/* deflate.c -- compress data using the deflation algorithm
* Copyright (C) 1995-2002 Jean-loup Gailly.
* For conditions of distribution and use, see copyright notice in zlib.h
*/
/*
* ALGORITHM
*
* The "deflation" process depends on being able to identify portions
* of the input text which are identical to earlier input (within a
* sliding window trailing behind the input currently being processed).
*
* The most straightforward technique turns out to be the fastest for
* most input files: try all possible matches and select the longest.
* The key feature of this algorithm is that insertions into the string
* dictionary are very simple and thus fast, and deletions are avoided
* completely. Insertions are performed at each input character, whereas
* string matches are performed only when the previous match ends. So it
* is preferable to spend more time in matches to allow very fast string
* insertions and avoid deletions. The matching algorithm for small
* strings is inspired from that of Rabin & Karp. A brute force approach
* is used to find longer strings when a small match has been found.
* A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
* (by Leonid Broukhis).
* A previous version of this file used a more sophisticated algorithm
* (by Fiala and Greene) which is guaranteed to run in linear amortized
* time, but has a larger average cost, uses more memory and is patented.
* However the F&G algorithm may be faster for some highly redundant
* files if the parameter max_chain_length (described below) is too large.
*
* ACKNOWLEDGEMENTS
*
* The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
* I found it in 'freeze' written by Leonid Broukhis.
* Thanks to many people for bug reports and testing.
*
* REFERENCES
*
* Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
* Available in ftp://ds.internic.net/rfc/rfc1951.txt
*
* A description of the Rabin and Karp algorithm is given in the book
* "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
*
* Fiala,E.R., and Greene,D.H.
* Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
*
*/
#include "deflate.h"
const char deflate_copyright[] =
" deflate 1.1.4 Copyright 1995-2002 Jean-loup Gailly ";
/*
If you use the zlib library in a product, an acknowledgment is welcome
in the documentation of your product. If for some reason you cannot
include such an acknowledgment, I would appreciate that you keep this
copyright string in the executable of your product.
*/
/* ===========================================================================
* Function prototypes.
*/
typedef enum {
} block_state;
/* Compression function. Returns the block state after the call. */
#ifdef ASMV
#else
#endif
#ifdef DEBUG
int length));
#endif
/* ===========================================================================
* Local data
*/
#define NIL 0
/* Tail of hash chains */
#ifndef TOO_FAR
#endif
/* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
/* Minimum amount of lookahead, except at the end of the input file.
* See deflate.c for comments about the MIN_MATCH+1.
*/
/* Values for max_lazy_match, good_match and max_chain_length, depending on
* the desired pack level (0..9). The values given below have been tuned to
* exclude worst case performance for pathological files. Better values may be
* found for specific files.
*/
typedef struct config_s {
} config;
/* good lazy nice chain */
/* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
* For deflate_fast() (levels <= 3) good is ignored and lazy has a different
* meaning.
*/
#define EQUAL 0
/* result of memcmp for equal strings */
/* ===========================================================================
* Update a hash value with the given input byte
* IN assertion: all calls to to UPDATE_HASH are made with consecutive
* input characters, so that a running hash key can be computed from the
* previous key instead of complete recalculation each time.
*/
/* ===========================================================================
* Insert string str in the dictionary and set match_head to the previous head
* of the hash chain (the most recent string with same hash key). Return
* the previous length of the hash chain.
* If this file is compiled with -DFASTEST, the compression level is forced
* to 1, and no hash chains are maintained.
* IN assertion: all calls to to INSERT_STRING are made with consecutive
* input characters and the first MIN_MATCH bytes of str are valid
* (except for the last MIN_MATCH-1 bytes of the input file).
*/
#ifdef FASTEST
#else
#endif
/* ===========================================================================
* Initialize the hash table (avoiding 64K overflow for 16 bit systems).
* prev[] will be initialized on the fly.
*/
#define CLEAR_HASH(s) \
/* ========================================================================= */
int level;
const char *version;
int stream_size;
{
/* To do: ignore strm->next_in if we use it as window */
}
/* ========================================================================= */
int level;
int method;
int windowBits;
int memLevel;
int strategy;
const char *version;
int stream_size;
{
deflate_state *s;
int noheader = 0;
/* We overlay pending_buf and d_buf+l_buf. This works since the average
* output size for (length,distance) codes is <= 24 bits.
*/
stream_size != sizeof(z_stream)) {
return Z_VERSION_ERROR;
}
}
#ifdef FASTEST
level = 1;
#endif
if (windowBits < 0) { /* undocumented feature: suppress zlib header */
noheader = 1;
windowBits = -windowBits;
}
return Z_STREAM_ERROR;
}
if (s == Z_NULL) return Z_MEM_ERROR;
s->w_bits = windowBits;
s->pending_buf == Z_NULL) {
deflateEnd (strm);
return Z_MEM_ERROR;
}
return deflateReset(strm);
}
/* ========================================================================= */
const Bytef *dictionary;
{
deflate_state *s;
uInt n;
#ifndef USE_DICT_HEAD
#endif
}
s->block_start = (long)length;
/* Insert all strings in the hash table (except for the last two bytes).
* s->lookahead stays null, so s->ins_h will be recomputed at the next
* call of fill_window.
*/
INSERT_STRING(s, n, hash_head);
}
return Z_OK;
}
/* ========================================================================= */
{
deflate_state *s;
s->pending = 0;
s->pending_out = s->pending_buf;
if (s->noheader < 0) {
s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
}
s->last_flush = Z_NO_FLUSH;
_tr_init(s);
lm_init(s);
return Z_OK;
}
/* ========================================================================= */
int level;
int strategy;
{
deflate_state *s;
if (level == Z_DEFAULT_COMPRESSION) {
level = 6;
}
return Z_STREAM_ERROR;
}
/* Flush the last buffer: */
}
}
return err;
}
/* =========================================================================
* Put a short in the pending buffer. The 16-bit value is put in MSB order.
* IN assertion: the stream state is correct and there is enough room in
* pending_buf.
*/
deflate_state *s;
uInt b;
{
}
/* =========================================================================
* Flush as much pending output as possible. All deflate() output goes
* through this function so some applications may wish to modify it
* to avoid allocating a large strm->next_out buffer and copying into it.
* (See also read_buf()).
*/
{
if (len == 0) return;
}
}
/* ========================================================================= */
int flush;
{
deflate_state *s;
return Z_STREAM_ERROR;
}
}
old_flush = s->last_flush;
s->last_flush = flush;
/* Write the zlib header */
if (s->status == INIT_STATE) {
s->status = BUSY_STATE;
putShortMSB(s, header);
/* Save the adler32 of the preset dictionary: */
if (s->strstart != 0) {
}
}
/* Flush as much pending output as possible */
if (s->pending != 0) {
/* Since avail_out is 0, deflate will be called again with
* more output space, but possibly with both pending and
* avail_in equal to zero. There won't be anything to do,
* but this is not an error situation so make sure we
* return OK instead of BUF_ERROR at next call of deflate:
*/
s->last_flush = -1;
return Z_OK;
}
/* Make sure there is something to do and avoid duplicate consecutive
* flushes. For repeated and useless calls with Z_FINISH, we keep
* returning Z_STREAM_END instead of Z_BUFF_ERROR.
*/
}
/* User must not provide more input after the first FINISH: */
}
/* Start a new block or continue the current one.
*/
s->status = FINISH_STATE;
}
}
return Z_OK;
/* If flush != Z_NO_FLUSH && avail_out == 0, the next call
* of deflate should use the same flush parameter to make sure
* that the flush is complete. So we don't have to output an
* empty block here, this will be done at next call. This also
* ensures that for a very small output buffer, we emit at most
* one empty block.
*/
}
if (bstate == block_done) {
if (flush == Z_PARTIAL_FLUSH) {
_tr_align(s);
} else { /* FULL_FLUSH or SYNC_FLUSH */
_tr_stored_block(s, (char*)0, 0L, 0);
/* For a full flush, this empty block will be recognized
* as a special marker by inflate_sync().
*/
if (flush == Z_FULL_FLUSH) {
CLEAR_HASH(s); /* forget history */
}
}
return Z_OK;
}
}
}
if (s->noheader) return Z_STREAM_END;
/* Write the zlib trailer (adler32) */
/* If avail_out is zero, the application will call deflate again
* to flush the rest.
*/
}
/* ========================================================================= */
{
int status;
status != FINISH_STATE) {
return Z_STREAM_ERROR;
}
/* Deallocate in reverse order of allocations: */
}
/* =========================================================================
* Copy the source state to the destination state.
* To simplify the source, this is not supported for 16-bit MSDOS (which
* doesn't have enough memory anyway to duplicate compression states).
*/
{
#ifdef MAXSEG_64K
return Z_STREAM_ERROR;
#else
return Z_STREAM_ERROR;
}
deflateEnd (dest);
return Z_MEM_ERROR;
}
/* following zmemcpy do not work for 16-bit MSDOS */
return Z_OK;
#endif
}
/* ===========================================================================
* Read a new buffer from the current input stream, update the adler32
* and total number of bytes read. All deflate() input goes through
* this function so some applications may wish to modify it to avoid
* allocating a large strm->next_in buffer and copying from it.
* (See also flush_pending()).
*/
unsigned size;
{
if (len == 0) return 0;
}
return (int)len;
}
/* ===========================================================================
* Initialize the "longest match" routines for a new zlib stream
*/
deflate_state *s;
{
CLEAR_HASH(s);
/* Set the default configuration parameters:
*/
s->strstart = 0;
s->block_start = 0L;
s->lookahead = 0;
s->match_available = 0;
s->ins_h = 0;
#ifdef ASMV
match_init(); /* initialize the asm code */
#endif
}
/* ===========================================================================
* Set match_start to the longest match starting at the given string and
* return its length. Matches shorter or equal to prev_length are discarded,
* in which case the result is equal to prev_length and match_start is
* garbage.
* IN assertions: cur_match is the head of the hash chain for the current
* string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
* OUT assertion: the match length is not greater than s->lookahead.
*/
#ifndef ASMV
/* For 80x86 and 680x0, an optimized version will be provided in match.asm or
* match.S. The code will be functionally equivalent.
*/
#ifndef FASTEST
deflate_state *s;
{
/* Stop when cur_match becomes <= limit. To simplify the code,
* we prevent matches with the string of window index 0.
*/
#ifdef UNALIGNED_OK
/* Compare two bytes at a time. Note: this is not always beneficial.
* Try with and without -DUNALIGNED_OK to check.
*/
#else
#endif
/* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
* It is easy to get rid of this optimization if necessary.
*/
/* Do not waste too much time if we already have a good match: */
if (s->prev_length >= s->good_match) {
chain_length >>= 2;
}
/* Do not look for matches beyond the end of the input. This is necessary
* to make deflate deterministic.
*/
do {
/* Skip to next match if the match length cannot increase
* or if the match length is less than 2:
*/
/* This code assumes sizeof(unsigned short) == 2. Do not use
* UNALIGNED_OK if your compiler uses a different size.
*/
/* It is not necessary to compare scan[2] and match[2] since they are
* always equal when the other bytes match, given that the hash keys
* are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
* strstart+3, +5, ... up to strstart+257. We check for insufficient
* lookahead only every 4th comparison; the 128th check will be made
* at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
* necessary to put more guard bytes at the end of the window, or
* to check more often for insufficient lookahead.
*/
do {
/* The funny "do {}" generates better code on most compilers */
/* Here, scan <= window+strstart+257 */
#else /* UNALIGNED_OK */
/* The check at best_len-1 can be removed because it will be made
* again later. (This heuristic is not always a win.)
* It is not necessary to compare scan[2] and match[2] since they
* are always equal when the other bytes match, given that
* the hash keys are equal and that HASH_BITS >= 8.
*/
/* We check for insufficient lookahead only every 8th comparison;
* the 256th check will be made at strstart+258.
*/
do {
#endif /* UNALIGNED_OK */
s->match_start = cur_match;
if (len >= nice_match) break;
#ifdef UNALIGNED_OK
#else
#endif
}
&& --chain_length != 0);
return s->lookahead;
}
#else /* FASTEST */
/* ---------------------------------------------------------------------------
* Optimized version for level == 1 only
*/
deflate_state *s;
{
/* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
* It is easy to get rid of this optimization if necessary.
*/
/* Return failure if the match length is less than 2:
*/
/* The check at best_len-1 can be removed because it will be made
* again later. (This heuristic is not always a win.)
* It is not necessary to compare scan[2] and match[2] since they
* are always equal when the other bytes match, given that
* the hash keys are equal and that HASH_BITS >= 8.
*/
/* We check for insufficient lookahead only every 8th comparison;
* the 256th check will be made at strstart+258.
*/
do {
s->match_start = cur_match;
}
#endif /* FASTEST */
#endif /* ASMV */
#ifdef DEBUG
/* ===========================================================================
* Check that the match at match_start is indeed a match.
*/
deflate_state *s;
int length;
{
/* check that the match is indeed a match */
do {
} while (--length != 0);
z_error("invalid match");
}
if (z_verbose > 1) {
}
}
#else
#endif
/* ===========================================================================
* Fill the window when the lookahead becomes insufficient.
* Updates strstart and lookahead.
*
* IN assertion: lookahead < MIN_LOOKAHEAD
* OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
* At least one byte has been read, or avail_in == 0; reads are
* performed for at least two bytes (required for the zip translate_eol
* option -- not supported here).
*/
deflate_state *s;
{
register unsigned n, m;
register Posf *p;
do {
/* Deal with !@#$% 64K limit: */
} else if (more == (unsigned)(-1)) {
/* Very unlikely, but possible on 16 bit machine if strstart == 0
* and lookahead == 1 (input done one byte at time)
*/
more--;
/* If the window is almost full and there is insufficient lookahead,
* move the upper half to the lower one to make room in the upper half.
*/
s->match_start -= wsize;
s->block_start -= (long) wsize;
/* Slide the hash table (could be avoided with 32 bit values
at the expense of memory usage). We slide even when level == 0
to keep the hash table consistent if we switch back to level > 0
later. (Using level 0 permanently is not an optimal usage of
zlib, so we don't care about this pathological case.)
*/
n = s->hash_size;
p = &s->head[n];
do {
m = *--p;
} while (--n);
n = wsize;
#ifndef FASTEST
p = &s->prev[n];
do {
m = *--p;
/* If n is not on any hash chain, prev[n] is garbage but
* its value will never be used.
*/
} while (--n);
#endif
}
/* If there was no sliding:
* strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
* more == window_size - lookahead - strstart
* => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
* => more >= window_size - 2*WSIZE + 2
* In the BIG_MEM or MMAP case (not yet supported),
* window_size == input_size + MIN_LOOKAHEAD &&
* strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
* Otherwise, window_size == 2*WSIZE so more >= 2.
* If there was sliding, more >= WSIZE. So in all cases, more >= 2.
*/
s->lookahead += n;
/* Initialize the hash value now that we have some input: */
#if MIN_MATCH != 3
#endif
}
/* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
* but this is not important since only literal bytes will be emitted.
*/
}
/* ===========================================================================
* Flush the current block, with given end-of-file flag.
* IN assertion: strstart is set to the end of the current match.
*/
_tr_flush_block(s, (s->block_start >= 0L ? \
(eof)); \
s->block_start = s->strstart; \
flush_pending(s->strm); \
}
/* Same but force premature exit if necessary. */
FLUSH_BLOCK_ONLY(s, eof); \
}
/* ===========================================================================
* Copy without compression as much as possible from the input stream, return
* the current block state.
* This function does not insert new strings in the dictionary since
* uncompressible data is probably not useful. This function is used
* only for the level=0 compression option.
* NOTE: this function should be optimized to avoid extra copying from
* window to pending_buf.
*/
deflate_state *s;
int flush;
{
/* Stored blocks are limited to 0xffff bytes, pending_buf is limited
* to pending_buf_size, and each stored block has a 5 byte header:
*/
}
/* Copy as much as possible from input to output: */
for (;;) {
/* Fill the window as much as possible: */
if (s->lookahead <= 1) {
fill_window(s);
if (s->lookahead == 0) break; /* flush the current block */
}
s->lookahead = 0;
/* Emit a stored block if pending_buf will be full: */
/* strstart == 0 is possible when wraparound on 16-bit machine */
FLUSH_BLOCK(s, 0);
}
/* Flush if we may have to slide, otherwise block_start may become
* negative and the data will be gone:
*/
FLUSH_BLOCK(s, 0);
}
}
}
/* ===========================================================================
* Compress as much as possible from the input stream, return the current
* block state.
* This function does not perform lazy evaluation of matches and inserts
* new strings in the dictionary only for unmatched strings or for short
* matches. It is used only for the fast compression options.
*/
deflate_state *s;
int flush;
{
for (;;) {
/* Make sure that we always have enough lookahead, except
* at the end of the input file. We need MAX_MATCH bytes
* for the next match, plus MIN_MATCH bytes to insert the
* string following the next match.
*/
if (s->lookahead < MIN_LOOKAHEAD) {
fill_window(s);
return need_more;
}
if (s->lookahead == 0) break; /* flush the current block */
}
/* Insert the string window[strstart .. strstart+2] in the
* dictionary, and set hash_head to the head of the hash chain:
*/
}
/* Find the longest match, discarding those <= prev_length.
* At this point we have always match_length < MIN_MATCH
*/
/* To simplify the code, we prevent matches with the string
* of window index 0 (in particular we have to avoid a match
* of the string with itself at the start of the input file).
*/
if (s->strategy != Z_HUFFMAN_ONLY) {
}
/* longest_match() sets match_start */
}
if (s->match_length >= MIN_MATCH) {
s->lookahead -= s->match_length;
/* Insert new strings in the hash table only if the match length
* is not too large. This saves time but degrades compression.
*/
#ifndef FASTEST
if (s->match_length <= s->max_insert_length &&
s->match_length--; /* string at strstart already in hash table */
do {
s->strstart++;
/* strstart never exceeds WSIZE-MAX_MATCH, so there are
* always MIN_MATCH bytes ahead.
*/
} while (--s->match_length != 0);
s->strstart++;
} else
#endif
{
s->strstart += s->match_length;
s->match_length = 0;
#if MIN_MATCH != 3
#endif
/* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
* matter since it will be recomputed at next deflate call.
*/
}
} else {
/* No match, output a literal byte */
s->lookahead--;
s->strstart++;
}
if (bflush) FLUSH_BLOCK(s, 0);
}
}
/* ===========================================================================
* Same as above, but achieves better compression. We use a lazy
* evaluation for matches: a match is finally adopted only if there is
* no better match at the next window position.
*/
deflate_state *s;
int flush;
{
/* Process the input block. */
for (;;) {
/* Make sure that we always have enough lookahead, except
* at the end of the input file. We need MAX_MATCH bytes
* for the next match, plus MIN_MATCH bytes to insert the
* string following the next match.
*/
if (s->lookahead < MIN_LOOKAHEAD) {
fill_window(s);
return need_more;
}
if (s->lookahead == 0) break; /* flush the current block */
}
/* Insert the string window[strstart .. strstart+2] in the
* dictionary, and set hash_head to the head of the hash chain:
*/
}
/* Find the longest match, discarding those <= prev_length.
*/
/* To simplify the code, we prevent matches with the string
* of window index 0 (in particular we have to avoid a match
* of the string with itself at the start of the input file).
*/
if (s->strategy != Z_HUFFMAN_ONLY) {
}
/* longest_match() sets match_start */
(s->match_length == MIN_MATCH &&
/* If prev_match is also MIN_MATCH, match_start is garbage
* but we will ignore the current match anyway.
*/
}
}
/* If there was a match at the previous step and the current
* match is not better, output the previous match:
*/
/* Do not insert strings in hash table beyond this. */
/* Insert in hash table all strings up to the end of the match.
* strstart-1 and strstart are already inserted. If there is not
* enough lookahead, the last two strings are not inserted in
* the hash table.
*/
s->prev_length -= 2;
do {
if (++s->strstart <= max_insert) {
}
} while (--s->prev_length != 0);
s->match_available = 0;
s->strstart++;
if (bflush) FLUSH_BLOCK(s, 0);
} else if (s->match_available) {
/* If there was no match at the previous position, output a
* single literal. If there was a match but the current match
* is longer, truncate the previous match to a single literal.
*/
if (bflush) {
FLUSH_BLOCK_ONLY(s, 0);
}
s->strstart++;
s->lookahead--;
} else {
/* There is no previous match to compare with, wait for
* the next step to decide.
*/
s->match_available = 1;
s->strstart++;
s->lookahead--;
}
}
if (s->match_available) {
s->match_available = 0;
}
}