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 source tree.
18 /* TODO: full support for -v for DESKTOP
19 * "/usr/bin/gzip -v a bogus aa" should say:
20 a: 85.1% -- replaced with a.gz
21 gzip: bogus: No such file or directory
22 aa: 85.1% -- replaced with aa.gz
25 //config: bool "gzip (19 kb)"
28 //config: gzip is used to compress files.
29 //config: It's probably the most widely used UNIX compression program.
31 //config:config FEATURE_GZIP_LONG_OPTIONS
32 //config: bool "Enable long options"
34 //config: depends on GZIP && LONG_OPTS
36 //config:config GZIP_FAST
37 //config: int "Trade memory for speed (0:small,slow - 2:fast,big)"
40 //config: depends on GZIP
42 //config: Enable big memory options for gzip.
43 //config: 0: small buffers, small hash-tables
44 //config: 1: larger buffers, larger hash-tables
45 //config: 2: larger buffers, largest hash-tables
46 //config: Larger models may give slightly better compression
48 //config:config FEATURE_GZIP_LEVELS
49 //config: bool "Enable compression levels"
51 //config: depends on GZIP
53 //config: Enable support for compression levels 4-9. The default level
54 //config: is 6. If levels 1-3 are specified, 4 is used.
55 //config: If this option is not selected, -N options are ignored and -9
58 //config:config FEATURE_GZIP_DECOMPRESS
59 //config: bool "Enable decompression"
61 //config: depends on GZIP || GUNZIP || ZCAT
63 //config: Enable -d (--decompress) and -t (--test) options for gzip.
64 //config: This will be automatically selected if gunzip or zcat is
67 //applet:IF_GZIP(APPLET(gzip, BB_DIR_BIN, BB_SUID_DROP))
69 //kbuild:lib-$(CONFIG_GZIP) += gzip.o
71 //usage:#define gzip_trivial_usage
72 //usage: "[-cfk" IF_FEATURE_GZIP_DECOMPRESS("dt") IF_FEATURE_GZIP_LEVELS("123456789") "] [FILE]..."
73 //usage:#define gzip_full_usage "\n\n"
74 //usage: "Compress FILEs (or stdin)\n"
75 //usage: IF_FEATURE_GZIP_LEVELS(
76 //usage: "\n -1..9 Compression level"
78 //usage: IF_FEATURE_GZIP_DECOMPRESS(
79 //usage: "\n -d Decompress"
80 //usage: "\n -t Test file integrity"
82 //usage: "\n -c Write to stdout"
83 //usage: "\n -f Force"
84 //usage: "\n -k Keep input files"
86 //usage:#define gzip_example_usage
87 //usage: "$ ls -la /tmp/busybox*\n"
88 //usage: "-rw-rw-r-- 1 andersen andersen 1761280 Apr 14 17:47 /tmp/busybox.tar\n"
89 //usage: "$ gzip /tmp/busybox.tar\n"
90 //usage: "$ ls -la /tmp/busybox*\n"
91 //usage: "-rw-rw-r-- 1 andersen andersen 554058 Apr 14 17:49 /tmp/busybox.tar.gz\n"
94 #include "bb_archive.h"
97 /* ===========================================================================
100 /* Diagnostic functions */
103 # define Assert(cond,msg) { if (!(cond)) bb_error_msg(msg); }
104 # define Trace(x) fprintf x
105 # define Tracev(x) {if (verbose) fprintf x; }
106 # define Tracevv(x) {if (verbose > 1) fprintf x; }
107 # define Tracec(c,x) {if (verbose && (c)) fprintf x; }
108 # define Tracecv(c,x) {if (verbose > 1 && (c)) fprintf x; }
110 # define Assert(cond,msg)
115 # define Tracecv(c,x)
119 /* ===========================================================================
121 #if CONFIG_GZIP_FAST == 0
123 #elif CONFIG_GZIP_FAST == 1
125 #elif CONFIG_GZIP_FAST == 2
128 # error "Invalid CONFIG_GZIP_FAST value"
133 # define INBUFSIZ 0x2000 /* input buffer size */
135 # define INBUFSIZ 0x8000 /* input buffer size */
141 # define OUTBUFSIZ 8192 /* output buffer size */
143 # define OUTBUFSIZ 16384 /* output buffer size */
149 # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */
151 # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */
156 #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */
157 #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
158 #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
159 #define ORIG_NAME 0x08 /* bit 3 set: original file name present */
160 #define COMMENT 0x10 /* bit 4 set: file comment present */
161 #define RESERVED 0xC0 /* bit 6,7: reserved */
163 /* internal file attribute */
164 #define UNKNOWN 0xffff
169 # define WSIZE 0x8000 /* window size--must be a power of two, and */
170 #endif /* at least 32K for zip's deflate method */
173 #define MAX_MATCH 258
174 /* The minimum and maximum match lengths */
176 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
177 /* Minimum amount of lookahead, except at the end of the input file.
178 * See deflate.c for comments about the MIN_MATCH+1.
181 #define MAX_DIST (WSIZE-MIN_LOOKAHEAD)
182 /* In order to simplify the code, particularly on 16 bit machines, match
183 * distances are limited to MAX_DIST instead of WSIZE.
187 # define MAX_PATH_LEN 1024 /* max pathname length */
190 #define seekable() 0 /* force sequential output */
191 #define translate_eol 0 /* no option -a yet */
196 #define INIT_BITS 9 /* Initial number of bits per code */
198 #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */
199 /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free.
200 * It's a pity that old uncompress does not check bit 0x20. That makes
201 * extension of the format actually undesirable because old compress
202 * would just crash on the new format instead of giving a meaningful
203 * error message. It does check the number of bits, but it's more
204 * helpful to say "unsupported format, get a new version" than
205 * "can only handle 16 bits".
209 # define MAX_SUFFIX MAX_EXT_CHARS
211 # define MAX_SUFFIX 30
215 /* ===========================================================================
216 * Compile with MEDIUM_MEM to reduce the memory requirements or
217 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
218 * entire input file can be held in memory (not possible on 16 bit systems).
219 * Warning: defining these symbols affects HASH_BITS (see below) and thus
220 * affects the compression ratio. The compressed output
221 * is still correct, and might even be smaller in some cases.
225 # define HASH_BITS 13 /* Number of bits used to hash strings */
228 # define HASH_BITS 14
231 # define HASH_BITS 15
232 /* For portability to 16 bit machines, do not use values above 15. */
235 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
236 #define HASH_MASK (HASH_SIZE-1)
237 #define WMASK (WSIZE-1)
238 /* HASH_SIZE and WSIZE must be powers of two */
240 # define TOO_FAR 4096
242 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
245 /* ===========================================================================
246 * These types are not really 'char', 'short' and 'long'
249 typedef uint16_t ush;
250 typedef uint32_t ulg;
254 typedef unsigned IPos;
255 /* A Pos is an index in the character window. We use short instead of int to
256 * save space in the various tables. IPos is used only for parameter passing.
260 WINDOW_SIZE = 2 * WSIZE,
261 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
262 * input file length plus MIN_LOOKAHEAD.
265 #if !ENABLE_FEATURE_GZIP_LEVELS
267 max_chain_length = 4096,
268 /* To speed up deflation, hash chains are never searched beyond this length.
269 * A higher limit improves compression ratio but degrades the speed.
272 max_lazy_match = 258,
273 /* Attempt to find a better match only when the current match is strictly
274 * smaller than this value. This mechanism is used only for compression
278 max_insert_length = max_lazy_match,
279 /* Insert new strings in the hash table only if the match length
280 * is not greater than this length. This saves time but degrades compression.
281 * max_insert_length is used only for compression levels <= 3.
285 /* Use a faster search when the previous match is longer than this */
287 /* Values for max_lazy_match, good_match and max_chain_length, depending on
288 * the desired pack level (0..9). The values given below have been tuned to
289 * exclude worst case performance for pathological files. Better values may be
290 * found for specific files.
293 nice_match = 258, /* Stop searching when current match exceeds this */
294 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
295 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
298 #endif /* ENABLE_FEATURE_GZIP_LEVELS */
304 #if ENABLE_FEATURE_GZIP_LEVELS
305 unsigned max_chain_length;
306 unsigned max_lazy_match;
309 #define max_chain_length (G1.max_chain_length)
310 #define max_lazy_match (G1.max_lazy_match)
311 #define good_match (G1.good_match)
312 #define nice_match (G1.nice_match)
315 /* window position at the beginning of the current output block. Gets
316 * negative when the window is moved backwards.
320 unsigned ins_h; /* hash index of string to be inserted */
322 /* Number of bits by which ins_h and del_h must be shifted at each
323 * input step. It must be such that after MIN_MATCH steps, the oldest
324 * byte no longer takes part in the hash key, that is:
325 * H_SHIFT * MIN_MATCH >= HASH_BITS
327 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1) / MIN_MATCH)
329 /* Length of the best match at previous step. Matches not greater than this
330 * are discarded. This is used in the lazy match evaluation.
332 unsigned prev_length;
334 unsigned strstart; /* start of string to insert */
335 unsigned match_start; /* start of matching string */
336 unsigned lookahead; /* number of valid bytes ahead in window */
338 /* number of input bytes */
339 ulg isize; /* only 32 bits stored in .gz file */
341 /* bbox always use stdin/stdout */
342 #define ifd STDIN_FILENO /* input file descriptor */
343 #define ofd STDOUT_FILENO /* output file descriptor */
346 unsigned insize; /* valid bytes in l_buf */
348 unsigned outcnt; /* bytes in output buffer */
349 smallint eofile; /* flag set at end of input file */
351 /* ===========================================================================
352 * Local data used by the "bit string" routines.
355 /* Output buffer. bits are inserted starting at the bottom (least significant
358 unsigned bi_buf; /* was unsigned short */
361 #define BUF_SIZE (int)(8 * sizeof(G1.bi_buf))
363 /* Number of bits used within bi_buf. (bi_buf might be implemented on
364 * more than 16 bits on some systems.)
369 ulg bits_sent; /* bit length of the compressed data */
372 /*uint32_t *crc_32_tab;*/
373 uint32_t crc; /* shift register contents */
375 /* ===========================================================================
377 #define DECLARE(type, array, size) \
379 #define ALLOC(type, array, size) \
380 array = xzalloc((size_t)(((size)+1L)/2) * 2*sizeof(type))
381 #define FREE(array) \
382 do { free(array); array = NULL; } while (0)
386 /* buffer for literals or lengths */
387 /* DECLARE(uch, l_buf, LIT_BUFSIZE); */
388 DECLARE(uch, l_buf, INBUFSIZ);
390 DECLARE(ush, d_buf, DIST_BUFSIZE);
391 DECLARE(uch, outbuf, OUTBUFSIZ);
393 /* Sliding window. Input bytes are read into the second half of the window,
394 * and move to the first half later to keep a dictionary of at least WSIZE
395 * bytes. With this organization, matches are limited to a distance of
396 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
397 * performed with a length multiple of the block size. Also, it limits
398 * the window size to 64K, which is quite useful on MSDOS.
399 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
400 * be less efficient).
402 DECLARE(uch, window, 2L * WSIZE);
404 /* Link to older string with same hash index. To limit the size of this
405 * array to 64K, this link is maintained only for the last 32K strings.
406 * An index in this array is thus a window index modulo 32K.
408 /* DECLARE(Pos, prev, WSIZE); */
409 DECLARE(ush, prev, 1L << BITS);
411 /* Heads of the hash chains or 0. */
412 /* DECLARE(Pos, head, 1<<HASH_BITS); */
413 #define head (G1.prev + WSIZE) /* hash head (see deflate.c) */
416 #define G1 (*(ptr_to_globals - 1))
419 /* ===========================================================================
420 * Write the output buffer outbuf[0..outcnt-1] and update bytes_out.
421 * (used for the compressed data only)
423 static void flush_outbuf(void)
428 xwrite(ofd, (char *) G1.outbuf, G1.outcnt);
433 /* ===========================================================================
435 /* put_8bit is used for the compressed output */
436 #define put_8bit(c) \
438 G1.outbuf[G1.outcnt++] = (c); \
439 if (G1.outcnt == OUTBUFSIZ) \
443 /* Output a 16 bit value, lsb first */
444 static void put_16bit(ush w)
446 /* GCC 4.2.1 won't optimize out redundant loads of G1.outcnt
447 * (probably because of fear of aliasing with G1.outbuf[]
448 * stores), do it explicitly:
450 unsigned outcnt = G1.outcnt;
451 uch *dst = &G1.outbuf[outcnt];
453 #if BB_UNALIGNED_MEMACCESS_OK && BB_LITTLE_ENDIAN
454 if (outcnt < OUTBUFSIZ-2) {
456 ush *dst16 = (void*) dst;
457 *dst16 = w; /* unaligned LSB 16-bit store */
458 G1.outcnt = outcnt + 2;
466 if (outcnt < OUTBUFSIZ-2) {
469 G1.outcnt = outcnt + 2;
474 /* Slowpath: we will need to do flush_outbuf() */
475 G1.outcnt = ++outcnt;
476 if (outcnt == OUTBUFSIZ)
481 #define OPTIMIZED_PUT_32BIT (CONFIG_GZIP_FAST > 0 && BB_UNALIGNED_MEMACCESS_OK && BB_LITTLE_ENDIAN)
482 static void put_32bit(ulg n)
484 #if OPTIMIZED_PUT_32BIT
485 unsigned outcnt = G1.outcnt;
486 if (outcnt < OUTBUFSIZ-4) {
488 ulg *dst32 = (void*) &G1.outbuf[outcnt];
489 *dst32 = n; /* unaligned LSB 32-bit store */
490 G1.outcnt = outcnt + 4;
498 /* ===========================================================================
499 * Run a set of bytes through the crc shift register. If s is a NULL
500 * pointer, then initialize the crc shift register contents instead.
501 * Return the current crc in either case.
503 static void updcrc(uch * s, unsigned n)
505 G1.crc = crc32_block_endian0(G1.crc, s, n, global_crc32_table /*G1.crc_32_tab*/);
509 /* ===========================================================================
510 * Read a new buffer from the current input file, perform end-of-line
511 * translation, and update the crc and input file size.
512 * IN assertion: size >= 2 (for end-of-line translation)
514 static unsigned file_read(void *buf, unsigned size)
518 Assert(G1.insize == 0, "l_buf not empty");
520 len = safe_read(ifd, buf, size);
521 if (len == (unsigned)(-1) || len == 0)
530 /* ===========================================================================
531 * Send a value on a given number of bits.
532 * IN assertion: length <= 16 and value fits in length bits.
534 static void send_bits(unsigned value, unsigned length)
539 Tracev((stderr, " l %2d v %4x ", length, value));
540 Assert(length > 0 && length <= 15, "invalid length");
541 G1.bits_sent += length;
543 BUILD_BUG_ON(BUF_SIZE != 32 && BUF_SIZE != 16);
545 new_buf = G1.bi_buf | (value << G1.bi_valid);
546 /* NB: the above may sometimes do "<< 32" shift (undefined)
547 * if check below is changed to "length > BUF_SIZE" instead of >= */
548 length += G1.bi_valid;
550 /* If bi_buf is full */
551 if (length >= BUF_SIZE) {
552 /* ...use (valid) bits from bi_buf and
553 * (BUF_SIZE - bi_valid) bits from value,
554 * leaving (width - (BUF_SIZE-bi_valid)) unused bits in value.
556 value >>= (BUF_SIZE - G1.bi_valid);
557 if (BUF_SIZE == 32) {
566 G1.bi_valid = length;
570 /* ===========================================================================
571 * Reverse the first len bits of a code, using straightforward code (a faster
572 * method would use a table)
573 * IN assertion: 1 <= len <= 15
575 static unsigned bi_reverse(unsigned code, int len)
581 if (--len <= 0) return res;
588 /* ===========================================================================
589 * Write out any remaining bits in an incomplete byte.
591 static void bi_windup(void)
593 unsigned bits = G1.bi_buf;
594 int cnt = G1.bi_valid;
604 G1.bits_sent = (G1.bits_sent + 7) & ~7;
609 /* ===========================================================================
610 * Copy a stored block to the zip file, storing first the length and its
611 * one's complement if requested.
613 static void copy_block(char *buf, unsigned len, int header)
615 bi_windup(); /* align on byte boundary */
621 G1.bits_sent += 2 * 16;
625 G1.bits_sent += (ulg) len << 3;
633 /* ===========================================================================
634 * Fill the window when the lookahead becomes insufficient.
635 * Updates strstart and lookahead, and sets eofile if end of input file.
636 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
637 * OUT assertions: at least one byte has been read, or eofile is set;
638 * file reads are performed for at least two bytes (required for the
639 * translate_eol option).
641 static void fill_window(void)
644 unsigned more = WINDOW_SIZE - G1.lookahead - G1.strstart;
645 /* Amount of free space at the end of the window. */
647 /* If the window is almost full and there is insufficient lookahead,
648 * move the upper half to the lower one to make room in the upper half.
650 if (more == (unsigned) -1) {
651 /* Very unlikely, but possible on 16 bit machine if strstart == 0
652 * and lookahead == 1 (input done one byte at time)
655 } else if (G1.strstart >= WSIZE + MAX_DIST) {
656 /* By the IN assertion, the window is not empty so we can't confuse
657 * more == 0 with more == 64K on a 16 bit machine.
659 Assert(WINDOW_SIZE == 2 * WSIZE, "no sliding with BIG_MEM");
661 memcpy(G1.window, G1.window + WSIZE, WSIZE);
662 G1.match_start -= WSIZE;
663 G1.strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
665 G1.block_start -= WSIZE;
667 for (n = 0; n < HASH_SIZE; n++) {
669 head[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
671 for (n = 0; n < WSIZE; n++) {
673 G1.prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
674 /* If n is not on any hash chain, prev[n] is garbage but
675 * its value will never be used.
680 /* At this point, more >= 2 */
682 n = file_read(G1.window + G1.strstart + G1.lookahead, more);
683 if (n == 0 || n == (unsigned) -1) {
690 /* Both users fill window with the same loop: */
691 static void fill_window_if_needed(void)
693 while (G1.lookahead < MIN_LOOKAHEAD && !G1.eofile)
698 /* ===========================================================================
699 * Set match_start to the longest match starting at the given string and
700 * return its length. Matches shorter or equal to prev_length are discarded,
701 * in which case the result is equal to prev_length and match_start is
703 * IN assertions: cur_match is the head of the hash chain for the current
704 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
707 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
708 * match.s. The code is functionally equivalent, so you can use the C version
711 static int longest_match(IPos cur_match)
713 unsigned chain_length = max_chain_length; /* max hash chain length */
714 uch *scan = G1.window + G1.strstart; /* current string */
715 uch *match; /* matched string */
716 int len; /* length of current match */
717 int best_len = G1.prev_length; /* best match length so far */
718 IPos limit = G1.strstart > (IPos) MAX_DIST ? G1.strstart - (IPos) MAX_DIST : 0;
719 /* Stop when cur_match becomes <= limit. To simplify the code,
720 * we prevent matches with the string of window index 0.
723 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
724 * It is easy to get rid of this optimization if necessary.
726 #if HASH_BITS < 8 || MAX_MATCH != 258
727 # error Code too clever
729 uch *strend = G1.window + G1.strstart + MAX_MATCH;
730 uch scan_end1 = scan[best_len - 1];
731 uch scan_end = scan[best_len];
733 /* Do not waste too much time if we already have a good match: */
734 if (G1.prev_length >= good_match) {
737 Assert(G1.strstart <= WINDOW_SIZE - MIN_LOOKAHEAD, "insufficient lookahead");
740 Assert(cur_match < G1.strstart, "no future");
741 match = G1.window + cur_match;
743 /* Skip to next match if the match length cannot increase
744 * or if the match length is less than 2:
746 if (match[best_len] != scan_end
747 || match[best_len - 1] != scan_end1
748 || *match != *scan || *++match != scan[1]
753 /* The check at best_len-1 can be removed because it will be made
754 * again later. (This heuristic is not always a win.)
755 * It is not necessary to compare scan[2] and match[2] since they
756 * are always equal when the other bytes match, given that
757 * the hash keys are equal and that HASH_BITS >= 8.
761 /* We check for insufficient lookahead only every 8th comparison;
762 * the 256th check will be made at strstart+258.
765 } while (*++scan == *++match && *++scan == *++match &&
766 *++scan == *++match && *++scan == *++match &&
767 *++scan == *++match && *++scan == *++match &&
768 *++scan == *++match && *++scan == *++match && scan < strend);
770 len = MAX_MATCH - (int) (strend - scan);
771 scan = strend - MAX_MATCH;
773 if (len > best_len) {
774 G1.match_start = cur_match;
776 if (len >= nice_match)
778 scan_end1 = scan[best_len - 1];
779 scan_end = scan[best_len];
781 } while ((cur_match = G1.prev[cur_match & WMASK]) > limit
782 && --chain_length != 0);
789 /* ===========================================================================
790 * Check that the match at match_start is indeed a match.
792 static void check_match(IPos start, IPos match, int length)
794 /* check that the match is indeed a match */
795 if (memcmp(G1.window + match, G1.window + start, length) != 0) {
796 bb_error_msg(" start %d, match %d, length %d", start, match, length);
797 bb_error_msg("invalid match");
800 bb_error_msg("\\[%d,%d]", start - match, length);
802 bb_putchar_stderr(G1.window[start++]);
803 } while (--length != 0);
807 # define check_match(start, match, length) ((void)0)
811 /* trees.c -- output deflated data using Huffman coding
812 * Copyright (C) 1992-1993 Jean-loup Gailly
813 * This is free software; you can redistribute it and/or modify it under the
814 * terms of the GNU General Public License, see the file COPYING.
818 * Encode various sets of source values using variable-length
822 * The PKZIP "deflation" process uses several Huffman trees. The more
823 * common source values are represented by shorter bit sequences.
825 * Each code tree is stored in the ZIP file in a compressed form
826 * which is itself a Huffman encoding of the lengths of
827 * all the code strings (in ascending order by source values).
828 * The actual code strings are reconstructed from the lengths in
829 * the UNZIP process, as described in the "application note"
830 * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program.
834 * Data Compression: Techniques and Applications, pp. 53-55.
835 * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7.
838 * Data Compression: Methods and Theory, pp. 49-50.
839 * Computer Science Press, 1988. ISBN 0-7167-8156-5.
843 * Addison-Wesley, 1983. ISBN 0-201-06672-6.
847 * Allocate the match buffer, initialize the various tables [and save
848 * the location of the internal file attribute (ascii/binary) and
849 * method (DEFLATE/STORE) -- deleted in bbox]
851 * void ct_tally(int dist, int lc);
852 * Save the match info and tally the frequency counts.
854 * ulg flush_block(char *buf, ulg stored_len, int eof)
855 * Determine the best encoding for the current block: dynamic trees,
856 * static trees or store, and output the encoded block to the zip
857 * file. Returns the total compressed length for the file so far.
861 /* All codes must not exceed MAX_BITS bits */
863 #define MAX_BL_BITS 7
864 /* Bit length codes must not exceed MAX_BL_BITS bits */
866 #define LENGTH_CODES 29
867 /* number of length codes, not counting the special END_BLOCK code */
870 /* number of literal bytes 0..255 */
872 #define END_BLOCK 256
873 /* end of block literal code */
875 #define L_CODES (LITERALS+1+LENGTH_CODES)
876 /* number of Literal or Length codes, including the END_BLOCK code */
879 /* number of distance codes */
882 /* number of codes used to transfer the bit lengths */
884 /* extra bits for each length code */
885 static const uint8_t extra_lbits[LENGTH_CODES] ALIGN1 = {
886 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4,
890 /* extra bits for each distance code */
891 static const uint8_t extra_dbits[D_CODES] ALIGN1 = {
892 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,
893 10, 10, 11, 11, 12, 12, 13, 13
896 /* extra bits for each bit length code */
897 static const uint8_t extra_blbits[BL_CODES] ALIGN1 = {
898 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 };
900 /* number of codes at each bit length for an optimal tree */
901 static const uint8_t bl_order[BL_CODES] ALIGN1 = {
902 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
904 #define STORED_BLOCK 0
905 #define STATIC_TREES 1
907 /* The three kinds of block type */
911 # define LIT_BUFSIZE 0x2000
914 # define LIT_BUFSIZE 0x4000
916 # define LIT_BUFSIZE 0x8000
921 # define DIST_BUFSIZE LIT_BUFSIZE
923 /* Sizes of match buffers for literals/lengths and distances. There are
924 * 4 reasons for limiting LIT_BUFSIZE to 64K:
925 * - frequencies can be kept in 16 bit counters
926 * - if compression is not successful for the first block, all input data is
927 * still in the window so we can still emit a stored block even when input
928 * comes from standard input. (This can also be done for all blocks if
929 * LIT_BUFSIZE is not greater than 32K.)
930 * - if compression is not successful for a file smaller than 64K, we can
931 * even emit a stored file instead of a stored block (saving 5 bytes).
932 * - creating new Huffman trees less frequently may not provide fast
933 * adaptation to changes in the input data statistics. (Take for
934 * example a binary file with poorly compressible code followed by
935 * a highly compressible string table.) Smaller buffer sizes give
936 * fast adaptation but have of course the overhead of transmitting trees
938 * - I can't count above 4
939 * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save
940 * memory at the expense of compression). Some optimizations would be possible
941 * if we rely on DIST_BUFSIZE == LIT_BUFSIZE.
944 /* repeat previous bit length 3-6 times (2 bits of repeat count) */
946 /* repeat a zero length 3-10 times (3 bits of repeat count) */
947 #define REPZ_11_138 18
948 /* repeat a zero length 11-138 times (7 bits of repeat count) */
950 /* ===========================================================================
952 /* Data structure describing a single value and its code string. */
953 typedef struct ct_data {
955 ush freq; /* frequency count */
956 ush code; /* bit string */
959 ush dad; /* father node in Huffman tree */
960 ush len; /* length of bit string */
969 #define HEAP_SIZE (2*L_CODES + 1)
970 /* maximum heap size */
972 typedef struct tree_desc {
973 ct_data *dyn_tree; /* the dynamic tree */
974 ct_data *static_tree; /* corresponding static tree or NULL */
975 const uint8_t *extra_bits; /* extra bits for each code or NULL */
976 int extra_base; /* base index for extra_bits */
977 int elems; /* max number of elements in the tree */
978 int max_length; /* max bit length for the codes */
979 int max_code; /* largest code with non zero frequency */
984 ush heap[HEAP_SIZE]; /* heap used to build the Huffman trees */
985 int heap_len; /* number of elements in the heap */
986 int heap_max; /* element of largest frequency */
988 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
989 * The same heap array is used to build all trees.
992 ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */
993 ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */
995 ct_data static_ltree[L_CODES + 2];
997 /* The static literal tree. Since the bit lengths are imposed, there is no
998 * need for the L_CODES extra codes used during heap construction. However
999 * The codes 286 and 287 are needed to build a canonical tree (see ct_init
1003 ct_data static_dtree[D_CODES];
1005 /* The static distance tree. (Actually a trivial tree since all codes use
1009 ct_data bl_tree[2 * BL_CODES + 1];
1011 /* Huffman tree for the bit lengths */
1017 ush bl_count[MAX_BITS + 1];
1019 /* The lengths of the bit length codes are sent in order of decreasing
1020 * probability, to avoid transmitting the lengths for unused bit length codes.
1023 uch depth[2 * L_CODES + 1];
1025 /* Depth of each subtree used as tie breaker for trees of equal frequency */
1027 uch length_code[MAX_MATCH - MIN_MATCH + 1];
1029 /* length code for each normalized match length (0 == MIN_MATCH) */
1033 /* distance codes. The first 256 values correspond to the distances
1034 * 3 .. 258, the last 256 values correspond to the top 8 bits of
1035 * the 15 bit distances.
1038 int base_length[LENGTH_CODES];
1040 /* First normalized length for each code (0 = MIN_MATCH) */
1042 int base_dist[D_CODES];
1044 /* First normalized distance for each code (0 = distance of 1) */
1046 uch flag_buf[LIT_BUFSIZE / 8];
1048 /* flag_buf is a bit array distinguishing literals from lengths in
1049 * l_buf, thus indicating the presence or absence of a distance.
1052 unsigned last_lit; /* running index in l_buf */
1053 unsigned last_dist; /* running index in d_buf */
1054 unsigned last_flags; /* running index in flag_buf */
1055 uch flags; /* current flags not yet saved in flag_buf */
1056 uch flag_bit; /* current bit used in flags */
1058 /* bits are filled in flags starting at bit 0 (least significant).
1059 * Note: these flags are overkill in the current code since we don't
1060 * take advantage of DIST_BUFSIZE == LIT_BUFSIZE.
1063 ulg opt_len; /* bit length of current block with optimal trees */
1064 ulg static_len; /* bit length of current block with static trees */
1066 ulg compressed_len; /* total bit length of compressed file */
1069 #define G2ptr ((struct globals2*)(ptr_to_globals))
1073 /* ===========================================================================
1075 static void gen_codes(ct_data * tree, int max_code);
1076 static void build_tree(tree_desc * desc);
1077 static void scan_tree(ct_data * tree, int max_code);
1078 static void send_tree(ct_data * tree, int max_code);
1079 static int build_bl_tree(void);
1080 static void send_all_trees(int lcodes, int dcodes, int blcodes);
1081 static void compress_block(ct_data * ltree, ct_data * dtree);
1085 /* Send a code of the given tree. c and tree must not have side effects */
1086 # define SEND_CODE(c, tree) send_bits(tree[c].Code, tree[c].Len)
1088 # define SEND_CODE(c, tree) \
1090 if (verbose > 1) bb_error_msg("\ncd %3d ", (c)); \
1091 send_bits(tree[c].Code, tree[c].Len); \
1095 #define D_CODE(dist) \
1096 ((dist) < 256 ? G2.dist_code[dist] : G2.dist_code[256 + ((dist)>>7)])
1097 /* Mapping from a distance to a distance code. dist is the distance - 1 and
1098 * must not have side effects. dist_code[256] and dist_code[257] are never
1100 * The arguments must not have side effects.
1104 /* ===========================================================================
1105 * Initialize a new block.
1107 static void init_block(void)
1109 int n; /* iterates over tree elements */
1111 /* Initialize the trees. */
1112 for (n = 0; n < L_CODES; n++)
1113 G2.dyn_ltree[n].Freq = 0;
1114 for (n = 0; n < D_CODES; n++)
1115 G2.dyn_dtree[n].Freq = 0;
1116 for (n = 0; n < BL_CODES; n++)
1117 G2.bl_tree[n].Freq = 0;
1119 G2.dyn_ltree[END_BLOCK].Freq = 1;
1120 G2.opt_len = G2.static_len = 0;
1121 G2.last_lit = G2.last_dist = G2.last_flags = 0;
1127 /* ===========================================================================
1128 * Restore the heap property by moving down the tree starting at node k,
1129 * exchanging a node with the smallest of its two sons if necessary, stopping
1130 * when the heap property is re-established (each father smaller than its
1134 /* Compares to subtrees, using the tree depth as tie breaker when
1135 * the subtrees have equal frequency. This minimizes the worst case length. */
1136 #define SMALLER(tree, n, m) \
1137 (tree[n].Freq < tree[m].Freq \
1138 || (tree[n].Freq == tree[m].Freq && G2.depth[n] <= G2.depth[m]))
1140 static void pqdownheap(ct_data * tree, int k)
1143 int j = k << 1; /* left son of k */
1145 while (j <= G2.heap_len) {
1146 /* Set j to the smallest of the two sons: */
1147 if (j < G2.heap_len && SMALLER(tree, G2.heap[j + 1], G2.heap[j]))
1150 /* Exit if v is smaller than both sons */
1151 if (SMALLER(tree, v, G2.heap[j]))
1154 /* Exchange v with the smallest son */
1155 G2.heap[k] = G2.heap[j];
1158 /* And continue down the tree, setting j to the left son of k */
1165 /* ===========================================================================
1166 * Compute the optimal bit lengths for a tree and update the total bit length
1167 * for the current block.
1168 * IN assertion: the fields freq and dad are set, heap[heap_max] and
1169 * above are the tree nodes sorted by increasing frequency.
1170 * OUT assertions: the field len is set to the optimal bit length, the
1171 * array bl_count contains the frequencies for each bit length.
1172 * The length opt_len is updated; static_len is also updated if stree is
1175 static void gen_bitlen(tree_desc * desc)
1177 ct_data *tree = desc->dyn_tree;
1178 const uint8_t *extra = desc->extra_bits;
1179 int base = desc->extra_base;
1180 int max_code = desc->max_code;
1181 int max_length = desc->max_length;
1182 ct_data *stree = desc->static_tree;
1183 int h; /* heap index */
1184 int n, m; /* iterate over the tree elements */
1185 int bits; /* bit length */
1186 int xbits; /* extra bits */
1187 ush f; /* frequency */
1188 int overflow = 0; /* number of elements with bit length too large */
1190 for (bits = 0; bits <= MAX_BITS; bits++)
1191 G2.bl_count[bits] = 0;
1193 /* In a first pass, compute the optimal bit lengths (which may
1194 * overflow in the case of the bit length tree).
1196 tree[G2.heap[G2.heap_max]].Len = 0; /* root of the heap */
1198 for (h = G2.heap_max + 1; h < HEAP_SIZE; h++) {
1200 bits = tree[tree[n].Dad].Len + 1;
1201 if (bits > max_length) {
1205 tree[n].Len = (ush) bits;
1206 /* We overwrite tree[n].Dad which is no longer needed */
1209 continue; /* not a leaf node */
1211 G2.bl_count[bits]++;
1214 xbits = extra[n - base];
1216 G2.opt_len += (ulg) f *(bits + xbits);
1219 G2.static_len += (ulg) f * (stree[n].Len + xbits);
1224 Trace((stderr, "\nbit length overflow\n"));
1225 /* This happens for example on obj2 and pic of the Calgary corpus */
1227 /* Find the first bit length which could increase: */
1229 bits = max_length - 1;
1230 while (G2.bl_count[bits] == 0)
1232 G2.bl_count[bits]--; /* move one leaf down the tree */
1233 G2.bl_count[bits + 1] += 2; /* move one overflow item as its brother */
1234 G2.bl_count[max_length]--;
1235 /* The brother of the overflow item also moves one step up,
1236 * but this does not affect bl_count[max_length]
1239 } while (overflow > 0);
1241 /* Now recompute all bit lengths, scanning in increasing frequency.
1242 * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all
1243 * lengths instead of fixing only the wrong ones. This idea is taken
1244 * from 'ar' written by Haruhiko Okumura.)
1246 for (bits = max_length; bits != 0; bits--) {
1247 n = G2.bl_count[bits];
1252 if (tree[m].Len != (unsigned) bits) {
1253 Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len, bits));
1254 G2.opt_len += ((int32_t) bits - tree[m].Len) * tree[m].Freq;
1263 /* ===========================================================================
1264 * Generate the codes for a given tree and bit counts (which need not be
1266 * IN assertion: the array bl_count contains the bit length statistics for
1267 * the given tree and the field len is set for all tree elements.
1268 * OUT assertion: the field code is set for all tree elements of non
1271 static void gen_codes(ct_data * tree, int max_code)
1273 ush next_code[MAX_BITS + 1]; /* next code value for each bit length */
1274 ush code = 0; /* running code value */
1275 int bits; /* bit index */
1276 int n; /* code index */
1278 /* The distribution counts are first used to generate the code values
1279 * without bit reversal.
1281 for (bits = 1; bits <= MAX_BITS; bits++) {
1282 next_code[bits] = code = (code + G2.bl_count[bits - 1]) << 1;
1284 /* Check that the bit counts in bl_count are consistent. The last code
1287 Assert(code + G2.bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1,
1288 "inconsistent bit counts");
1289 Tracev((stderr, "\ngen_codes: max_code %d ", max_code));
1291 for (n = 0; n <= max_code; n++) {
1292 int len = tree[n].Len;
1296 /* Now reverse the bits */
1297 tree[n].Code = bi_reverse(next_code[len]++, len);
1299 Tracec(tree != G2.static_ltree,
1300 (stderr, "\nn %3d %c l %2d c %4x (%x) ", n,
1301 (n > ' ' ? n : ' '), len, tree[n].Code,
1302 next_code[len] - 1));
1307 /* ===========================================================================
1308 * Construct one Huffman tree and assigns the code bit strings and lengths.
1309 * Update the total bit length for the current block.
1310 * IN assertion: the field freq is set for all tree elements.
1311 * OUT assertions: the fields len and code are set to the optimal bit length
1312 * and corresponding code. The length opt_len is updated; static_len is
1313 * also updated if stree is not null. The field max_code is set.
1316 /* Remove the smallest element from the heap and recreate the heap with
1317 * one less element. Updates heap and heap_len. */
1320 /* Index within the heap array of least frequent node in the Huffman tree */
1322 #define PQREMOVE(tree, top) \
1324 top = G2.heap[SMALLEST]; \
1325 G2.heap[SMALLEST] = G2.heap[G2.heap_len--]; \
1326 pqdownheap(tree, SMALLEST); \
1329 static void build_tree(tree_desc * desc)
1331 ct_data *tree = desc->dyn_tree;
1332 ct_data *stree = desc->static_tree;
1333 int elems = desc->elems;
1334 int n, m; /* iterate over heap elements */
1335 int max_code = -1; /* largest code with non zero frequency */
1336 int node = elems; /* next internal node of the tree */
1338 /* Construct the initial heap, with least frequent element in
1339 * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1].
1340 * heap[0] is not used.
1343 G2.heap_max = HEAP_SIZE;
1345 for (n = 0; n < elems; n++) {
1346 if (tree[n].Freq != 0) {
1347 G2.heap[++G2.heap_len] = max_code = n;
1354 /* The pkzip format requires that at least one distance code exists,
1355 * and that at least one bit should be sent even if there is only one
1356 * possible code. So to avoid special checks later on we force at least
1357 * two codes of non zero frequency.
1359 while (G2.heap_len < 2) {
1360 int new = G2.heap[++G2.heap_len] = (max_code < 2 ? ++max_code : 0);
1366 G2.static_len -= stree[new].Len;
1367 /* new is 0 or 1 so it does not have extra bits */
1369 desc->max_code = max_code;
1371 /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree,
1372 * establish sub-heaps of increasing lengths:
1374 for (n = G2.heap_len / 2; n >= 1; n--)
1375 pqdownheap(tree, n);
1377 /* Construct the Huffman tree by repeatedly combining the least two
1381 PQREMOVE(tree, n); /* n = node of least frequency */
1382 m = G2.heap[SMALLEST]; /* m = node of next least frequency */
1384 G2.heap[--G2.heap_max] = n; /* keep the nodes sorted by frequency */
1385 G2.heap[--G2.heap_max] = m;
1387 /* Create a new node father of n and m */
1388 tree[node].Freq = tree[n].Freq + tree[m].Freq;
1389 G2.depth[node] = MAX(G2.depth[n], G2.depth[m]) + 1;
1390 tree[n].Dad = tree[m].Dad = (ush) node;
1392 if (tree == G2.bl_tree) {
1393 bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)",
1394 node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq);
1397 /* and insert the new node in the heap */
1398 G2.heap[SMALLEST] = node++;
1399 pqdownheap(tree, SMALLEST);
1400 } while (G2.heap_len >= 2);
1402 G2.heap[--G2.heap_max] = G2.heap[SMALLEST];
1404 /* At this point, the fields freq and dad are set. We can now
1405 * generate the bit lengths.
1407 gen_bitlen((tree_desc *) desc);
1409 /* The field len is now set, we can generate the bit codes */
1410 gen_codes((ct_data *) tree, max_code);
1414 /* ===========================================================================
1415 * Scan a literal or distance tree to determine the frequencies of the codes
1416 * in the bit length tree. Updates opt_len to take into account the repeat
1417 * counts. (The contribution of the bit length codes will be added later
1418 * during the construction of bl_tree.)
1420 static void scan_tree(ct_data * tree, int max_code)
1422 int n; /* iterates over all tree elements */
1423 int prevlen = -1; /* last emitted length */
1424 int curlen; /* length of current code */
1425 int nextlen = tree[0].Len; /* length of next code */
1426 int count = 0; /* repeat count of the current code */
1427 int max_count = 7; /* max repeat count */
1428 int min_count = 4; /* min repeat count */
1434 tree[max_code + 1].Len = 0xffff; /* guard */
1436 for (n = 0; n <= max_code; n++) {
1438 nextlen = tree[n + 1].Len;
1439 if (++count < max_count && curlen == nextlen)
1442 if (count < min_count) {
1443 G2.bl_tree[curlen].Freq += count;
1444 } else if (curlen != 0) {
1445 if (curlen != prevlen)
1446 G2.bl_tree[curlen].Freq++;
1447 G2.bl_tree[REP_3_6].Freq++;
1448 } else if (count <= 10) {
1449 G2.bl_tree[REPZ_3_10].Freq++;
1451 G2.bl_tree[REPZ_11_138].Freq++;
1461 } else if (curlen == nextlen) {
1469 /* ===========================================================================
1470 * Send a literal or distance tree in compressed form, using the codes in
1473 static void send_tree(ct_data * tree, int max_code)
1475 int n; /* iterates over all tree elements */
1476 int prevlen = -1; /* last emitted length */
1477 int curlen; /* length of current code */
1478 int nextlen = tree[0].Len; /* length of next code */
1479 int count = 0; /* repeat count of the current code */
1480 int max_count = 7; /* max repeat count */
1481 int min_count = 4; /* min repeat count */
1483 /* tree[max_code+1].Len = -1; *//* guard already set */
1485 max_count = 138, min_count = 3;
1487 for (n = 0; n <= max_code; n++) {
1489 nextlen = tree[n + 1].Len;
1490 if (++count < max_count && curlen == nextlen) {
1492 } else if (count < min_count) {
1494 SEND_CODE(curlen, G2.bl_tree);
1496 } else if (curlen != 0) {
1497 if (curlen != prevlen) {
1498 SEND_CODE(curlen, G2.bl_tree);
1501 Assert(count >= 3 && count <= 6, " 3_6?");
1502 SEND_CODE(REP_3_6, G2.bl_tree);
1503 send_bits(count - 3, 2);
1504 } else if (count <= 10) {
1505 SEND_CODE(REPZ_3_10, G2.bl_tree);
1506 send_bits(count - 3, 3);
1508 SEND_CODE(REPZ_11_138, G2.bl_tree);
1509 send_bits(count - 11, 7);
1516 } else if (curlen == nextlen) {
1527 /* ===========================================================================
1528 * Construct the Huffman tree for the bit lengths and return the index in
1529 * bl_order of the last bit length code to send.
1531 static int build_bl_tree(void)
1533 int max_blindex; /* index of last bit length code of non zero freq */
1535 /* Determine the bit length frequencies for literal and distance trees */
1536 scan_tree(G2.dyn_ltree, G2.l_desc.max_code);
1537 scan_tree(G2.dyn_dtree, G2.d_desc.max_code);
1539 /* Build the bit length tree: */
1540 build_tree(&G2.bl_desc);
1541 /* opt_len now includes the length of the tree representations, except
1542 * the lengths of the bit lengths codes and the 5+5+4 bits for the counts.
1545 /* Determine the number of bit length codes to send. The pkzip format
1546 * requires that at least 4 bit length codes be sent. (appnote.txt says
1547 * 3 but the actual value used is 4.)
1549 for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) {
1550 if (G2.bl_tree[bl_order[max_blindex]].Len != 0)
1553 /* Update opt_len to include the bit length tree and counts */
1554 G2.opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4;
1555 Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", (long)G2.opt_len, (long)G2.static_len));
1561 /* ===========================================================================
1562 * Send the header for a block using dynamic Huffman trees: the counts, the
1563 * lengths of the bit length codes, the literal tree and the distance tree.
1564 * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4.
1566 static void send_all_trees(int lcodes, int dcodes, int blcodes)
1568 int rank; /* index in bl_order */
1570 Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes");
1571 Assert(lcodes <= L_CODES && dcodes <= D_CODES
1572 && blcodes <= BL_CODES, "too many codes");
1573 Tracev((stderr, "\nbl counts: "));
1574 send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */
1575 send_bits(dcodes - 1, 5);
1576 send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */
1577 for (rank = 0; rank < blcodes; rank++) {
1578 Tracev((stderr, "\nbl code %2d ", bl_order[rank]));
1579 send_bits(G2.bl_tree[bl_order[rank]].Len, 3);
1581 Tracev((stderr, "\nbl tree: sent %ld", (long)G1.bits_sent));
1583 send_tree((ct_data *) G2.dyn_ltree, lcodes - 1); /* send the literal tree */
1584 Tracev((stderr, "\nlit tree: sent %ld", (long)G1.bits_sent));
1586 send_tree((ct_data *) G2.dyn_dtree, dcodes - 1); /* send the distance tree */
1587 Tracev((stderr, "\ndist tree: sent %ld", (long)G1.bits_sent));
1591 /* ===========================================================================
1592 * Save the match info and tally the frequency counts. Return true if
1593 * the current block must be flushed.
1595 static int ct_tally(int dist, int lc)
1597 G1.l_buf[G2.last_lit++] = lc;
1599 /* lc is the unmatched char */
1600 G2.dyn_ltree[lc].Freq++;
1602 /* Here, lc is the match length - MIN_MATCH */
1603 dist--; /* dist = match distance - 1 */
1604 Assert((ush) dist < (ush) MAX_DIST
1605 && (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH)
1606 && (ush) D_CODE(dist) < (ush) D_CODES, "ct_tally: bad match"
1609 G2.dyn_ltree[G2.length_code[lc] + LITERALS + 1].Freq++;
1610 G2.dyn_dtree[D_CODE(dist)].Freq++;
1612 G1.d_buf[G2.last_dist++] = dist;
1613 G2.flags |= G2.flag_bit;
1617 /* Output the flags if they fill a byte: */
1618 if ((G2.last_lit & 7) == 0) {
1619 G2.flag_buf[G2.last_flags++] = G2.flags;
1623 /* Try to guess if it is profitable to stop the current block here */
1624 if ((G2.last_lit & 0xfff) == 0) {
1625 /* Compute an upper bound for the compressed length */
1626 ulg out_length = G2.last_lit * 8L;
1627 ulg in_length = (ulg) G1.strstart - G1.block_start;
1630 for (dcode = 0; dcode < D_CODES; dcode++) {
1631 out_length += G2.dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]);
1635 "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ",
1636 G2.last_lit, G2.last_dist,
1637 (long)in_length, (long)out_length,
1638 100L - out_length * 100L / in_length));
1639 if (G2.last_dist < G2.last_lit / 2 && out_length < in_length / 2)
1642 return (G2.last_lit == LIT_BUFSIZE - 1 || G2.last_dist == DIST_BUFSIZE);
1643 /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K
1644 * on 16 bit machines and because stored blocks are restricted to
1649 /* ===========================================================================
1650 * Send the block data compressed using the given Huffman trees
1652 static void compress_block(ct_data * ltree, ct_data * dtree)
1654 unsigned dist; /* distance of matched string */
1655 int lc; /* match length or unmatched char (if dist == 0) */
1656 unsigned lx = 0; /* running index in l_buf */
1657 unsigned dx = 0; /* running index in d_buf */
1658 unsigned fx = 0; /* running index in flag_buf */
1659 uch flag = 0; /* current flags */
1660 unsigned code; /* the code to send */
1661 int extra; /* number of extra bits to send */
1663 if (G2.last_lit != 0) do {
1665 flag = G2.flag_buf[fx++];
1666 lc = G1.l_buf[lx++];
1667 if ((flag & 1) == 0) {
1668 SEND_CODE(lc, ltree); /* send a literal byte */
1669 Tracecv(lc > ' ', (stderr, " '%c' ", lc));
1671 /* Here, lc is the match length - MIN_MATCH */
1672 code = G2.length_code[lc];
1673 SEND_CODE(code + LITERALS + 1, ltree); /* send the length code */
1674 extra = extra_lbits[code];
1676 lc -= G2.base_length[code];
1677 send_bits(lc, extra); /* send the extra length bits */
1679 dist = G1.d_buf[dx++];
1680 /* Here, dist is the match distance - 1 */
1681 code = D_CODE(dist);
1682 Assert(code < D_CODES, "bad d_code");
1684 SEND_CODE(code, dtree); /* send the distance code */
1685 extra = extra_dbits[code];
1687 dist -= G2.base_dist[code];
1688 send_bits(dist, extra); /* send the extra distance bits */
1690 } /* literal or match pair ? */
1692 } while (lx < G2.last_lit);
1694 SEND_CODE(END_BLOCK, ltree);
1698 /* ===========================================================================
1699 * Determine the best encoding for the current block: dynamic trees, static
1700 * trees or store, and output the encoded block to the zip file. This function
1701 * returns the total compressed length for the file so far.
1703 static ulg flush_block(char *buf, ulg stored_len, int eof)
1705 ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
1706 int max_blindex; /* index of last bit length code of non zero freq */
1708 G2.flag_buf[G2.last_flags] = G2.flags; /* Save the flags for the last 8 items */
1710 /* Construct the literal and distance trees */
1711 build_tree(&G2.l_desc);
1712 Tracev((stderr, "\nlit data: dyn %ld, stat %ld", (long)G2.opt_len, (long)G2.static_len));
1714 build_tree(&G2.d_desc);
1715 Tracev((stderr, "\ndist data: dyn %ld, stat %ld", (long)G2.opt_len, (long)G2.static_len));
1716 /* At this point, opt_len and static_len are the total bit lengths of
1717 * the compressed block data, excluding the tree representations.
1720 /* Build the bit length tree for the above two trees, and get the index
1721 * in bl_order of the last bit length code to send.
1723 max_blindex = build_bl_tree();
1725 /* Determine the best encoding. Compute first the block length in bytes */
1726 opt_lenb = (G2.opt_len + 3 + 7) >> 3;
1727 static_lenb = (G2.static_len + 3 + 7) >> 3;
1730 "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ",
1731 (unsigned long)opt_lenb, (unsigned long)G2.opt_len,
1732 (unsigned long)static_lenb, (unsigned long)G2.static_len,
1733 (unsigned long)stored_len,
1734 G2.last_lit, G2.last_dist));
1736 if (static_lenb <= opt_lenb)
1737 opt_lenb = static_lenb;
1739 /* If compression failed and this is the first and last block,
1740 * and if the zip file can be seeked (to rewrite the local header),
1741 * the whole file is transformed into a stored file:
1743 if (stored_len <= opt_lenb && eof && G2.compressed_len == 0L && seekable()) {
1744 /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */
1746 bb_error_msg("block vanished");
1748 copy_block(buf, (unsigned) stored_len, 0); /* without header */
1749 G2.compressed_len = stored_len << 3;
1750 } else if (stored_len + 4 <= opt_lenb && buf != NULL) {
1751 /* 4: two words for the lengths */
1752 /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE.
1753 * Otherwise we can't have processed more than WSIZE input bytes since
1754 * the last block flush, because compression would have been
1755 * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to
1756 * transform a block into a stored block.
1758 send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */
1759 G2.compressed_len = (G2.compressed_len + 3 + 7) & ~7L;
1760 G2.compressed_len += (stored_len + 4) << 3;
1762 copy_block(buf, (unsigned) stored_len, 1); /* with header */
1763 } else if (static_lenb == opt_lenb) {
1764 send_bits((STATIC_TREES << 1) + eof, 3);
1765 compress_block((ct_data *) G2.static_ltree, (ct_data *) G2.static_dtree);
1766 G2.compressed_len += 3 + G2.static_len;
1768 send_bits((DYN_TREES << 1) + eof, 3);
1769 send_all_trees(G2.l_desc.max_code + 1, G2.d_desc.max_code + 1,
1771 compress_block((ct_data *) G2.dyn_ltree, (ct_data *) G2.dyn_dtree);
1772 G2.compressed_len += 3 + G2.opt_len;
1774 Assert(G2.compressed_len == G1.bits_sent, "bad compressed size");
1779 G2.compressed_len += 7; /* align on byte boundary */
1781 Tracev((stderr, "\ncomprlen %lu(%lu) ",
1782 (unsigned long)G2.compressed_len >> 3,
1783 (unsigned long)G2.compressed_len - 7 * eof));
1785 return G2.compressed_len >> 3;
1789 /* ===========================================================================
1790 * Update a hash value with the given input byte
1791 * IN assertion: all calls to UPDATE_HASH are made with consecutive
1792 * input characters, so that a running hash key can be computed from the
1793 * previous key instead of complete recalculation each time.
1795 #define UPDATE_HASH(h, c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
1798 /* ===========================================================================
1799 * Same as above, but achieves better compression. We use a lazy
1800 * evaluation for matches: a match is finally adopted only if there is
1801 * no better match at the next window position.
1803 * Processes a new input file and return its compressed length. Sets
1804 * the compressed length, crc, deflate flags and internal file
1808 /* Flush the current block, with given end-of-file flag.
1809 * IN assertion: strstart is set to the end of the current match. */
1810 #define FLUSH_BLOCK(eof) \
1812 G1.block_start >= 0L \
1813 ? (char*)&G1.window[(unsigned)G1.block_start] \
1815 (ulg)G1.strstart - G1.block_start, \
1819 /* Insert string s in the dictionary and set match_head to the previous head
1820 * of the hash chain (the most recent string with same hash key). Return
1821 * the previous length of the hash chain.
1822 * IN assertion: all calls to INSERT_STRING are made with consecutive
1823 * input characters and the first MIN_MATCH bytes of s are valid
1824 * (except for the last MIN_MATCH-1 bytes of the input file). */
1825 #define INSERT_STRING(s, match_head) \
1827 UPDATE_HASH(G1.ins_h, G1.window[(s) + MIN_MATCH-1]); \
1828 G1.prev[(s) & WMASK] = match_head = head[G1.ins_h]; \
1829 head[G1.ins_h] = (s); \
1832 static NOINLINE ulg deflate(void)
1834 IPos hash_head; /* head of hash chain */
1835 IPos prev_match; /* previous match */
1836 int flush; /* set if current block must be flushed */
1837 int match_available = 0; /* set if previous match exists */
1838 unsigned match_length = MIN_MATCH - 1; /* length of best match */
1840 /* Process the input block. */
1841 while (G1.lookahead != 0) {
1842 /* Insert the string window[strstart .. strstart+2] in the
1843 * dictionary, and set hash_head to the head of the hash chain:
1845 INSERT_STRING(G1.strstart, hash_head);
1847 /* Find the longest match, discarding those <= prev_length.
1849 G1.prev_length = match_length;
1850 prev_match = G1.match_start;
1851 match_length = MIN_MATCH - 1;
1853 if (hash_head != 0 && G1.prev_length < max_lazy_match
1854 && G1.strstart - hash_head <= MAX_DIST
1856 /* To simplify the code, we prevent matches with the string
1857 * of window index 0 (in particular we have to avoid a match
1858 * of the string with itself at the start of the input file).
1860 match_length = longest_match(hash_head);
1861 /* longest_match() sets match_start */
1862 if (match_length > G1.lookahead)
1863 match_length = G1.lookahead;
1865 /* Ignore a length 3 match if it is too distant: */
1866 if (match_length == MIN_MATCH && G1.strstart - G1.match_start > TOO_FAR) {
1867 /* If prev_match is also MIN_MATCH, G1.match_start is garbage
1868 * but we will ignore the current match anyway.
1873 /* If there was a match at the previous step and the current
1874 * match is not better, output the previous match:
1876 if (G1.prev_length >= MIN_MATCH && match_length <= G1.prev_length) {
1877 check_match(G1.strstart - 1, prev_match, G1.prev_length);
1878 flush = ct_tally(G1.strstart - 1 - prev_match, G1.prev_length - MIN_MATCH);
1880 /* Insert in hash table all strings up to the end of the match.
1881 * strstart-1 and strstart are already inserted.
1883 G1.lookahead -= G1.prev_length - 1;
1884 G1.prev_length -= 2;
1887 INSERT_STRING(G1.strstart, hash_head);
1888 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1889 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
1890 * these bytes are garbage, but it does not matter since the
1891 * next lookahead bytes will always be emitted as literals.
1893 } while (--G1.prev_length != 0);
1894 match_available = 0;
1895 match_length = MIN_MATCH - 1;
1899 G1.block_start = G1.strstart;
1901 } else if (match_available) {
1902 /* If there was no match at the previous position, output a
1903 * single literal. If there was a match but the current match
1904 * is longer, truncate the previous match to a single literal.
1906 Tracevv((stderr, "%c", G1.window[G1.strstart - 1]));
1907 if (ct_tally(0, G1.window[G1.strstart - 1])) {
1909 G1.block_start = G1.strstart;
1914 /* There is no previous match to compare with, wait for
1915 * the next step to decide.
1917 match_available = 1;
1921 Assert(G1.strstart <= G1.isize && G1.lookahead <= G1.isize, "a bit too far");
1923 /* Make sure that we always have enough lookahead, except
1924 * at the end of the input file. We need MAX_MATCH bytes
1925 * for the next match, plus MIN_MATCH bytes to insert the
1926 * string following the next match.
1928 fill_window_if_needed();
1930 if (match_available)
1931 ct_tally(0, G1.window[G1.strstart - 1]);
1933 return FLUSH_BLOCK(1); /* eof */
1937 /* ===========================================================================
1938 * Initialize the bit string routines.
1940 static void bi_init(void)
1942 //G1.bi_buf = 0; // globals are zeroed in pack_gzip()
1943 //G1.bi_valid = 0; // globals are zeroed in pack_gzip()
1945 //G1.bits_sent = 0L; // globals are zeroed in pack_gzip()
1950 /* ===========================================================================
1951 * Initialize the "longest match" routines for a new file
1953 static void lm_init(unsigned *flags16p)
1957 /* Initialize the hash table. */
1958 memset(head, 0, HASH_SIZE * sizeof(*head));
1959 /* prev will be initialized on the fly */
1961 /* speed options for the general purpose bit flag */
1962 *flags16p |= 2; /* FAST 4, SLOW 2 */
1963 /* ??? reduce max_chain_length for binary files */
1965 //G1.strstart = 0; // globals are zeroed in pack_gzip()
1966 //G1.block_start = 0L; // globals are zeroed in pack_gzip()
1968 G1.lookahead = file_read(G1.window,
1969 sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE);
1971 if (G1.lookahead == 0 || G1.lookahead == (unsigned) -1) {
1976 //G1.eofile = 0; // globals are zeroed in pack_gzip()
1978 /* Make sure that we always have enough lookahead. This is important
1979 * if input comes from a device such as a tty.
1981 fill_window_if_needed();
1983 //G1.ins_h = 0; // globals are zeroed in pack_gzip()
1984 for (j = 0; j < MIN_MATCH - 1; j++)
1985 UPDATE_HASH(G1.ins_h, G1.window[j]);
1986 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
1987 * not important since only literal bytes will be emitted.
1992 /* ===========================================================================
1993 * Allocate the match buffer, initialize the various tables and save the
1994 * location of the internal file attribute (ascii/binary) and method
1996 * One callsite in zip()
1998 static void ct_init(void)
2000 int n; /* iterates over tree elements */
2001 int length; /* length value */
2002 int code; /* code value */
2003 int dist; /* distance index */
2005 //G2.compressed_len = 0L; // globals are zeroed in pack_gzip()
2008 if (G2.static_dtree[0].Len != 0)
2009 return; /* ct_init already called */
2012 /* Initialize the mapping length (0..255) -> length code (0..28) */
2014 for (code = 0; code < LENGTH_CODES - 1; code++) {
2015 G2.base_length[code] = length;
2016 for (n = 0; n < (1 << extra_lbits[code]); n++) {
2017 G2.length_code[length++] = code;
2020 Assert(length == 256, "ct_init: length != 256");
2021 /* Note that the length 255 (match length 258) can be represented
2022 * in two different ways: code 284 + 5 bits or code 285, so we
2023 * overwrite length_code[255] to use the best encoding:
2025 G2.length_code[length - 1] = code;
2027 /* Initialize the mapping dist (0..32K) -> dist code (0..29) */
2029 for (code = 0; code < 16; code++) {
2030 G2.base_dist[code] = dist;
2031 for (n = 0; n < (1 << extra_dbits[code]); n++) {
2032 G2.dist_code[dist++] = code;
2035 Assert(dist == 256, "ct_init: dist != 256");
2036 dist >>= 7; /* from now on, all distances are divided by 128 */
2037 for (; code < D_CODES; code++) {
2038 G2.base_dist[code] = dist << 7;
2039 for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) {
2040 G2.dist_code[256 + dist++] = code;
2043 Assert(dist == 256, "ct_init: 256+dist != 512");
2045 /* Construct the codes of the static literal tree */
2046 //for (n = 0; n <= MAX_BITS; n++) // globals are zeroed in pack_gzip()
2047 // G2.bl_count[n] = 0;
2051 G2.static_ltree[n++].Len = 8;
2055 G2.static_ltree[n++].Len = 9;
2059 G2.static_ltree[n++].Len = 7;
2063 G2.static_ltree[n++].Len = 8;
2066 /* Codes 286 and 287 do not exist, but we must include them in the
2067 * tree construction to get a canonical Huffman tree (longest code
2070 gen_codes((ct_data *) G2.static_ltree, L_CODES + 1);
2072 /* The static distance tree is trivial: */
2073 for (n = 0; n < D_CODES; n++) {
2074 G2.static_dtree[n].Len = 5;
2075 G2.static_dtree[n].Code = bi_reverse(n, 5);
2078 /* Initialize the first block of the first file: */
2083 /* ===========================================================================
2084 * Deflate in to out.
2085 * IN assertions: the input and output buffers are cleared.
2087 static void zip(void)
2089 unsigned deflate_flags;
2091 //G1.outcnt = 0; // globals are zeroed in pack_gzip()
2093 /* Write the header to the gzip file. See algorithm.doc for the format */
2094 /* magic header for gzip files: 1F 8B */
2095 /* compression method: 8 (DEFLATED) */
2096 /* general flags: 0 */
2097 put_32bit(0x00088b1f);
2098 put_32bit(0); /* Unix timestamp */
2100 /* Write deflated file to zip file */
2105 deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */
2106 lm_init(&deflate_flags);
2108 put_16bit(deflate_flags | 0x300); /* extra flags. OS id = 3 (Unix) */
2110 #if OPTIMIZED_PUT_32BIT
2111 /* put_32bit() performs 32bit stores. If we use it in send_bits()... */
2113 /* then all stores are misaligned, unless we flush the buffer now */
2119 /* Write the crc and uncompressed size */
2121 put_32bit(G1.isize);
2127 /* ======================================================================== */
2129 IF_DESKTOP(long long) int FAST_FUNC pack_gzip(transformer_state_t *xstate UNUSED_PARAM)
2131 /* Reinit G1.xxx except pointers to allocated buffers */
2132 memset(&G1, 0, offsetof(struct globals, l_buf));
2134 /* Clear input and output buffers */
2142 memset(&G2, 0, sizeof(G2));
2143 G2.l_desc.dyn_tree = G2.dyn_ltree;
2144 G2.l_desc.static_tree = G2.static_ltree;
2145 G2.l_desc.extra_bits = extra_lbits;
2146 G2.l_desc.extra_base = LITERALS + 1;
2147 G2.l_desc.elems = L_CODES;
2148 G2.l_desc.max_length = MAX_BITS;
2149 //G2.l_desc.max_code = 0;
2150 G2.d_desc.dyn_tree = G2.dyn_dtree;
2151 G2.d_desc.static_tree = G2.static_dtree;
2152 G2.d_desc.extra_bits = extra_dbits;
2153 //G2.d_desc.extra_base = 0;
2154 G2.d_desc.elems = D_CODES;
2155 G2.d_desc.max_length = MAX_BITS;
2156 //G2.d_desc.max_code = 0;
2157 G2.bl_desc.dyn_tree = G2.bl_tree;
2158 //G2.bl_desc.static_tree = NULL;
2159 G2.bl_desc.extra_bits = extra_blbits,
2160 //G2.bl_desc.extra_base = 0;
2161 G2.bl_desc.elems = BL_CODES;
2162 G2.bl_desc.max_length = MAX_BL_BITS;
2163 //G2.bl_desc.max_code = 0;
2166 /* Saving of timestamp is disabled. Why?
2167 * - it is not Y2038-safe.
2168 * - some people want deterministic results
2169 * (normally they'd use -n, but our -n is a nop).
2171 * Per RFC 1952, gzfile.time=0 is "no timestamp".
2172 * If users will demand this to be reinstated,
2173 * implement -n "don't save timestamp".
2177 fstat(STDIN_FILENO, &s);
2185 #if ENABLE_FEATURE_GZIP_LONG_OPTIONS
2186 static const char gzip_longopts[] ALIGN1 =
2187 "stdout\0" No_argument "c"
2188 "to-stdout\0" No_argument "c"
2189 "force\0" No_argument "f"
2190 "verbose\0" No_argument "v"
2191 #if ENABLE_FEATURE_GZIP_DECOMPRESS
2192 "decompress\0" No_argument "d"
2193 "uncompress\0" No_argument "d"
2194 "test\0" No_argument "t"
2196 "quiet\0" No_argument "q"
2197 "fast\0" No_argument "1"
2198 "best\0" No_argument "9"
2199 "no-name\0" No_argument "n"
2204 * Linux kernel build uses gzip -d -n. We accept and ignore -n.
2207 * gzip: do not save the original file name and time stamp.
2208 * (The original name is always saved if the name had to be truncated.)
2209 * gunzip: do not restore the original file name/time even if present
2210 * (remove only the gzip suffix from the compressed file name).
2211 * This option is the default when decompressing.
2213 * gzip: always save the original file name and time stamp (this is the default)
2214 * gunzip: restore the original file name and time stamp if present.
2217 int gzip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
2218 #if ENABLE_FEATURE_GZIP_DECOMPRESS
2219 int gzip_main(int argc, char **argv)
2221 int gzip_main(int argc UNUSED_PARAM, char **argv)
2225 #if ENABLE_FEATURE_GZIP_LEVELS
2226 static const struct {
2228 uint8_t chain_shift;
2231 } gzip_level_config[6] = {
2232 {4, 4, 4/2, 16/2}, /* Level 4 */
2233 {8, 5, 16/2, 32/2}, /* Level 5 */
2234 {8, 7, 16/2, 128/2}, /* Level 6 */
2235 {8, 8, 32/2, 128/2}, /* Level 7 */
2236 {32, 10, 128/2, 258/2}, /* Level 8 */
2237 {32, 12, 258/2, 258/2}, /* Level 9 */
2241 SET_PTR_TO_GLOBALS((char *)xzalloc(sizeof(struct globals)+sizeof(struct globals2))
2242 + sizeof(struct globals));
2244 /* Must match bbunzip's constants OPT_STDOUT, OPT_FORCE! */
2245 #if ENABLE_FEATURE_GZIP_LONG_OPTIONS
2246 opt = getopt32long(argv, "cfkv" IF_FEATURE_GZIP_DECOMPRESS("dt") "qn123456789", gzip_longopts);
2248 opt = getopt32(argv, "cfkv" IF_FEATURE_GZIP_DECOMPRESS("dt") "qn123456789");
2250 #if ENABLE_FEATURE_GZIP_DECOMPRESS /* gunzip_main may not be visible... */
2251 if (opt & 0x30) // -d and/or -t
2252 return gunzip_main(argc, argv);
2254 #if ENABLE_FEATURE_GZIP_LEVELS
2255 opt >>= ENABLE_FEATURE_GZIP_DECOMPRESS ? 8 : 6; /* drop cfkv[dt]qn bits */
2257 opt = 1 << 6; /* default: 6 */
2258 opt = ffs(opt >> 4); /* Maps -1..-4 to [0], -5 to [1] ... -9 to [5] */
2259 max_chain_length = 1 << gzip_level_config[opt].chain_shift;
2260 good_match = gzip_level_config[opt].good;
2261 max_lazy_match = gzip_level_config[opt].lazy2 * 2;
2262 nice_match = gzip_level_config[opt].nice2 * 2;
2264 option_mask32 &= 0xf; /* retain only -cfkv */
2266 /* Allocate all global buffers (for DYN_ALLOC option) */
2267 ALLOC(uch, G1.l_buf, INBUFSIZ);
2268 ALLOC(uch, G1.outbuf, OUTBUFSIZ);
2269 ALLOC(ush, G1.d_buf, DIST_BUFSIZE);
2270 ALLOC(uch, G1.window, 2L * WSIZE);
2271 ALLOC(ush, G1.prev, 1L << BITS);
2273 /* Initialize the CRC32 table */
2274 global_crc32_table = crc32_filltable(NULL, 0);
2277 return bbunpack(argv, pack_gzip, append_ext, "gz");