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 /* big objects in bss:
20 * 00000074 b base_length
21 * 00000078 b base_dist
22 * 00000078 b static_dtree
24 * 000000f4 b dyn_dtree
25 * 00000100 b length_code
26 * 00000200 b dist_code
30 * 00000480 b static_ltree
31 * 000008f4 b dyn_ltree
33 /* TODO: full support for -v for DESKTOP
34 * "/usr/bin/gzip -v a bogus aa" should say:
35 a: 85.1% -- replaced with a.gz
36 gzip: bogus: No such file or directory
37 aa: 85.1% -- replaced with aa.gz
41 //config: bool "gzip (19 kb)"
44 //config: gzip is used to compress files.
45 //config: It's probably the most widely used UNIX compression program.
47 //config:config FEATURE_GZIP_LONG_OPTIONS
48 //config: bool "Enable long options"
50 //config: depends on GZIP && LONG_OPTS
52 //config:config GZIP_FAST
53 //config: int "Trade memory for speed (0:small,slow - 2:fast,big)"
56 //config: depends on GZIP
58 //config: Enable big memory options for gzip.
59 //config: 0: small buffers, small hash-tables
60 //config: 1: larger buffers, larger hash-tables
61 //config: 2: larger buffers, largest hash-tables
62 //config: Larger models may give slightly better compression
64 //config:config FEATURE_GZIP_LEVELS
65 //config: bool "Enable compression levels"
67 //config: depends on GZIP
69 //config: Enable support for compression levels 4-9. The default level
70 //config: is 6. If levels 1-3 are specified, 4 is used.
71 //config: If this option is not selected, -N options are ignored and -9
74 //config:config FEATURE_GZIP_DECOMPRESS
75 //config: bool "Enable decompression"
77 //config: depends on GZIP || GUNZIP || ZCAT
79 //config: Enable -d (--decompress) and -t (--test) options for gzip.
80 //config: This will be automatically selected if gunzip or zcat is
83 //applet:IF_GZIP(APPLET(gzip, BB_DIR_BIN, BB_SUID_DROP))
84 //kbuild:lib-$(CONFIG_GZIP) += gzip.o
86 //usage:#define gzip_trivial_usage
87 //usage: "[-cfk" IF_FEATURE_GZIP_DECOMPRESS("dt") IF_FEATURE_GZIP_LEVELS("123456789") "] [FILE]..."
88 //usage:#define gzip_full_usage "\n\n"
89 //usage: "Compress FILEs (or stdin)\n"
90 //usage: IF_FEATURE_GZIP_LEVELS(
91 //usage: "\n -1..9 Compression level"
93 //usage: IF_FEATURE_GZIP_DECOMPRESS(
94 //usage: "\n -d Decompress"
95 //usage: "\n -t Test file integrity"
97 //usage: "\n -c Write to stdout"
98 //usage: "\n -f Force"
99 //usage: "\n -k Keep input files"
101 //usage:#define gzip_example_usage
102 //usage: "$ ls -la /tmp/busybox*\n"
103 //usage: "-rw-rw-r-- 1 andersen andersen 1761280 Apr 14 17:47 /tmp/busybox.tar\n"
104 //usage: "$ gzip /tmp/busybox.tar\n"
105 //usage: "$ ls -la /tmp/busybox*\n"
106 //usage: "-rw-rw-r-- 1 andersen andersen 554058 Apr 14 17:49 /tmp/busybox.tar.gz\n"
109 #include "bb_archive.h"
112 /* ===========================================================================
115 /* Diagnostic functions */
117 # define Assert(cond,msg) { if (!(cond)) bb_error_msg(msg); }
118 # define Trace(x) fprintf x
119 # define Tracev(x) {if (verbose) fprintf x; }
120 # define Tracevv(x) {if (verbose > 1) fprintf x; }
121 # define Tracec(c,x) {if (verbose && (c)) fprintf x; }
122 # define Tracecv(c,x) {if (verbose > 1 && (c)) fprintf x; }
124 # define Assert(cond,msg)
129 # define Tracecv(c,x)
133 /* ===========================================================================
135 #if CONFIG_GZIP_FAST == 0
137 #elif CONFIG_GZIP_FAST == 1
139 #elif CONFIG_GZIP_FAST == 2
142 # error "Invalid CONFIG_GZIP_FAST value"
147 # define INBUFSIZ 0x2000 /* input buffer size */
149 # define INBUFSIZ 0x8000 /* input buffer size */
155 # define OUTBUFSIZ 8192 /* output buffer size */
157 # define OUTBUFSIZ 16384 /* output buffer size */
163 # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */
165 # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */
170 #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */
171 #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
172 #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
173 #define ORIG_NAME 0x08 /* bit 3 set: original file name present */
174 #define COMMENT 0x10 /* bit 4 set: file comment present */
175 #define RESERVED 0xC0 /* bit 6,7: reserved */
177 /* internal file attribute */
178 #define UNKNOWN 0xffff
183 # define WSIZE 0x8000 /* window size--must be a power of two, and */
184 #endif /* at least 32K for zip's deflate method */
187 #define MAX_MATCH 258
188 /* The minimum and maximum match lengths */
190 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
191 /* Minimum amount of lookahead, except at the end of the input file.
192 * See deflate.c for comments about the MIN_MATCH+1.
195 #define MAX_DIST (WSIZE-MIN_LOOKAHEAD)
196 /* In order to simplify the code, particularly on 16 bit machines, match
197 * distances are limited to MAX_DIST instead of WSIZE.
201 # define MAX_PATH_LEN 1024 /* max pathname length */
204 #define seekable() 0 /* force sequential output */
205 #define translate_eol 0 /* no option -a yet */
210 #define INIT_BITS 9 /* Initial number of bits per code */
212 #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */
213 /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free.
214 * It's a pity that old uncompress does not check bit 0x20. That makes
215 * extension of the format actually undesirable because old compress
216 * would just crash on the new format instead of giving a meaningful
217 * error message. It does check the number of bits, but it's more
218 * helpful to say "unsupported format, get a new version" than
219 * "can only handle 16 bits".
223 # define MAX_SUFFIX MAX_EXT_CHARS
225 # define MAX_SUFFIX 30
229 /* ===========================================================================
230 * Compile with MEDIUM_MEM to reduce the memory requirements or
231 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
232 * entire input file can be held in memory (not possible on 16 bit systems).
233 * Warning: defining these symbols affects HASH_BITS (see below) and thus
234 * affects the compression ratio. The compressed output
235 * is still correct, and might even be smaller in some cases.
239 # define HASH_BITS 13 /* Number of bits used to hash strings */
242 # define HASH_BITS 14
245 # define HASH_BITS 15
246 /* For portability to 16 bit machines, do not use values above 15. */
249 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
250 #define HASH_MASK (HASH_SIZE-1)
251 #define WMASK (WSIZE-1)
252 /* HASH_SIZE and WSIZE must be powers of two */
254 # define TOO_FAR 4096
256 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
259 /* ===========================================================================
260 * These types are not really 'char', 'short' and 'long'
263 typedef uint16_t ush;
264 typedef uint32_t ulg;
268 typedef unsigned IPos;
269 /* A Pos is an index in the character window. We use short instead of int to
270 * save space in the various tables. IPos is used only for parameter passing.
274 WINDOW_SIZE = 2 * WSIZE,
275 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
276 * input file length plus MIN_LOOKAHEAD.
279 #if !ENABLE_FEATURE_GZIP_LEVELS
281 max_chain_length = 4096,
282 /* To speed up deflation, hash chains are never searched beyond this length.
283 * A higher limit improves compression ratio but degrades the speed.
286 max_lazy_match = 258,
287 /* Attempt to find a better match only when the current match is strictly
288 * smaller than this value. This mechanism is used only for compression
292 max_insert_length = max_lazy_match,
293 /* Insert new strings in the hash table only if the match length
294 * is not greater than this length. This saves time but degrades compression.
295 * max_insert_length is used only for compression levels <= 3.
299 /* Use a faster search when the previous match is longer than this */
301 /* Values for max_lazy_match, good_match and max_chain_length, depending on
302 * the desired pack level (0..9). The values given below have been tuned to
303 * exclude worst case performance for pathological files. Better values may be
304 * found for specific files.
307 nice_match = 258, /* Stop searching when current match exceeds this */
308 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
309 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
312 #endif /* ENABLE_FEATURE_GZIP_LEVELS */
318 #if ENABLE_FEATURE_GZIP_LEVELS
319 unsigned max_chain_length;
320 unsigned max_lazy_match;
323 #define max_chain_length (G1.max_chain_length)
324 #define max_lazy_match (G1.max_lazy_match)
325 #define good_match (G1.good_match)
326 #define nice_match (G1.nice_match)
331 /* window position at the beginning of the current output block. Gets
332 * negative when the window is moved backwards.
334 unsigned ins_h; /* hash index of string to be inserted */
336 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1) / MIN_MATCH)
337 /* Number of bits by which ins_h and del_h must be shifted at each
338 * input step. It must be such that after MIN_MATCH steps, the oldest
339 * byte no longer takes part in the hash key, that is:
340 * H_SHIFT * MIN_MATCH >= HASH_BITS
343 unsigned prev_length;
345 /* Length of the best match at previous step. Matches not greater than this
346 * are discarded. This is used in the lazy match evaluation.
349 unsigned strstart; /* start of string to insert */
350 unsigned match_start; /* start of matching string */
351 unsigned lookahead; /* number of valid bytes ahead in window */
353 /* ===========================================================================
355 #define DECLARE(type, array, size) \
357 #define ALLOC(type, array, size) \
358 array = xzalloc((size_t)(((size)+1L)/2) * 2*sizeof(type))
359 #define FREE(array) \
360 do { free(array); array = NULL; } while (0)
364 /* buffer for literals or lengths */
365 /* DECLARE(uch, l_buf, LIT_BUFSIZE); */
366 DECLARE(uch, l_buf, INBUFSIZ);
368 DECLARE(ush, d_buf, DIST_BUFSIZE);
369 DECLARE(uch, outbuf, OUTBUFSIZ);
371 /* Sliding window. Input bytes are read into the second half of the window,
372 * and move to the first half later to keep a dictionary of at least WSIZE
373 * bytes. With this organization, matches are limited to a distance of
374 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
375 * performed with a length multiple of the block size. Also, it limits
376 * the window size to 64K, which is quite useful on MSDOS.
377 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
378 * be less efficient).
380 DECLARE(uch, window, 2L * WSIZE);
382 /* Link to older string with same hash index. To limit the size of this
383 * array to 64K, this link is maintained only for the last 32K strings.
384 * An index in this array is thus a window index modulo 32K.
386 /* DECLARE(Pos, prev, WSIZE); */
387 DECLARE(ush, prev, 1L << BITS);
389 /* Heads of the hash chains or 0. */
390 /* DECLARE(Pos, head, 1<<HASH_BITS); */
391 #define head (G1.prev + WSIZE) /* hash head (see deflate.c) */
393 /* number of input bytes */
394 ulg isize; /* only 32 bits stored in .gz file */
396 /* bbox always use stdin/stdout */
397 #define ifd STDIN_FILENO /* input file descriptor */
398 #define ofd STDOUT_FILENO /* output file descriptor */
401 unsigned insize; /* valid bytes in l_buf */
403 unsigned outcnt; /* bytes in output buffer */
405 smallint eofile; /* flag set at end of input file */
407 /* ===========================================================================
408 * Local data used by the "bit string" routines.
411 unsigned short bi_buf;
413 /* Output buffer. bits are inserted starting at the bottom (least significant
418 #define BUF_SIZE (8 * sizeof(G1.bi_buf))
419 /* Number of bits used within bi_buf. (bi_buf might be implemented on
420 * more than 16 bits on some systems.)
425 /* Current input function. Set to mem_read for in-memory compression */
428 ulg bits_sent; /* bit length of the compressed data */
431 /*uint32_t *crc_32_tab;*/
432 uint32_t crc; /* shift register contents */
435 #define G1 (*(ptr_to_globals - 1))
438 /* ===========================================================================
439 * Write the output buffer outbuf[0..outcnt-1] and update bytes_out.
440 * (used for the compressed data only)
442 static void flush_outbuf(void)
447 xwrite(ofd, (char *) G1.outbuf, G1.outcnt);
452 /* ===========================================================================
454 /* put_8bit is used for the compressed output */
455 #define put_8bit(c) \
457 G1.outbuf[G1.outcnt++] = (c); \
458 if (G1.outcnt == OUTBUFSIZ) \
462 /* Output a 16 bit value, lsb first */
463 static void put_16bit(ush w)
465 /* GCC 4.2.1 won't optimize out redundant loads of G1.outcnt
466 * (probably because of fear of aliasing with G1.outbuf[]
467 * stores), do it explicitly:
469 unsigned outcnt = G1.outcnt;
470 uch *dst = &G1.outbuf[outcnt];
472 #if BB_UNALIGNED_MEMACCESS_OK && BB_LITTLE_ENDIAN
473 if (outcnt < OUTBUFSIZ-2) {
475 ush *dst16 = (void*) dst;
476 *dst16 = w; /* unalinged LSB 16-bit store */
477 G1.outcnt = outcnt + 2;
485 if (outcnt < OUTBUFSIZ-2) {
488 G1.outcnt = outcnt + 2;
493 /* Slowpath: we will need to do flush_outbuf() */
494 G1.outcnt = ++outcnt;
495 if (outcnt == OUTBUFSIZ)
500 static void put_32bit(ulg n)
506 /* ===========================================================================
507 * Run a set of bytes through the crc shift register. If s is a NULL
508 * pointer, then initialize the crc shift register contents instead.
509 * Return the current crc in either case.
511 static void updcrc(uch * s, unsigned n)
513 G1.crc = crc32_block_endian0(G1.crc, s, n, global_crc32_table /*G1.crc_32_tab*/);
517 /* ===========================================================================
518 * Read a new buffer from the current input file, perform end-of-line
519 * translation, and update the crc and input file size.
520 * IN assertion: size >= 2 (for end-of-line translation)
522 static unsigned file_read(void *buf, unsigned size)
526 Assert(G1.insize == 0, "l_buf not empty");
528 len = safe_read(ifd, buf, size);
529 if (len == (unsigned)(-1) || len == 0)
538 /* ===========================================================================
539 * Send a value on a given number of bits.
540 * IN assertion: length <= 16 and value fits in length bits.
542 static void send_bits(int value, int length)
545 Tracev((stderr, " l %2d v %4x ", length, value));
546 Assert(length > 0 && length <= 15, "invalid length");
547 G1.bits_sent += length;
549 /* If not enough room in bi_buf, use (valid) bits from bi_buf and
550 * (16 - bi_valid) bits from value, leaving (width - (16-bi_valid))
551 * unused bits in value.
553 if (G1.bi_valid > (int) BUF_SIZE - length) {
554 G1.bi_buf |= (value << G1.bi_valid);
555 put_16bit(G1.bi_buf);
556 G1.bi_buf = (ush) value >> (BUF_SIZE - G1.bi_valid);
557 G1.bi_valid += length - BUF_SIZE;
559 G1.bi_buf |= value << G1.bi_valid;
560 G1.bi_valid += length;
565 /* ===========================================================================
566 * Reverse the first len bits of a code, using straightforward code (a faster
567 * method would use a table)
568 * IN assertion: 1 <= len <= 15
570 static unsigned bi_reverse(unsigned code, int len)
576 if (--len <= 0) return res;
583 /* ===========================================================================
584 * Write out any remaining bits in an incomplete byte.
586 static void bi_windup(void)
588 if (G1.bi_valid > 8) {
589 put_16bit(G1.bi_buf);
590 } else if (G1.bi_valid > 0) {
596 G1.bits_sent = (G1.bits_sent + 7) & ~7;
601 /* ===========================================================================
602 * Copy a stored block to the zip file, storing first the length and its
603 * one's complement if requested.
605 static void copy_block(char *buf, unsigned len, int header)
607 bi_windup(); /* align on byte boundary */
613 G1.bits_sent += 2 * 16;
617 G1.bits_sent += (ulg) len << 3;
625 /* ===========================================================================
626 * Fill the window when the lookahead becomes insufficient.
627 * Updates strstart and lookahead, and sets eofile if end of input file.
628 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
629 * OUT assertions: at least one byte has been read, or eofile is set;
630 * file reads are performed for at least two bytes (required for the
631 * translate_eol option).
633 static void fill_window(void)
636 unsigned more = WINDOW_SIZE - G1.lookahead - G1.strstart;
637 /* Amount of free space at the end of the window. */
639 /* If the window is almost full and there is insufficient lookahead,
640 * move the upper half to the lower one to make room in the upper half.
642 if (more == (unsigned) -1) {
643 /* Very unlikely, but possible on 16 bit machine if strstart == 0
644 * and lookahead == 1 (input done one byte at time)
647 } else if (G1.strstart >= WSIZE + MAX_DIST) {
648 /* By the IN assertion, the window is not empty so we can't confuse
649 * more == 0 with more == 64K on a 16 bit machine.
651 Assert(WINDOW_SIZE == 2 * WSIZE, "no sliding with BIG_MEM");
653 memcpy(G1.window, G1.window + WSIZE, WSIZE);
654 G1.match_start -= WSIZE;
655 G1.strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
657 G1.block_start -= WSIZE;
659 for (n = 0; n < HASH_SIZE; n++) {
661 head[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
663 for (n = 0; n < WSIZE; n++) {
665 G1.prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
666 /* If n is not on any hash chain, prev[n] is garbage but
667 * its value will never be used.
672 /* At this point, more >= 2 */
674 n = file_read(G1.window + G1.strstart + G1.lookahead, more);
675 if (n == 0 || n == (unsigned) -1) {
684 /* ===========================================================================
685 * Set match_start to the longest match starting at the given string and
686 * return its length. Matches shorter or equal to prev_length are discarded,
687 * in which case the result is equal to prev_length and match_start is
689 * IN assertions: cur_match is the head of the hash chain for the current
690 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
693 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
694 * match.s. The code is functionally equivalent, so you can use the C version
697 static int longest_match(IPos cur_match)
699 unsigned chain_length = max_chain_length; /* max hash chain length */
700 uch *scan = G1.window + G1.strstart; /* current string */
701 uch *match; /* matched string */
702 int len; /* length of current match */
703 int best_len = G1.prev_length; /* best match length so far */
704 IPos limit = G1.strstart > (IPos) MAX_DIST ? G1.strstart - (IPos) MAX_DIST : 0;
705 /* Stop when cur_match becomes <= limit. To simplify the code,
706 * we prevent matches with the string of window index 0.
709 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
710 * It is easy to get rid of this optimization if necessary.
712 #if HASH_BITS < 8 || MAX_MATCH != 258
713 # error Code too clever
715 uch *strend = G1.window + G1.strstart + MAX_MATCH;
716 uch scan_end1 = scan[best_len - 1];
717 uch scan_end = scan[best_len];
719 /* Do not waste too much time if we already have a good match: */
720 if (G1.prev_length >= good_match) {
723 Assert(G1.strstart <= WINDOW_SIZE - MIN_LOOKAHEAD, "insufficient lookahead");
726 Assert(cur_match < G1.strstart, "no future");
727 match = G1.window + cur_match;
729 /* Skip to next match if the match length cannot increase
730 * or if the match length is less than 2:
732 if (match[best_len] != scan_end
733 || match[best_len - 1] != scan_end1
734 || *match != *scan || *++match != scan[1]
739 /* The check at best_len-1 can be removed because it will be made
740 * again later. (This heuristic is not always a win.)
741 * It is not necessary to compare scan[2] and match[2] since they
742 * are always equal when the other bytes match, given that
743 * the hash keys are equal and that HASH_BITS >= 8.
747 /* We check for insufficient lookahead only every 8th comparison;
748 * the 256th check will be made at strstart+258.
751 } while (*++scan == *++match && *++scan == *++match &&
752 *++scan == *++match && *++scan == *++match &&
753 *++scan == *++match && *++scan == *++match &&
754 *++scan == *++match && *++scan == *++match && scan < strend);
756 len = MAX_MATCH - (int) (strend - scan);
757 scan = strend - MAX_MATCH;
759 if (len > best_len) {
760 G1.match_start = cur_match;
762 if (len >= nice_match)
764 scan_end1 = scan[best_len - 1];
765 scan_end = scan[best_len];
767 } while ((cur_match = G1.prev[cur_match & WMASK]) > limit
768 && --chain_length != 0);
775 /* ===========================================================================
776 * Check that the match at match_start is indeed a match.
778 static void check_match(IPos start, IPos match, int length)
780 /* check that the match is indeed a match */
781 if (memcmp(G1.window + match, G1.window + start, length) != 0) {
782 bb_error_msg(" start %d, match %d, length %d", start, match, length);
783 bb_error_msg("invalid match");
786 bb_error_msg("\\[%d,%d]", start - match, length);
788 bb_putchar_stderr(G1.window[start++]);
789 } while (--length != 0);
793 # define check_match(start, match, length) ((void)0)
797 /* trees.c -- output deflated data using Huffman coding
798 * Copyright (C) 1992-1993 Jean-loup Gailly
799 * This is free software; you can redistribute it and/or modify it under the
800 * terms of the GNU General Public License, see the file COPYING.
804 * Encode various sets of source values using variable-length
808 * The PKZIP "deflation" process uses several Huffman trees. The more
809 * common source values are represented by shorter bit sequences.
811 * Each code tree is stored in the ZIP file in a compressed form
812 * which is itself a Huffman encoding of the lengths of
813 * all the code strings (in ascending order by source values).
814 * The actual code strings are reconstructed from the lengths in
815 * the UNZIP process, as described in the "application note"
816 * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program.
820 * Data Compression: Techniques and Applications, pp. 53-55.
821 * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7.
824 * Data Compression: Methods and Theory, pp. 49-50.
825 * Computer Science Press, 1988. ISBN 0-7167-8156-5.
829 * Addison-Wesley, 1983. ISBN 0-201-06672-6.
833 * Allocate the match buffer, initialize the various tables [and save
834 * the location of the internal file attribute (ascii/binary) and
835 * method (DEFLATE/STORE) -- deleted in bbox]
837 * void ct_tally(int dist, int lc);
838 * Save the match info and tally the frequency counts.
840 * ulg flush_block(char *buf, ulg stored_len, int eof)
841 * Determine the best encoding for the current block: dynamic trees,
842 * static trees or store, and output the encoded block to the zip
843 * file. Returns the total compressed length for the file so far.
847 /* All codes must not exceed MAX_BITS bits */
849 #define MAX_BL_BITS 7
850 /* Bit length codes must not exceed MAX_BL_BITS bits */
852 #define LENGTH_CODES 29
853 /* number of length codes, not counting the special END_BLOCK code */
856 /* number of literal bytes 0..255 */
858 #define END_BLOCK 256
859 /* end of block literal code */
861 #define L_CODES (LITERALS+1+LENGTH_CODES)
862 /* number of Literal or Length codes, including the END_BLOCK code */
865 /* number of distance codes */
868 /* number of codes used to transfer the bit lengths */
870 /* extra bits for each length code */
871 static const uint8_t extra_lbits[LENGTH_CODES] ALIGN1 = {
872 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4,
876 /* extra bits for each distance code */
877 static const uint8_t extra_dbits[D_CODES] ALIGN1 = {
878 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,
879 10, 10, 11, 11, 12, 12, 13, 13
882 /* extra bits for each bit length code */
883 static const uint8_t extra_blbits[BL_CODES] ALIGN1 = {
884 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 };
886 /* number of codes at each bit length for an optimal tree */
887 static const uint8_t bl_order[BL_CODES] ALIGN1 = {
888 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
890 #define STORED_BLOCK 0
891 #define STATIC_TREES 1
893 /* The three kinds of block type */
897 # define LIT_BUFSIZE 0x2000
900 # define LIT_BUFSIZE 0x4000
902 # define LIT_BUFSIZE 0x8000
907 # define DIST_BUFSIZE LIT_BUFSIZE
909 /* Sizes of match buffers for literals/lengths and distances. There are
910 * 4 reasons for limiting LIT_BUFSIZE to 64K:
911 * - frequencies can be kept in 16 bit counters
912 * - if compression is not successful for the first block, all input data is
913 * still in the window so we can still emit a stored block even when input
914 * comes from standard input. (This can also be done for all blocks if
915 * LIT_BUFSIZE is not greater than 32K.)
916 * - if compression is not successful for a file smaller than 64K, we can
917 * even emit a stored file instead of a stored block (saving 5 bytes).
918 * - creating new Huffman trees less frequently may not provide fast
919 * adaptation to changes in the input data statistics. (Take for
920 * example a binary file with poorly compressible code followed by
921 * a highly compressible string table.) Smaller buffer sizes give
922 * fast adaptation but have of course the overhead of transmitting trees
924 * - I can't count above 4
925 * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save
926 * memory at the expense of compression). Some optimizations would be possible
927 * if we rely on DIST_BUFSIZE == LIT_BUFSIZE.
930 /* repeat previous bit length 3-6 times (2 bits of repeat count) */
932 /* repeat a zero length 3-10 times (3 bits of repeat count) */
933 #define REPZ_11_138 18
934 /* repeat a zero length 11-138 times (7 bits of repeat count) */
936 /* ===========================================================================
938 /* Data structure describing a single value and its code string. */
939 typedef struct ct_data {
941 ush freq; /* frequency count */
942 ush code; /* bit string */
945 ush dad; /* father node in Huffman tree */
946 ush len; /* length of bit string */
955 #define HEAP_SIZE (2*L_CODES + 1)
956 /* maximum heap size */
958 typedef struct tree_desc {
959 ct_data *dyn_tree; /* the dynamic tree */
960 ct_data *static_tree; /* corresponding static tree or NULL */
961 const uint8_t *extra_bits; /* extra bits for each code or NULL */
962 int extra_base; /* base index for extra_bits */
963 int elems; /* max number of elements in the tree */
964 int max_length; /* max bit length for the codes */
965 int max_code; /* largest code with non zero frequency */
970 ush heap[HEAP_SIZE]; /* heap used to build the Huffman trees */
971 int heap_len; /* number of elements in the heap */
972 int heap_max; /* element of largest frequency */
974 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
975 * The same heap array is used to build all trees.
978 ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */
979 ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */
981 ct_data static_ltree[L_CODES + 2];
983 /* The static literal tree. Since the bit lengths are imposed, there is no
984 * need for the L_CODES extra codes used during heap construction. However
985 * The codes 286 and 287 are needed to build a canonical tree (see ct_init
989 ct_data static_dtree[D_CODES];
991 /* The static distance tree. (Actually a trivial tree since all codes use
995 ct_data bl_tree[2 * BL_CODES + 1];
997 /* Huffman tree for the bit lengths */
1003 ush bl_count[MAX_BITS + 1];
1005 /* The lengths of the bit length codes are sent in order of decreasing
1006 * probability, to avoid transmitting the lengths for unused bit length codes.
1009 uch depth[2 * L_CODES + 1];
1011 /* Depth of each subtree used as tie breaker for trees of equal frequency */
1013 uch length_code[MAX_MATCH - MIN_MATCH + 1];
1015 /* length code for each normalized match length (0 == MIN_MATCH) */
1019 /* distance codes. The first 256 values correspond to the distances
1020 * 3 .. 258, the last 256 values correspond to the top 8 bits of
1021 * the 15 bit distances.
1024 int base_length[LENGTH_CODES];
1026 /* First normalized length for each code (0 = MIN_MATCH) */
1028 int base_dist[D_CODES];
1030 /* First normalized distance for each code (0 = distance of 1) */
1032 uch flag_buf[LIT_BUFSIZE / 8];
1034 /* flag_buf is a bit array distinguishing literals from lengths in
1035 * l_buf, thus indicating the presence or absence of a distance.
1038 unsigned last_lit; /* running index in l_buf */
1039 unsigned last_dist; /* running index in d_buf */
1040 unsigned last_flags; /* running index in flag_buf */
1041 uch flags; /* current flags not yet saved in flag_buf */
1042 uch flag_bit; /* current bit used in flags */
1044 /* bits are filled in flags starting at bit 0 (least significant).
1045 * Note: these flags are overkill in the current code since we don't
1046 * take advantage of DIST_BUFSIZE == LIT_BUFSIZE.
1049 ulg opt_len; /* bit length of current block with optimal trees */
1050 ulg static_len; /* bit length of current block with static trees */
1052 ulg compressed_len; /* total bit length of compressed file */
1055 #define G2ptr ((struct globals2*)(ptr_to_globals))
1059 /* ===========================================================================
1061 static void gen_codes(ct_data * tree, int max_code);
1062 static void build_tree(tree_desc * desc);
1063 static void scan_tree(ct_data * tree, int max_code);
1064 static void send_tree(ct_data * tree, int max_code);
1065 static int build_bl_tree(void);
1066 static void send_all_trees(int lcodes, int dcodes, int blcodes);
1067 static void compress_block(ct_data * ltree, ct_data * dtree);
1071 /* Send a code of the given tree. c and tree must not have side effects */
1072 # define SEND_CODE(c, tree) send_bits(tree[c].Code, tree[c].Len)
1074 # define SEND_CODE(c, tree) \
1076 if (verbose > 1) bb_error_msg("\ncd %3d ", (c)); \
1077 send_bits(tree[c].Code, tree[c].Len); \
1081 #define D_CODE(dist) \
1082 ((dist) < 256 ? G2.dist_code[dist] : G2.dist_code[256 + ((dist)>>7)])
1083 /* Mapping from a distance to a distance code. dist is the distance - 1 and
1084 * must not have side effects. dist_code[256] and dist_code[257] are never
1086 * The arguments must not have side effects.
1090 /* ===========================================================================
1091 * Initialize a new block.
1093 static void init_block(void)
1095 int n; /* iterates over tree elements */
1097 /* Initialize the trees. */
1098 for (n = 0; n < L_CODES; n++)
1099 G2.dyn_ltree[n].Freq = 0;
1100 for (n = 0; n < D_CODES; n++)
1101 G2.dyn_dtree[n].Freq = 0;
1102 for (n = 0; n < BL_CODES; n++)
1103 G2.bl_tree[n].Freq = 0;
1105 G2.dyn_ltree[END_BLOCK].Freq = 1;
1106 G2.opt_len = G2.static_len = 0;
1107 G2.last_lit = G2.last_dist = G2.last_flags = 0;
1113 /* ===========================================================================
1114 * Restore the heap property by moving down the tree starting at node k,
1115 * exchanging a node with the smallest of its two sons if necessary, stopping
1116 * when the heap property is re-established (each father smaller than its
1120 /* Compares to subtrees, using the tree depth as tie breaker when
1121 * the subtrees have equal frequency. This minimizes the worst case length. */
1122 #define SMALLER(tree, n, m) \
1123 (tree[n].Freq < tree[m].Freq \
1124 || (tree[n].Freq == tree[m].Freq && G2.depth[n] <= G2.depth[m]))
1126 static void pqdownheap(ct_data * tree, int k)
1129 int j = k << 1; /* left son of k */
1131 while (j <= G2.heap_len) {
1132 /* Set j to the smallest of the two sons: */
1133 if (j < G2.heap_len && SMALLER(tree, G2.heap[j + 1], G2.heap[j]))
1136 /* Exit if v is smaller than both sons */
1137 if (SMALLER(tree, v, G2.heap[j]))
1140 /* Exchange v with the smallest son */
1141 G2.heap[k] = G2.heap[j];
1144 /* And continue down the tree, setting j to the left son of k */
1151 /* ===========================================================================
1152 * Compute the optimal bit lengths for a tree and update the total bit length
1153 * for the current block.
1154 * IN assertion: the fields freq and dad are set, heap[heap_max] and
1155 * above are the tree nodes sorted by increasing frequency.
1156 * OUT assertions: the field len is set to the optimal bit length, the
1157 * array bl_count contains the frequencies for each bit length.
1158 * The length opt_len is updated; static_len is also updated if stree is
1161 static void gen_bitlen(tree_desc * desc)
1163 ct_data *tree = desc->dyn_tree;
1164 const uint8_t *extra = desc->extra_bits;
1165 int base = desc->extra_base;
1166 int max_code = desc->max_code;
1167 int max_length = desc->max_length;
1168 ct_data *stree = desc->static_tree;
1169 int h; /* heap index */
1170 int n, m; /* iterate over the tree elements */
1171 int bits; /* bit length */
1172 int xbits; /* extra bits */
1173 ush f; /* frequency */
1174 int overflow = 0; /* number of elements with bit length too large */
1176 for (bits = 0; bits <= MAX_BITS; bits++)
1177 G2.bl_count[bits] = 0;
1179 /* In a first pass, compute the optimal bit lengths (which may
1180 * overflow in the case of the bit length tree).
1182 tree[G2.heap[G2.heap_max]].Len = 0; /* root of the heap */
1184 for (h = G2.heap_max + 1; h < HEAP_SIZE; h++) {
1186 bits = tree[tree[n].Dad].Len + 1;
1187 if (bits > max_length) {
1191 tree[n].Len = (ush) bits;
1192 /* We overwrite tree[n].Dad which is no longer needed */
1195 continue; /* not a leaf node */
1197 G2.bl_count[bits]++;
1200 xbits = extra[n - base];
1202 G2.opt_len += (ulg) f *(bits + xbits);
1205 G2.static_len += (ulg) f * (stree[n].Len + xbits);
1210 Trace((stderr, "\nbit length overflow\n"));
1211 /* This happens for example on obj2 and pic of the Calgary corpus */
1213 /* Find the first bit length which could increase: */
1215 bits = max_length - 1;
1216 while (G2.bl_count[bits] == 0)
1218 G2.bl_count[bits]--; /* move one leaf down the tree */
1219 G2.bl_count[bits + 1] += 2; /* move one overflow item as its brother */
1220 G2.bl_count[max_length]--;
1221 /* The brother of the overflow item also moves one step up,
1222 * but this does not affect bl_count[max_length]
1225 } while (overflow > 0);
1227 /* Now recompute all bit lengths, scanning in increasing frequency.
1228 * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all
1229 * lengths instead of fixing only the wrong ones. This idea is taken
1230 * from 'ar' written by Haruhiko Okumura.)
1232 for (bits = max_length; bits != 0; bits--) {
1233 n = G2.bl_count[bits];
1238 if (tree[m].Len != (unsigned) bits) {
1239 Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len, bits));
1240 G2.opt_len += ((int32_t) bits - tree[m].Len) * tree[m].Freq;
1249 /* ===========================================================================
1250 * Generate the codes for a given tree and bit counts (which need not be
1252 * IN assertion: the array bl_count contains the bit length statistics for
1253 * the given tree and the field len is set for all tree elements.
1254 * OUT assertion: the field code is set for all tree elements of non
1257 static void gen_codes(ct_data * tree, int max_code)
1259 ush next_code[MAX_BITS + 1]; /* next code value for each bit length */
1260 ush code = 0; /* running code value */
1261 int bits; /* bit index */
1262 int n; /* code index */
1264 /* The distribution counts are first used to generate the code values
1265 * without bit reversal.
1267 for (bits = 1; bits <= MAX_BITS; bits++) {
1268 next_code[bits] = code = (code + G2.bl_count[bits - 1]) << 1;
1270 /* Check that the bit counts in bl_count are consistent. The last code
1273 Assert(code + G2.bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1,
1274 "inconsistent bit counts");
1275 Tracev((stderr, "\ngen_codes: max_code %d ", max_code));
1277 for (n = 0; n <= max_code; n++) {
1278 int len = tree[n].Len;
1282 /* Now reverse the bits */
1283 tree[n].Code = bi_reverse(next_code[len]++, len);
1285 Tracec(tree != G2.static_ltree,
1286 (stderr, "\nn %3d %c l %2d c %4x (%x) ", n,
1287 (n > ' ' ? n : ' '), len, tree[n].Code,
1288 next_code[len] - 1));
1293 /* ===========================================================================
1294 * Construct one Huffman tree and assigns the code bit strings and lengths.
1295 * Update the total bit length for the current block.
1296 * IN assertion: the field freq is set for all tree elements.
1297 * OUT assertions: the fields len and code are set to the optimal bit length
1298 * and corresponding code. The length opt_len is updated; static_len is
1299 * also updated if stree is not null. The field max_code is set.
1302 /* Remove the smallest element from the heap and recreate the heap with
1303 * one less element. Updates heap and heap_len. */
1306 /* Index within the heap array of least frequent node in the Huffman tree */
1308 #define PQREMOVE(tree, top) \
1310 top = G2.heap[SMALLEST]; \
1311 G2.heap[SMALLEST] = G2.heap[G2.heap_len--]; \
1312 pqdownheap(tree, SMALLEST); \
1315 static void build_tree(tree_desc * desc)
1317 ct_data *tree = desc->dyn_tree;
1318 ct_data *stree = desc->static_tree;
1319 int elems = desc->elems;
1320 int n, m; /* iterate over heap elements */
1321 int max_code = -1; /* largest code with non zero frequency */
1322 int node = elems; /* next internal node of the tree */
1324 /* Construct the initial heap, with least frequent element in
1325 * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1].
1326 * heap[0] is not used.
1329 G2.heap_max = HEAP_SIZE;
1331 for (n = 0; n < elems; n++) {
1332 if (tree[n].Freq != 0) {
1333 G2.heap[++G2.heap_len] = max_code = n;
1340 /* The pkzip format requires that at least one distance code exists,
1341 * and that at least one bit should be sent even if there is only one
1342 * possible code. So to avoid special checks later on we force at least
1343 * two codes of non zero frequency.
1345 while (G2.heap_len < 2) {
1346 int new = G2.heap[++G2.heap_len] = (max_code < 2 ? ++max_code : 0);
1352 G2.static_len -= stree[new].Len;
1353 /* new is 0 or 1 so it does not have extra bits */
1355 desc->max_code = max_code;
1357 /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree,
1358 * establish sub-heaps of increasing lengths:
1360 for (n = G2.heap_len / 2; n >= 1; n--)
1361 pqdownheap(tree, n);
1363 /* Construct the Huffman tree by repeatedly combining the least two
1367 PQREMOVE(tree, n); /* n = node of least frequency */
1368 m = G2.heap[SMALLEST]; /* m = node of next least frequency */
1370 G2.heap[--G2.heap_max] = n; /* keep the nodes sorted by frequency */
1371 G2.heap[--G2.heap_max] = m;
1373 /* Create a new node father of n and m */
1374 tree[node].Freq = tree[n].Freq + tree[m].Freq;
1375 G2.depth[node] = MAX(G2.depth[n], G2.depth[m]) + 1;
1376 tree[n].Dad = tree[m].Dad = (ush) node;
1378 if (tree == G2.bl_tree) {
1379 bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)",
1380 node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq);
1383 /* and insert the new node in the heap */
1384 G2.heap[SMALLEST] = node++;
1385 pqdownheap(tree, SMALLEST);
1386 } while (G2.heap_len >= 2);
1388 G2.heap[--G2.heap_max] = G2.heap[SMALLEST];
1390 /* At this point, the fields freq and dad are set. We can now
1391 * generate the bit lengths.
1393 gen_bitlen((tree_desc *) desc);
1395 /* The field len is now set, we can generate the bit codes */
1396 gen_codes((ct_data *) tree, max_code);
1400 /* ===========================================================================
1401 * Scan a literal or distance tree to determine the frequencies of the codes
1402 * in the bit length tree. Updates opt_len to take into account the repeat
1403 * counts. (The contribution of the bit length codes will be added later
1404 * during the construction of bl_tree.)
1406 static void scan_tree(ct_data * tree, int max_code)
1408 int n; /* iterates over all tree elements */
1409 int prevlen = -1; /* last emitted length */
1410 int curlen; /* length of current code */
1411 int nextlen = tree[0].Len; /* length of next code */
1412 int count = 0; /* repeat count of the current code */
1413 int max_count = 7; /* max repeat count */
1414 int min_count = 4; /* min repeat count */
1420 tree[max_code + 1].Len = 0xffff; /* guard */
1422 for (n = 0; n <= max_code; n++) {
1424 nextlen = tree[n + 1].Len;
1425 if (++count < max_count && curlen == nextlen)
1428 if (count < min_count) {
1429 G2.bl_tree[curlen].Freq += count;
1430 } else if (curlen != 0) {
1431 if (curlen != prevlen)
1432 G2.bl_tree[curlen].Freq++;
1433 G2.bl_tree[REP_3_6].Freq++;
1434 } else if (count <= 10) {
1435 G2.bl_tree[REPZ_3_10].Freq++;
1437 G2.bl_tree[REPZ_11_138].Freq++;
1447 } else if (curlen == nextlen) {
1455 /* ===========================================================================
1456 * Send a literal or distance tree in compressed form, using the codes in
1459 static void send_tree(ct_data * tree, int max_code)
1461 int n; /* iterates over all tree elements */
1462 int prevlen = -1; /* last emitted length */
1463 int curlen; /* length of current code */
1464 int nextlen = tree[0].Len; /* length of next code */
1465 int count = 0; /* repeat count of the current code */
1466 int max_count = 7; /* max repeat count */
1467 int min_count = 4; /* min repeat count */
1469 /* tree[max_code+1].Len = -1; *//* guard already set */
1471 max_count = 138, min_count = 3;
1473 for (n = 0; n <= max_code; n++) {
1475 nextlen = tree[n + 1].Len;
1476 if (++count < max_count && curlen == nextlen) {
1478 } else if (count < min_count) {
1480 SEND_CODE(curlen, G2.bl_tree);
1482 } else if (curlen != 0) {
1483 if (curlen != prevlen) {
1484 SEND_CODE(curlen, G2.bl_tree);
1487 Assert(count >= 3 && count <= 6, " 3_6?");
1488 SEND_CODE(REP_3_6, G2.bl_tree);
1489 send_bits(count - 3, 2);
1490 } else if (count <= 10) {
1491 SEND_CODE(REPZ_3_10, G2.bl_tree);
1492 send_bits(count - 3, 3);
1494 SEND_CODE(REPZ_11_138, G2.bl_tree);
1495 send_bits(count - 11, 7);
1502 } else if (curlen == nextlen) {
1513 /* ===========================================================================
1514 * Construct the Huffman tree for the bit lengths and return the index in
1515 * bl_order of the last bit length code to send.
1517 static int build_bl_tree(void)
1519 int max_blindex; /* index of last bit length code of non zero freq */
1521 /* Determine the bit length frequencies for literal and distance trees */
1522 scan_tree(G2.dyn_ltree, G2.l_desc.max_code);
1523 scan_tree(G2.dyn_dtree, G2.d_desc.max_code);
1525 /* Build the bit length tree: */
1526 build_tree(&G2.bl_desc);
1527 /* opt_len now includes the length of the tree representations, except
1528 * the lengths of the bit lengths codes and the 5+5+4 bits for the counts.
1531 /* Determine the number of bit length codes to send. The pkzip format
1532 * requires that at least 4 bit length codes be sent. (appnote.txt says
1533 * 3 but the actual value used is 4.)
1535 for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) {
1536 if (G2.bl_tree[bl_order[max_blindex]].Len != 0)
1539 /* Update opt_len to include the bit length tree and counts */
1540 G2.opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4;
1541 Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", G2.opt_len, G2.static_len));
1547 /* ===========================================================================
1548 * Send the header for a block using dynamic Huffman trees: the counts, the
1549 * lengths of the bit length codes, the literal tree and the distance tree.
1550 * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4.
1552 static void send_all_trees(int lcodes, int dcodes, int blcodes)
1554 int rank; /* index in bl_order */
1556 Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes");
1557 Assert(lcodes <= L_CODES && dcodes <= D_CODES
1558 && blcodes <= BL_CODES, "too many codes");
1559 Tracev((stderr, "\nbl counts: "));
1560 send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */
1561 send_bits(dcodes - 1, 5);
1562 send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */
1563 for (rank = 0; rank < blcodes; rank++) {
1564 Tracev((stderr, "\nbl code %2d ", bl_order[rank]));
1565 send_bits(G2.bl_tree[bl_order[rank]].Len, 3);
1567 Tracev((stderr, "\nbl tree: sent %ld", G1.bits_sent));
1569 send_tree((ct_data *) G2.dyn_ltree, lcodes - 1); /* send the literal tree */
1570 Tracev((stderr, "\nlit tree: sent %ld", G1.bits_sent));
1572 send_tree((ct_data *) G2.dyn_dtree, dcodes - 1); /* send the distance tree */
1573 Tracev((stderr, "\ndist tree: sent %ld", G1.bits_sent));
1577 /* ===========================================================================
1578 * Save the match info and tally the frequency counts. Return true if
1579 * the current block must be flushed.
1581 static int ct_tally(int dist, int lc)
1583 G1.l_buf[G2.last_lit++] = lc;
1585 /* lc is the unmatched char */
1586 G2.dyn_ltree[lc].Freq++;
1588 /* Here, lc is the match length - MIN_MATCH */
1589 dist--; /* dist = match distance - 1 */
1590 Assert((ush) dist < (ush) MAX_DIST
1591 && (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH)
1592 && (ush) D_CODE(dist) < (ush) D_CODES, "ct_tally: bad match"
1595 G2.dyn_ltree[G2.length_code[lc] + LITERALS + 1].Freq++;
1596 G2.dyn_dtree[D_CODE(dist)].Freq++;
1598 G1.d_buf[G2.last_dist++] = dist;
1599 G2.flags |= G2.flag_bit;
1603 /* Output the flags if they fill a byte: */
1604 if ((G2.last_lit & 7) == 0) {
1605 G2.flag_buf[G2.last_flags++] = G2.flags;
1609 /* Try to guess if it is profitable to stop the current block here */
1610 if ((G2.last_lit & 0xfff) == 0) {
1611 /* Compute an upper bound for the compressed length */
1612 ulg out_length = G2.last_lit * 8L;
1613 ulg in_length = (ulg) G1.strstart - G1.block_start;
1616 for (dcode = 0; dcode < D_CODES; dcode++) {
1617 out_length += G2.dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]);
1621 "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ",
1622 G2.last_lit, G2.last_dist, in_length, out_length,
1623 100L - out_length * 100L / in_length));
1624 if (G2.last_dist < G2.last_lit / 2 && out_length < in_length / 2)
1627 return (G2.last_lit == LIT_BUFSIZE - 1 || G2.last_dist == DIST_BUFSIZE);
1628 /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K
1629 * on 16 bit machines and because stored blocks are restricted to
1634 /* ===========================================================================
1635 * Send the block data compressed using the given Huffman trees
1637 static void compress_block(ct_data * ltree, ct_data * dtree)
1639 unsigned dist; /* distance of matched string */
1640 int lc; /* match length or unmatched char (if dist == 0) */
1641 unsigned lx = 0; /* running index in l_buf */
1642 unsigned dx = 0; /* running index in d_buf */
1643 unsigned fx = 0; /* running index in flag_buf */
1644 uch flag = 0; /* current flags */
1645 unsigned code; /* the code to send */
1646 int extra; /* number of extra bits to send */
1648 if (G2.last_lit != 0) do {
1650 flag = G2.flag_buf[fx++];
1651 lc = G1.l_buf[lx++];
1652 if ((flag & 1) == 0) {
1653 SEND_CODE(lc, ltree); /* send a literal byte */
1654 Tracecv(lc > ' ', (stderr, " '%c' ", lc));
1656 /* Here, lc is the match length - MIN_MATCH */
1657 code = G2.length_code[lc];
1658 SEND_CODE(code + LITERALS + 1, ltree); /* send the length code */
1659 extra = extra_lbits[code];
1661 lc -= G2.base_length[code];
1662 send_bits(lc, extra); /* send the extra length bits */
1664 dist = G1.d_buf[dx++];
1665 /* Here, dist is the match distance - 1 */
1666 code = D_CODE(dist);
1667 Assert(code < D_CODES, "bad d_code");
1669 SEND_CODE(code, dtree); /* send the distance code */
1670 extra = extra_dbits[code];
1672 dist -= G2.base_dist[code];
1673 send_bits(dist, extra); /* send the extra distance bits */
1675 } /* literal or match pair ? */
1677 } while (lx < G2.last_lit);
1679 SEND_CODE(END_BLOCK, ltree);
1683 /* ===========================================================================
1684 * Determine the best encoding for the current block: dynamic trees, static
1685 * trees or store, and output the encoded block to the zip file. This function
1686 * returns the total compressed length for the file so far.
1688 static ulg flush_block(char *buf, ulg stored_len, int eof)
1690 ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
1691 int max_blindex; /* index of last bit length code of non zero freq */
1693 G2.flag_buf[G2.last_flags] = G2.flags; /* Save the flags for the last 8 items */
1695 /* Construct the literal and distance trees */
1696 build_tree(&G2.l_desc);
1697 Tracev((stderr, "\nlit data: dyn %ld, stat %ld", G2.opt_len, G2.static_len));
1699 build_tree(&G2.d_desc);
1700 Tracev((stderr, "\ndist data: dyn %ld, stat %ld", G2.opt_len, G2.static_len));
1701 /* At this point, opt_len and static_len are the total bit lengths of
1702 * the compressed block data, excluding the tree representations.
1705 /* Build the bit length tree for the above two trees, and get the index
1706 * in bl_order of the last bit length code to send.
1708 max_blindex = build_bl_tree();
1710 /* Determine the best encoding. Compute first the block length in bytes */
1711 opt_lenb = (G2.opt_len + 3 + 7) >> 3;
1712 static_lenb = (G2.static_len + 3 + 7) >> 3;
1715 "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ",
1716 opt_lenb, G2.opt_len, static_lenb, G2.static_len, stored_len,
1717 G2.last_lit, G2.last_dist));
1719 if (static_lenb <= opt_lenb)
1720 opt_lenb = static_lenb;
1722 /* If compression failed and this is the first and last block,
1723 * and if the zip file can be seeked (to rewrite the local header),
1724 * the whole file is transformed into a stored file:
1726 if (stored_len <= opt_lenb && eof && G2.compressed_len == 0L && seekable()) {
1727 /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */
1729 bb_error_msg("block vanished");
1731 copy_block(buf, (unsigned) stored_len, 0); /* without header */
1732 G2.compressed_len = stored_len << 3;
1733 } else if (stored_len + 4 <= opt_lenb && buf != NULL) {
1734 /* 4: two words for the lengths */
1735 /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE.
1736 * Otherwise we can't have processed more than WSIZE input bytes since
1737 * the last block flush, because compression would have been
1738 * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to
1739 * transform a block into a stored block.
1741 send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */
1742 G2.compressed_len = (G2.compressed_len + 3 + 7) & ~7L;
1743 G2.compressed_len += (stored_len + 4) << 3;
1745 copy_block(buf, (unsigned) stored_len, 1); /* with header */
1746 } else if (static_lenb == opt_lenb) {
1747 send_bits((STATIC_TREES << 1) + eof, 3);
1748 compress_block((ct_data *) G2.static_ltree, (ct_data *) G2.static_dtree);
1749 G2.compressed_len += 3 + G2.static_len;
1751 send_bits((DYN_TREES << 1) + eof, 3);
1752 send_all_trees(G2.l_desc.max_code + 1, G2.d_desc.max_code + 1,
1754 compress_block((ct_data *) G2.dyn_ltree, (ct_data *) G2.dyn_dtree);
1755 G2.compressed_len += 3 + G2.opt_len;
1757 Assert(G2.compressed_len == G1.bits_sent, "bad compressed size");
1762 G2.compressed_len += 7; /* align on byte boundary */
1764 Tracev((stderr, "\ncomprlen %lu(%lu) ", G2.compressed_len >> 3,
1765 G2.compressed_len - 7 * eof));
1767 return G2.compressed_len >> 3;
1771 /* ===========================================================================
1772 * Update a hash value with the given input byte
1773 * IN assertion: all calls to UPDATE_HASH are made with consecutive
1774 * input characters, so that a running hash key can be computed from the
1775 * previous key instead of complete recalculation each time.
1777 #define UPDATE_HASH(h, c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
1780 /* ===========================================================================
1781 * Same as above, but achieves better compression. We use a lazy
1782 * evaluation for matches: a match is finally adopted only if there is
1783 * no better match at the next window position.
1785 * Processes a new input file and return its compressed length. Sets
1786 * the compressed length, crc, deflate flags and internal file
1790 /* Flush the current block, with given end-of-file flag.
1791 * IN assertion: strstart is set to the end of the current match. */
1792 #define FLUSH_BLOCK(eof) \
1794 G1.block_start >= 0L \
1795 ? (char*)&G1.window[(unsigned)G1.block_start] \
1797 (ulg)G1.strstart - G1.block_start, \
1801 /* Insert string s in the dictionary and set match_head to the previous head
1802 * of the hash chain (the most recent string with same hash key). Return
1803 * the previous length of the hash chain.
1804 * IN assertion: all calls to INSERT_STRING are made with consecutive
1805 * input characters and the first MIN_MATCH bytes of s are valid
1806 * (except for the last MIN_MATCH-1 bytes of the input file). */
1807 #define INSERT_STRING(s, match_head) \
1809 UPDATE_HASH(G1.ins_h, G1.window[(s) + MIN_MATCH-1]); \
1810 G1.prev[(s) & WMASK] = match_head = head[G1.ins_h]; \
1811 head[G1.ins_h] = (s); \
1814 static ulg deflate(void)
1816 IPos hash_head; /* head of hash chain */
1817 IPos prev_match; /* previous match */
1818 int flush; /* set if current block must be flushed */
1819 int match_available = 0; /* set if previous match exists */
1820 unsigned match_length = MIN_MATCH - 1; /* length of best match */
1822 /* Process the input block. */
1823 while (G1.lookahead != 0) {
1824 /* Insert the string window[strstart .. strstart+2] in the
1825 * dictionary, and set hash_head to the head of the hash chain:
1827 INSERT_STRING(G1.strstart, hash_head);
1829 /* Find the longest match, discarding those <= prev_length.
1831 G1.prev_length = match_length;
1832 prev_match = G1.match_start;
1833 match_length = MIN_MATCH - 1;
1835 if (hash_head != 0 && G1.prev_length < max_lazy_match
1836 && G1.strstart - hash_head <= MAX_DIST
1838 /* To simplify the code, we prevent matches with the string
1839 * of window index 0 (in particular we have to avoid a match
1840 * of the string with itself at the start of the input file).
1842 match_length = longest_match(hash_head);
1843 /* longest_match() sets match_start */
1844 if (match_length > G1.lookahead)
1845 match_length = G1.lookahead;
1847 /* Ignore a length 3 match if it is too distant: */
1848 if (match_length == MIN_MATCH && G1.strstart - G1.match_start > TOO_FAR) {
1849 /* If prev_match is also MIN_MATCH, G1.match_start is garbage
1850 * but we will ignore the current match anyway.
1855 /* If there was a match at the previous step and the current
1856 * match is not better, output the previous match:
1858 if (G1.prev_length >= MIN_MATCH && match_length <= G1.prev_length) {
1859 check_match(G1.strstart - 1, prev_match, G1.prev_length);
1860 flush = ct_tally(G1.strstart - 1 - prev_match, G1.prev_length - MIN_MATCH);
1862 /* Insert in hash table all strings up to the end of the match.
1863 * strstart-1 and strstart are already inserted.
1865 G1.lookahead -= G1.prev_length - 1;
1866 G1.prev_length -= 2;
1869 INSERT_STRING(G1.strstart, hash_head);
1870 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1871 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
1872 * these bytes are garbage, but it does not matter since the
1873 * next lookahead bytes will always be emitted as literals.
1875 } while (--G1.prev_length != 0);
1876 match_available = 0;
1877 match_length = MIN_MATCH - 1;
1881 G1.block_start = G1.strstart;
1883 } else if (match_available) {
1884 /* If there was no match at the previous position, output a
1885 * single literal. If there was a match but the current match
1886 * is longer, truncate the previous match to a single literal.
1888 Tracevv((stderr, "%c", G1.window[G1.strstart - 1]));
1889 if (ct_tally(0, G1.window[G1.strstart - 1])) {
1891 G1.block_start = G1.strstart;
1896 /* There is no previous match to compare with, wait for
1897 * the next step to decide.
1899 match_available = 1;
1903 Assert(G1.strstart <= G1.isize && lookahead <= G1.isize, "a bit too far");
1905 /* Make sure that we always have enough lookahead, except
1906 * at the end of the input file. We need MAX_MATCH bytes
1907 * for the next match, plus MIN_MATCH bytes to insert the
1908 * string following the next match.
1910 while (G1.lookahead < MIN_LOOKAHEAD && !G1.eofile)
1913 if (match_available)
1914 ct_tally(0, G1.window[G1.strstart - 1]);
1916 return FLUSH_BLOCK(1); /* eof */
1920 /* ===========================================================================
1921 * Initialize the bit string routines.
1923 static void bi_init(void)
1933 /* ===========================================================================
1934 * Initialize the "longest match" routines for a new file
1936 static void lm_init(ush * flagsp)
1940 /* Initialize the hash table. */
1941 memset(head, 0, HASH_SIZE * sizeof(*head));
1942 /* prev will be initialized on the fly */
1944 /* speed options for the general purpose bit flag */
1945 *flagsp |= 2; /* FAST 4, SLOW 2 */
1946 /* ??? reduce max_chain_length for binary files */
1949 G1.block_start = 0L;
1951 G1.lookahead = file_read(G1.window,
1952 sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE);
1954 if (G1.lookahead == 0 || G1.lookahead == (unsigned) -1) {
1960 /* Make sure that we always have enough lookahead. This is important
1961 * if input comes from a device such as a tty.
1963 while (G1.lookahead < MIN_LOOKAHEAD && !G1.eofile)
1967 for (j = 0; j < MIN_MATCH - 1; j++)
1968 UPDATE_HASH(G1.ins_h, G1.window[j]);
1969 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
1970 * not important since only literal bytes will be emitted.
1975 /* ===========================================================================
1976 * Allocate the match buffer, initialize the various tables and save the
1977 * location of the internal file attribute (ascii/binary) and method
1979 * One callsite in zip()
1981 static void ct_init(void)
1983 int n; /* iterates over tree elements */
1984 int length; /* length value */
1985 int code; /* code value */
1986 int dist; /* distance index */
1988 G2.compressed_len = 0L;
1991 if (G2.static_dtree[0].Len != 0)
1992 return; /* ct_init already called */
1995 /* Initialize the mapping length (0..255) -> length code (0..28) */
1997 for (code = 0; code < LENGTH_CODES - 1; code++) {
1998 G2.base_length[code] = length;
1999 for (n = 0; n < (1 << extra_lbits[code]); n++) {
2000 G2.length_code[length++] = code;
2003 Assert(length == 256, "ct_init: length != 256");
2004 /* Note that the length 255 (match length 258) can be represented
2005 * in two different ways: code 284 + 5 bits or code 285, so we
2006 * overwrite length_code[255] to use the best encoding:
2008 G2.length_code[length - 1] = code;
2010 /* Initialize the mapping dist (0..32K) -> dist code (0..29) */
2012 for (code = 0; code < 16; code++) {
2013 G2.base_dist[code] = dist;
2014 for (n = 0; n < (1 << extra_dbits[code]); n++) {
2015 G2.dist_code[dist++] = code;
2018 Assert(dist == 256, "ct_init: dist != 256");
2019 dist >>= 7; /* from now on, all distances are divided by 128 */
2020 for (; code < D_CODES; code++) {
2021 G2.base_dist[code] = dist << 7;
2022 for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) {
2023 G2.dist_code[256 + dist++] = code;
2026 Assert(dist == 256, "ct_init: 256+dist != 512");
2028 /* Construct the codes of the static literal tree */
2029 /* already zeroed - it's in bss
2030 for (n = 0; n <= MAX_BITS; n++)
2031 G2.bl_count[n] = 0; */
2035 G2.static_ltree[n++].Len = 8;
2039 G2.static_ltree[n++].Len = 9;
2043 G2.static_ltree[n++].Len = 7;
2047 G2.static_ltree[n++].Len = 8;
2050 /* Codes 286 and 287 do not exist, but we must include them in the
2051 * tree construction to get a canonical Huffman tree (longest code
2054 gen_codes((ct_data *) G2.static_ltree, L_CODES + 1);
2056 /* The static distance tree is trivial: */
2057 for (n = 0; n < D_CODES; n++) {
2058 G2.static_dtree[n].Len = 5;
2059 G2.static_dtree[n].Code = bi_reverse(n, 5);
2062 /* Initialize the first block of the first file: */
2067 /* ===========================================================================
2068 * Deflate in to out.
2069 * IN assertions: the input and output buffers are cleared.
2072 static void zip(void)
2074 ush deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */
2078 /* Write the header to the gzip file. See algorithm.doc for the format */
2079 /* magic header for gzip files: 1F 8B */
2080 /* compression method: 8 (DEFLATED) */
2081 /* general flags: 0 */
2082 put_32bit(0x00088b1f);
2083 put_32bit(0); /* Unix timestamp */
2085 /* Write deflated file to zip file */
2090 lm_init(&deflate_flags);
2092 put_8bit(deflate_flags); /* extra flags */
2093 put_8bit(3); /* OS identifier = 3 (Unix) */
2097 /* Write the crc and uncompressed size */
2099 put_32bit(G1.isize);
2105 /* ======================================================================== */
2107 IF_DESKTOP(long long) int FAST_FUNC pack_gzip(transformer_state_t *xstate UNUSED_PARAM)
2109 /* Clear input and output buffers */
2117 memset(&G2, 0, sizeof(G2));
2118 G2.l_desc.dyn_tree = G2.dyn_ltree;
2119 G2.l_desc.static_tree = G2.static_ltree;
2120 G2.l_desc.extra_bits = extra_lbits;
2121 G2.l_desc.extra_base = LITERALS + 1;
2122 G2.l_desc.elems = L_CODES;
2123 G2.l_desc.max_length = MAX_BITS;
2124 //G2.l_desc.max_code = 0;
2125 G2.d_desc.dyn_tree = G2.dyn_dtree;
2126 G2.d_desc.static_tree = G2.static_dtree;
2127 G2.d_desc.extra_bits = extra_dbits;
2128 //G2.d_desc.extra_base = 0;
2129 G2.d_desc.elems = D_CODES;
2130 G2.d_desc.max_length = MAX_BITS;
2131 //G2.d_desc.max_code = 0;
2132 G2.bl_desc.dyn_tree = G2.bl_tree;
2133 //G2.bl_desc.static_tree = NULL;
2134 G2.bl_desc.extra_bits = extra_blbits,
2135 //G2.bl_desc.extra_base = 0;
2136 G2.bl_desc.elems = BL_CODES;
2137 G2.bl_desc.max_length = MAX_BL_BITS;
2138 //G2.bl_desc.max_code = 0;
2141 /* Saving of timestamp is disabled. Why?
2142 * - it is not Y2038-safe.
2143 * - some people want deterministic results
2144 * (normally they'd use -n, but our -n is a nop).
2146 * Per RFC 1952, gzfile.time=0 is "no timestamp".
2147 * If users will demand this to be reinstated,
2148 * implement -n "don't save timestamp".
2152 fstat(STDIN_FILENO, &s);
2160 #if ENABLE_FEATURE_GZIP_LONG_OPTIONS
2161 static const char gzip_longopts[] ALIGN1 =
2162 "stdout\0" No_argument "c"
2163 "to-stdout\0" No_argument "c"
2164 "force\0" No_argument "f"
2165 "verbose\0" No_argument "v"
2166 #if ENABLE_FEATURE_GZIP_DECOMPRESS
2167 "decompress\0" No_argument "d"
2168 "uncompress\0" No_argument "d"
2169 "test\0" No_argument "t"
2171 "quiet\0" No_argument "q"
2172 "fast\0" No_argument "1"
2173 "best\0" No_argument "9"
2174 "no-name\0" No_argument "n"
2179 * Linux kernel build uses gzip -d -n. We accept and ignore -n.
2182 * gzip: do not save the original file name and time stamp.
2183 * (The original name is always saved if the name had to be truncated.)
2184 * gunzip: do not restore the original file name/time even if present
2185 * (remove only the gzip suffix from the compressed file name).
2186 * This option is the default when decompressing.
2188 * gzip: always save the original file name and time stamp (this is the default)
2189 * gunzip: restore the original file name and time stamp if present.
2192 int gzip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
2193 #if ENABLE_FEATURE_GZIP_DECOMPRESS
2194 int gzip_main(int argc, char **argv)
2196 int gzip_main(int argc UNUSED_PARAM, char **argv)
2200 #if ENABLE_FEATURE_GZIP_LEVELS
2201 static const struct {
2203 uint8_t chain_shift;
2206 } gzip_level_config[6] = {
2207 {4, 4, 4/2, 16/2}, /* Level 4 */
2208 {8, 5, 16/2, 32/2}, /* Level 5 */
2209 {8, 7, 16/2, 128/2}, /* Level 6 */
2210 {8, 8, 32/2, 128/2}, /* Level 7 */
2211 {32, 10, 128/2, 258/2}, /* Level 8 */
2212 {32, 12, 258/2, 258/2}, /* Level 9 */
2216 SET_PTR_TO_GLOBALS((char *)xzalloc(sizeof(struct globals)+sizeof(struct globals2))
2217 + sizeof(struct globals));
2219 /* Must match bbunzip's constants OPT_STDOUT, OPT_FORCE! */
2220 #if ENABLE_FEATURE_GZIP_LONG_OPTIONS
2221 opt = getopt32long(argv, "cfkv" IF_FEATURE_GZIP_DECOMPRESS("dt") "qn123456789", gzip_longopts);
2223 opt = getopt32(argv, "cfkv" IF_FEATURE_GZIP_DECOMPRESS("dt") "qn123456789");
2225 #if ENABLE_FEATURE_GZIP_DECOMPRESS /* gunzip_main may not be visible... */
2226 if (opt & 0x30) // -d and/or -t
2227 return gunzip_main(argc, argv);
2229 #if ENABLE_FEATURE_GZIP_LEVELS
2230 opt >>= ENABLE_FEATURE_GZIP_DECOMPRESS ? 8 : 6; /* drop cfkv[dt]qn bits */
2232 opt = 1 << 6; /* default: 6 */
2233 opt = ffs(opt >> 4); /* Maps -1..-4 to [0], -5 to [1] ... -9 to [5] */
2234 max_chain_length = 1 << gzip_level_config[opt].chain_shift;
2235 good_match = gzip_level_config[opt].good;
2236 max_lazy_match = gzip_level_config[opt].lazy2 * 2;
2237 nice_match = gzip_level_config[opt].nice2 * 2;
2239 option_mask32 &= 0xf; /* retain only -cfkv */
2241 /* Allocate all global buffers (for DYN_ALLOC option) */
2242 ALLOC(uch, G1.l_buf, INBUFSIZ);
2243 ALLOC(uch, G1.outbuf, OUTBUFSIZ);
2244 ALLOC(ush, G1.d_buf, DIST_BUFSIZE);
2245 ALLOC(uch, G1.window, 2L * WSIZE);
2246 ALLOC(ush, G1.prev, 1L << BITS);
2248 /* Initialize the CRC32 table */
2249 global_crc32_table = crc32_filltable(NULL, 0);
2252 return bbunpack(argv, pack_gzip, append_ext, "gz");