1 /* vi: set sw=4 ts=4: */
3 * Gzip implementation for busybox
5 * Based on GNU gzip Copyright (C) 1992-1993 Jean-loup Gailly.
7 * Originally adjusted for busybox by Charles P. Wright <cpw@unix.asb.com>
8 * "this is a stripped down version of gzip I put into busybox, it does
9 * only standard in to standard out with -9 compression. It also requires
10 * the zcat module for some important functions."
12 * Adjusted further by Erik Andersen <andersen@codepoet.org> to support
13 * files as well as stdin/stdout, and to generally behave itself wrt
14 * command line handling.
16 * Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
26 #include <sys/types.h>
30 #include <sys/types.h>
37 typedef unsigned char uch;
38 typedef unsigned short ush;
39 typedef unsigned long ulg;
41 /* Return codes from gzip */
46 /* Compression methods (see algorithm.doc) */
47 /* Only STORED and DEFLATED are supported by this BusyBox module */
49 /* methods 4 to 7 reserved */
52 /* To save memory for 16 bit systems, some arrays are overlaid between
53 * the various modules:
54 * deflate: prev+head window d_buf l_buf outbuf
55 * unlzw: tab_prefix tab_suffix stack inbuf outbuf
56 * For compression, input is done in window[]. For decompression, output
57 * is done in window except for unlzw.
62 # define INBUFSIZ 0x2000 /* input buffer size */
64 # define INBUFSIZ 0x8000 /* input buffer size */
67 #define INBUF_EXTRA 64 /* required by unlzw() */
71 # define OUTBUFSIZ 8192 /* output buffer size */
73 # define OUTBUFSIZ 16384 /* output buffer size */
76 #define OUTBUF_EXTRA 2048 /* required by unlzw() */
80 # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */
82 # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */
86 # define DECLARE(type, array, size) static type * array
87 # define ALLOC(type, array, size) { \
88 array = (type*)xzalloc((size_t)(((size)+1L)/2) * 2*sizeof(type)); \
90 # define FREE(array) {free(array), array=NULL;}
92 #define tab_suffix window
93 #define tab_prefix prev /* hash link (see deflate.c) */
94 #define head (prev+WSIZE) /* hash head (see deflate.c) */
96 static long bytes_in; /* number of input bytes */
98 #define isize bytes_in
99 /* for compatibility with old zip sources (to be cleaned) */
101 typedef int file_t; /* Do not use stdio */
103 #define NO_FILE (-1) /* in memory compression */
106 #define PACK_MAGIC "\037\036" /* Magic header for packed files */
107 #define GZIP_MAGIC "\037\213" /* Magic header for gzip files, 1F 8B */
108 #define OLD_GZIP_MAGIC "\037\236" /* Magic header for gzip 0.5 = freeze 1.x */
109 #define LZH_MAGIC "\037\240" /* Magic header for SCO LZH Compress files */
110 #define PKZIP_MAGIC "\120\113\003\004" /* Magic header for pkzip files */
113 #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */
114 #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
115 #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
116 #define ORIG_NAME 0x08 /* bit 3 set: original file name present */
117 #define COMMENT 0x10 /* bit 4 set: file comment present */
118 #define RESERVED 0xC0 /* bit 6,7: reserved */
120 /* internal file attribute */
121 #define UNKNOWN 0xffff
126 # define WSIZE 0x8000 /* window size--must be a power of two, and */
127 #endif /* at least 32K for zip's deflate method */
130 #define MAX_MATCH 258
131 /* The minimum and maximum match lengths */
133 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
134 /* Minimum amount of lookahead, except at the end of the input file.
135 * See deflate.c for comments about the MIN_MATCH+1.
138 #define MAX_DIST (WSIZE-MIN_LOOKAHEAD)
139 /* In order to simplify the code, particularly on 16 bit machines, match
140 * distances are limited to MAX_DIST instead of WSIZE.
143 /* put_byte is used for the compressed output */
144 #define put_byte(c) {outbuf[outcnt++]=(uch)(c); if (outcnt==OUTBUFSIZ)\
148 /* Output a 32 bit value to the bit stream, lsb first */
150 #define put_long(n) { \
151 put_short((n) & 0xffff); \
152 put_short(((ulg)(n)) >> 16); \
156 #define seekable() 0 /* force sequential output */
157 #define translate_eol 0 /* no option -a yet */
159 /* Diagnostic functions */
161 # define Assert(cond,msg) {if(!(cond)) bb_error_msg(msg);}
162 # define Trace(x) fprintf x
163 # define Tracev(x) {if (verbose) fprintf x ;}
164 # define Tracevv(x) {if (verbose>1) fprintf x ;}
165 # define Tracec(c,x) {if (verbose && (c)) fprintf x ;}
166 # define Tracecv(c,x) {if (verbose>1 && (c)) fprintf x ;}
168 # define Assert(cond,msg)
173 # define Tracecv(c,x)
176 #define WARN(msg) {if (!quiet) fprintf msg ; \
177 if (exit_code == OK) exit_code = WARNING;}
180 # define MAX_PATH_LEN 1024 /* max pathname length */
185 static int zip(int in, int out);
186 static int file_read(char *buf, unsigned size);
189 static void lm_init(ush * flags);
190 static ulg deflate(void);
193 static void ct_init(ush * attr, int *methodp);
194 static int ct_tally(int dist, int lc);
195 static ulg flush_block(char *buf, ulg stored_len, int eof);
198 static void bi_init(file_t zipfile);
199 static void send_bits(int value, int length);
200 static unsigned bi_reverse(unsigned value, int length);
201 static void bi_windup(void);
202 static void copy_block(char *buf, unsigned len, int header);
203 static int (*read_buf) (char *buf, unsigned size);
206 static void flush_outbuf(void);
208 /* lzw.h -- define the lzw functions.
209 * Copyright (C) 1992-1993 Jean-loup Gailly.
210 * This is free software; you can redistribute it and/or modify it under the
211 * terms of the GNU General Public License, see the file COPYING.
217 #define INIT_BITS 9 /* Initial number of bits per code */
219 #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */
220 /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free.
221 * It's a pity that old uncompress does not check bit 0x20. That makes
222 * extension of the format actually undesirable because old compress
223 * would just crash on the new format instead of giving a meaningful
224 * error message. It does check the number of bits, but it's more
225 * helpful to say "unsupported format, get a new version" than
226 * "can only handle 16 bits".
229 /* tailor.h -- target dependent definitions
230 * Copyright (C) 1992-1993 Jean-loup Gailly.
231 * This is free software; you can redistribute it and/or modify it under the
232 * terms of the GNU General Public License, see the file COPYING.
235 /* The target dependent definitions should be defined here only.
236 * The target dependent functions should be defined in tailor.c.
240 /* Common defaults */
243 # define OS_CODE 0x03 /* assume Unix */
247 # define PATH_SEP '/'
251 # define OPTIONS_VAR "GZIP"
255 # define Z_SUFFIX ".gz"
259 # define MAX_SUFFIX MAX_EXT_CHARS
261 # define MAX_SUFFIX 30
266 DECLARE(uch, inbuf, INBUFSIZ + INBUF_EXTRA);
267 DECLARE(uch, outbuf, OUTBUFSIZ + OUTBUF_EXTRA);
268 DECLARE(ush, d_buf, DIST_BUFSIZE);
269 DECLARE(uch, window, 2L * WSIZE);
270 DECLARE(ush, tab_prefix, 1L << BITS);
272 static int foreground; /* set if program run in foreground */
273 static int method = DEFLATED; /* compression method */
274 static int exit_code = OK; /* program exit code */
275 static int part_nb; /* number of parts in .gz file */
276 static long time_stamp; /* original time stamp (modification time) */
277 static long ifile_size; /* input file size, -1 for devices (debug only) */
278 static char z_suffix[MAX_SUFFIX + 1]; /* default suffix (can be set with --suffix) */
279 static int z_len; /* strlen(z_suffix) */
281 static int ifd; /* input file descriptor */
282 static int ofd; /* output file descriptor */
283 static unsigned insize; /* valid bytes in inbuf */
284 static unsigned outcnt; /* bytes in output buffer */
286 static uint32_t *crc_32_tab;
288 /* Output a 16 bit value, lsb first */
289 static void put_short(ush w)
291 if (outcnt < OUTBUFSIZ - 2) {
292 outbuf[outcnt++] = (uch) ((w) & 0xff);
293 outbuf[outcnt++] = (uch) ((ush) (w) >> 8);
295 put_byte((uch) ((w) & 0xff));
296 put_byte((uch) ((ush) (w) >> 8));
300 /* ========================================================================
301 * Signal and error handler.
303 static void abort_gzip(int ATTRIBUTE_UNUSED ignored)
308 /* ===========================================================================
309 * Clear input and output buffers
311 static void clear_bufs(void)
318 /* ===========================================================================
319 * Does the same as write(), but also handles partial pipe writes and checks
322 static void write_buf(int fd, void *buf, unsigned cnt)
326 while ((n = write(fd, buf, cnt)) != cnt) {
327 if (n == (unsigned) (-1)) bb_error_msg_and_die(bb_msg_write_error);
329 buf = (void *) ((char *) buf + n);
333 /* ===========================================================================
334 * Run a set of bytes through the crc shift register. If s is a NULL
335 * pointer, then initialize the crc shift register contents instead.
336 * Return the current crc in either case.
338 static uint32_t updcrc(uch * s, unsigned n)
340 static uint32_t crc = ~0; /* shift register contents */
341 uint32_t c; /* temporary variable */
349 c = crc_32_tab[((int) c ^ (*s++)) & 0xff] ^ (c >> 8);
356 /* bits.c -- output variable-length bit strings
357 * Copyright (C) 1992-1993 Jean-loup Gailly
358 * This is free software; you can redistribute it and/or modify it under the
359 * terms of the GNU General Public License, see the file COPYING.
366 * Output variable-length bit strings. Compression can be done
367 * to a file or to memory. (The latter is not supported in this version.)
371 * The PKZIP "deflate" file format interprets compressed file data
372 * as a sequence of bits. Multi-bit strings in the file may cross
373 * byte boundaries without restriction.
375 * The first bit of each byte is the low-order bit.
377 * The routines in this file allow a variable-length bit value to
378 * be output right-to-left (useful for literal values). For
379 * left-to-right output (useful for code strings from the tree routines),
380 * the bits must have been reversed first with bi_reverse().
382 * For in-memory compression, the compressed bit stream goes directly
383 * into the requested output buffer. The input data is read in blocks
384 * by the mem_read() function. The buffer is limited to 64K on 16 bit
389 * void bi_init (FILE *zipfile)
390 * Initialize the bit string routines.
392 * void send_bits (int value, int length)
393 * Write out a bit string, taking the source bits right to
396 * int bi_reverse (int value, int length)
397 * Reverse the bits of a bit string, taking the source bits left to
398 * right and emitting them right to left.
400 * void bi_windup (void)
401 * Write out any remaining bits in an incomplete byte.
403 * void copy_block(char *buf, unsigned len, int header)
404 * Copy a stored block to the zip file, storing first the length and
405 * its one's complement if requested.
409 /* ===========================================================================
410 * Local data used by the "bit string" routines.
413 static file_t zfile; /* output gzip file */
415 static unsigned short bi_buf;
417 /* Output buffer. bits are inserted starting at the bottom (least significant
421 #define Buf_size (8 * 2*sizeof(char))
422 /* Number of bits used within bi_buf. (bi_buf might be implemented on
423 * more than 16 bits on some systems.)
428 /* Current input function. Set to mem_read for in-memory compression */
431 ulg bits_sent; /* bit length of the compressed data */
434 /* ===========================================================================
435 * Initialize the bit string routines.
437 static void bi_init(file_t zipfile)
446 /* Set the defaults for file compression. They are set by memcompress
447 * for in-memory compression.
449 if (zfile != NO_FILE) {
450 read_buf = file_read;
454 /* ===========================================================================
455 * Send a value on a given number of bits.
456 * IN assertion: length <= 16 and value fits in length bits.
458 static void send_bits(int value, int length)
461 Tracev((stderr, " l %2d v %4x ", length, value));
462 Assert(length > 0 && length <= 15, "invalid length");
463 bits_sent += (ulg) length;
465 /* If not enough room in bi_buf, use (valid) bits from bi_buf and
466 * (16 - bi_valid) bits from value, leaving (width - (16-bi_valid))
467 * unused bits in value.
469 if (bi_valid > (int) Buf_size - length) {
470 bi_buf |= (value << bi_valid);
472 bi_buf = (ush) value >> (Buf_size - bi_valid);
473 bi_valid += length - Buf_size;
475 bi_buf |= value << bi_valid;
480 /* ===========================================================================
481 * Reverse the first len bits of a code, using straightforward code (a faster
482 * method would use a table)
483 * IN assertion: 1 <= len <= 15
485 static unsigned bi_reverse(unsigned code, int len)
487 register unsigned res = 0;
491 code >>= 1, res <<= 1;
496 /* ===========================================================================
497 * Write out any remaining bits in an incomplete byte.
499 static void bi_windup(void)
503 } else if (bi_valid > 0) {
509 bits_sent = (bits_sent + 7) & ~7;
513 /* ===========================================================================
514 * Copy a stored block to the zip file, storing first the length and its
515 * one's complement if requested.
517 static void copy_block(char *buf, unsigned len, int header)
519 bi_windup(); /* align on byte boundary */
522 put_short((ush) len);
523 put_short((ush) ~ len);
529 bits_sent += (ulg) len << 3;
536 /* deflate.c -- compress data using the deflation algorithm
537 * Copyright (C) 1992-1993 Jean-loup Gailly
538 * This is free software; you can redistribute it and/or modify it under the
539 * terms of the GNU General Public License, see the file COPYING.
545 * Identify new text as repetitions of old text within a fixed-
546 * length sliding window trailing behind the new text.
550 * The "deflation" process depends on being able to identify portions
551 * of the input text which are identical to earlier input (within a
552 * sliding window trailing behind the input currently being processed).
554 * The most straightforward technique turns out to be the fastest for
555 * most input files: try all possible matches and select the longest.
556 * The key feature of this algorithm is that insertions into the string
557 * dictionary are very simple and thus fast, and deletions are avoided
558 * completely. Insertions are performed at each input character, whereas
559 * string matches are performed only when the previous match ends. So it
560 * is preferable to spend more time in matches to allow very fast string
561 * insertions and avoid deletions. The matching algorithm for small
562 * strings is inspired from that of Rabin & Karp. A brute force approach
563 * is used to find longer strings when a small match has been found.
564 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
565 * (by Leonid Broukhis).
566 * A previous version of this file used a more sophisticated algorithm
567 * (by Fiala and Greene) which is guaranteed to run in linear amortized
568 * time, but has a larger average cost, uses more memory and is patented.
569 * However the F&G algorithm may be faster for some highly redundant
570 * files if the parameter max_chain_length (described below) is too large.
574 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
575 * I found it in 'freeze' written by Leonid Broukhis.
576 * Thanks to many info-zippers for bug reports and testing.
580 * APPNOTE.TXT documentation file in PKZIP 1.93a distribution.
582 * A description of the Rabin and Karp algorithm is given in the book
583 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
585 * Fiala,E.R., and Greene,D.H.
586 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
590 * void lm_init (int pack_level, ush *flags)
591 * Initialize the "longest match" routines for a new file
594 * Processes a new input file and return its compressed length. Sets
595 * the compressed length, crc, deflate flags and internal file
600 /* ===========================================================================
601 * Configuration parameters
604 /* Compile with MEDIUM_MEM to reduce the memory requirements or
605 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
606 * entire input file can be held in memory (not possible on 16 bit systems).
607 * Warning: defining these symbols affects HASH_BITS (see below) and thus
608 * affects the compression ratio. The compressed output
609 * is still correct, and might even be smaller in some cases.
613 # define HASH_BITS 13 /* Number of bits used to hash strings */
616 # define HASH_BITS 14
619 # define HASH_BITS 15
620 /* For portability to 16 bit machines, do not use values above 15. */
623 /* To save space (see unlzw.c), we overlay prev+head with tab_prefix and
624 * window with tab_suffix. Check that we can do this:
626 #if (WSIZE<<1) > (1<<BITS)
627 # error cannot overlay window with tab_suffix and prev with tab_prefix0
629 #if HASH_BITS > BITS-1
630 # error cannot overlay head with tab_prefix1
632 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
633 #define HASH_MASK (HASH_SIZE-1)
634 #define WMASK (WSIZE-1)
635 /* HASH_SIZE and WSIZE must be powers of two */
637 /* Tail of hash chains */
640 /* speed options for the general purpose bit flag */
642 # define TOO_FAR 4096
644 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
645 /* ===========================================================================
646 * Local data used by the "longest match" routines.
649 typedef unsigned IPos;
651 /* A Pos is an index in the character window. We use short instead of int to
652 * save space in the various tables. IPos is used only for parameter passing.
655 /* DECLARE(uch, window, 2L*WSIZE); */
656 /* Sliding window. Input bytes are read into the second half of the window,
657 * and move to the first half later to keep a dictionary of at least WSIZE
658 * bytes. With this organization, matches are limited to a distance of
659 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
660 * performed with a length multiple of the block size. Also, it limits
661 * the window size to 64K, which is quite useful on MSDOS.
662 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
663 * be less efficient).
666 /* DECLARE(Pos, prev, WSIZE); */
667 /* Link to older string with same hash index. To limit the size of this
668 * array to 64K, this link is maintained only for the last 32K strings.
669 * An index in this array is thus a window index modulo 32K.
672 /* DECLARE(Pos, head, 1<<HASH_BITS); */
673 /* Heads of the hash chains or NIL. */
675 static const ulg window_size = (ulg) 2 * WSIZE;
677 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
678 * input file length plus MIN_LOOKAHEAD.
681 static long block_start;
683 /* window position at the beginning of the current output block. Gets
684 * negative when the window is moved backwards.
687 static unsigned ins_h; /* hash index of string to be inserted */
689 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1)/MIN_MATCH)
690 /* Number of bits by which ins_h and del_h must be shifted at each
691 * input step. It must be such that after MIN_MATCH steps, the oldest
692 * byte no longer takes part in the hash key, that is:
693 * H_SHIFT * MIN_MATCH >= HASH_BITS
696 static unsigned int prev_length;
698 /* Length of the best match at previous step. Matches not greater than this
699 * are discarded. This is used in the lazy match evaluation.
702 static unsigned strstart; /* start of string to insert */
703 static unsigned match_start; /* start of matching string */
704 static int eofile; /* flag set at end of input file */
705 static unsigned lookahead; /* number of valid bytes ahead in window */
708 max_chain_length = 4096,
710 /* To speed up deflation, hash chains are never searched beyond this length.
711 * A higher limit improves compression ratio but degrades the speed.
714 max_lazy_match = 258,
716 /* Attempt to find a better match only when the current match is strictly
717 * smaller than this value. This mechanism is used only for compression
720 max_insert_length = max_lazy_match,
721 /* Insert new strings in the hash table only if the match length
722 * is not greater than this length. This saves time but degrades compression.
723 * max_insert_length is used only for compression levels <= 3.
728 /* Use a faster search when the previous match is longer than this */
731 /* Values for max_lazy_match, good_match and max_chain_length, depending on
732 * the desired pack level (0..9). The values given below have been tuned to
733 * exclude worst case performance for pathological files. Better values may be
734 * found for specific files.
737 nice_match = 258 /* Stop searching when current match exceeds this */
739 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
740 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
746 /* result of memcmp for equal strings */
748 /* ===========================================================================
749 * Prototypes for local functions.
751 static void fill_window(void);
753 static int longest_match(IPos cur_match);
756 static void check_match(IPos start, IPos match, int length);
759 /* ===========================================================================
760 * Update a hash value with the given input byte
761 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
762 * input characters, so that a running hash key can be computed from the
763 * previous key instead of complete recalculation each time.
765 #define UPDATE_HASH(h,c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
767 /* ===========================================================================
768 * Insert string s in the dictionary and set match_head to the previous head
769 * of the hash chain (the most recent string with same hash key). Return
770 * the previous length of the hash chain.
771 * IN assertion: all calls to to INSERT_STRING are made with consecutive
772 * input characters and the first MIN_MATCH bytes of s are valid
773 * (except for the last MIN_MATCH-1 bytes of the input file).
775 #define INSERT_STRING(s, match_head) \
776 (UPDATE_HASH(ins_h, window[(s) + MIN_MATCH-1]), \
777 prev[(s) & WMASK] = match_head = head[ins_h], \
780 /* ===========================================================================
781 * Initialize the "longest match" routines for a new file
783 static void lm_init(ush * flags)
787 /* Initialize the hash table. */
788 memset(head, 0, HASH_SIZE * sizeof(*head));
789 /* prev will be initialized on the fly */
792 /* ??? reduce max_chain_length for binary files */
797 lookahead = read_buf((char *) window,
798 sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE);
800 if (lookahead == 0 || lookahead == (unsigned) EOF) {
801 eofile = 1, lookahead = 0;
805 /* Make sure that we always have enough lookahead. This is important
806 * if input comes from a device such as a tty.
808 while (lookahead < MIN_LOOKAHEAD && !eofile)
812 for (j = 0; j < MIN_MATCH - 1; j++)
813 UPDATE_HASH(ins_h, window[j]);
814 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
815 * not important since only literal bytes will be emitted.
819 /* ===========================================================================
820 * Set match_start to the longest match starting at the given string and
821 * return its length. Matches shorter or equal to prev_length are discarded,
822 * in which case the result is equal to prev_length and match_start is
824 * IN assertions: cur_match is the head of the hash chain for the current
825 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
828 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
829 * match.s. The code is functionally equivalent, so you can use the C version
832 static int longest_match(IPos cur_match)
834 unsigned chain_length = max_chain_length; /* max hash chain length */
835 register uch *scan = window + strstart; /* current string */
836 register uch *match; /* matched string */
837 register int len; /* length of current match */
838 int best_len = prev_length; /* best match length so far */
840 strstart > (IPos) MAX_DIST ? strstart - (IPos) MAX_DIST : NIL;
841 /* Stop when cur_match becomes <= limit. To simplify the code,
842 * we prevent matches with the string of window index 0.
845 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
846 * It is easy to get rid of this optimization if necessary.
848 #if HASH_BITS < 8 || MAX_MATCH != 258
849 # error Code too clever
851 register uch *strend = window + strstart + MAX_MATCH;
852 register uch scan_end1 = scan[best_len - 1];
853 register uch scan_end = scan[best_len];
855 /* Do not waste too much time if we already have a good match: */
856 if (prev_length >= good_match) {
859 Assert(strstart <= window_size - MIN_LOOKAHEAD, "insufficient lookahead");
862 Assert(cur_match < strstart, "no future");
863 match = window + cur_match;
865 /* Skip to next match if the match length cannot increase
866 * or if the match length is less than 2:
868 if (match[best_len] != scan_end ||
869 match[best_len - 1] != scan_end1 ||
870 *match != *scan || *++match != scan[1])
873 /* The check at best_len-1 can be removed because it will be made
874 * again later. (This heuristic is not always a win.)
875 * It is not necessary to compare scan[2] and match[2] since they
876 * are always equal when the other bytes match, given that
877 * the hash keys are equal and that HASH_BITS >= 8.
881 /* We check for insufficient lookahead only every 8th comparison;
882 * the 256th check will be made at strstart+258.
885 } while (*++scan == *++match && *++scan == *++match &&
886 *++scan == *++match && *++scan == *++match &&
887 *++scan == *++match && *++scan == *++match &&
888 *++scan == *++match && *++scan == *++match && scan < strend);
890 len = MAX_MATCH - (int) (strend - scan);
891 scan = strend - MAX_MATCH;
893 if (len > best_len) {
894 match_start = cur_match;
896 if (len >= nice_match)
898 scan_end1 = scan[best_len - 1];
899 scan_end = scan[best_len];
901 } while ((cur_match = prev[cur_match & WMASK]) > limit
902 && --chain_length != 0);
908 /* ===========================================================================
909 * Check that the match at match_start is indeed a match.
911 static void check_match(IPos start, IPos match, int length)
913 /* check that the match is indeed a match */
914 if (memcmp((char *) window + match,
915 (char *) window + start, length) != EQUAL) {
916 bb_error_msg(" start %d, match %d, length %d", start, match, length);
917 bb_error_msg("invalid match");
920 bb_error_msg("\\[%d,%d]", start - match, length);
922 putc(window[start++], stderr);
923 } while (--length != 0);
927 # define check_match(start, match, length)
930 /* ===========================================================================
931 * Fill the window when the lookahead becomes insufficient.
932 * Updates strstart and lookahead, and sets eofile if end of input file.
933 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
934 * OUT assertions: at least one byte has been read, or eofile is set;
935 * file reads are performed for at least two bytes (required for the
936 * translate_eol option).
938 static void fill_window(void)
940 register unsigned n, m;
942 (unsigned) (window_size - (ulg) lookahead - (ulg) strstart);
943 /* Amount of free space at the end of the window. */
945 /* If the window is almost full and there is insufficient lookahead,
946 * move the upper half to the lower one to make room in the upper half.
948 if (more == (unsigned) EOF) {
949 /* Very unlikely, but possible on 16 bit machine if strstart == 0
950 * and lookahead == 1 (input done one byte at time)
953 } else if (strstart >= WSIZE + MAX_DIST) {
954 /* By the IN assertion, the window is not empty so we can't confuse
955 * more == 0 with more == 64K on a 16 bit machine.
957 Assert(window_size == (ulg) 2 * WSIZE, "no sliding with BIG_MEM");
959 memcpy((char *) window, (char *) window + WSIZE, (unsigned) WSIZE);
960 match_start -= WSIZE;
961 strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
963 block_start -= (long) WSIZE;
965 for (n = 0; n < HASH_SIZE; n++) {
967 head[n] = (Pos) (m >= WSIZE ? m - WSIZE : NIL);
969 for (n = 0; n < WSIZE; n++) {
971 prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : NIL);
972 /* If n is not on any hash chain, prev[n] is garbage but
973 * its value will never be used.
978 /* At this point, more >= 2 */
980 n = read_buf((char *) window + strstart + lookahead, more);
981 if (n == 0 || n == (unsigned) EOF) {
989 /* ===========================================================================
990 * Flush the current block, with given end-of-file flag.
991 * IN assertion: strstart is set to the end of the current match.
993 #define FLUSH_BLOCK(eof) \
994 flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \
995 (char*)NULL, (long)strstart - block_start, (eof))
997 /* ===========================================================================
998 * Same as above, but achieves better compression. We use a lazy
999 * evaluation for matches: a match is finally adopted only if there is
1000 * no better match at the next window position.
1002 static ulg deflate(void)
1004 IPos hash_head; /* head of hash chain */
1005 IPos prev_match; /* previous match */
1006 int flush; /* set if current block must be flushed */
1007 int match_available = 0; /* set if previous match exists */
1008 register unsigned match_length = MIN_MATCH - 1; /* length of best match */
1010 /* Process the input block. */
1011 while (lookahead != 0) {
1012 /* Insert the string window[strstart .. strstart+2] in the
1013 * dictionary, and set hash_head to the head of the hash chain:
1015 INSERT_STRING(strstart, hash_head);
1017 /* Find the longest match, discarding those <= prev_length.
1019 prev_length = match_length, prev_match = match_start;
1020 match_length = MIN_MATCH - 1;
1022 if (hash_head != NIL && prev_length < max_lazy_match &&
1023 strstart - hash_head <= MAX_DIST) {
1024 /* To simplify the code, we prevent matches with the string
1025 * of window index 0 (in particular we have to avoid a match
1026 * of the string with itself at the start of the input file).
1028 match_length = longest_match(hash_head);
1029 /* longest_match() sets match_start */
1030 if (match_length > lookahead)
1031 match_length = lookahead;
1033 /* Ignore a length 3 match if it is too distant: */
1034 if (match_length == MIN_MATCH && strstart - match_start > TOO_FAR) {
1035 /* If prev_match is also MIN_MATCH, match_start is garbage
1036 * but we will ignore the current match anyway.
1041 /* If there was a match at the previous step and the current
1042 * match is not better, output the previous match:
1044 if (prev_length >= MIN_MATCH && match_length <= prev_length) {
1046 check_match(strstart - 1, prev_match, prev_length);
1049 ct_tally(strstart - 1 - prev_match, prev_length - MIN_MATCH);
1051 /* Insert in hash table all strings up to the end of the match.
1052 * strstart-1 and strstart are already inserted.
1054 lookahead -= prev_length - 1;
1058 INSERT_STRING(strstart, hash_head);
1059 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1060 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
1061 * these bytes are garbage, but it does not matter since the
1062 * next lookahead bytes will always be emitted as literals.
1064 } while (--prev_length != 0);
1065 match_available = 0;
1066 match_length = MIN_MATCH - 1;
1069 FLUSH_BLOCK(0), block_start = strstart;
1071 } else if (match_available) {
1072 /* If there was no match at the previous position, output a
1073 * single literal. If there was a match but the current match
1074 * is longer, truncate the previous match to a single literal.
1076 Tracevv((stderr, "%c", window[strstart - 1]));
1077 if (ct_tally(0, window[strstart - 1])) {
1078 FLUSH_BLOCK(0), block_start = strstart;
1083 /* There is no previous match to compare with, wait for
1084 * the next step to decide.
1086 match_available = 1;
1090 Assert(strstart <= isize && lookahead <= isize, "a bit too far");
1092 /* Make sure that we always have enough lookahead, except
1093 * at the end of the input file. We need MAX_MATCH bytes
1094 * for the next match, plus MIN_MATCH bytes to insert the
1095 * string following the next match.
1097 while (lookahead < MIN_LOOKAHEAD && !eofile)
1100 if (match_available)
1101 ct_tally(0, window[strstart - 1]);
1103 return FLUSH_BLOCK(1); /* eof */
1106 /* gzip (GNU zip) -- compress files with zip algorithm and 'compress' interface
1107 * Copyright (C) 1992-1993 Jean-loup Gailly
1108 * The unzip code was written and put in the public domain by Mark Adler.
1109 * Portions of the lzw code are derived from the public domain 'compress'
1110 * written by Spencer Thomas, Joe Orost, James Woods, Jim McKie, Steve Davies,
1111 * Ken Turkowski, Dave Mack and Peter Jannesen.
1113 * See the license_msg below and the file COPYING for the software license.
1114 * See the file algorithm.doc for the compression algorithms and file formats.
1117 /* Compress files with zip algorithm and 'compress' interface.
1118 * See usage() and help() functions below for all options.
1120 * file.gz: compressed file with same mode, owner, and utimes
1121 * or stdout with -c option or if stdin used as input.
1122 * If the output file name had to be truncated, the original name is kept
1123 * in the compressed file.
1128 typedef struct dirent dir_type;
1130 /* ======================================================================== */
1131 int gzip_main(int argc, char **argv)
1136 struct stat statBuf;
1142 while ((opt = getopt(argc, argv, "cf123456789dq")) != -1) {
1150 /* Ignore 1-9 (compression level) options */
1163 #ifdef CONFIG_GUNZIP
1166 return gunzip_main(argc, argv);
1173 foreground = signal(SIGINT, SIG_IGN) != SIG_IGN;
1175 (void) signal(SIGINT, abort_gzip);
1178 if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
1179 (void) signal(SIGTERM, abort_gzip);
1183 if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
1184 (void) signal(SIGHUP, abort_gzip);
1188 strncpy(z_suffix, Z_SUFFIX, sizeof(z_suffix) - 1);
1189 z_len = strlen(z_suffix);
1191 /* Allocate all global buffers (for DYN_ALLOC option) */
1192 ALLOC(uch, inbuf, INBUFSIZ + INBUF_EXTRA);
1193 ALLOC(uch, outbuf, OUTBUFSIZ + OUTBUF_EXTRA);
1194 ALLOC(ush, d_buf, DIST_BUFSIZE);
1195 ALLOC(uch, window, 2L * WSIZE);
1196 ALLOC(ush, tab_prefix, 1L << BITS);
1198 /* Initialise the CRC32 table */
1199 crc_32_tab = bb_crc32_filltable(0);
1204 if (optind == argc) {
1207 zip(STDIN_FILENO, STDOUT_FILENO);
1211 for (i = optind; i < argc; i++) {
1215 if (strcmp(argv[i], "-") == 0) {
1218 inFileNum = STDIN_FILENO;
1219 outFileNum = STDOUT_FILENO;
1221 inFileNum = bb_xopen3(argv[i], O_RDONLY, 0);
1222 if (fstat(inFileNum, &statBuf) < 0)
1223 bb_perror_msg_and_die("%s", argv[i]);
1224 time_stamp = statBuf.st_ctime;
1225 ifile_size = statBuf.st_size;
1228 path = xmalloc(strlen(argv[i]) + 4);
1229 strcpy(path, argv[i]);
1230 strcat(path, ".gz");
1232 /* Open output file */
1233 #if (__GLIBC__ >= 2) && (__GLIBC_MINOR__ >= 1) && defined O_NOFOLLOW
1235 open(path, O_RDWR | O_CREAT | O_EXCL | O_NOFOLLOW);
1237 outFileNum = open(path, O_RDWR | O_CREAT | O_EXCL);
1239 if (outFileNum < 0) {
1240 bb_perror_msg("%s", path);
1245 /* Set permissions on the file */
1246 fchmod(outFileNum, statBuf.st_mode);
1248 outFileNum = STDOUT_FILENO;
1251 if (path == NULL && isatty(outFileNum) && force == 0) {
1253 ("compressed data not written to a terminal. Use -f to force compression.");
1258 result = zip(inFileNum, outFileNum);
1264 /* Delete the original file */
1266 delFileName = argv[i];
1270 if (unlink(delFileName) < 0)
1271 bb_perror_msg("%s", delFileName);
1281 /* trees.c -- output deflated data using Huffman coding
1282 * Copyright (C) 1992-1993 Jean-loup Gailly
1283 * This is free software; you can redistribute it and/or modify it under the
1284 * terms of the GNU General Public License, see the file COPYING.
1290 * Encode various sets of source values using variable-length
1291 * binary code trees.
1295 * The PKZIP "deflation" process uses several Huffman trees. The more
1296 * common source values are represented by shorter bit sequences.
1298 * Each code tree is stored in the ZIP file in a compressed form
1299 * which is itself a Huffman encoding of the lengths of
1300 * all the code strings (in ascending order by source values).
1301 * The actual code strings are reconstructed from the lengths in
1302 * the UNZIP process, as described in the "application note"
1303 * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program.
1308 * Data Compression: Techniques and Applications, pp. 53-55.
1309 * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7.
1312 * Data Compression: Methods and Theory, pp. 49-50.
1313 * Computer Science Press, 1988. ISBN 0-7167-8156-5.
1317 * Addison-Wesley, 1983. ISBN 0-201-06672-6.
1321 * void ct_init (ush *attr, int *methodp)
1322 * Allocate the match buffer, initialize the various tables and save
1323 * the location of the internal file attribute (ascii/binary) and
1324 * method (DEFLATE/STORE)
1326 * void ct_tally (int dist, int lc);
1327 * Save the match info and tally the frequency counts.
1329 * long flush_block (char *buf, ulg stored_len, int eof)
1330 * Determine the best encoding for the current block: dynamic trees,
1331 * static trees or store, and output the encoded block to the zip
1332 * file. Returns the total compressed length for the file so far.
1336 /* ===========================================================================
1341 /* All codes must not exceed MAX_BITS bits */
1343 #define MAX_BL_BITS 7
1344 /* Bit length codes must not exceed MAX_BL_BITS bits */
1346 #define LENGTH_CODES 29
1347 /* number of length codes, not counting the special END_BLOCK code */
1349 #define LITERALS 256
1350 /* number of literal bytes 0..255 */
1352 #define END_BLOCK 256
1353 /* end of block literal code */
1355 #define L_CODES (LITERALS+1+LENGTH_CODES)
1356 /* number of Literal or Length codes, including the END_BLOCK code */
1359 /* number of distance codes */
1362 /* number of codes used to transfer the bit lengths */
1364 typedef uch extra_bits_t;
1366 /* extra bits for each length code */
1367 static const extra_bits_t extra_lbits[LENGTH_CODES]
1368 = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4,
1372 /* extra bits for each distance code */
1373 static const extra_bits_t extra_dbits[D_CODES]
1374 = { 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,
1375 10, 10, 11, 11, 12, 12, 13, 13
1378 /* extra bits for each bit length code */
1379 static const extra_bits_t extra_blbits[BL_CODES]
1380 = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 };
1382 #define STORED_BLOCK 0
1383 #define STATIC_TREES 1
1385 /* The three kinds of block type */
1389 # define LIT_BUFSIZE 0x2000
1392 # define LIT_BUFSIZE 0x4000
1394 # define LIT_BUFSIZE 0x8000
1398 #ifndef DIST_BUFSIZE
1399 # define DIST_BUFSIZE LIT_BUFSIZE
1401 /* Sizes of match buffers for literals/lengths and distances. There are
1402 * 4 reasons for limiting LIT_BUFSIZE to 64K:
1403 * - frequencies can be kept in 16 bit counters
1404 * - if compression is not successful for the first block, all input data is
1405 * still in the window so we can still emit a stored block even when input
1406 * comes from standard input. (This can also be done for all blocks if
1407 * LIT_BUFSIZE is not greater than 32K.)
1408 * - if compression is not successful for a file smaller than 64K, we can
1409 * even emit a stored file instead of a stored block (saving 5 bytes).
1410 * - creating new Huffman trees less frequently may not provide fast
1411 * adaptation to changes in the input data statistics. (Take for
1412 * example a binary file with poorly compressible code followed by
1413 * a highly compressible string table.) Smaller buffer sizes give
1414 * fast adaptation but have of course the overhead of transmitting trees
1416 * - I can't count above 4
1417 * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save
1418 * memory at the expense of compression). Some optimizations would be possible
1419 * if we rely on DIST_BUFSIZE == LIT_BUFSIZE.
1421 #if LIT_BUFSIZE > INBUFSIZ
1422 #error cannot overlay l_buf and inbuf
1425 /* repeat previous bit length 3-6 times (2 bits of repeat count) */
1426 #define REPZ_3_10 17
1427 /* repeat a zero length 3-10 times (3 bits of repeat count) */
1428 #define REPZ_11_138 18
1429 /* repeat a zero length 11-138 times (7 bits of repeat count) */
1431 /* ===========================================================================
1435 /* Data structure describing a single value and its code string. */
1436 typedef struct ct_data {
1438 ush freq; /* frequency count */
1439 ush code; /* bit string */
1442 ush dad; /* father node in Huffman tree */
1443 ush len; /* length of bit string */
1447 #define Freq fc.freq
1448 #define Code fc.code
1452 #define HEAP_SIZE (2*L_CODES+1)
1453 /* maximum heap size */
1455 static ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */
1456 static ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */
1458 static ct_data static_ltree[L_CODES + 2];
1460 /* The static literal tree. Since the bit lengths are imposed, there is no
1461 * need for the L_CODES extra codes used during heap construction. However
1462 * The codes 286 and 287 are needed to build a canonical tree (see ct_init
1466 static ct_data static_dtree[D_CODES];
1468 /* The static distance tree. (Actually a trivial tree since all codes use
1472 static ct_data bl_tree[2 * BL_CODES + 1];
1474 /* Huffman tree for the bit lengths */
1476 typedef struct tree_desc {
1477 ct_data *dyn_tree; /* the dynamic tree */
1478 ct_data *static_tree; /* corresponding static tree or NULL */
1479 const extra_bits_t *extra_bits; /* extra bits for each code or NULL */
1480 int extra_base; /* base index for extra_bits */
1481 int elems; /* max number of elements in the tree */
1482 int max_length; /* max bit length for the codes */
1483 int max_code; /* largest code with non zero frequency */
1486 static tree_desc l_desc =
1487 { dyn_ltree, static_ltree, extra_lbits, LITERALS + 1, L_CODES,
1491 static tree_desc d_desc =
1492 { dyn_dtree, static_dtree, extra_dbits, 0, D_CODES, MAX_BITS, 0 };
1494 static tree_desc bl_desc =
1495 { bl_tree, (ct_data *) 0, extra_blbits, 0, BL_CODES, MAX_BL_BITS,
1500 static ush bl_count[MAX_BITS + 1];
1502 /* number of codes at each bit length for an optimal tree */
1504 static const uch bl_order[BL_CODES]
1505 = { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
1507 /* The lengths of the bit length codes are sent in order of decreasing
1508 * probability, to avoid transmitting the lengths for unused bit length codes.
1511 static int heap[2 * L_CODES + 1]; /* heap used to build the Huffman trees */
1512 static int heap_len; /* number of elements in the heap */
1513 static int heap_max; /* element of largest frequency */
1515 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
1516 * The same heap array is used to build all trees.
1519 static uch depth[2 * L_CODES + 1];
1521 /* Depth of each subtree used as tie breaker for trees of equal frequency */
1523 static uch length_code[MAX_MATCH - MIN_MATCH + 1];
1525 /* length code for each normalized match length (0 == MIN_MATCH) */
1527 static uch dist_code[512];
1529 /* distance codes. The first 256 values correspond to the distances
1530 * 3 .. 258, the last 256 values correspond to the top 8 bits of
1531 * the 15 bit distances.
1534 static int base_length[LENGTH_CODES];
1536 /* First normalized length for each code (0 = MIN_MATCH) */
1538 static int base_dist[D_CODES];
1540 /* First normalized distance for each code (0 = distance of 1) */
1543 /* DECLARE(uch, l_buf, LIT_BUFSIZE); buffer for literals or lengths */
1545 /* DECLARE(ush, d_buf, DIST_BUFSIZE); buffer for distances */
1547 static uch flag_buf[(LIT_BUFSIZE / 8)];
1549 /* flag_buf is a bit array distinguishing literals from lengths in
1550 * l_buf, thus indicating the presence or absence of a distance.
1553 static unsigned last_lit; /* running index in l_buf */
1554 static unsigned last_dist; /* running index in d_buf */
1555 static unsigned last_flags; /* running index in flag_buf */
1556 static uch flags; /* current flags not yet saved in flag_buf */
1557 static uch flag_bit; /* current bit used in flags */
1559 /* bits are filled in flags starting at bit 0 (least significant).
1560 * Note: these flags are overkill in the current code since we don't
1561 * take advantage of DIST_BUFSIZE == LIT_BUFSIZE.
1564 static ulg opt_len; /* bit length of current block with optimal trees */
1565 static ulg static_len; /* bit length of current block with static trees */
1567 static ulg compressed_len; /* total bit length of compressed file */
1570 static ush *file_type; /* pointer to UNKNOWN, BINARY or ASCII */
1571 static int *file_method; /* pointer to DEFLATE or STORE */
1573 /* ===========================================================================
1574 * Local (static) routines in this file.
1577 static void init_block(void);
1578 static void pqdownheap(ct_data * tree, int k);
1579 static void gen_bitlen(tree_desc * desc);
1580 static void gen_codes(ct_data * tree, int max_code);
1581 static void build_tree(tree_desc * desc);
1582 static void scan_tree(ct_data * tree, int max_code);
1583 static void send_tree(ct_data * tree, int max_code);
1584 static int build_bl_tree(void);
1585 static void send_all_trees(int lcodes, int dcodes, int blcodes);
1586 static void compress_block(ct_data * ltree, ct_data * dtree);
1587 static void set_file_type(void);
1591 # define send_code(c, tree) send_bits(tree[c].Code, tree[c].Len)
1592 /* Send a code of the given tree. c and tree must not have side effects */
1595 # define send_code(c, tree) \
1596 { if (verbose>1) bb_error_msg("\ncd %3d ",(c)); \
1597 send_bits(tree[c].Code, tree[c].Len); }
1600 #define d_code(dist) \
1601 ((dist) < 256 ? dist_code[dist] : dist_code[256+((dist)>>7)])
1602 /* Mapping from a distance to a distance code. dist is the distance - 1 and
1603 * must not have side effects. dist_code[256] and dist_code[257] are never
1607 /* the arguments must not have side effects */
1609 /* ===========================================================================
1610 * Allocate the match buffer, initialize the various tables and save the
1611 * location of the internal file attribute (ascii/binary) and method
1614 static void ct_init(ush * attr, int *methodp)
1616 int n; /* iterates over tree elements */
1617 int bits; /* bit counter */
1618 int length; /* length value */
1619 int code; /* code value */
1620 int dist; /* distance index */
1623 file_method = methodp;
1624 compressed_len = 0L;
1626 if (static_dtree[0].Len != 0)
1627 return; /* ct_init already called */
1629 /* Initialize the mapping length (0..255) -> length code (0..28) */
1631 for (code = 0; code < LENGTH_CODES - 1; code++) {
1632 base_length[code] = length;
1633 for (n = 0; n < (1 << extra_lbits[code]); n++) {
1634 length_code[length++] = (uch) code;
1637 Assert(length == 256, "ct_init: length != 256");
1638 /* Note that the length 255 (match length 258) can be represented
1639 * in two different ways: code 284 + 5 bits or code 285, so we
1640 * overwrite length_code[255] to use the best encoding:
1642 length_code[length - 1] = (uch) code;
1644 /* Initialize the mapping dist (0..32K) -> dist code (0..29) */
1646 for (code = 0; code < 16; code++) {
1647 base_dist[code] = dist;
1648 for (n = 0; n < (1 << extra_dbits[code]); n++) {
1649 dist_code[dist++] = (uch) code;
1652 Assert(dist == 256, "ct_init: dist != 256");
1653 dist >>= 7; /* from now on, all distances are divided by 128 */
1654 for (; code < D_CODES; code++) {
1655 base_dist[code] = dist << 7;
1656 for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) {
1657 dist_code[256 + dist++] = (uch) code;
1660 Assert(dist == 256, "ct_init: 256+dist != 512");
1662 /* Construct the codes of the static literal tree */
1663 for (bits = 0; bits <= MAX_BITS; bits++)
1667 static_ltree[n++].Len = 8, bl_count[8]++;
1669 static_ltree[n++].Len = 9, bl_count[9]++;
1671 static_ltree[n++].Len = 7, bl_count[7]++;
1673 static_ltree[n++].Len = 8, bl_count[8]++;
1674 /* Codes 286 and 287 do not exist, but we must include them in the
1675 * tree construction to get a canonical Huffman tree (longest code
1678 gen_codes((ct_data *) static_ltree, L_CODES + 1);
1680 /* The static distance tree is trivial: */
1681 for (n = 0; n < D_CODES; n++) {
1682 static_dtree[n].Len = 5;
1683 static_dtree[n].Code = bi_reverse(n, 5);
1686 /* Initialize the first block of the first file: */
1690 /* ===========================================================================
1691 * Initialize a new block.
1693 static void init_block(void)
1695 int n; /* iterates over tree elements */
1697 /* Initialize the trees. */
1698 for (n = 0; n < L_CODES; n++)
1699 dyn_ltree[n].Freq = 0;
1700 for (n = 0; n < D_CODES; n++)
1701 dyn_dtree[n].Freq = 0;
1702 for (n = 0; n < BL_CODES; n++)
1703 bl_tree[n].Freq = 0;
1705 dyn_ltree[END_BLOCK].Freq = 1;
1706 opt_len = static_len = 0L;
1707 last_lit = last_dist = last_flags = 0;
1713 /* Index within the heap array of least frequent node in the Huffman tree */
1716 /* ===========================================================================
1717 * Remove the smallest element from the heap and recreate the heap with
1718 * one less element. Updates heap and heap_len.
1720 #define pqremove(tree, top) \
1722 top = heap[SMALLEST]; \
1723 heap[SMALLEST] = heap[heap_len--]; \
1724 pqdownheap(tree, SMALLEST); \
1727 /* ===========================================================================
1728 * Compares to subtrees, using the tree depth as tie breaker when
1729 * the subtrees have equal frequency. This minimizes the worst case length.
1731 #define smaller(tree, n, m) \
1732 (tree[n].Freq < tree[m].Freq || \
1733 (tree[n].Freq == tree[m].Freq && depth[n] <= depth[m]))
1735 /* ===========================================================================
1736 * Restore the heap property by moving down the tree starting at node k,
1737 * exchanging a node with the smallest of its two sons if necessary, stopping
1738 * when the heap property is re-established (each father smaller than its
1741 static void pqdownheap(ct_data * tree, int k)
1744 int j = k << 1; /* left son of k */
1746 while (j <= heap_len) {
1747 /* Set j to the smallest of the two sons: */
1748 if (j < heap_len && smaller(tree, heap[j + 1], heap[j]))
1751 /* Exit if v is smaller than both sons */
1752 if (smaller(tree, v, heap[j]))
1755 /* Exchange v with the smallest son */
1759 /* And continue down the tree, setting j to the left son of k */
1765 /* ===========================================================================
1766 * Compute the optimal bit lengths for a tree and update the total bit length
1767 * for the current block.
1768 * IN assertion: the fields freq and dad are set, heap[heap_max] and
1769 * above are the tree nodes sorted by increasing frequency.
1770 * OUT assertions: the field len is set to the optimal bit length, the
1771 * array bl_count contains the frequencies for each bit length.
1772 * The length opt_len is updated; static_len is also updated if stree is
1775 static void gen_bitlen(tree_desc * desc)
1777 ct_data *tree = desc->dyn_tree;
1778 const extra_bits_t *extra = desc->extra_bits;
1779 int base = desc->extra_base;
1780 int max_code = desc->max_code;
1781 int max_length = desc->max_length;
1782 ct_data *stree = desc->static_tree;
1783 int h; /* heap index */
1784 int n, m; /* iterate over the tree elements */
1785 int bits; /* bit length */
1786 int xbits; /* extra bits */
1787 ush f; /* frequency */
1788 int overflow = 0; /* number of elements with bit length too large */
1790 for (bits = 0; bits <= MAX_BITS; bits++)
1793 /* In a first pass, compute the optimal bit lengths (which may
1794 * overflow in the case of the bit length tree).
1796 tree[heap[heap_max]].Len = 0; /* root of the heap */
1798 for (h = heap_max + 1; h < HEAP_SIZE; h++) {
1800 bits = tree[tree[n].Dad].Len + 1;
1801 if (bits > max_length)
1802 bits = max_length, overflow++;
1803 tree[n].Len = (ush) bits;
1804 /* We overwrite tree[n].Dad which is no longer needed */
1807 continue; /* not a leaf node */
1812 xbits = extra[n - base];
1814 opt_len += (ulg) f *(bits + xbits);
1817 static_len += (ulg) f *(stree[n].Len + xbits);
1822 Trace((stderr, "\nbit length overflow\n"));
1823 /* This happens for example on obj2 and pic of the Calgary corpus */
1825 /* Find the first bit length which could increase: */
1827 bits = max_length - 1;
1828 while (bl_count[bits] == 0)
1830 bl_count[bits]--; /* move one leaf down the tree */
1831 bl_count[bits + 1] += 2; /* move one overflow item as its brother */
1832 bl_count[max_length]--;
1833 /* The brother of the overflow item also moves one step up,
1834 * but this does not affect bl_count[max_length]
1837 } while (overflow > 0);
1839 /* Now recompute all bit lengths, scanning in increasing frequency.
1840 * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all
1841 * lengths instead of fixing only the wrong ones. This idea is taken
1842 * from 'ar' written by Haruhiko Okumura.)
1844 for (bits = max_length; bits != 0; bits--) {
1850 if (tree[m].Len != (unsigned) bits) {
1851 Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len,
1854 ((long) bits - (long) tree[m].Len) * (long) tree[m].Freq;
1855 tree[m].Len = (ush) bits;
1862 /* ===========================================================================
1863 * Generate the codes for a given tree and bit counts (which need not be
1865 * IN assertion: the array bl_count contains the bit length statistics for
1866 * the given tree and the field len is set for all tree elements.
1867 * OUT assertion: the field code is set for all tree elements of non
1870 static void gen_codes(ct_data * tree, int max_code)
1872 ush next_code[MAX_BITS + 1]; /* next code value for each bit length */
1873 ush code = 0; /* running code value */
1874 int bits; /* bit index */
1875 int n; /* code index */
1877 /* The distribution counts are first used to generate the code values
1878 * without bit reversal.
1880 for (bits = 1; bits <= MAX_BITS; bits++) {
1881 next_code[bits] = code = (code + bl_count[bits - 1]) << 1;
1883 /* Check that the bit counts in bl_count are consistent. The last code
1886 Assert(code + bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1,
1887 "inconsistent bit counts");
1888 Tracev((stderr, "\ngen_codes: max_code %d ", max_code));
1890 for (n = 0; n <= max_code; n++) {
1891 int len = tree[n].Len;
1895 /* Now reverse the bits */
1896 tree[n].Code = bi_reverse(next_code[len]++, len);
1898 Tracec(tree != static_ltree,
1899 (stderr, "\nn %3d %c l %2d c %4x (%x) ", n,
1900 (isgraph(n) ? n : ' '), len, tree[n].Code,
1901 next_code[len] - 1));
1905 /* ===========================================================================
1906 * Construct one Huffman tree and assigns the code bit strings and lengths.
1907 * Update the total bit length for the current block.
1908 * IN assertion: the field freq is set for all tree elements.
1909 * OUT assertions: the fields len and code are set to the optimal bit length
1910 * and corresponding code. The length opt_len is updated; static_len is
1911 * also updated if stree is not null. The field max_code is set.
1913 static void build_tree(tree_desc * desc)
1915 ct_data *tree = desc->dyn_tree;
1916 ct_data *stree = desc->static_tree;
1917 int elems = desc->elems;
1918 int n, m; /* iterate over heap elements */
1919 int max_code = -1; /* largest code with non zero frequency */
1920 int node = elems; /* next internal node of the tree */
1922 /* Construct the initial heap, with least frequent element in
1923 * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1].
1924 * heap[0] is not used.
1926 heap_len = 0, heap_max = HEAP_SIZE;
1928 for (n = 0; n < elems; n++) {
1929 if (tree[n].Freq != 0) {
1930 heap[++heap_len] = max_code = n;
1937 /* The pkzip format requires that at least one distance code exists,
1938 * and that at least one bit should be sent even if there is only one
1939 * possible code. So to avoid special checks later on we force at least
1940 * two codes of non zero frequency.
1942 while (heap_len < 2) {
1943 int new = heap[++heap_len] = (max_code < 2 ? ++max_code : 0);
1949 static_len -= stree[new].Len;
1950 /* new is 0 or 1 so it does not have extra bits */
1952 desc->max_code = max_code;
1954 /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree,
1955 * establish sub-heaps of increasing lengths:
1957 for (n = heap_len / 2; n >= 1; n--)
1958 pqdownheap(tree, n);
1960 /* Construct the Huffman tree by repeatedly combining the least two
1964 pqremove(tree, n); /* n = node of least frequency */
1965 m = heap[SMALLEST]; /* m = node of next least frequency */
1967 heap[--heap_max] = n; /* keep the nodes sorted by frequency */
1968 heap[--heap_max] = m;
1970 /* Create a new node father of n and m */
1971 tree[node].Freq = tree[n].Freq + tree[m].Freq;
1972 depth[node] = (uch) (MAX(depth[n], depth[m]) + 1);
1973 tree[n].Dad = tree[m].Dad = (ush) node;
1975 if (tree == bl_tree) {
1976 bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)",
1977 node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq);
1980 /* and insert the new node in the heap */
1981 heap[SMALLEST] = node++;
1982 pqdownheap(tree, SMALLEST);
1984 } while (heap_len >= 2);
1986 heap[--heap_max] = heap[SMALLEST];
1988 /* At this point, the fields freq and dad are set. We can now
1989 * generate the bit lengths.
1991 gen_bitlen((tree_desc *) desc);
1993 /* The field len is now set, we can generate the bit codes */
1994 gen_codes((ct_data *) tree, max_code);
1997 /* ===========================================================================
1998 * Scan a literal or distance tree to determine the frequencies of the codes
1999 * in the bit length tree. Updates opt_len to take into account the repeat
2000 * counts. (The contribution of the bit length codes will be added later
2001 * during the construction of bl_tree.)
2003 static void scan_tree(ct_data * tree, int max_code)
2005 int n; /* iterates over all tree elements */
2006 int prevlen = -1; /* last emitted length */
2007 int curlen; /* length of current code */
2008 int nextlen = tree[0].Len; /* length of next code */
2009 int count = 0; /* repeat count of the current code */
2010 int max_count = 7; /* max repeat count */
2011 int min_count = 4; /* min repeat count */
2014 max_count = 138, min_count = 3;
2015 tree[max_code + 1].Len = (ush) 0xffff; /* guard */
2017 for (n = 0; n <= max_code; n++) {
2019 nextlen = tree[n + 1].Len;
2020 if (++count < max_count && curlen == nextlen) {
2022 } else if (count < min_count) {
2023 bl_tree[curlen].Freq += count;
2024 } else if (curlen != 0) {
2025 if (curlen != prevlen)
2026 bl_tree[curlen].Freq++;
2027 bl_tree[REP_3_6].Freq++;
2028 } else if (count <= 10) {
2029 bl_tree[REPZ_3_10].Freq++;
2031 bl_tree[REPZ_11_138].Freq++;
2036 max_count = 138, min_count = 3;
2037 } else if (curlen == nextlen) {
2038 max_count = 6, min_count = 3;
2040 max_count = 7, min_count = 4;
2045 /* ===========================================================================
2046 * Send a literal or distance tree in compressed form, using the codes in
2049 static void send_tree(ct_data * tree, int max_code)
2051 int n; /* iterates over all tree elements */
2052 int prevlen = -1; /* last emitted length */
2053 int curlen; /* length of current code */
2054 int nextlen = tree[0].Len; /* length of next code */
2055 int count = 0; /* repeat count of the current code */
2056 int max_count = 7; /* max repeat count */
2057 int min_count = 4; /* min repeat count */
2059 /* tree[max_code+1].Len = -1; *//* guard already set */
2061 max_count = 138, min_count = 3;
2063 for (n = 0; n <= max_code; n++) {
2065 nextlen = tree[n + 1].Len;
2066 if (++count < max_count && curlen == nextlen) {
2068 } else if (count < min_count) {
2070 send_code(curlen, bl_tree);
2071 } while (--count != 0);
2073 } else if (curlen != 0) {
2074 if (curlen != prevlen) {
2075 send_code(curlen, bl_tree);
2078 Assert(count >= 3 && count <= 6, " 3_6?");
2079 send_code(REP_3_6, bl_tree);
2080 send_bits(count - 3, 2);
2082 } else if (count <= 10) {
2083 send_code(REPZ_3_10, bl_tree);
2084 send_bits(count - 3, 3);
2087 send_code(REPZ_11_138, bl_tree);
2088 send_bits(count - 11, 7);
2093 max_count = 138, min_count = 3;
2094 } else if (curlen == nextlen) {
2095 max_count = 6, min_count = 3;
2097 max_count = 7, min_count = 4;
2102 /* ===========================================================================
2103 * Construct the Huffman tree for the bit lengths and return the index in
2104 * bl_order of the last bit length code to send.
2106 static int build_bl_tree(void)
2108 int max_blindex; /* index of last bit length code of non zero freq */
2110 /* Determine the bit length frequencies for literal and distance trees */
2111 scan_tree((ct_data *) dyn_ltree, l_desc.max_code);
2112 scan_tree((ct_data *) dyn_dtree, d_desc.max_code);
2114 /* Build the bit length tree: */
2115 build_tree((tree_desc *) (&bl_desc));
2116 /* opt_len now includes the length of the tree representations, except
2117 * the lengths of the bit lengths codes and the 5+5+4 bits for the counts.
2120 /* Determine the number of bit length codes to send. The pkzip format
2121 * requires that at least 4 bit length codes be sent. (appnote.txt says
2122 * 3 but the actual value used is 4.)
2124 for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) {
2125 if (bl_tree[bl_order[max_blindex]].Len != 0)
2128 /* Update opt_len to include the bit length tree and counts */
2129 opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4;
2130 Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", opt_len, static_len));
2135 /* ===========================================================================
2136 * Send the header for a block using dynamic Huffman trees: the counts, the
2137 * lengths of the bit length codes, the literal tree and the distance tree.
2138 * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4.
2140 static void send_all_trees(int lcodes, int dcodes, int blcodes)
2142 int rank; /* index in bl_order */
2144 Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes");
2145 Assert(lcodes <= L_CODES && dcodes <= D_CODES
2146 && blcodes <= BL_CODES, "too many codes");
2147 Tracev((stderr, "\nbl counts: "));
2148 send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */
2149 send_bits(dcodes - 1, 5);
2150 send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */
2151 for (rank = 0; rank < blcodes; rank++) {
2152 Tracev((stderr, "\nbl code %2d ", bl_order[rank]));
2153 send_bits(bl_tree[bl_order[rank]].Len, 3);
2155 Tracev((stderr, "\nbl tree: sent %ld", bits_sent));
2157 send_tree((ct_data *) dyn_ltree, lcodes - 1); /* send the literal tree */
2158 Tracev((stderr, "\nlit tree: sent %ld", bits_sent));
2160 send_tree((ct_data *) dyn_dtree, dcodes - 1); /* send the distance tree */
2161 Tracev((stderr, "\ndist tree: sent %ld", bits_sent));
2164 /* ===========================================================================
2165 * Determine the best encoding for the current block: dynamic trees, static
2166 * trees or store, and output the encoded block to the zip file. This function
2167 * returns the total compressed length for the file so far.
2169 static ulg flush_block(char *buf, ulg stored_len, int eof)
2171 ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
2172 int max_blindex; /* index of last bit length code of non zero freq */
2174 flag_buf[last_flags] = flags; /* Save the flags for the last 8 items */
2176 /* Check if the file is ascii or binary */
2177 if (*file_type == (ush) UNKNOWN)
2180 /* Construct the literal and distance trees */
2181 build_tree((tree_desc *) (&l_desc));
2182 Tracev((stderr, "\nlit data: dyn %ld, stat %ld", opt_len, static_len));
2184 build_tree((tree_desc *) (&d_desc));
2185 Tracev((stderr, "\ndist data: dyn %ld, stat %ld", opt_len, static_len));
2186 /* At this point, opt_len and static_len are the total bit lengths of
2187 * the compressed block data, excluding the tree representations.
2190 /* Build the bit length tree for the above two trees, and get the index
2191 * in bl_order of the last bit length code to send.
2193 max_blindex = build_bl_tree();
2195 /* Determine the best encoding. Compute first the block length in bytes */
2196 opt_lenb = (opt_len + 3 + 7) >> 3;
2197 static_lenb = (static_len + 3 + 7) >> 3;
2200 "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ",
2201 opt_lenb, opt_len, static_lenb, static_len, stored_len,
2202 last_lit, last_dist));
2204 if (static_lenb <= opt_lenb)
2205 opt_lenb = static_lenb;
2207 /* If compression failed and this is the first and last block,
2208 * and if the zip file can be seeked (to rewrite the local header),
2209 * the whole file is transformed into a stored file:
2211 if (stored_len <= opt_lenb && eof && compressed_len == 0L && seekable()) {
2212 /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */
2213 if (buf == (char *) 0)
2214 bb_error_msg("block vanished");
2216 copy_block(buf, (unsigned) stored_len, 0); /* without header */
2217 compressed_len = stored_len << 3;
2218 *file_method = STORED;
2220 } else if (stored_len + 4 <= opt_lenb && buf != (char *) 0) {
2221 /* 4: two words for the lengths */
2222 /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE.
2223 * Otherwise we can't have processed more than WSIZE input bytes since
2224 * the last block flush, because compression would have been
2225 * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to
2226 * transform a block into a stored block.
2228 send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */
2229 compressed_len = (compressed_len + 3 + 7) & ~7L;
2230 compressed_len += (stored_len + 4) << 3;
2232 copy_block(buf, (unsigned) stored_len, 1); /* with header */
2234 } else if (static_lenb == opt_lenb) {
2235 send_bits((STATIC_TREES << 1) + eof, 3);
2236 compress_block((ct_data *) static_ltree, (ct_data *) static_dtree);
2237 compressed_len += 3 + static_len;
2239 send_bits((DYN_TREES << 1) + eof, 3);
2240 send_all_trees(l_desc.max_code + 1, d_desc.max_code + 1,
2242 compress_block((ct_data *) dyn_ltree, (ct_data *) dyn_dtree);
2243 compressed_len += 3 + opt_len;
2245 Assert(compressed_len == bits_sent, "bad compressed size");
2250 compressed_len += 7; /* align on byte boundary */
2252 Tracev((stderr, "\ncomprlen %lu(%lu) ", compressed_len >> 3,
2253 compressed_len - 7 * eof));
2255 return compressed_len >> 3;
2258 /* ===========================================================================
2259 * Save the match info and tally the frequency counts. Return true if
2260 * the current block must be flushed.
2262 static int ct_tally(int dist, int lc)
2264 l_buf[last_lit++] = (uch) lc;
2266 /* lc is the unmatched char */
2267 dyn_ltree[lc].Freq++;
2269 /* Here, lc is the match length - MIN_MATCH */
2270 dist--; /* dist = match distance - 1 */
2271 Assert((ush) dist < (ush) MAX_DIST &&
2272 (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH) &&
2273 (ush) d_code(dist) < (ush) D_CODES, "ct_tally: bad match");
2275 dyn_ltree[length_code[lc] + LITERALS + 1].Freq++;
2276 dyn_dtree[d_code(dist)].Freq++;
2278 d_buf[last_dist++] = (ush) dist;
2283 /* Output the flags if they fill a byte: */
2284 if ((last_lit & 7) == 0) {
2285 flag_buf[last_flags++] = flags;
2286 flags = 0, flag_bit = 1;
2288 /* Try to guess if it is profitable to stop the current block here */
2289 if ((last_lit & 0xfff) == 0) {
2290 /* Compute an upper bound for the compressed length */
2291 ulg out_length = (ulg) last_lit * 8L;
2292 ulg in_length = (ulg) strstart - block_start;
2295 for (dcode = 0; dcode < D_CODES; dcode++) {
2297 (ulg) dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]);
2301 "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ",
2302 last_lit, last_dist, in_length, out_length,
2303 100L - out_length * 100L / in_length));
2304 if (last_dist < last_lit / 2 && out_length < in_length / 2)
2307 return (last_lit == LIT_BUFSIZE - 1 || last_dist == DIST_BUFSIZE);
2308 /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K
2309 * on 16 bit machines and because stored blocks are restricted to
2314 /* ===========================================================================
2315 * Send the block data compressed using the given Huffman trees
2317 static void compress_block(ct_data * ltree, ct_data * dtree)
2319 unsigned dist; /* distance of matched string */
2320 int lc; /* match length or unmatched char (if dist == 0) */
2321 unsigned lx = 0; /* running index in l_buf */
2322 unsigned dx = 0; /* running index in d_buf */
2323 unsigned fx = 0; /* running index in flag_buf */
2324 uch flag = 0; /* current flags */
2325 unsigned code; /* the code to send */
2326 int extra; /* number of extra bits to send */
2331 flag = flag_buf[fx++];
2333 if ((flag & 1) == 0) {
2334 send_code(lc, ltree); /* send a literal byte */
2335 Tracecv(isgraph(lc), (stderr, " '%c' ", lc));
2337 /* Here, lc is the match length - MIN_MATCH */
2338 code = length_code[lc];
2339 send_code(code + LITERALS + 1, ltree); /* send the length code */
2340 extra = extra_lbits[code];
2342 lc -= base_length[code];
2343 send_bits(lc, extra); /* send the extra length bits */
2346 /* Here, dist is the match distance - 1 */
2347 code = d_code(dist);
2348 Assert(code < D_CODES, "bad d_code");
2350 send_code(code, dtree); /* send the distance code */
2351 extra = extra_dbits[code];
2353 dist -= base_dist[code];
2354 send_bits(dist, extra); /* send the extra distance bits */
2356 } /* literal or match pair ? */
2358 } while (lx < last_lit);
2360 send_code(END_BLOCK, ltree);
2363 /* ===========================================================================
2364 * Set the file type to ASCII or BINARY, using a crude approximation:
2365 * binary if more than 20% of the bytes are <= 6 or >= 128, ascii otherwise.
2366 * IN assertion: the fields freq of dyn_ltree are set and the total of all
2367 * frequencies does not exceed 64K (to fit in an int on 16 bit machines).
2369 static void set_file_type(void)
2372 unsigned ascii_freq = 0;
2373 unsigned bin_freq = 0;
2376 bin_freq += dyn_ltree[n++].Freq;
2378 ascii_freq += dyn_ltree[n++].Freq;
2379 while (n < LITERALS)
2380 bin_freq += dyn_ltree[n++].Freq;
2381 *file_type = bin_freq > (ascii_freq >> 2) ? BINARY : ASCII;
2382 if (*file_type == BINARY && translate_eol) {
2383 bb_error_msg("-l used on binary file");
2387 /* zip.c -- compress files to the gzip or pkzip format
2388 * Copyright (C) 1992-1993 Jean-loup Gailly
2389 * This is free software; you can redistribute it and/or modify it under the
2390 * terms of the GNU General Public License, see the file COPYING.
2394 static uint32_t crc; /* crc on uncompressed file data */
2395 static long header_bytes; /* number of bytes in gzip header */
2397 static void put_long(ulg n)
2399 put_short((n) & 0xffff);
2400 put_short(((ulg) (n)) >> 16);
2403 /* put_header_byte is used for the compressed output
2404 * - for the initial 4 bytes that can't overflow the buffer.
2406 #define put_header_byte(c) {outbuf[outcnt++]=(uch)(c);}
2408 /* ===========================================================================
2409 * Deflate in to out.
2410 * IN assertions: the input and output buffers are cleared.
2411 * The variables time_stamp and save_orig_name are initialized.
2413 static int zip(int in, int out)
2415 uch my_flags = 0; /* general purpose bit flags */
2416 ush attr = 0; /* ascii/binary flag */
2417 ush deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */
2423 /* Write the header to the gzip file. See algorithm.doc for the format */
2427 put_header_byte(GZIP_MAGIC[0]); /* magic header */
2428 put_header_byte(GZIP_MAGIC[1]);
2429 put_header_byte(DEFLATED); /* compression method */
2431 put_header_byte(my_flags); /* general flags */
2432 put_long(time_stamp);
2434 /* Write deflated file to zip file */
2438 ct_init(&attr, &method);
2439 lm_init(&deflate_flags);
2441 put_byte((uch) deflate_flags); /* extra flags */
2442 put_byte(OS_CODE); /* OS identifier */
2444 header_bytes = (long) outcnt;
2448 /* Write the crc and uncompressed size */
2451 header_bytes += 2 * sizeof(long);
2458 /* ===========================================================================
2459 * Read a new buffer from the current input file, perform end-of-line
2460 * translation, and update the crc and input file size.
2461 * IN assertion: size >= 2 (for end-of-line translation)
2463 static int file_read(char *buf, unsigned size)
2467 Assert(insize == 0, "inbuf not empty");
2469 len = read(ifd, buf, size);
2470 if (len == (unsigned) (-1) || len == 0)
2473 crc = updcrc((uch *) buf, len);
2478 /* ===========================================================================
2479 * Write the output buffer outbuf[0..outcnt-1] and update bytes_out.
2480 * (used for the compressed data only)
2482 static void flush_outbuf(void)
2487 write_buf(ofd, (char *) outbuf, outcnt);