@89: * strstart-1 and strstart are already inserted. If there is not cannam@89: * enough lookahead, the last two strings are not inserted in cannam@89: * the hash table. cannam@89: */ cannam@89: s->lookahead -= s->prev_length-1; cannam@89: s->prev_length -= 2; cannam@89: do { cannam@89: if (++s->strstart <= max_insert) { cannam@89: INSERT_STRING(s, s->strstart, hash_head); cannam@89: } cannam@89: } while (--s->prev_length != 0); cannam@89: s->match_available = 0; cannam@89: s->match_length = MIN_MATCH-1; cannam@89: s->strstart++; cannam@89: cannam@89: if (bflush) FLUSH_BLOCK(s, 0); cannam@89: cannam@89: } else if (s->match_available) { cannam@89: /* If there was no match at the previous position, output a cannam@89: * single literal. If there was a match but the current match cannam@89: * is longer, truncate the previous match to a single literal. cannam@89: */ cannam@89: Tracevv((stderr,"%c", s->window[s->strstart-1])); cannam@89: _tr_tally_lit(s, s->window[s->strstart-1], bflush); cannam@89: if (bflush) { cannam@89: FLUSH_BLOCK_ONLY(s, 0); cannam@89: } cannam@89: s->strstart++; cannam@89: s->lookahead--; cannam@89: if (s->strm->avail_out == 0) return need_more; cannam@89: } else { cannam@89: /* There is no previous match to compare with, wait for cannam@89: * the next step to decide. cannam@89: */ cannam@89: s->match_available = 1; cannam@89: s->strstart++; cannam@89: s->lookahead--; cannam@89: } cannam@89: } cannam@89: Assert (flush != Z_NO_FLUSH, "no flush?"); cannam@89: if (s->match_available) { cannam@89: Tracevv((stderr,"%c", s->window[s->strstart-1])); cannam@89: _tr_tally_lit(s, s->window[s->strstart-1], bflush); cannam@89: s->match_available = 0; cannam@89: } cannam@89: s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1; cannam@89: if (flush == Z_FINISH) { cannam@89: FLUSH_BLOCK(s, 1); cannam@89: return finish_done; cannam@89: } cannam@89: if (s->last_lit) cannam@89: FLUSH_BLOCK(s, 0); cannam@89: return block_done; cannam@89: } cannam@89: #endif /* FASTEST */ cannam@89: cannam@89: /* =========================================================================== cannam@89: * For Z_RLE, simply look for runs of bytes, generate matches only of distance cannam@89: * one. Do not maintain a hash table. (It will be regenerated if this run of cannam@89: * deflate switches away from Z_RLE.) cannam@89: */ cannam@89: local block_state deflate_rle(s, flush) cannam@89: deflate_state *s; cannam@89: int flush; cannam@89: { cannam@89: int bflush; /* set if current block must be flushed */ cannam@89: uInt prev; /* byte at distance one to match */ cannam@89: Bytef *scan, *strend; /* scan goes up to strend for length of run */ cannam@89: cannam@89: for (;;) { cannam@89: /* Make sure that we always have enough lookahead, except cannam@89: * at the end of the input file. We need MAX_MATCH bytes cannam@89: * for the longest run, plus one for the unrolled loop. cannam@89: */ cannam@89: if (s->lookahead <= MAX_MATCH) { cannam@89: fill_window(s); cannam@89: if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) { cannam@89: return need_more; cannam@89: } cannam@89: if (s->lookahead == 0) break; /* flush the current block */ cannam@89: } cannam@89: cannam@89: /* See how many times the previous byte repeats */ cannam@89: s->match_length = 0; cannam@89: if (s->lookahead >= MIN_MATCH && s->strstart > 0) { cannam@89: scan = s->window + s->strstart - 1; cannam@89: prev = *scan; cannam@89: if (prev == *++scan && prev == *++scan && prev == *++scan) { cannam@89: strend = s->window + s->strstart + MAX_MATCH; cannam@89: do { cannam@89: } while (prev == *++scan && prev == *++scan && cannam@89: prev == *++scan && prev == *++scan && cannam@89: prev == *++scan && prev == *++scan && cannam@89: prev == *++scan && prev == *++scan && cannam@89: scan < strend); cannam@89: s->match_length = MAX_MATCH - (int)(strend - scan); cannam@89: if (s->match_length > s->lookahead) cannam@89: s->match_length = s->lookahead; cannam@89: } cannam@89: Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan"); cannam@89: } cannam@89: cannam@89: /* Emit match if have run of MIN_MATCH or longer, else emit literal */ cannam@89: if (s->match_length >= MIN_MATCH) { cannam@89: check_match(s, s->strstart, s->strstart - 1, s->match_length); cannam@89: cannam@89: _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush); cannam@89: cannam@89: s->lookahead -= s->match_length; cannam@89: s->strstart += s->match_length; cannam@89: s->match_length = 0; cannam@89: } else { cannam@89: /* No match, output a literal byte */ cannam@89: Tracevv((stderr,"%c", s->window[s->strstart])); cannam@89: _tr_tally_lit (s, s->window[s->strstart], bflush); cannam@89: s->lookahead--; cannam@89: s->strstart++; cannam@89: } cannam@89: if (bflush) FLUSH_BLOCK(s, 0); cannam@89: } cannam@89: s->insert = 0; cannam@89: if (flush == Z_FINISH) { cannam@89: FLUSH_BLOCK(s, 1); cannam@89: return finish_done; cannam@89: } cannam@89: if (s->last_lit) cannam@89: FLUSH_BLOCK(s, 0); cannam@89: return block_done; cannam@89: } cannam@89: cannam@89: /* =========================================================================== cannam@89: * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table. cannam@89: * (It will be regenerated if this run of deflate switches away from Huffman.) cannam@89: */ cannam@89: local block_state deflate_huff(s, flush) cannam@89: deflate_state *s; cannam@89: int flush; cannam@89: { cannam@89: int bflush; /* set if current block must be flushed */ cannam@89: cannam@89: for (;;) { cannam@89: /* Make sure that we have a literal to write. */ cannam@89: if (s->lookahead == 0) { cannam@89: fill_window(s); cannam@89: if (s->lookahead == 0) { cannam@89: if (flush == Z_NO_FLUSH) cannam@89: return need_more; cannam@89: break; /* flush the current block */ cannam@89: } cannam@89: } cannam@89: cannam@89: /* Output a literal byte */ cannam@89: s->match_length = 0; cannam@89: Tracevv((stderr,"%c", s->window[s->strstart])); cannam@89: _tr_tally_lit (s, s->window[s->strstart], bflush); cannam@89: s->lookahead--; cannam@89: s->strstart++; cannam@89: if (bflush) FLUSH_BLOCK(s, 0); cannam@89: } cannam@89: s->insert = 0; cannam@89: if (flush == Z_FINISH) { cannam@89: FLUSH_BLOCK(s, 1); cannam@89: return finish_done; cannam@89: } cannam@89: if (s->last_lit) cannam@89: FLUSH_BLOCK(s, 0); cannam@89: return block_done; cannam@89: }