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.
19 /* These defines are very important for BusyBox. Without these,
20 * huge chunks of ram are pre-allocated making the BusyBox bss
21 * size Freaking Huge(tm), which is a bad thing.*/
30 #include <sys/types.h>
34 #include <sys/types.h>
41 typedef unsigned char uch;
42 typedef unsigned short ush;
43 typedef unsigned long ulg;
45 /* Return codes from gzip */
50 /* Compression methods (see algorithm.doc) */
51 /* Only STORED and DEFLATED are supported by this BusyBox module */
53 /* methods 4 to 7 reserved */
56 /* To save memory for 16 bit systems, some arrays are overlaid between
57 * the various modules:
58 * deflate: prev+head window d_buf l_buf outbuf
59 * unlzw: tab_prefix tab_suffix stack inbuf outbuf
60 * For compression, input is done in window[]. For decompression, output
61 * is done in window except for unlzw.
66 # define INBUFSIZ 0x2000 /* input buffer size */
68 # define INBUFSIZ 0x8000 /* input buffer size */
71 #define INBUF_EXTRA 64 /* required by unlzw() */
75 # define OUTBUFSIZ 8192 /* output buffer size */
77 # define OUTBUFSIZ 16384 /* output buffer size */
80 #define OUTBUF_EXTRA 2048 /* required by unlzw() */
84 # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */
86 # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */
91 # define DECLARE(type, array, size) static type * array
92 # define ALLOC(type, array, size) { \
93 array = (type*)xcalloc((size_t)(((size)+1L)/2), 2*sizeof(type)); \
95 # define FREE(array) {free(array), array=NULL;}
97 # define DECLARE(type, array, size) static type array[size]
98 # define ALLOC(type, array, size)
102 #define tab_suffix window
103 #define tab_prefix prev /* hash link (see deflate.c) */
104 #define head (prev+WSIZE) /* hash head (see deflate.c) */
106 static long bytes_in; /* number of input bytes */
108 #define isize bytes_in
109 /* for compatibility with old zip sources (to be cleaned) */
111 typedef int file_t; /* Do not use stdio */
113 #define NO_FILE (-1) /* in memory compression */
116 #define PACK_MAGIC "\037\036" /* Magic header for packed files */
117 #define GZIP_MAGIC "\037\213" /* Magic header for gzip files, 1F 8B */
118 #define OLD_GZIP_MAGIC "\037\236" /* Magic header for gzip 0.5 = freeze 1.x */
119 #define LZH_MAGIC "\037\240" /* Magic header for SCO LZH Compress files */
120 #define PKZIP_MAGIC "\120\113\003\004" /* Magic header for pkzip files */
123 #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */
124 #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
125 #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
126 #define ORIG_NAME 0x08 /* bit 3 set: original file name present */
127 #define COMMENT 0x10 /* bit 4 set: file comment present */
128 #define RESERVED 0xC0 /* bit 6,7: reserved */
130 /* internal file attribute */
131 #define UNKNOWN 0xffff
136 # define WSIZE 0x8000 /* window size--must be a power of two, and */
137 #endif /* at least 32K for zip's deflate method */
140 #define MAX_MATCH 258
141 /* The minimum and maximum match lengths */
143 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
144 /* Minimum amount of lookahead, except at the end of the input file.
145 * See deflate.c for comments about the MIN_MATCH+1.
148 #define MAX_DIST (WSIZE-MIN_LOOKAHEAD)
149 /* In order to simplify the code, particularly on 16 bit machines, match
150 * distances are limited to MAX_DIST instead of WSIZE.
153 /* put_byte is used for the compressed output */
154 #define put_byte(c) {outbuf[outcnt++]=(uch)(c); if (outcnt==OUTBUFSIZ)\
158 /* Output a 32 bit value to the bit stream, lsb first */
160 #define put_long(n) { \
161 put_short((n) & 0xffff); \
162 put_short(((ulg)(n)) >> 16); \
166 #define seekable() 0 /* force sequential output */
167 #define translate_eol 0 /* no option -a yet */
169 /* Diagnostic functions */
171 # define Assert(cond,msg) {if(!(cond)) bb_error_msg(msg);}
172 # define Trace(x) fprintf x
173 # define Tracev(x) {if (verbose) fprintf x ;}
174 # define Tracevv(x) {if (verbose>1) fprintf x ;}
175 # define Tracec(c,x) {if (verbose && (c)) fprintf x ;}
176 # define Tracecv(c,x) {if (verbose>1 && (c)) fprintf x ;}
178 # define Assert(cond,msg)
183 # define Tracecv(c,x)
186 #define WARN(msg) {if (!quiet) fprintf msg ; \
187 if (exit_code == OK) exit_code = WARNING;}
190 # define MAX_PATH_LEN 1024 /* max pathname length */
195 static int zip(int in, int out);
196 static int file_read(char *buf, unsigned size);
199 static void lm_init(ush * flags);
200 static ulg deflate(void);
203 static void ct_init(ush * attr, int *methodp);
204 static int ct_tally(int dist, int lc);
205 static ulg flush_block(char *buf, ulg stored_len, int eof);
208 static void bi_init(file_t zipfile);
209 static void send_bits(int value, int length);
210 static unsigned bi_reverse(unsigned value, int length);
211 static void bi_windup(void);
212 static void copy_block(char *buf, unsigned len, int header);
213 static int (*read_buf) (char *buf, unsigned size);
216 static void flush_outbuf(void);
218 /* lzw.h -- define the lzw functions.
219 * Copyright (C) 1992-1993 Jean-loup Gailly.
220 * This is free software; you can redistribute it and/or modify it under the
221 * terms of the GNU General Public License, see the file COPYING.
227 #define INIT_BITS 9 /* Initial number of bits per code */
229 #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */
230 /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free.
231 * It's a pity that old uncompress does not check bit 0x20. That makes
232 * extension of the format actually undesirable because old compress
233 * would just crash on the new format instead of giving a meaningful
234 * error message. It does check the number of bits, but it's more
235 * helpful to say "unsupported format, get a new version" than
236 * "can only handle 16 bits".
239 /* tailor.h -- target dependent definitions
240 * Copyright (C) 1992-1993 Jean-loup Gailly.
241 * This is free software; you can redistribute it and/or modify it under the
242 * terms of the GNU General Public License, see the file COPYING.
245 /* The target dependent definitions should be defined here only.
246 * The target dependent functions should be defined in tailor.c.
250 /* Common defaults */
253 # define OS_CODE 0x03 /* assume Unix */
257 # define PATH_SEP '/'
261 # define OPTIONS_VAR "GZIP"
265 # define Z_SUFFIX ".gz"
269 # define MAX_SUFFIX MAX_EXT_CHARS
271 # define MAX_SUFFIX 30
276 DECLARE(uch, inbuf, INBUFSIZ + INBUF_EXTRA);
277 DECLARE(uch, outbuf, OUTBUFSIZ + OUTBUF_EXTRA);
278 DECLARE(ush, d_buf, DIST_BUFSIZE);
279 DECLARE(uch, window, 2L * WSIZE);
280 DECLARE(ush, tab_prefix, 1L << BITS);
282 static int foreground; /* set if program run in foreground */
283 static int method = DEFLATED; /* compression method */
284 static int exit_code = OK; /* program exit code */
285 static int part_nb; /* number of parts in .gz file */
286 static long time_stamp; /* original time stamp (modification time) */
287 static long ifile_size; /* input file size, -1 for devices (debug only) */
288 static char z_suffix[MAX_SUFFIX + 1]; /* default suffix (can be set with --suffix) */
289 static int z_len; /* strlen(z_suffix) */
291 static int ifd; /* input file descriptor */
292 static int ofd; /* output file descriptor */
293 static unsigned insize; /* valid bytes in inbuf */
294 static unsigned outcnt; /* bytes in output buffer */
297 /* Output a 16 bit value, lsb first */
298 static void put_short(ush w)
300 if (outcnt < OUTBUFSIZ - 2) {
301 outbuf[outcnt++] = (uch) ((w) & 0xff);
302 outbuf[outcnt++] = (uch) ((ush) (w) >> 8);
304 put_byte((uch) ((w) & 0xff));
305 put_byte((uch) ((ush) (w) >> 8));
309 /* ========================================================================
310 * Signal and error handler.
312 static void abort_gzip(int ATTRIBUTE_UNUSED ignored)
317 /* ===========================================================================
318 * Clear input and output buffers
320 static void clear_bufs(void)
327 /* ===========================================================================
328 * Does the same as write(), but also handles partial pipe writes and checks
331 static void write_buf(int fd, void *buf, unsigned cnt)
335 while ((n = write(fd, buf, cnt)) != cnt) {
336 if (n == (unsigned) (-1)) bb_error_msg_and_die("can't write");
338 buf = (void *) ((char *) buf + n);
342 /* ===========================================================================
343 * Run a set of bytes through the crc shift register. If s is a NULL
344 * pointer, then initialize the crc shift register contents instead.
345 * Return the current crc in either case.
347 static ulg updcrc(uch * s, unsigned n)
349 static ulg crc = (ulg) 0xffffffffL; /* shift register contents */
350 register ulg c; /* temporary variable */
351 static unsigned long crc_32_tab[256];
353 if (crc_32_tab[1] == 0x00000000L) {
354 unsigned long csr; /* crc shift register */
355 const unsigned long e = 0xedb88320L; /* polynomial exclusive-or pattern */
356 int i; /* counter for all possible eight bit values */
357 int k; /* byte being shifted into crc apparatus */
359 /* Compute table of CRC's. */
360 for (i = 1; i < 256; i++) {
362 /* The idea to initialize the register with the byte instead of
363 * zero was stolen from Haruhiko Okumura's ar002
366 csr = csr & 1 ? (csr >> 1) ^ e : csr >> 1;
377 c = crc_32_tab[((int) c ^ (*s++)) & 0xff] ^ (c >> 8);
381 return c ^ 0xffffffffL; /* (instead of ~c for 64-bit machines) */
384 /* bits.c -- output variable-length bit strings
385 * Copyright (C) 1992-1993 Jean-loup Gailly
386 * This is free software; you can redistribute it and/or modify it under the
387 * terms of the GNU General Public License, see the file COPYING.
394 * Output variable-length bit strings. Compression can be done
395 * to a file or to memory. (The latter is not supported in this version.)
399 * The PKZIP "deflate" file format interprets compressed file data
400 * as a sequence of bits. Multi-bit strings in the file may cross
401 * byte boundaries without restriction.
403 * The first bit of each byte is the low-order bit.
405 * The routines in this file allow a variable-length bit value to
406 * be output right-to-left (useful for literal values). For
407 * left-to-right output (useful for code strings from the tree routines),
408 * the bits must have been reversed first with bi_reverse().
410 * For in-memory compression, the compressed bit stream goes directly
411 * into the requested output buffer. The input data is read in blocks
412 * by the mem_read() function. The buffer is limited to 64K on 16 bit
417 * void bi_init (FILE *zipfile)
418 * Initialize the bit string routines.
420 * void send_bits (int value, int length)
421 * Write out a bit string, taking the source bits right to
424 * int bi_reverse (int value, int length)
425 * Reverse the bits of a bit string, taking the source bits left to
426 * right and emitting them right to left.
428 * void bi_windup (void)
429 * Write out any remaining bits in an incomplete byte.
431 * void copy_block(char *buf, unsigned len, int header)
432 * Copy a stored block to the zip file, storing first the length and
433 * its one's complement if requested.
437 /* ===========================================================================
438 * Local data used by the "bit string" routines.
441 static file_t zfile; /* output gzip file */
443 static unsigned short bi_buf;
445 /* Output buffer. bits are inserted starting at the bottom (least significant
449 #define Buf_size (8 * 2*sizeof(char))
450 /* Number of bits used within bi_buf. (bi_buf might be implemented on
451 * more than 16 bits on some systems.)
456 /* Current input function. Set to mem_read for in-memory compression */
459 ulg bits_sent; /* bit length of the compressed data */
462 /* ===========================================================================
463 * Initialize the bit string routines.
465 static void bi_init(file_t zipfile)
474 /* Set the defaults for file compression. They are set by memcompress
475 * for in-memory compression.
477 if (zfile != NO_FILE) {
478 read_buf = file_read;
482 /* ===========================================================================
483 * Send a value on a given number of bits.
484 * IN assertion: length <= 16 and value fits in length bits.
486 static void send_bits(int value, int length)
489 Tracev((stderr, " l %2d v %4x ", length, value));
490 Assert(length > 0 && length <= 15, "invalid length");
491 bits_sent += (ulg) length;
493 /* If not enough room in bi_buf, use (valid) bits from bi_buf and
494 * (16 - bi_valid) bits from value, leaving (width - (16-bi_valid))
495 * unused bits in value.
497 if (bi_valid > (int) Buf_size - length) {
498 bi_buf |= (value << bi_valid);
500 bi_buf = (ush) value >> (Buf_size - bi_valid);
501 bi_valid += length - Buf_size;
503 bi_buf |= value << bi_valid;
508 /* ===========================================================================
509 * Reverse the first len bits of a code, using straightforward code (a faster
510 * method would use a table)
511 * IN assertion: 1 <= len <= 15
513 static unsigned bi_reverse(unsigned code, int len)
515 register unsigned res = 0;
519 code >>= 1, res <<= 1;
524 /* ===========================================================================
525 * Write out any remaining bits in an incomplete byte.
527 static void bi_windup(void)
531 } else if (bi_valid > 0) {
537 bits_sent = (bits_sent + 7) & ~7;
541 /* ===========================================================================
542 * Copy a stored block to the zip file, storing first the length and its
543 * one's complement if requested.
545 static void copy_block(char *buf, unsigned len, int header)
547 bi_windup(); /* align on byte boundary */
550 put_short((ush) len);
551 put_short((ush) ~ len);
557 bits_sent += (ulg) len << 3;
564 /* deflate.c -- compress data using the deflation algorithm
565 * Copyright (C) 1992-1993 Jean-loup Gailly
566 * This is free software; you can redistribute it and/or modify it under the
567 * terms of the GNU General Public License, see the file COPYING.
573 * Identify new text as repetitions of old text within a fixed-
574 * length sliding window trailing behind the new text.
578 * The "deflation" process depends on being able to identify portions
579 * of the input text which are identical to earlier input (within a
580 * sliding window trailing behind the input currently being processed).
582 * The most straightforward technique turns out to be the fastest for
583 * most input files: try all possible matches and select the longest.
584 * The key feature of this algorithm is that insertions into the string
585 * dictionary are very simple and thus fast, and deletions are avoided
586 * completely. Insertions are performed at each input character, whereas
587 * string matches are performed only when the previous match ends. So it
588 * is preferable to spend more time in matches to allow very fast string
589 * insertions and avoid deletions. The matching algorithm for small
590 * strings is inspired from that of Rabin & Karp. A brute force approach
591 * is used to find longer strings when a small match has been found.
592 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
593 * (by Leonid Broukhis).
594 * A previous version of this file used a more sophisticated algorithm
595 * (by Fiala and Greene) which is guaranteed to run in linear amortized
596 * time, but has a larger average cost, uses more memory and is patented.
597 * However the F&G algorithm may be faster for some highly redundant
598 * files if the parameter max_chain_length (described below) is too large.
602 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
603 * I found it in 'freeze' written by Leonid Broukhis.
604 * Thanks to many info-zippers for bug reports and testing.
608 * APPNOTE.TXT documentation file in PKZIP 1.93a distribution.
610 * A description of the Rabin and Karp algorithm is given in the book
611 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
613 * Fiala,E.R., and Greene,D.H.
614 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
618 * void lm_init (int pack_level, ush *flags)
619 * Initialize the "longest match" routines for a new file
622 * Processes a new input file and return its compressed length. Sets
623 * the compressed length, crc, deflate flags and internal file
628 /* ===========================================================================
629 * Configuration parameters
632 /* Compile with MEDIUM_MEM to reduce the memory requirements or
633 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
634 * entire input file can be held in memory (not possible on 16 bit systems).
635 * Warning: defining these symbols affects HASH_BITS (see below) and thus
636 * affects the compression ratio. The compressed output
637 * is still correct, and might even be smaller in some cases.
641 # define HASH_BITS 13 /* Number of bits used to hash strings */
644 # define HASH_BITS 14
647 # define HASH_BITS 15
648 /* For portability to 16 bit machines, do not use values above 15. */
651 /* To save space (see unlzw.c), we overlay prev+head with tab_prefix and
652 * window with tab_suffix. Check that we can do this:
654 #if (WSIZE<<1) > (1<<BITS)
655 # error cannot overlay window with tab_suffix and prev with tab_prefix0
657 #if HASH_BITS > BITS-1
658 # error cannot overlay head with tab_prefix1
660 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
661 #define HASH_MASK (HASH_SIZE-1)
662 #define WMASK (WSIZE-1)
663 /* HASH_SIZE and WSIZE must be powers of two */
665 /* Tail of hash chains */
668 /* speed options for the general purpose bit flag */
670 # define TOO_FAR 4096
672 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
673 /* ===========================================================================
674 * Local data used by the "longest match" routines.
677 typedef unsigned IPos;
679 /* A Pos is an index in the character window. We use short instead of int to
680 * save space in the various tables. IPos is used only for parameter passing.
683 /* DECLARE(uch, window, 2L*WSIZE); */
684 /* Sliding window. Input bytes are read into the second half of the window,
685 * and move to the first half later to keep a dictionary of at least WSIZE
686 * bytes. With this organization, matches are limited to a distance of
687 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
688 * performed with a length multiple of the block size. Also, it limits
689 * the window size to 64K, which is quite useful on MSDOS.
690 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
691 * be less efficient).
694 /* DECLARE(Pos, prev, WSIZE); */
695 /* Link to older string with same hash index. To limit the size of this
696 * array to 64K, this link is maintained only for the last 32K strings.
697 * An index in this array is thus a window index modulo 32K.
700 /* DECLARE(Pos, head, 1<<HASH_BITS); */
701 /* Heads of the hash chains or NIL. */
703 static const ulg window_size = (ulg) 2 * WSIZE;
705 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
706 * input file length plus MIN_LOOKAHEAD.
709 static long block_start;
711 /* window position at the beginning of the current output block. Gets
712 * negative when the window is moved backwards.
715 static unsigned ins_h; /* hash index of string to be inserted */
717 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1)/MIN_MATCH)
718 /* Number of bits by which ins_h and del_h must be shifted at each
719 * input step. It must be such that after MIN_MATCH steps, the oldest
720 * byte no longer takes part in the hash key, that is:
721 * H_SHIFT * MIN_MATCH >= HASH_BITS
724 static unsigned int prev_length;
726 /* Length of the best match at previous step. Matches not greater than this
727 * are discarded. This is used in the lazy match evaluation.
730 static unsigned strstart; /* start of string to insert */
731 static unsigned match_start; /* start of matching string */
732 static int eofile; /* flag set at end of input file */
733 static unsigned lookahead; /* number of valid bytes ahead in window */
736 max_chain_length = 4096,
738 /* To speed up deflation, hash chains are never searched beyond this length.
739 * A higher limit improves compression ratio but degrades the speed.
742 max_lazy_match = 258,
744 /* Attempt to find a better match only when the current match is strictly
745 * smaller than this value. This mechanism is used only for compression
748 max_insert_length = max_lazy_match,
749 /* Insert new strings in the hash table only if the match length
750 * is not greater than this length. This saves time but degrades compression.
751 * max_insert_length is used only for compression levels <= 3.
756 /* Use a faster search when the previous match is longer than this */
759 /* Values for max_lazy_match, good_match and max_chain_length, depending on
760 * the desired pack level (0..9). The values given below have been tuned to
761 * exclude worst case performance for pathological files. Better values may be
762 * found for specific files.
765 nice_match = 258 /* Stop searching when current match exceeds this */
767 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
768 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
774 /* result of memcmp for equal strings */
776 /* ===========================================================================
777 * Prototypes for local functions.
779 static void fill_window(void);
781 static int longest_match(IPos cur_match);
784 static void check_match(IPos start, IPos match, int length);
787 /* ===========================================================================
788 * Update a hash value with the given input byte
789 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
790 * input characters, so that a running hash key can be computed from the
791 * previous key instead of complete recalculation each time.
793 #define UPDATE_HASH(h,c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
795 /* ===========================================================================
796 * Insert string s in the dictionary and set match_head to the previous head
797 * of the hash chain (the most recent string with same hash key). Return
798 * the previous length of the hash chain.
799 * IN assertion: all calls to to INSERT_STRING are made with consecutive
800 * input characters and the first MIN_MATCH bytes of s are valid
801 * (except for the last MIN_MATCH-1 bytes of the input file).
803 #define INSERT_STRING(s, match_head) \
804 (UPDATE_HASH(ins_h, window[(s) + MIN_MATCH-1]), \
805 prev[(s) & WMASK] = match_head = head[ins_h], \
808 /* ===========================================================================
809 * Initialize the "longest match" routines for a new file
811 static void lm_init(ush * flags)
815 /* Initialize the hash table. */
816 memset(head, 0, HASH_SIZE * sizeof(*head));
817 /* prev will be initialized on the fly */
820 /* ??? reduce max_chain_length for binary files */
825 lookahead = read_buf((char *) window,
826 sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE);
828 if (lookahead == 0 || lookahead == (unsigned) EOF) {
829 eofile = 1, lookahead = 0;
833 /* Make sure that we always have enough lookahead. This is important
834 * if input comes from a device such as a tty.
836 while (lookahead < MIN_LOOKAHEAD && !eofile)
840 for (j = 0; j < MIN_MATCH - 1; j++)
841 UPDATE_HASH(ins_h, window[j]);
842 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
843 * not important since only literal bytes will be emitted.
847 /* ===========================================================================
848 * Set match_start to the longest match starting at the given string and
849 * return its length. Matches shorter or equal to prev_length are discarded,
850 * in which case the result is equal to prev_length and match_start is
852 * IN assertions: cur_match is the head of the hash chain for the current
853 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
856 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
857 * match.s. The code is functionally equivalent, so you can use the C version
860 static int longest_match(IPos cur_match)
862 unsigned chain_length = max_chain_length; /* max hash chain length */
863 register uch *scan = window + strstart; /* current string */
864 register uch *match; /* matched string */
865 register int len; /* length of current match */
866 int best_len = prev_length; /* best match length so far */
868 strstart > (IPos) MAX_DIST ? strstart - (IPos) MAX_DIST : NIL;
869 /* Stop when cur_match becomes <= limit. To simplify the code,
870 * we prevent matches with the string of window index 0.
873 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
874 * It is easy to get rid of this optimization if necessary.
876 #if HASH_BITS < 8 || MAX_MATCH != 258
877 # error Code too clever
879 register uch *strend = window + strstart + MAX_MATCH;
880 register uch scan_end1 = scan[best_len - 1];
881 register uch scan_end = scan[best_len];
883 /* Do not waste too much time if we already have a good match: */
884 if (prev_length >= good_match) {
887 Assert(strstart <= window_size - MIN_LOOKAHEAD, "insufficient lookahead");
890 Assert(cur_match < strstart, "no future");
891 match = window + cur_match;
893 /* Skip to next match if the match length cannot increase
894 * or if the match length is less than 2:
896 if (match[best_len] != scan_end ||
897 match[best_len - 1] != scan_end1 ||
898 *match != *scan || *++match != scan[1])
901 /* The check at best_len-1 can be removed because it will be made
902 * again later. (This heuristic is not always a win.)
903 * It is not necessary to compare scan[2] and match[2] since they
904 * are always equal when the other bytes match, given that
905 * the hash keys are equal and that HASH_BITS >= 8.
909 /* We check for insufficient lookahead only every 8th comparison;
910 * the 256th check will be made at strstart+258.
913 } while (*++scan == *++match && *++scan == *++match &&
914 *++scan == *++match && *++scan == *++match &&
915 *++scan == *++match && *++scan == *++match &&
916 *++scan == *++match && *++scan == *++match && scan < strend);
918 len = MAX_MATCH - (int) (strend - scan);
919 scan = strend - MAX_MATCH;
921 if (len > best_len) {
922 match_start = cur_match;
924 if (len >= nice_match)
926 scan_end1 = scan[best_len - 1];
927 scan_end = scan[best_len];
929 } while ((cur_match = prev[cur_match & WMASK]) > limit
930 && --chain_length != 0);
936 /* ===========================================================================
937 * Check that the match at match_start is indeed a match.
939 static void check_match(IPos start, IPos match, int length)
941 /* check that the match is indeed a match */
942 if (memcmp((char *) window + match,
943 (char *) window + start, length) != EQUAL) {
944 bb_error_msg(" start %d, match %d, length %d", start, match, length);
945 bb_error_msg("invalid match");
948 bb_error_msg("\\[%d,%d]", start - match, length);
950 putc(window[start++], stderr);
951 } while (--length != 0);
955 # define check_match(start, match, length)
958 /* ===========================================================================
959 * Fill the window when the lookahead becomes insufficient.
960 * Updates strstart and lookahead, and sets eofile if end of input file.
961 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
962 * OUT assertions: at least one byte has been read, or eofile is set;
963 * file reads are performed for at least two bytes (required for the
964 * translate_eol option).
966 static void fill_window(void)
968 register unsigned n, m;
970 (unsigned) (window_size - (ulg) lookahead - (ulg) strstart);
971 /* Amount of free space at the end of the window. */
973 /* If the window is almost full and there is insufficient lookahead,
974 * move the upper half to the lower one to make room in the upper half.
976 if (more == (unsigned) EOF) {
977 /* Very unlikely, but possible on 16 bit machine if strstart == 0
978 * and lookahead == 1 (input done one byte at time)
981 } else if (strstart >= WSIZE + MAX_DIST) {
982 /* By the IN assertion, the window is not empty so we can't confuse
983 * more == 0 with more == 64K on a 16 bit machine.
985 Assert(window_size == (ulg) 2 * WSIZE, "no sliding with BIG_MEM");
987 memcpy((char *) window, (char *) window + WSIZE, (unsigned) WSIZE);
988 match_start -= WSIZE;
989 strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
991 block_start -= (long) WSIZE;
993 for (n = 0; n < HASH_SIZE; n++) {
995 head[n] = (Pos) (m >= WSIZE ? m - WSIZE : NIL);
997 for (n = 0; n < WSIZE; n++) {
999 prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : NIL);
1000 /* If n is not on any hash chain, prev[n] is garbage but
1001 * its value will never be used.
1006 /* At this point, more >= 2 */
1008 n = read_buf((char *) window + strstart + lookahead, more);
1009 if (n == 0 || n == (unsigned) EOF) {
1017 /* ===========================================================================
1018 * Flush the current block, with given end-of-file flag.
1019 * IN assertion: strstart is set to the end of the current match.
1021 #define FLUSH_BLOCK(eof) \
1022 flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \
1023 (char*)NULL, (long)strstart - block_start, (eof))
1025 /* ===========================================================================
1026 * Same as above, but achieves better compression. We use a lazy
1027 * evaluation for matches: a match is finally adopted only if there is
1028 * no better match at the next window position.
1030 static ulg deflate(void)
1032 IPos hash_head; /* head of hash chain */
1033 IPos prev_match; /* previous match */
1034 int flush; /* set if current block must be flushed */
1035 int match_available = 0; /* set if previous match exists */
1036 register unsigned match_length = MIN_MATCH - 1; /* length of best match */
1038 /* Process the input block. */
1039 while (lookahead != 0) {
1040 /* Insert the string window[strstart .. strstart+2] in the
1041 * dictionary, and set hash_head to the head of the hash chain:
1043 INSERT_STRING(strstart, hash_head);
1045 /* Find the longest match, discarding those <= prev_length.
1047 prev_length = match_length, prev_match = match_start;
1048 match_length = MIN_MATCH - 1;
1050 if (hash_head != NIL && prev_length < max_lazy_match &&
1051 strstart - hash_head <= MAX_DIST) {
1052 /* To simplify the code, we prevent matches with the string
1053 * of window index 0 (in particular we have to avoid a match
1054 * of the string with itself at the start of the input file).
1056 match_length = longest_match(hash_head);
1057 /* longest_match() sets match_start */
1058 if (match_length > lookahead)
1059 match_length = lookahead;
1061 /* Ignore a length 3 match if it is too distant: */
1062 if (match_length == MIN_MATCH && strstart - match_start > TOO_FAR) {
1063 /* If prev_match is also MIN_MATCH, match_start is garbage
1064 * but we will ignore the current match anyway.
1069 /* If there was a match at the previous step and the current
1070 * match is not better, output the previous match:
1072 if (prev_length >= MIN_MATCH && match_length <= prev_length) {
1074 check_match(strstart - 1, prev_match, prev_length);
1077 ct_tally(strstart - 1 - prev_match, prev_length - MIN_MATCH);
1079 /* Insert in hash table all strings up to the end of the match.
1080 * strstart-1 and strstart are already inserted.
1082 lookahead -= prev_length - 1;
1086 INSERT_STRING(strstart, hash_head);
1087 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1088 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
1089 * these bytes are garbage, but it does not matter since the
1090 * next lookahead bytes will always be emitted as literals.
1092 } while (--prev_length != 0);
1093 match_available = 0;
1094 match_length = MIN_MATCH - 1;
1097 FLUSH_BLOCK(0), block_start = strstart;
1099 } else if (match_available) {
1100 /* If there was no match at the previous position, output a
1101 * single literal. If there was a match but the current match
1102 * is longer, truncate the previous match to a single literal.
1104 Tracevv((stderr, "%c", window[strstart - 1]));
1105 if (ct_tally(0, window[strstart - 1])) {
1106 FLUSH_BLOCK(0), block_start = strstart;
1111 /* There is no previous match to compare with, wait for
1112 * the next step to decide.
1114 match_available = 1;
1118 Assert(strstart <= isize && lookahead <= isize, "a bit too far");
1120 /* Make sure that we always have enough lookahead, except
1121 * at the end of the input file. We need MAX_MATCH bytes
1122 * for the next match, plus MIN_MATCH bytes to insert the
1123 * string following the next match.
1125 while (lookahead < MIN_LOOKAHEAD && !eofile)
1128 if (match_available)
1129 ct_tally(0, window[strstart - 1]);
1131 return FLUSH_BLOCK(1); /* eof */
1134 /* gzip (GNU zip) -- compress files with zip algorithm and 'compress' interface
1135 * Copyright (C) 1992-1993 Jean-loup Gailly
1136 * The unzip code was written and put in the public domain by Mark Adler.
1137 * Portions of the lzw code are derived from the public domain 'compress'
1138 * written by Spencer Thomas, Joe Orost, James Woods, Jim McKie, Steve Davies,
1139 * Ken Turkowski, Dave Mack and Peter Jannesen.
1141 * See the license_msg below and the file COPYING for the software license.
1142 * See the file algorithm.doc for the compression algorithms and file formats.
1145 /* Compress files with zip algorithm and 'compress' interface.
1146 * See usage() and help() functions below for all options.
1148 * file.gz: compressed file with same mode, owner, and utimes
1149 * or stdout with -c option or if stdin used as input.
1150 * If the output file name had to be truncated, the original name is kept
1151 * in the compressed file.
1156 typedef struct dirent dir_type;
1158 /* ======================================================================== */
1159 int gzip_main(int argc, char **argv)
1164 struct stat statBuf;
1170 while ((opt = getopt(argc, argv, "cf123456789dq")) != -1) {
1178 /* Ignore 1-9 (compression level) options */
1191 #ifdef CONFIG_GUNZIP
1194 return gunzip_main(argc, argv);
1201 foreground = signal(SIGINT, SIG_IGN) != SIG_IGN;
1203 (void) signal(SIGINT, abort_gzip);
1206 if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
1207 (void) signal(SIGTERM, abort_gzip);
1211 if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
1212 (void) signal(SIGHUP, abort_gzip);
1216 strncpy(z_suffix, Z_SUFFIX, sizeof(z_suffix) - 1);
1217 z_len = strlen(z_suffix);
1219 /* Allocate all global buffers (for DYN_ALLOC option) */
1220 ALLOC(uch, inbuf, INBUFSIZ + INBUF_EXTRA);
1221 ALLOC(uch, outbuf, OUTBUFSIZ + OUTBUF_EXTRA);
1222 ALLOC(ush, d_buf, DIST_BUFSIZE);
1223 ALLOC(uch, window, 2L * WSIZE);
1224 ALLOC(ush, tab_prefix, 1L << BITS);
1229 if (optind == argc) {
1232 zip(STDIN_FILENO, STDOUT_FILENO);
1236 for (i = optind; i < argc; i++) {
1240 if (strcmp(argv[i], "-") == 0) {
1243 inFileNum = STDIN_FILENO;
1244 outFileNum = STDOUT_FILENO;
1246 inFileNum = open(argv[i], O_RDONLY);
1247 if (inFileNum < 0 || fstat(inFileNum, &statBuf) < 0)
1248 bb_perror_msg_and_die("%s", argv[i]);
1249 time_stamp = statBuf.st_ctime;
1250 ifile_size = statBuf.st_size;
1253 path = xmalloc(strlen(argv[i]) + 4);
1254 strcpy(path, argv[i]);
1255 strcat(path, ".gz");
1257 /* Open output file */
1258 #if (__GLIBC__ >= 2) && (__GLIBC_MINOR__ >= 1) && defined O_NOFOLLOW
1260 open(path, O_RDWR | O_CREAT | O_EXCL | O_NOFOLLOW);
1262 outFileNum = open(path, O_RDWR | O_CREAT | O_EXCL);
1264 if (outFileNum < 0) {
1265 bb_perror_msg("%s", path);
1270 /* Set permissions on the file */
1271 fchmod(outFileNum, statBuf.st_mode);
1273 outFileNum = STDOUT_FILENO;
1276 if (path == NULL && isatty(outFileNum) && force == 0) {
1278 ("compressed data not written to a terminal. Use -f to force compression.");
1283 result = zip(inFileNum, outFileNum);
1289 /* Delete the original file */
1291 delFileName = argv[i];
1295 if (unlink(delFileName) < 0)
1296 bb_perror_msg("%s", delFileName);
1306 /* trees.c -- output deflated data using Huffman coding
1307 * Copyright (C) 1992-1993 Jean-loup Gailly
1308 * This is free software; you can redistribute it and/or modify it under the
1309 * terms of the GNU General Public License, see the file COPYING.
1315 * Encode various sets of source values using variable-length
1316 * binary code trees.
1320 * The PKZIP "deflation" process uses several Huffman trees. The more
1321 * common source values are represented by shorter bit sequences.
1323 * Each code tree is stored in the ZIP file in a compressed form
1324 * which is itself a Huffman encoding of the lengths of
1325 * all the code strings (in ascending order by source values).
1326 * The actual code strings are reconstructed from the lengths in
1327 * the UNZIP process, as described in the "application note"
1328 * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program.
1333 * Data Compression: Techniques and Applications, pp. 53-55.
1334 * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7.
1337 * Data Compression: Methods and Theory, pp. 49-50.
1338 * Computer Science Press, 1988. ISBN 0-7167-8156-5.
1342 * Addison-Wesley, 1983. ISBN 0-201-06672-6.
1346 * void ct_init (ush *attr, int *methodp)
1347 * Allocate the match buffer, initialize the various tables and save
1348 * the location of the internal file attribute (ascii/binary) and
1349 * method (DEFLATE/STORE)
1351 * void ct_tally (int dist, int lc);
1352 * Save the match info and tally the frequency counts.
1354 * long flush_block (char *buf, ulg stored_len, int eof)
1355 * Determine the best encoding for the current block: dynamic trees,
1356 * static trees or store, and output the encoded block to the zip
1357 * file. Returns the total compressed length for the file so far.
1361 /* ===========================================================================
1366 /* All codes must not exceed MAX_BITS bits */
1368 #define MAX_BL_BITS 7
1369 /* Bit length codes must not exceed MAX_BL_BITS bits */
1371 #define LENGTH_CODES 29
1372 /* number of length codes, not counting the special END_BLOCK code */
1374 #define LITERALS 256
1375 /* number of literal bytes 0..255 */
1377 #define END_BLOCK 256
1378 /* end of block literal code */
1380 #define L_CODES (LITERALS+1+LENGTH_CODES)
1381 /* number of Literal or Length codes, including the END_BLOCK code */
1384 /* number of distance codes */
1387 /* number of codes used to transfer the bit lengths */
1389 typedef uch extra_bits_t;
1391 /* extra bits for each length code */
1392 static const extra_bits_t extra_lbits[LENGTH_CODES]
1393 = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4,
1397 /* extra bits for each distance code */
1398 static const extra_bits_t extra_dbits[D_CODES]
1399 = { 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,
1400 10, 10, 11, 11, 12, 12, 13, 13
1403 /* extra bits for each bit length code */
1404 static const extra_bits_t extra_blbits[BL_CODES]
1405 = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 };
1407 #define STORED_BLOCK 0
1408 #define STATIC_TREES 1
1410 /* The three kinds of block type */
1414 # define LIT_BUFSIZE 0x2000
1417 # define LIT_BUFSIZE 0x4000
1419 # define LIT_BUFSIZE 0x8000
1423 #ifndef DIST_BUFSIZE
1424 # define DIST_BUFSIZE LIT_BUFSIZE
1426 /* Sizes of match buffers for literals/lengths and distances. There are
1427 * 4 reasons for limiting LIT_BUFSIZE to 64K:
1428 * - frequencies can be kept in 16 bit counters
1429 * - if compression is not successful for the first block, all input data is
1430 * still in the window so we can still emit a stored block even when input
1431 * comes from standard input. (This can also be done for all blocks if
1432 * LIT_BUFSIZE is not greater than 32K.)
1433 * - if compression is not successful for a file smaller than 64K, we can
1434 * even emit a stored file instead of a stored block (saving 5 bytes).
1435 * - creating new Huffman trees less frequently may not provide fast
1436 * adaptation to changes in the input data statistics. (Take for
1437 * example a binary file with poorly compressible code followed by
1438 * a highly compressible string table.) Smaller buffer sizes give
1439 * fast adaptation but have of course the overhead of transmitting trees
1441 * - I can't count above 4
1442 * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save
1443 * memory at the expense of compression). Some optimizations would be possible
1444 * if we rely on DIST_BUFSIZE == LIT_BUFSIZE.
1446 #if LIT_BUFSIZE > INBUFSIZ
1447 #error cannot overlay l_buf and inbuf
1450 /* repeat previous bit length 3-6 times (2 bits of repeat count) */
1451 #define REPZ_3_10 17
1452 /* repeat a zero length 3-10 times (3 bits of repeat count) */
1453 #define REPZ_11_138 18
1454 /* repeat a zero length 11-138 times (7 bits of repeat count) */
1456 /* ===========================================================================
1460 /* Data structure describing a single value and its code string. */
1461 typedef struct ct_data {
1463 ush freq; /* frequency count */
1464 ush code; /* bit string */
1467 ush dad; /* father node in Huffman tree */
1468 ush len; /* length of bit string */
1472 #define Freq fc.freq
1473 #define Code fc.code
1477 #define HEAP_SIZE (2*L_CODES+1)
1478 /* maximum heap size */
1480 static ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */
1481 static ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */
1483 static ct_data static_ltree[L_CODES + 2];
1485 /* The static literal tree. Since the bit lengths are imposed, there is no
1486 * need for the L_CODES extra codes used during heap construction. However
1487 * The codes 286 and 287 are needed to build a canonical tree (see ct_init
1491 static ct_data static_dtree[D_CODES];
1493 /* The static distance tree. (Actually a trivial tree since all codes use
1497 static ct_data bl_tree[2 * BL_CODES + 1];
1499 /* Huffman tree for the bit lengths */
1501 typedef struct tree_desc {
1502 ct_data *dyn_tree; /* the dynamic tree */
1503 ct_data *static_tree; /* corresponding static tree or NULL */
1504 const extra_bits_t *extra_bits; /* extra bits for each code or NULL */
1505 int extra_base; /* base index for extra_bits */
1506 int elems; /* max number of elements in the tree */
1507 int max_length; /* max bit length for the codes */
1508 int max_code; /* largest code with non zero frequency */
1511 static tree_desc l_desc =
1512 { dyn_ltree, static_ltree, extra_lbits, LITERALS + 1, L_CODES,
1516 static tree_desc d_desc =
1517 { dyn_dtree, static_dtree, extra_dbits, 0, D_CODES, MAX_BITS, 0 };
1519 static tree_desc bl_desc =
1520 { bl_tree, (ct_data *) 0, extra_blbits, 0, BL_CODES, MAX_BL_BITS,
1525 static ush bl_count[MAX_BITS + 1];
1527 /* number of codes at each bit length for an optimal tree */
1529 static const uch bl_order[BL_CODES]
1530 = { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
1532 /* The lengths of the bit length codes are sent in order of decreasing
1533 * probability, to avoid transmitting the lengths for unused bit length codes.
1536 static int heap[2 * L_CODES + 1]; /* heap used to build the Huffman trees */
1537 static int heap_len; /* number of elements in the heap */
1538 static int heap_max; /* element of largest frequency */
1540 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
1541 * The same heap array is used to build all trees.
1544 static uch depth[2 * L_CODES + 1];
1546 /* Depth of each subtree used as tie breaker for trees of equal frequency */
1548 static uch length_code[MAX_MATCH - MIN_MATCH + 1];
1550 /* length code for each normalized match length (0 == MIN_MATCH) */
1552 static uch dist_code[512];
1554 /* distance codes. The first 256 values correspond to the distances
1555 * 3 .. 258, the last 256 values correspond to the top 8 bits of
1556 * the 15 bit distances.
1559 static int base_length[LENGTH_CODES];
1561 /* First normalized length for each code (0 = MIN_MATCH) */
1563 static int base_dist[D_CODES];
1565 /* First normalized distance for each code (0 = distance of 1) */
1568 /* DECLARE(uch, l_buf, LIT_BUFSIZE); buffer for literals or lengths */
1570 /* DECLARE(ush, d_buf, DIST_BUFSIZE); buffer for distances */
1572 static uch flag_buf[(LIT_BUFSIZE / 8)];
1574 /* flag_buf is a bit array distinguishing literals from lengths in
1575 * l_buf, thus indicating the presence or absence of a distance.
1578 static unsigned last_lit; /* running index in l_buf */
1579 static unsigned last_dist; /* running index in d_buf */
1580 static unsigned last_flags; /* running index in flag_buf */
1581 static uch flags; /* current flags not yet saved in flag_buf */
1582 static uch flag_bit; /* current bit used in flags */
1584 /* bits are filled in flags starting at bit 0 (least significant).
1585 * Note: these flags are overkill in the current code since we don't
1586 * take advantage of DIST_BUFSIZE == LIT_BUFSIZE.
1589 static ulg opt_len; /* bit length of current block with optimal trees */
1590 static ulg static_len; /* bit length of current block with static trees */
1592 static ulg compressed_len; /* total bit length of compressed file */
1595 static ush *file_type; /* pointer to UNKNOWN, BINARY or ASCII */
1596 static int *file_method; /* pointer to DEFLATE or STORE */
1598 /* ===========================================================================
1599 * Local (static) routines in this file.
1602 static void init_block(void);
1603 static void pqdownheap(ct_data * tree, int k);
1604 static void gen_bitlen(tree_desc * desc);
1605 static void gen_codes(ct_data * tree, int max_code);
1606 static void build_tree(tree_desc * desc);
1607 static void scan_tree(ct_data * tree, int max_code);
1608 static void send_tree(ct_data * tree, int max_code);
1609 static int build_bl_tree(void);
1610 static void send_all_trees(int lcodes, int dcodes, int blcodes);
1611 static void compress_block(ct_data * ltree, ct_data * dtree);
1612 static void set_file_type(void);
1616 # define send_code(c, tree) send_bits(tree[c].Code, tree[c].Len)
1617 /* Send a code of the given tree. c and tree must not have side effects */
1620 # define send_code(c, tree) \
1621 { if (verbose>1) bb_error_msg("\ncd %3d ",(c)); \
1622 send_bits(tree[c].Code, tree[c].Len); }
1625 #define d_code(dist) \
1626 ((dist) < 256 ? dist_code[dist] : dist_code[256+((dist)>>7)])
1627 /* Mapping from a distance to a distance code. dist is the distance - 1 and
1628 * must not have side effects. dist_code[256] and dist_code[257] are never
1632 /* the arguments must not have side effects */
1634 /* ===========================================================================
1635 * Allocate the match buffer, initialize the various tables and save the
1636 * location of the internal file attribute (ascii/binary) and method
1639 static void ct_init(ush * attr, int *methodp)
1641 int n; /* iterates over tree elements */
1642 int bits; /* bit counter */
1643 int length; /* length value */
1644 int code; /* code value */
1645 int dist; /* distance index */
1648 file_method = methodp;
1649 compressed_len = 0L;
1651 if (static_dtree[0].Len != 0)
1652 return; /* ct_init already called */
1654 /* Initialize the mapping length (0..255) -> length code (0..28) */
1656 for (code = 0; code < LENGTH_CODES - 1; code++) {
1657 base_length[code] = length;
1658 for (n = 0; n < (1 << extra_lbits[code]); n++) {
1659 length_code[length++] = (uch) code;
1662 Assert(length == 256, "ct_init: length != 256");
1663 /* Note that the length 255 (match length 258) can be represented
1664 * in two different ways: code 284 + 5 bits or code 285, so we
1665 * overwrite length_code[255] to use the best encoding:
1667 length_code[length - 1] = (uch) code;
1669 /* Initialize the mapping dist (0..32K) -> dist code (0..29) */
1671 for (code = 0; code < 16; code++) {
1672 base_dist[code] = dist;
1673 for (n = 0; n < (1 << extra_dbits[code]); n++) {
1674 dist_code[dist++] = (uch) code;
1677 Assert(dist == 256, "ct_init: dist != 256");
1678 dist >>= 7; /* from now on, all distances are divided by 128 */
1679 for (; code < D_CODES; code++) {
1680 base_dist[code] = dist << 7;
1681 for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) {
1682 dist_code[256 + dist++] = (uch) code;
1685 Assert(dist == 256, "ct_init: 256+dist != 512");
1687 /* Construct the codes of the static literal tree */
1688 for (bits = 0; bits <= MAX_BITS; bits++)
1692 static_ltree[n++].Len = 8, bl_count[8]++;
1694 static_ltree[n++].Len = 9, bl_count[9]++;
1696 static_ltree[n++].Len = 7, bl_count[7]++;
1698 static_ltree[n++].Len = 8, bl_count[8]++;
1699 /* Codes 286 and 287 do not exist, but we must include them in the
1700 * tree construction to get a canonical Huffman tree (longest code
1703 gen_codes((ct_data *) static_ltree, L_CODES + 1);
1705 /* The static distance tree is trivial: */
1706 for (n = 0; n < D_CODES; n++) {
1707 static_dtree[n].Len = 5;
1708 static_dtree[n].Code = bi_reverse(n, 5);
1711 /* Initialize the first block of the first file: */
1715 /* ===========================================================================
1716 * Initialize a new block.
1718 static void init_block(void)
1720 int n; /* iterates over tree elements */
1722 /* Initialize the trees. */
1723 for (n = 0; n < L_CODES; n++)
1724 dyn_ltree[n].Freq = 0;
1725 for (n = 0; n < D_CODES; n++)
1726 dyn_dtree[n].Freq = 0;
1727 for (n = 0; n < BL_CODES; n++)
1728 bl_tree[n].Freq = 0;
1730 dyn_ltree[END_BLOCK].Freq = 1;
1731 opt_len = static_len = 0L;
1732 last_lit = last_dist = last_flags = 0;
1738 /* Index within the heap array of least frequent node in the Huffman tree */
1741 /* ===========================================================================
1742 * Remove the smallest element from the heap and recreate the heap with
1743 * one less element. Updates heap and heap_len.
1745 #define pqremove(tree, top) \
1747 top = heap[SMALLEST]; \
1748 heap[SMALLEST] = heap[heap_len--]; \
1749 pqdownheap(tree, SMALLEST); \
1752 /* ===========================================================================
1753 * Compares to subtrees, using the tree depth as tie breaker when
1754 * the subtrees have equal frequency. This minimizes the worst case length.
1756 #define smaller(tree, n, m) \
1757 (tree[n].Freq < tree[m].Freq || \
1758 (tree[n].Freq == tree[m].Freq && depth[n] <= depth[m]))
1760 /* ===========================================================================
1761 * Restore the heap property by moving down the tree starting at node k,
1762 * exchanging a node with the smallest of its two sons if necessary, stopping
1763 * when the heap property is re-established (each father smaller than its
1766 static void pqdownheap(ct_data * tree, int k)
1769 int j = k << 1; /* left son of k */
1771 while (j <= heap_len) {
1772 /* Set j to the smallest of the two sons: */
1773 if (j < heap_len && smaller(tree, heap[j + 1], heap[j]))
1776 /* Exit if v is smaller than both sons */
1777 if (smaller(tree, v, heap[j]))
1780 /* Exchange v with the smallest son */
1784 /* And continue down the tree, setting j to the left son of k */
1790 /* ===========================================================================
1791 * Compute the optimal bit lengths for a tree and update the total bit length
1792 * for the current block.
1793 * IN assertion: the fields freq and dad are set, heap[heap_max] and
1794 * above are the tree nodes sorted by increasing frequency.
1795 * OUT assertions: the field len is set to the optimal bit length, the
1796 * array bl_count contains the frequencies for each bit length.
1797 * The length opt_len is updated; static_len is also updated if stree is
1800 static void gen_bitlen(tree_desc * desc)
1802 ct_data *tree = desc->dyn_tree;
1803 const extra_bits_t *extra = desc->extra_bits;
1804 int base = desc->extra_base;
1805 int max_code = desc->max_code;
1806 int max_length = desc->max_length;
1807 ct_data *stree = desc->static_tree;
1808 int h; /* heap index */
1809 int n, m; /* iterate over the tree elements */
1810 int bits; /* bit length */
1811 int xbits; /* extra bits */
1812 ush f; /* frequency */
1813 int overflow = 0; /* number of elements with bit length too large */
1815 for (bits = 0; bits <= MAX_BITS; bits++)
1818 /* In a first pass, compute the optimal bit lengths (which may
1819 * overflow in the case of the bit length tree).
1821 tree[heap[heap_max]].Len = 0; /* root of the heap */
1823 for (h = heap_max + 1; h < HEAP_SIZE; h++) {
1825 bits = tree[tree[n].Dad].Len + 1;
1826 if (bits > max_length)
1827 bits = max_length, overflow++;
1828 tree[n].Len = (ush) bits;
1829 /* We overwrite tree[n].Dad which is no longer needed */
1832 continue; /* not a leaf node */
1837 xbits = extra[n - base];
1839 opt_len += (ulg) f *(bits + xbits);
1842 static_len += (ulg) f *(stree[n].Len + xbits);
1847 Trace((stderr, "\nbit length overflow\n"));
1848 /* This happens for example on obj2 and pic of the Calgary corpus */
1850 /* Find the first bit length which could increase: */
1852 bits = max_length - 1;
1853 while (bl_count[bits] == 0)
1855 bl_count[bits]--; /* move one leaf down the tree */
1856 bl_count[bits + 1] += 2; /* move one overflow item as its brother */
1857 bl_count[max_length]--;
1858 /* The brother of the overflow item also moves one step up,
1859 * but this does not affect bl_count[max_length]
1862 } while (overflow > 0);
1864 /* Now recompute all bit lengths, scanning in increasing frequency.
1865 * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all
1866 * lengths instead of fixing only the wrong ones. This idea is taken
1867 * from 'ar' written by Haruhiko Okumura.)
1869 for (bits = max_length; bits != 0; bits--) {
1875 if (tree[m].Len != (unsigned) bits) {
1876 Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len,
1879 ((long) bits - (long) tree[m].Len) * (long) tree[m].Freq;
1880 tree[m].Len = (ush) bits;
1887 /* ===========================================================================
1888 * Generate the codes for a given tree and bit counts (which need not be
1890 * IN assertion: the array bl_count contains the bit length statistics for
1891 * the given tree and the field len is set for all tree elements.
1892 * OUT assertion: the field code is set for all tree elements of non
1895 static void gen_codes(ct_data * tree, int max_code)
1897 ush next_code[MAX_BITS + 1]; /* next code value for each bit length */
1898 ush code = 0; /* running code value */
1899 int bits; /* bit index */
1900 int n; /* code index */
1902 /* The distribution counts are first used to generate the code values
1903 * without bit reversal.
1905 for (bits = 1; bits <= MAX_BITS; bits++) {
1906 next_code[bits] = code = (code + bl_count[bits - 1]) << 1;
1908 /* Check that the bit counts in bl_count are consistent. The last code
1911 Assert(code + bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1,
1912 "inconsistent bit counts");
1913 Tracev((stderr, "\ngen_codes: max_code %d ", max_code));
1915 for (n = 0; n <= max_code; n++) {
1916 int len = tree[n].Len;
1920 /* Now reverse the bits */
1921 tree[n].Code = bi_reverse(next_code[len]++, len);
1923 Tracec(tree != static_ltree,
1924 (stderr, "\nn %3d %c l %2d c %4x (%x) ", n,
1925 (isgraph(n) ? n : ' '), len, tree[n].Code,
1926 next_code[len] - 1));
1930 /* ===========================================================================
1931 * Construct one Huffman tree and assigns the code bit strings and lengths.
1932 * Update the total bit length for the current block.
1933 * IN assertion: the field freq is set for all tree elements.
1934 * OUT assertions: the fields len and code are set to the optimal bit length
1935 * and corresponding code. The length opt_len is updated; static_len is
1936 * also updated if stree is not null. The field max_code is set.
1938 static void build_tree(tree_desc * desc)
1940 ct_data *tree = desc->dyn_tree;
1941 ct_data *stree = desc->static_tree;
1942 int elems = desc->elems;
1943 int n, m; /* iterate over heap elements */
1944 int max_code = -1; /* largest code with non zero frequency */
1945 int node = elems; /* next internal node of the tree */
1947 /* Construct the initial heap, with least frequent element in
1948 * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1].
1949 * heap[0] is not used.
1951 heap_len = 0, heap_max = HEAP_SIZE;
1953 for (n = 0; n < elems; n++) {
1954 if (tree[n].Freq != 0) {
1955 heap[++heap_len] = max_code = n;
1962 /* The pkzip format requires that at least one distance code exists,
1963 * and that at least one bit should be sent even if there is only one
1964 * possible code. So to avoid special checks later on we force at least
1965 * two codes of non zero frequency.
1967 while (heap_len < 2) {
1968 int new = heap[++heap_len] = (max_code < 2 ? ++max_code : 0);
1974 static_len -= stree[new].Len;
1975 /* new is 0 or 1 so it does not have extra bits */
1977 desc->max_code = max_code;
1979 /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree,
1980 * establish sub-heaps of increasing lengths:
1982 for (n = heap_len / 2; n >= 1; n--)
1983 pqdownheap(tree, n);
1985 /* Construct the Huffman tree by repeatedly combining the least two
1989 pqremove(tree, n); /* n = node of least frequency */
1990 m = heap[SMALLEST]; /* m = node of next least frequency */
1992 heap[--heap_max] = n; /* keep the nodes sorted by frequency */
1993 heap[--heap_max] = m;
1995 /* Create a new node father of n and m */
1996 tree[node].Freq = tree[n].Freq + tree[m].Freq;
1997 depth[node] = (uch) (MAX(depth[n], depth[m]) + 1);
1998 tree[n].Dad = tree[m].Dad = (ush) node;
2000 if (tree == bl_tree) {
2001 bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)",
2002 node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq);
2005 /* and insert the new node in the heap */
2006 heap[SMALLEST] = node++;
2007 pqdownheap(tree, SMALLEST);
2009 } while (heap_len >= 2);
2011 heap[--heap_max] = heap[SMALLEST];
2013 /* At this point, the fields freq and dad are set. We can now
2014 * generate the bit lengths.
2016 gen_bitlen((tree_desc *) desc);
2018 /* The field len is now set, we can generate the bit codes */
2019 gen_codes((ct_data *) tree, max_code);
2022 /* ===========================================================================
2023 * Scan a literal or distance tree to determine the frequencies of the codes
2024 * in the bit length tree. Updates opt_len to take into account the repeat
2025 * counts. (The contribution of the bit length codes will be added later
2026 * during the construction of bl_tree.)
2028 static void scan_tree(ct_data * tree, int max_code)
2030 int n; /* iterates over all tree elements */
2031 int prevlen = -1; /* last emitted length */
2032 int curlen; /* length of current code */
2033 int nextlen = tree[0].Len; /* length of next code */
2034 int count = 0; /* repeat count of the current code */
2035 int max_count = 7; /* max repeat count */
2036 int min_count = 4; /* min repeat count */
2039 max_count = 138, min_count = 3;
2040 tree[max_code + 1].Len = (ush) 0xffff; /* guard */
2042 for (n = 0; n <= max_code; n++) {
2044 nextlen = tree[n + 1].Len;
2045 if (++count < max_count && curlen == nextlen) {
2047 } else if (count < min_count) {
2048 bl_tree[curlen].Freq += count;
2049 } else if (curlen != 0) {
2050 if (curlen != prevlen)
2051 bl_tree[curlen].Freq++;
2052 bl_tree[REP_3_6].Freq++;
2053 } else if (count <= 10) {
2054 bl_tree[REPZ_3_10].Freq++;
2056 bl_tree[REPZ_11_138].Freq++;
2061 max_count = 138, min_count = 3;
2062 } else if (curlen == nextlen) {
2063 max_count = 6, min_count = 3;
2065 max_count = 7, min_count = 4;
2070 /* ===========================================================================
2071 * Send a literal or distance tree in compressed form, using the codes in
2074 static void send_tree(ct_data * tree, int max_code)
2076 int n; /* iterates over all tree elements */
2077 int prevlen = -1; /* last emitted length */
2078 int curlen; /* length of current code */
2079 int nextlen = tree[0].Len; /* length of next code */
2080 int count = 0; /* repeat count of the current code */
2081 int max_count = 7; /* max repeat count */
2082 int min_count = 4; /* min repeat count */
2084 /* tree[max_code+1].Len = -1; *//* guard already set */
2086 max_count = 138, min_count = 3;
2088 for (n = 0; n <= max_code; n++) {
2090 nextlen = tree[n + 1].Len;
2091 if (++count < max_count && curlen == nextlen) {
2093 } else if (count < min_count) {
2095 send_code(curlen, bl_tree);
2096 } while (--count != 0);
2098 } else if (curlen != 0) {
2099 if (curlen != prevlen) {
2100 send_code(curlen, bl_tree);
2103 Assert(count >= 3 && count <= 6, " 3_6?");
2104 send_code(REP_3_6, bl_tree);
2105 send_bits(count - 3, 2);
2107 } else if (count <= 10) {
2108 send_code(REPZ_3_10, bl_tree);
2109 send_bits(count - 3, 3);
2112 send_code(REPZ_11_138, bl_tree);
2113 send_bits(count - 11, 7);
2118 max_count = 138, min_count = 3;
2119 } else if (curlen == nextlen) {
2120 max_count = 6, min_count = 3;
2122 max_count = 7, min_count = 4;
2127 /* ===========================================================================
2128 * Construct the Huffman tree for the bit lengths and return the index in
2129 * bl_order of the last bit length code to send.
2131 static int build_bl_tree(void)
2133 int max_blindex; /* index of last bit length code of non zero freq */
2135 /* Determine the bit length frequencies for literal and distance trees */
2136 scan_tree((ct_data *) dyn_ltree, l_desc.max_code);
2137 scan_tree((ct_data *) dyn_dtree, d_desc.max_code);
2139 /* Build the bit length tree: */
2140 build_tree((tree_desc *) (&bl_desc));
2141 /* opt_len now includes the length of the tree representations, except
2142 * the lengths of the bit lengths codes and the 5+5+4 bits for the counts.
2145 /* Determine the number of bit length codes to send. The pkzip format
2146 * requires that at least 4 bit length codes be sent. (appnote.txt says
2147 * 3 but the actual value used is 4.)
2149 for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) {
2150 if (bl_tree[bl_order[max_blindex]].Len != 0)
2153 /* Update opt_len to include the bit length tree and counts */
2154 opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4;
2155 Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", opt_len, static_len));
2160 /* ===========================================================================
2161 * Send the header for a block using dynamic Huffman trees: the counts, the
2162 * lengths of the bit length codes, the literal tree and the distance tree.
2163 * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4.
2165 static void send_all_trees(int lcodes, int dcodes, int blcodes)
2167 int rank; /* index in bl_order */
2169 Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes");
2170 Assert(lcodes <= L_CODES && dcodes <= D_CODES
2171 && blcodes <= BL_CODES, "too many codes");
2172 Tracev((stderr, "\nbl counts: "));
2173 send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */
2174 send_bits(dcodes - 1, 5);
2175 send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */
2176 for (rank = 0; rank < blcodes; rank++) {
2177 Tracev((stderr, "\nbl code %2d ", bl_order[rank]));
2178 send_bits(bl_tree[bl_order[rank]].Len, 3);
2180 Tracev((stderr, "\nbl tree: sent %ld", bits_sent));
2182 send_tree((ct_data *) dyn_ltree, lcodes - 1); /* send the literal tree */
2183 Tracev((stderr, "\nlit tree: sent %ld", bits_sent));
2185 send_tree((ct_data *) dyn_dtree, dcodes - 1); /* send the distance tree */
2186 Tracev((stderr, "\ndist tree: sent %ld", bits_sent));
2189 /* ===========================================================================
2190 * Determine the best encoding for the current block: dynamic trees, static
2191 * trees or store, and output the encoded block to the zip file. This function
2192 * returns the total compressed length for the file so far.
2194 static ulg flush_block(char *buf, ulg stored_len, int eof)
2196 ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
2197 int max_blindex; /* index of last bit length code of non zero freq */
2199 flag_buf[last_flags] = flags; /* Save the flags for the last 8 items */
2201 /* Check if the file is ascii or binary */
2202 if (*file_type == (ush) UNKNOWN)
2205 /* Construct the literal and distance trees */
2206 build_tree((tree_desc *) (&l_desc));
2207 Tracev((stderr, "\nlit data: dyn %ld, stat %ld", opt_len, static_len));
2209 build_tree((tree_desc *) (&d_desc));
2210 Tracev((stderr, "\ndist data: dyn %ld, stat %ld", opt_len, static_len));
2211 /* At this point, opt_len and static_len are the total bit lengths of
2212 * the compressed block data, excluding the tree representations.
2215 /* Build the bit length tree for the above two trees, and get the index
2216 * in bl_order of the last bit length code to send.
2218 max_blindex = build_bl_tree();
2220 /* Determine the best encoding. Compute first the block length in bytes */
2221 opt_lenb = (opt_len + 3 + 7) >> 3;
2222 static_lenb = (static_len + 3 + 7) >> 3;
2225 "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ",
2226 opt_lenb, opt_len, static_lenb, static_len, stored_len,
2227 last_lit, last_dist));
2229 if (static_lenb <= opt_lenb)
2230 opt_lenb = static_lenb;
2232 /* If compression failed and this is the first and last block,
2233 * and if the zip file can be seeked (to rewrite the local header),
2234 * the whole file is transformed into a stored file:
2236 if (stored_len <= opt_lenb && eof && compressed_len == 0L && seekable()) {
2237 /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */
2238 if (buf == (char *) 0)
2239 bb_error_msg("block vanished");
2241 copy_block(buf, (unsigned) stored_len, 0); /* without header */
2242 compressed_len = stored_len << 3;
2243 *file_method = STORED;
2245 } else if (stored_len + 4 <= opt_lenb && buf != (char *) 0) {
2246 /* 4: two words for the lengths */
2247 /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE.
2248 * Otherwise we can't have processed more than WSIZE input bytes since
2249 * the last block flush, because compression would have been
2250 * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to
2251 * transform a block into a stored block.
2253 send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */
2254 compressed_len = (compressed_len + 3 + 7) & ~7L;
2255 compressed_len += (stored_len + 4) << 3;
2257 copy_block(buf, (unsigned) stored_len, 1); /* with header */
2259 } else if (static_lenb == opt_lenb) {
2260 send_bits((STATIC_TREES << 1) + eof, 3);
2261 compress_block((ct_data *) static_ltree, (ct_data *) static_dtree);
2262 compressed_len += 3 + static_len;
2264 send_bits((DYN_TREES << 1) + eof, 3);
2265 send_all_trees(l_desc.max_code + 1, d_desc.max_code + 1,
2267 compress_block((ct_data *) dyn_ltree, (ct_data *) dyn_dtree);
2268 compressed_len += 3 + opt_len;
2270 Assert(compressed_len == bits_sent, "bad compressed size");
2275 compressed_len += 7; /* align on byte boundary */
2277 Tracev((stderr, "\ncomprlen %lu(%lu) ", compressed_len >> 3,
2278 compressed_len - 7 * eof));
2280 return compressed_len >> 3;
2283 /* ===========================================================================
2284 * Save the match info and tally the frequency counts. Return true if
2285 * the current block must be flushed.
2287 static int ct_tally(int dist, int lc)
2289 l_buf[last_lit++] = (uch) lc;
2291 /* lc is the unmatched char */
2292 dyn_ltree[lc].Freq++;
2294 /* Here, lc is the match length - MIN_MATCH */
2295 dist--; /* dist = match distance - 1 */
2296 Assert((ush) dist < (ush) MAX_DIST &&
2297 (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH) &&
2298 (ush) d_code(dist) < (ush) D_CODES, "ct_tally: bad match");
2300 dyn_ltree[length_code[lc] + LITERALS + 1].Freq++;
2301 dyn_dtree[d_code(dist)].Freq++;
2303 d_buf[last_dist++] = (ush) dist;
2308 /* Output the flags if they fill a byte: */
2309 if ((last_lit & 7) == 0) {
2310 flag_buf[last_flags++] = flags;
2311 flags = 0, flag_bit = 1;
2313 /* Try to guess if it is profitable to stop the current block here */
2314 if ((last_lit & 0xfff) == 0) {
2315 /* Compute an upper bound for the compressed length */
2316 ulg out_length = (ulg) last_lit * 8L;
2317 ulg in_length = (ulg) strstart - block_start;
2320 for (dcode = 0; dcode < D_CODES; dcode++) {
2322 (ulg) dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]);
2326 "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ",
2327 last_lit, last_dist, in_length, out_length,
2328 100L - out_length * 100L / in_length));
2329 if (last_dist < last_lit / 2 && out_length < in_length / 2)
2332 return (last_lit == LIT_BUFSIZE - 1 || last_dist == DIST_BUFSIZE);
2333 /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K
2334 * on 16 bit machines and because stored blocks are restricted to
2339 /* ===========================================================================
2340 * Send the block data compressed using the given Huffman trees
2342 static void compress_block(ct_data * ltree, ct_data * dtree)
2344 unsigned dist; /* distance of matched string */
2345 int lc; /* match length or unmatched char (if dist == 0) */
2346 unsigned lx = 0; /* running index in l_buf */
2347 unsigned dx = 0; /* running index in d_buf */
2348 unsigned fx = 0; /* running index in flag_buf */
2349 uch flag = 0; /* current flags */
2350 unsigned code; /* the code to send */
2351 int extra; /* number of extra bits to send */
2356 flag = flag_buf[fx++];
2358 if ((flag & 1) == 0) {
2359 send_code(lc, ltree); /* send a literal byte */
2360 Tracecv(isgraph(lc), (stderr, " '%c' ", lc));
2362 /* Here, lc is the match length - MIN_MATCH */
2363 code = length_code[lc];
2364 send_code(code + LITERALS + 1, ltree); /* send the length code */
2365 extra = extra_lbits[code];
2367 lc -= base_length[code];
2368 send_bits(lc, extra); /* send the extra length bits */
2371 /* Here, dist is the match distance - 1 */
2372 code = d_code(dist);
2373 Assert(code < D_CODES, "bad d_code");
2375 send_code(code, dtree); /* send the distance code */
2376 extra = extra_dbits[code];
2378 dist -= base_dist[code];
2379 send_bits(dist, extra); /* send the extra distance bits */
2381 } /* literal or match pair ? */
2383 } while (lx < last_lit);
2385 send_code(END_BLOCK, ltree);
2388 /* ===========================================================================
2389 * Set the file type to ASCII or BINARY, using a crude approximation:
2390 * binary if more than 20% of the bytes are <= 6 or >= 128, ascii otherwise.
2391 * IN assertion: the fields freq of dyn_ltree are set and the total of all
2392 * frequencies does not exceed 64K (to fit in an int on 16 bit machines).
2394 static void set_file_type(void)
2397 unsigned ascii_freq = 0;
2398 unsigned bin_freq = 0;
2401 bin_freq += dyn_ltree[n++].Freq;
2403 ascii_freq += dyn_ltree[n++].Freq;
2404 while (n < LITERALS)
2405 bin_freq += dyn_ltree[n++].Freq;
2406 *file_type = bin_freq > (ascii_freq >> 2) ? BINARY : ASCII;
2407 if (*file_type == BINARY && translate_eol) {
2408 bb_error_msg("-l used on binary file");
2412 /* zip.c -- compress files to the gzip or pkzip format
2413 * Copyright (C) 1992-1993 Jean-loup Gailly
2414 * This is free software; you can redistribute it and/or modify it under the
2415 * terms of the GNU General Public License, see the file COPYING.
2419 static ulg crc; /* crc on uncompressed file data */
2420 static long header_bytes; /* number of bytes in gzip header */
2422 static void put_long(ulg n)
2424 put_short((n) & 0xffff);
2425 put_short(((ulg) (n)) >> 16);
2428 /* put_header_byte is used for the compressed output
2429 * - for the initial 4 bytes that can't overflow the buffer.
2431 #define put_header_byte(c) {outbuf[outcnt++]=(uch)(c);}
2433 /* ===========================================================================
2434 * Deflate in to out.
2435 * IN assertions: the input and output buffers are cleared.
2436 * The variables time_stamp and save_orig_name are initialized.
2438 static int zip(int in, int out)
2440 uch my_flags = 0; /* general purpose bit flags */
2441 ush attr = 0; /* ascii/binary flag */
2442 ush deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */
2448 /* Write the header to the gzip file. See algorithm.doc for the format */
2452 put_header_byte(GZIP_MAGIC[0]); /* magic header */
2453 put_header_byte(GZIP_MAGIC[1]);
2454 put_header_byte(DEFLATED); /* compression method */
2456 put_header_byte(my_flags); /* general flags */
2457 put_long(time_stamp);
2459 /* Write deflated file to zip file */
2463 ct_init(&attr, &method);
2464 lm_init(&deflate_flags);
2466 put_byte((uch) deflate_flags); /* extra flags */
2467 put_byte(OS_CODE); /* OS identifier */
2469 header_bytes = (long) outcnt;
2473 /* Write the crc and uncompressed size */
2476 header_bytes += 2 * sizeof(long);
2483 /* ===========================================================================
2484 * Read a new buffer from the current input file, perform end-of-line
2485 * translation, and update the crc and input file size.
2486 * IN assertion: size >= 2 (for end-of-line translation)
2488 static int file_read(char *buf, unsigned size)
2492 Assert(insize == 0, "inbuf not empty");
2494 len = read(ifd, buf, size);
2495 if (len == (unsigned) (-1) || len == 0)
2498 crc = updcrc((uch *) buf, len);
2503 /* ===========================================================================
2504 * Write the output buffer outbuf[0..outcnt-1] and update bytes_out.
2505 * (used for the compressed data only)
2507 static void flush_outbuf(void)
2512 write_buf(ofd, (char *) outbuf, outcnt);