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.
19 /* big objects in bss:
21 * 00000074 b base_length
22 * 00000078 b base_dist
23 * 00000078 b static_dtree
25 * 000000f4 b dyn_dtree
26 * 00000100 b length_code
27 * 00000200 b dist_code
31 * 00000480 b static_ltree
32 * 000008f4 b dyn_ltree
35 /* TODO: full support for -v for DESKTOP
36 * "/usr/bin/gzip -v a bogus aa" should say:
37 a: 85.1% -- replaced with a.gz
38 gzip: bogus: No such file or directory
39 aa: 85.1% -- replaced with aa.gz
42 //kbuild:lib-$(CONFIG_GZIP) += gzip.o
44 //usage:#define gzip_trivial_usage
45 //usage: "[-cfd] [FILE]..."
46 //usage:#define gzip_full_usage "\n\n"
47 //usage: "Compress FILEs (or stdin)\n"
48 //usage: "\n -d Decompress"
49 //usage: "\n -c Write to stdout"
50 //usage: "\n -f Force"
52 //usage:#define gzip_example_usage
53 //usage: "$ ls -la /tmp/busybox*\n"
54 //usage: "-rw-rw-r-- 1 andersen andersen 1761280 Apr 14 17:47 /tmp/busybox.tar\n"
55 //usage: "$ gzip /tmp/busybox.tar\n"
56 //usage: "$ ls -la /tmp/busybox*\n"
57 //usage: "-rw-rw-r-- 1 andersen andersen 554058 Apr 14 17:49 /tmp/busybox.tar.gz\n"
60 #include "bb_archive.h"
63 /* ===========================================================================
66 /* Diagnostic functions */
68 # define Assert(cond,msg) { if (!(cond)) bb_error_msg(msg); }
69 # define Trace(x) fprintf x
70 # define Tracev(x) {if (verbose) fprintf x; }
71 # define Tracevv(x) {if (verbose > 1) fprintf x; }
72 # define Tracec(c,x) {if (verbose && (c)) fprintf x; }
73 # define Tracecv(c,x) {if (verbose > 1 && (c)) fprintf x; }
75 # define Assert(cond,msg)
84 /* ===========================================================================
86 #if CONFIG_GZIP_FAST == 0
88 #elif CONFIG_GZIP_FAST == 1
90 #elif CONFIG_GZIP_FAST == 2
93 # error "Invalid CONFIG_GZIP_FAST value"
98 # define INBUFSIZ 0x2000 /* input buffer size */
100 # define INBUFSIZ 0x8000 /* input buffer size */
106 # define OUTBUFSIZ 8192 /* output buffer size */
108 # define OUTBUFSIZ 16384 /* output buffer size */
114 # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */
116 # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */
121 #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */
122 #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
123 #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
124 #define ORIG_NAME 0x08 /* bit 3 set: original file name present */
125 #define COMMENT 0x10 /* bit 4 set: file comment present */
126 #define RESERVED 0xC0 /* bit 6,7: reserved */
128 /* internal file attribute */
129 #define UNKNOWN 0xffff
134 # define WSIZE 0x8000 /* window size--must be a power of two, and */
135 #endif /* at least 32K for zip's deflate method */
138 #define MAX_MATCH 258
139 /* The minimum and maximum match lengths */
141 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
142 /* Minimum amount of lookahead, except at the end of the input file.
143 * See deflate.c for comments about the MIN_MATCH+1.
146 #define MAX_DIST (WSIZE-MIN_LOOKAHEAD)
147 /* In order to simplify the code, particularly on 16 bit machines, match
148 * distances are limited to MAX_DIST instead of WSIZE.
152 # define MAX_PATH_LEN 1024 /* max pathname length */
155 #define seekable() 0 /* force sequential output */
156 #define translate_eol 0 /* no option -a yet */
161 #define INIT_BITS 9 /* Initial number of bits per code */
163 #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */
164 /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free.
165 * It's a pity that old uncompress does not check bit 0x20. That makes
166 * extension of the format actually undesirable because old compress
167 * would just crash on the new format instead of giving a meaningful
168 * error message. It does check the number of bits, but it's more
169 * helpful to say "unsupported format, get a new version" than
170 * "can only handle 16 bits".
174 # define MAX_SUFFIX MAX_EXT_CHARS
176 # define MAX_SUFFIX 30
180 /* ===========================================================================
181 * Compile with MEDIUM_MEM to reduce the memory requirements or
182 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
183 * entire input file can be held in memory (not possible on 16 bit systems).
184 * Warning: defining these symbols affects HASH_BITS (see below) and thus
185 * affects the compression ratio. The compressed output
186 * is still correct, and might even be smaller in some cases.
190 # define HASH_BITS 13 /* Number of bits used to hash strings */
193 # define HASH_BITS 14
196 # define HASH_BITS 15
197 /* For portability to 16 bit machines, do not use values above 15. */
200 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
201 #define HASH_MASK (HASH_SIZE-1)
202 #define WMASK (WSIZE-1)
203 /* HASH_SIZE and WSIZE must be powers of two */
205 # define TOO_FAR 4096
207 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
210 /* ===========================================================================
211 * These types are not really 'char', 'short' and 'long'
214 typedef uint16_t ush;
215 typedef uint32_t ulg;
219 typedef unsigned IPos;
220 /* A Pos is an index in the character window. We use short instead of int to
221 * save space in the various tables. IPos is used only for parameter passing.
225 WINDOW_SIZE = 2 * WSIZE,
226 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
227 * input file length plus MIN_LOOKAHEAD.
230 max_chain_length = 4096,
231 /* To speed up deflation, hash chains are never searched beyond this length.
232 * A higher limit improves compression ratio but degrades the speed.
235 max_lazy_match = 258,
236 /* Attempt to find a better match only when the current match is strictly
237 * smaller than this value. This mechanism is used only for compression
241 max_insert_length = max_lazy_match,
242 /* Insert new strings in the hash table only if the match length
243 * is not greater than this length. This saves time but degrades compression.
244 * max_insert_length is used only for compression levels <= 3.
248 /* Use a faster search when the previous match is longer than this */
250 /* Values for max_lazy_match, good_match and max_chain_length, depending on
251 * the desired pack level (0..9). The values given below have been tuned to
252 * exclude worst case performance for pathological files. Better values may be
253 * found for specific files.
256 nice_match = 258, /* Stop searching when current match exceeds this */
257 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
258 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
268 /* window position at the beginning of the current output block. Gets
269 * negative when the window is moved backwards.
271 unsigned ins_h; /* hash index of string to be inserted */
273 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1) / MIN_MATCH)
274 /* Number of bits by which ins_h and del_h must be shifted at each
275 * input step. It must be such that after MIN_MATCH steps, the oldest
276 * byte no longer takes part in the hash key, that is:
277 * H_SHIFT * MIN_MATCH >= HASH_BITS
280 unsigned prev_length;
282 /* Length of the best match at previous step. Matches not greater than this
283 * are discarded. This is used in the lazy match evaluation.
286 unsigned strstart; /* start of string to insert */
287 unsigned match_start; /* start of matching string */
288 unsigned lookahead; /* number of valid bytes ahead in window */
290 /* ===========================================================================
292 #define DECLARE(type, array, size) \
294 #define ALLOC(type, array, size) \
295 array = xzalloc((size_t)(((size)+1L)/2) * 2*sizeof(type))
296 #define FREE(array) \
297 do { free(array); array = NULL; } while (0)
301 /* buffer for literals or lengths */
302 /* DECLARE(uch, l_buf, LIT_BUFSIZE); */
303 DECLARE(uch, l_buf, INBUFSIZ);
305 DECLARE(ush, d_buf, DIST_BUFSIZE);
306 DECLARE(uch, outbuf, OUTBUFSIZ);
308 /* Sliding window. Input bytes are read into the second half of the window,
309 * and move to the first half later to keep a dictionary of at least WSIZE
310 * bytes. With this organization, matches are limited to a distance of
311 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
312 * performed with a length multiple of the block size. Also, it limits
313 * the window size to 64K, which is quite useful on MSDOS.
314 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
315 * be less efficient).
317 DECLARE(uch, window, 2L * WSIZE);
319 /* Link to older string with same hash index. To limit the size of this
320 * array to 64K, this link is maintained only for the last 32K strings.
321 * An index in this array is thus a window index modulo 32K.
323 /* DECLARE(Pos, prev, WSIZE); */
324 DECLARE(ush, prev, 1L << BITS);
326 /* Heads of the hash chains or 0. */
327 /* DECLARE(Pos, head, 1<<HASH_BITS); */
328 #define head (G1.prev + WSIZE) /* hash head (see deflate.c) */
330 /* number of input bytes */
331 ulg isize; /* only 32 bits stored in .gz file */
333 /* bbox always use stdin/stdout */
334 #define ifd STDIN_FILENO /* input file descriptor */
335 #define ofd STDOUT_FILENO /* output file descriptor */
338 unsigned insize; /* valid bytes in l_buf */
340 unsigned outcnt; /* bytes in output buffer */
342 smallint eofile; /* flag set at end of input file */
344 /* ===========================================================================
345 * Local data used by the "bit string" routines.
348 unsigned short bi_buf;
350 /* Output buffer. bits are inserted starting at the bottom (least significant
355 #define BUF_SIZE (8 * sizeof(G1.bi_buf))
356 /* Number of bits used within bi_buf. (bi_buf might be implemented on
357 * more than 16 bits on some systems.)
362 /* Current input function. Set to mem_read for in-memory compression */
365 ulg bits_sent; /* bit length of the compressed data */
368 /*uint32_t *crc_32_tab;*/
369 uint32_t crc; /* shift register contents */
372 #define G1 (*(ptr_to_globals - 1))
375 /* ===========================================================================
376 * Write the output buffer outbuf[0..outcnt-1] and update bytes_out.
377 * (used for the compressed data only)
379 static void flush_outbuf(void)
384 xwrite(ofd, (char *) G1.outbuf, G1.outcnt);
389 /* ===========================================================================
391 /* put_8bit is used for the compressed output */
392 #define put_8bit(c) \
394 G1.outbuf[G1.outcnt++] = (c); \
395 if (G1.outcnt == OUTBUFSIZ) flush_outbuf(); \
398 /* Output a 16 bit value, lsb first */
399 static void put_16bit(ush w)
401 if (G1.outcnt < OUTBUFSIZ - 2) {
402 G1.outbuf[G1.outcnt++] = w;
403 G1.outbuf[G1.outcnt++] = w >> 8;
410 static void put_32bit(ulg n)
416 /* ===========================================================================
417 * Run a set of bytes through the crc shift register. If s is a NULL
418 * pointer, then initialize the crc shift register contents instead.
419 * Return the current crc in either case.
421 static void updcrc(uch * s, unsigned n)
423 G1.crc = crc32_block_endian0(G1.crc, s, n, global_crc32_table /*G1.crc_32_tab*/);
427 /* ===========================================================================
428 * Read a new buffer from the current input file, perform end-of-line
429 * translation, and update the crc and input file size.
430 * IN assertion: size >= 2 (for end-of-line translation)
432 static unsigned file_read(void *buf, unsigned size)
436 Assert(G1.insize == 0, "l_buf not empty");
438 len = safe_read(ifd, buf, size);
439 if (len == (unsigned)(-1) || len == 0)
448 /* ===========================================================================
449 * Send a value on a given number of bits.
450 * IN assertion: length <= 16 and value fits in length bits.
452 static void send_bits(int value, int length)
455 Tracev((stderr, " l %2d v %4x ", length, value));
456 Assert(length > 0 && length <= 15, "invalid length");
457 G1.bits_sent += length;
459 /* If not enough room in bi_buf, use (valid) bits from bi_buf and
460 * (16 - bi_valid) bits from value, leaving (width - (16-bi_valid))
461 * unused bits in value.
463 if (G1.bi_valid > (int) BUF_SIZE - length) {
464 G1.bi_buf |= (value << G1.bi_valid);
465 put_16bit(G1.bi_buf);
466 G1.bi_buf = (ush) value >> (BUF_SIZE - G1.bi_valid);
467 G1.bi_valid += length - BUF_SIZE;
469 G1.bi_buf |= value << G1.bi_valid;
470 G1.bi_valid += length;
475 /* ===========================================================================
476 * Reverse the first len bits of a code, using straightforward code (a faster
477 * method would use a table)
478 * IN assertion: 1 <= len <= 15
480 static unsigned bi_reverse(unsigned code, int len)
486 if (--len <= 0) return res;
493 /* ===========================================================================
494 * Write out any remaining bits in an incomplete byte.
496 static void bi_windup(void)
498 if (G1.bi_valid > 8) {
499 put_16bit(G1.bi_buf);
500 } else if (G1.bi_valid > 0) {
506 G1.bits_sent = (G1.bits_sent + 7) & ~7;
511 /* ===========================================================================
512 * Copy a stored block to the zip file, storing first the length and its
513 * one's complement if requested.
515 static void copy_block(char *buf, unsigned len, int header)
517 bi_windup(); /* align on byte boundary */
523 G1.bits_sent += 2 * 16;
527 G1.bits_sent += (ulg) len << 3;
535 /* ===========================================================================
536 * Fill the window when the lookahead becomes insufficient.
537 * Updates strstart and lookahead, and sets eofile if end of input file.
538 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
539 * OUT assertions: at least one byte has been read, or eofile is set;
540 * file reads are performed for at least two bytes (required for the
541 * translate_eol option).
543 static void fill_window(void)
546 unsigned more = WINDOW_SIZE - G1.lookahead - G1.strstart;
547 /* Amount of free space at the end of the window. */
549 /* If the window is almost full and there is insufficient lookahead,
550 * move the upper half to the lower one to make room in the upper half.
552 if (more == (unsigned) -1) {
553 /* Very unlikely, but possible on 16 bit machine if strstart == 0
554 * and lookahead == 1 (input done one byte at time)
557 } else if (G1.strstart >= WSIZE + MAX_DIST) {
558 /* By the IN assertion, the window is not empty so we can't confuse
559 * more == 0 with more == 64K on a 16 bit machine.
561 Assert(WINDOW_SIZE == 2 * WSIZE, "no sliding with BIG_MEM");
563 memcpy(G1.window, G1.window + WSIZE, WSIZE);
564 G1.match_start -= WSIZE;
565 G1.strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
567 G1.block_start -= WSIZE;
569 for (n = 0; n < HASH_SIZE; n++) {
571 head[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
573 for (n = 0; n < WSIZE; n++) {
575 G1.prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
576 /* If n is not on any hash chain, prev[n] is garbage but
577 * its value will never be used.
582 /* At this point, more >= 2 */
584 n = file_read(G1.window + G1.strstart + G1.lookahead, more);
585 if (n == 0 || n == (unsigned) -1) {
594 /* ===========================================================================
595 * Set match_start to the longest match starting at the given string and
596 * return its length. Matches shorter or equal to prev_length are discarded,
597 * in which case the result is equal to prev_length and match_start is
599 * IN assertions: cur_match is the head of the hash chain for the current
600 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
603 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
604 * match.s. The code is functionally equivalent, so you can use the C version
607 static int longest_match(IPos cur_match)
609 unsigned chain_length = max_chain_length; /* max hash chain length */
610 uch *scan = G1.window + G1.strstart; /* current string */
611 uch *match; /* matched string */
612 int len; /* length of current match */
613 int best_len = G1.prev_length; /* best match length so far */
614 IPos limit = G1.strstart > (IPos) MAX_DIST ? G1.strstart - (IPos) MAX_DIST : 0;
615 /* Stop when cur_match becomes <= limit. To simplify the code,
616 * we prevent matches with the string of window index 0.
619 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
620 * It is easy to get rid of this optimization if necessary.
622 #if HASH_BITS < 8 || MAX_MATCH != 258
623 # error Code too clever
625 uch *strend = G1.window + G1.strstart + MAX_MATCH;
626 uch scan_end1 = scan[best_len - 1];
627 uch scan_end = scan[best_len];
629 /* Do not waste too much time if we already have a good match: */
630 if (G1.prev_length >= good_match) {
633 Assert(G1.strstart <= WINDOW_SIZE - MIN_LOOKAHEAD, "insufficient lookahead");
636 Assert(cur_match < G1.strstart, "no future");
637 match = G1.window + cur_match;
639 /* Skip to next match if the match length cannot increase
640 * or if the match length is less than 2:
642 if (match[best_len] != scan_end
643 || match[best_len - 1] != scan_end1
644 || *match != *scan || *++match != scan[1]
649 /* The check at best_len-1 can be removed because it will be made
650 * again later. (This heuristic is not always a win.)
651 * It is not necessary to compare scan[2] and match[2] since they
652 * are always equal when the other bytes match, given that
653 * the hash keys are equal and that HASH_BITS >= 8.
657 /* We check for insufficient lookahead only every 8th comparison;
658 * the 256th check will be made at strstart+258.
661 } while (*++scan == *++match && *++scan == *++match &&
662 *++scan == *++match && *++scan == *++match &&
663 *++scan == *++match && *++scan == *++match &&
664 *++scan == *++match && *++scan == *++match && scan < strend);
666 len = MAX_MATCH - (int) (strend - scan);
667 scan = strend - MAX_MATCH;
669 if (len > best_len) {
670 G1.match_start = cur_match;
672 if (len >= nice_match)
674 scan_end1 = scan[best_len - 1];
675 scan_end = scan[best_len];
677 } while ((cur_match = G1.prev[cur_match & WMASK]) > limit
678 && --chain_length != 0);
685 /* ===========================================================================
686 * Check that the match at match_start is indeed a match.
688 static void check_match(IPos start, IPos match, int length)
690 /* check that the match is indeed a match */
691 if (memcmp(G1.window + match, G1.window + start, length) != 0) {
692 bb_error_msg(" start %d, match %d, length %d", start, match, length);
693 bb_error_msg("invalid match");
696 bb_error_msg("\\[%d,%d]", start - match, length);
698 bb_putchar_stderr(G1.window[start++]);
699 } while (--length != 0);
703 # define check_match(start, match, length) ((void)0)
707 /* trees.c -- output deflated data using Huffman coding
708 * Copyright (C) 1992-1993 Jean-loup Gailly
709 * This is free software; you can redistribute it and/or modify it under the
710 * terms of the GNU General Public License, see the file COPYING.
714 * Encode various sets of source values using variable-length
718 * The PKZIP "deflation" process uses several Huffman trees. The more
719 * common source values are represented by shorter bit sequences.
721 * Each code tree is stored in the ZIP file in a compressed form
722 * which is itself a Huffman encoding of the lengths of
723 * all the code strings (in ascending order by source values).
724 * The actual code strings are reconstructed from the lengths in
725 * the UNZIP process, as described in the "application note"
726 * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program.
730 * Data Compression: Techniques and Applications, pp. 53-55.
731 * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7.
734 * Data Compression: Methods and Theory, pp. 49-50.
735 * Computer Science Press, 1988. ISBN 0-7167-8156-5.
739 * Addison-Wesley, 1983. ISBN 0-201-06672-6.
743 * Allocate the match buffer, initialize the various tables [and save
744 * the location of the internal file attribute (ascii/binary) and
745 * method (DEFLATE/STORE) -- deleted in bbox]
747 * void ct_tally(int dist, int lc);
748 * Save the match info and tally the frequency counts.
750 * ulg flush_block(char *buf, ulg stored_len, int eof)
751 * Determine the best encoding for the current block: dynamic trees,
752 * static trees or store, and output the encoded block to the zip
753 * file. Returns the total compressed length for the file so far.
757 /* All codes must not exceed MAX_BITS bits */
759 #define MAX_BL_BITS 7
760 /* Bit length codes must not exceed MAX_BL_BITS bits */
762 #define LENGTH_CODES 29
763 /* number of length codes, not counting the special END_BLOCK code */
766 /* number of literal bytes 0..255 */
768 #define END_BLOCK 256
769 /* end of block literal code */
771 #define L_CODES (LITERALS+1+LENGTH_CODES)
772 /* number of Literal or Length codes, including the END_BLOCK code */
775 /* number of distance codes */
778 /* number of codes used to transfer the bit lengths */
780 /* extra bits for each length code */
781 static const uint8_t extra_lbits[LENGTH_CODES] ALIGN1 = {
782 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4,
786 /* extra bits for each distance code */
787 static const uint8_t extra_dbits[D_CODES] ALIGN1 = {
788 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,
789 10, 10, 11, 11, 12, 12, 13, 13
792 /* extra bits for each bit length code */
793 static const uint8_t extra_blbits[BL_CODES] ALIGN1 = {
794 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 };
796 /* number of codes at each bit length for an optimal tree */
797 static const uint8_t bl_order[BL_CODES] ALIGN1 = {
798 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
800 #define STORED_BLOCK 0
801 #define STATIC_TREES 1
803 /* The three kinds of block type */
807 # define LIT_BUFSIZE 0x2000
810 # define LIT_BUFSIZE 0x4000
812 # define LIT_BUFSIZE 0x8000
817 # define DIST_BUFSIZE LIT_BUFSIZE
819 /* Sizes of match buffers for literals/lengths and distances. There are
820 * 4 reasons for limiting LIT_BUFSIZE to 64K:
821 * - frequencies can be kept in 16 bit counters
822 * - if compression is not successful for the first block, all input data is
823 * still in the window so we can still emit a stored block even when input
824 * comes from standard input. (This can also be done for all blocks if
825 * LIT_BUFSIZE is not greater than 32K.)
826 * - if compression is not successful for a file smaller than 64K, we can
827 * even emit a stored file instead of a stored block (saving 5 bytes).
828 * - creating new Huffman trees less frequently may not provide fast
829 * adaptation to changes in the input data statistics. (Take for
830 * example a binary file with poorly compressible code followed by
831 * a highly compressible string table.) Smaller buffer sizes give
832 * fast adaptation but have of course the overhead of transmitting trees
834 * - I can't count above 4
835 * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save
836 * memory at the expense of compression). Some optimizations would be possible
837 * if we rely on DIST_BUFSIZE == LIT_BUFSIZE.
840 /* repeat previous bit length 3-6 times (2 bits of repeat count) */
842 /* repeat a zero length 3-10 times (3 bits of repeat count) */
843 #define REPZ_11_138 18
844 /* repeat a zero length 11-138 times (7 bits of repeat count) */
846 /* ===========================================================================
848 /* Data structure describing a single value and its code string. */
849 typedef struct ct_data {
851 ush freq; /* frequency count */
852 ush code; /* bit string */
855 ush dad; /* father node in Huffman tree */
856 ush len; /* length of bit string */
865 #define HEAP_SIZE (2*L_CODES + 1)
866 /* maximum heap size */
868 typedef struct tree_desc {
869 ct_data *dyn_tree; /* the dynamic tree */
870 ct_data *static_tree; /* corresponding static tree or NULL */
871 const uint8_t *extra_bits; /* extra bits for each code or NULL */
872 int extra_base; /* base index for extra_bits */
873 int elems; /* max number of elements in the tree */
874 int max_length; /* max bit length for the codes */
875 int max_code; /* largest code with non zero frequency */
880 ush heap[HEAP_SIZE]; /* heap used to build the Huffman trees */
881 int heap_len; /* number of elements in the heap */
882 int heap_max; /* element of largest frequency */
884 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
885 * The same heap array is used to build all trees.
888 ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */
889 ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */
891 ct_data static_ltree[L_CODES + 2];
893 /* The static literal tree. Since the bit lengths are imposed, there is no
894 * need for the L_CODES extra codes used during heap construction. However
895 * The codes 286 and 287 are needed to build a canonical tree (see ct_init
899 ct_data static_dtree[D_CODES];
901 /* The static distance tree. (Actually a trivial tree since all codes use
905 ct_data bl_tree[2 * BL_CODES + 1];
907 /* Huffman tree for the bit lengths */
913 ush bl_count[MAX_BITS + 1];
915 /* The lengths of the bit length codes are sent in order of decreasing
916 * probability, to avoid transmitting the lengths for unused bit length codes.
919 uch depth[2 * L_CODES + 1];
921 /* Depth of each subtree used as tie breaker for trees of equal frequency */
923 uch length_code[MAX_MATCH - MIN_MATCH + 1];
925 /* length code for each normalized match length (0 == MIN_MATCH) */
929 /* distance codes. The first 256 values correspond to the distances
930 * 3 .. 258, the last 256 values correspond to the top 8 bits of
931 * the 15 bit distances.
934 int base_length[LENGTH_CODES];
936 /* First normalized length for each code (0 = MIN_MATCH) */
938 int base_dist[D_CODES];
940 /* First normalized distance for each code (0 = distance of 1) */
942 uch flag_buf[LIT_BUFSIZE / 8];
944 /* flag_buf is a bit array distinguishing literals from lengths in
945 * l_buf, thus indicating the presence or absence of a distance.
948 unsigned last_lit; /* running index in l_buf */
949 unsigned last_dist; /* running index in d_buf */
950 unsigned last_flags; /* running index in flag_buf */
951 uch flags; /* current flags not yet saved in flag_buf */
952 uch flag_bit; /* current bit used in flags */
954 /* bits are filled in flags starting at bit 0 (least significant).
955 * Note: these flags are overkill in the current code since we don't
956 * take advantage of DIST_BUFSIZE == LIT_BUFSIZE.
959 ulg opt_len; /* bit length of current block with optimal trees */
960 ulg static_len; /* bit length of current block with static trees */
962 ulg compressed_len; /* total bit length of compressed file */
965 #define G2ptr ((struct globals2*)(ptr_to_globals))
969 /* ===========================================================================
971 static void gen_codes(ct_data * tree, int max_code);
972 static void build_tree(tree_desc * desc);
973 static void scan_tree(ct_data * tree, int max_code);
974 static void send_tree(ct_data * tree, int max_code);
975 static int build_bl_tree(void);
976 static void send_all_trees(int lcodes, int dcodes, int blcodes);
977 static void compress_block(ct_data * ltree, ct_data * dtree);
981 /* Send a code of the given tree. c and tree must not have side effects */
982 # define SEND_CODE(c, tree) send_bits(tree[c].Code, tree[c].Len)
984 # define SEND_CODE(c, tree) \
986 if (verbose > 1) bb_error_msg("\ncd %3d ", (c)); \
987 send_bits(tree[c].Code, tree[c].Len); \
991 #define D_CODE(dist) \
992 ((dist) < 256 ? G2.dist_code[dist] : G2.dist_code[256 + ((dist)>>7)])
993 /* Mapping from a distance to a distance code. dist is the distance - 1 and
994 * must not have side effects. dist_code[256] and dist_code[257] are never
996 * The arguments must not have side effects.
1000 /* ===========================================================================
1001 * Initialize a new block.
1003 static void init_block(void)
1005 int n; /* iterates over tree elements */
1007 /* Initialize the trees. */
1008 for (n = 0; n < L_CODES; n++)
1009 G2.dyn_ltree[n].Freq = 0;
1010 for (n = 0; n < D_CODES; n++)
1011 G2.dyn_dtree[n].Freq = 0;
1012 for (n = 0; n < BL_CODES; n++)
1013 G2.bl_tree[n].Freq = 0;
1015 G2.dyn_ltree[END_BLOCK].Freq = 1;
1016 G2.opt_len = G2.static_len = 0;
1017 G2.last_lit = G2.last_dist = G2.last_flags = 0;
1023 /* ===========================================================================
1024 * Restore the heap property by moving down the tree starting at node k,
1025 * exchanging a node with the smallest of its two sons if necessary, stopping
1026 * when the heap property is re-established (each father smaller than its
1030 /* Compares to subtrees, using the tree depth as tie breaker when
1031 * the subtrees have equal frequency. This minimizes the worst case length. */
1032 #define SMALLER(tree, n, m) \
1033 (tree[n].Freq < tree[m].Freq \
1034 || (tree[n].Freq == tree[m].Freq && G2.depth[n] <= G2.depth[m]))
1036 static void pqdownheap(ct_data * tree, int k)
1039 int j = k << 1; /* left son of k */
1041 while (j <= G2.heap_len) {
1042 /* Set j to the smallest of the two sons: */
1043 if (j < G2.heap_len && SMALLER(tree, G2.heap[j + 1], G2.heap[j]))
1046 /* Exit if v is smaller than both sons */
1047 if (SMALLER(tree, v, G2.heap[j]))
1050 /* Exchange v with the smallest son */
1051 G2.heap[k] = G2.heap[j];
1054 /* And continue down the tree, setting j to the left son of k */
1061 /* ===========================================================================
1062 * Compute the optimal bit lengths for a tree and update the total bit length
1063 * for the current block.
1064 * IN assertion: the fields freq and dad are set, heap[heap_max] and
1065 * above are the tree nodes sorted by increasing frequency.
1066 * OUT assertions: the field len is set to the optimal bit length, the
1067 * array bl_count contains the frequencies for each bit length.
1068 * The length opt_len is updated; static_len is also updated if stree is
1071 static void gen_bitlen(tree_desc * desc)
1073 ct_data *tree = desc->dyn_tree;
1074 const uint8_t *extra = desc->extra_bits;
1075 int base = desc->extra_base;
1076 int max_code = desc->max_code;
1077 int max_length = desc->max_length;
1078 ct_data *stree = desc->static_tree;
1079 int h; /* heap index */
1080 int n, m; /* iterate over the tree elements */
1081 int bits; /* bit length */
1082 int xbits; /* extra bits */
1083 ush f; /* frequency */
1084 int overflow = 0; /* number of elements with bit length too large */
1086 for (bits = 0; bits <= MAX_BITS; bits++)
1087 G2.bl_count[bits] = 0;
1089 /* In a first pass, compute the optimal bit lengths (which may
1090 * overflow in the case of the bit length tree).
1092 tree[G2.heap[G2.heap_max]].Len = 0; /* root of the heap */
1094 for (h = G2.heap_max + 1; h < HEAP_SIZE; h++) {
1096 bits = tree[tree[n].Dad].Len + 1;
1097 if (bits > max_length) {
1101 tree[n].Len = (ush) bits;
1102 /* We overwrite tree[n].Dad which is no longer needed */
1105 continue; /* not a leaf node */
1107 G2.bl_count[bits]++;
1110 xbits = extra[n - base];
1112 G2.opt_len += (ulg) f *(bits + xbits);
1115 G2.static_len += (ulg) f * (stree[n].Len + xbits);
1120 Trace((stderr, "\nbit length overflow\n"));
1121 /* This happens for example on obj2 and pic of the Calgary corpus */
1123 /* Find the first bit length which could increase: */
1125 bits = max_length - 1;
1126 while (G2.bl_count[bits] == 0)
1128 G2.bl_count[bits]--; /* move one leaf down the tree */
1129 G2.bl_count[bits + 1] += 2; /* move one overflow item as its brother */
1130 G2.bl_count[max_length]--;
1131 /* The brother of the overflow item also moves one step up,
1132 * but this does not affect bl_count[max_length]
1135 } while (overflow > 0);
1137 /* Now recompute all bit lengths, scanning in increasing frequency.
1138 * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all
1139 * lengths instead of fixing only the wrong ones. This idea is taken
1140 * from 'ar' written by Haruhiko Okumura.)
1142 for (bits = max_length; bits != 0; bits--) {
1143 n = G2.bl_count[bits];
1148 if (tree[m].Len != (unsigned) bits) {
1149 Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len, bits));
1150 G2.opt_len += ((int32_t) bits - tree[m].Len) * tree[m].Freq;
1159 /* ===========================================================================
1160 * Generate the codes for a given tree and bit counts (which need not be
1162 * IN assertion: the array bl_count contains the bit length statistics for
1163 * the given tree and the field len is set for all tree elements.
1164 * OUT assertion: the field code is set for all tree elements of non
1167 static void gen_codes(ct_data * tree, int max_code)
1169 ush next_code[MAX_BITS + 1]; /* next code value for each bit length */
1170 ush code = 0; /* running code value */
1171 int bits; /* bit index */
1172 int n; /* code index */
1174 /* The distribution counts are first used to generate the code values
1175 * without bit reversal.
1177 for (bits = 1; bits <= MAX_BITS; bits++) {
1178 next_code[bits] = code = (code + G2.bl_count[bits - 1]) << 1;
1180 /* Check that the bit counts in bl_count are consistent. The last code
1183 Assert(code + G2.bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1,
1184 "inconsistent bit counts");
1185 Tracev((stderr, "\ngen_codes: max_code %d ", max_code));
1187 for (n = 0; n <= max_code; n++) {
1188 int len = tree[n].Len;
1192 /* Now reverse the bits */
1193 tree[n].Code = bi_reverse(next_code[len]++, len);
1195 Tracec(tree != G2.static_ltree,
1196 (stderr, "\nn %3d %c l %2d c %4x (%x) ", n,
1197 (n > ' ' ? n : ' '), len, tree[n].Code,
1198 next_code[len] - 1));
1203 /* ===========================================================================
1204 * Construct one Huffman tree and assigns the code bit strings and lengths.
1205 * Update the total bit length for the current block.
1206 * IN assertion: the field freq is set for all tree elements.
1207 * OUT assertions: the fields len and code are set to the optimal bit length
1208 * and corresponding code. The length opt_len is updated; static_len is
1209 * also updated if stree is not null. The field max_code is set.
1212 /* Remove the smallest element from the heap and recreate the heap with
1213 * one less element. Updates heap and heap_len. */
1216 /* Index within the heap array of least frequent node in the Huffman tree */
1218 #define PQREMOVE(tree, top) \
1220 top = G2.heap[SMALLEST]; \
1221 G2.heap[SMALLEST] = G2.heap[G2.heap_len--]; \
1222 pqdownheap(tree, SMALLEST); \
1225 static void build_tree(tree_desc * desc)
1227 ct_data *tree = desc->dyn_tree;
1228 ct_data *stree = desc->static_tree;
1229 int elems = desc->elems;
1230 int n, m; /* iterate over heap elements */
1231 int max_code = -1; /* largest code with non zero frequency */
1232 int node = elems; /* next internal node of the tree */
1234 /* Construct the initial heap, with least frequent element in
1235 * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1].
1236 * heap[0] is not used.
1239 G2.heap_max = HEAP_SIZE;
1241 for (n = 0; n < elems; n++) {
1242 if (tree[n].Freq != 0) {
1243 G2.heap[++G2.heap_len] = max_code = n;
1250 /* The pkzip format requires that at least one distance code exists,
1251 * and that at least one bit should be sent even if there is only one
1252 * possible code. So to avoid special checks later on we force at least
1253 * two codes of non zero frequency.
1255 while (G2.heap_len < 2) {
1256 int new = G2.heap[++G2.heap_len] = (max_code < 2 ? ++max_code : 0);
1262 G2.static_len -= stree[new].Len;
1263 /* new is 0 or 1 so it does not have extra bits */
1265 desc->max_code = max_code;
1267 /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree,
1268 * establish sub-heaps of increasing lengths:
1270 for (n = G2.heap_len / 2; n >= 1; n--)
1271 pqdownheap(tree, n);
1273 /* Construct the Huffman tree by repeatedly combining the least two
1277 PQREMOVE(tree, n); /* n = node of least frequency */
1278 m = G2.heap[SMALLEST]; /* m = node of next least frequency */
1280 G2.heap[--G2.heap_max] = n; /* keep the nodes sorted by frequency */
1281 G2.heap[--G2.heap_max] = m;
1283 /* Create a new node father of n and m */
1284 tree[node].Freq = tree[n].Freq + tree[m].Freq;
1285 G2.depth[node] = MAX(G2.depth[n], G2.depth[m]) + 1;
1286 tree[n].Dad = tree[m].Dad = (ush) node;
1288 if (tree == G2.bl_tree) {
1289 bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)",
1290 node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq);
1293 /* and insert the new node in the heap */
1294 G2.heap[SMALLEST] = node++;
1295 pqdownheap(tree, SMALLEST);
1297 } while (G2.heap_len >= 2);
1299 G2.heap[--G2.heap_max] = G2.heap[SMALLEST];
1301 /* At this point, the fields freq and dad are set. We can now
1302 * generate the bit lengths.
1304 gen_bitlen((tree_desc *) desc);
1306 /* The field len is now set, we can generate the bit codes */
1307 gen_codes((ct_data *) tree, max_code);
1311 /* ===========================================================================
1312 * Scan a literal or distance tree to determine the frequencies of the codes
1313 * in the bit length tree. Updates opt_len to take into account the repeat
1314 * counts. (The contribution of the bit length codes will be added later
1315 * during the construction of bl_tree.)
1317 static void scan_tree(ct_data * tree, int max_code)
1319 int n; /* iterates over all tree elements */
1320 int prevlen = -1; /* last emitted length */
1321 int curlen; /* length of current code */
1322 int nextlen = tree[0].Len; /* length of next code */
1323 int count = 0; /* repeat count of the current code */
1324 int max_count = 7; /* max repeat count */
1325 int min_count = 4; /* min repeat count */
1331 tree[max_code + 1].Len = 0xffff; /* guard */
1333 for (n = 0; n <= max_code; n++) {
1335 nextlen = tree[n + 1].Len;
1336 if (++count < max_count && curlen == nextlen)
1339 if (count < min_count) {
1340 G2.bl_tree[curlen].Freq += count;
1341 } else if (curlen != 0) {
1342 if (curlen != prevlen)
1343 G2.bl_tree[curlen].Freq++;
1344 G2.bl_tree[REP_3_6].Freq++;
1345 } else if (count <= 10) {
1346 G2.bl_tree[REPZ_3_10].Freq++;
1348 G2.bl_tree[REPZ_11_138].Freq++;
1358 } else if (curlen == nextlen) {
1366 /* ===========================================================================
1367 * Send a literal or distance tree in compressed form, using the codes in
1370 static void send_tree(ct_data * tree, int max_code)
1372 int n; /* iterates over all tree elements */
1373 int prevlen = -1; /* last emitted length */
1374 int curlen; /* length of current code */
1375 int nextlen = tree[0].Len; /* length of next code */
1376 int count = 0; /* repeat count of the current code */
1377 int max_count = 7; /* max repeat count */
1378 int min_count = 4; /* min repeat count */
1380 /* tree[max_code+1].Len = -1; *//* guard already set */
1382 max_count = 138, min_count = 3;
1384 for (n = 0; n <= max_code; n++) {
1386 nextlen = tree[n + 1].Len;
1387 if (++count < max_count && curlen == nextlen) {
1389 } else if (count < min_count) {
1391 SEND_CODE(curlen, G2.bl_tree);
1393 } else if (curlen != 0) {
1394 if (curlen != prevlen) {
1395 SEND_CODE(curlen, G2.bl_tree);
1398 Assert(count >= 3 && count <= 6, " 3_6?");
1399 SEND_CODE(REP_3_6, G2.bl_tree);
1400 send_bits(count - 3, 2);
1401 } else if (count <= 10) {
1402 SEND_CODE(REPZ_3_10, G2.bl_tree);
1403 send_bits(count - 3, 3);
1405 SEND_CODE(REPZ_11_138, G2.bl_tree);
1406 send_bits(count - 11, 7);
1413 } else if (curlen == nextlen) {
1424 /* ===========================================================================
1425 * Construct the Huffman tree for the bit lengths and return the index in
1426 * bl_order of the last bit length code to send.
1428 static int build_bl_tree(void)
1430 int max_blindex; /* index of last bit length code of non zero freq */
1432 /* Determine the bit length frequencies for literal and distance trees */
1433 scan_tree(G2.dyn_ltree, G2.l_desc.max_code);
1434 scan_tree(G2.dyn_dtree, G2.d_desc.max_code);
1436 /* Build the bit length tree: */
1437 build_tree(&G2.bl_desc);
1438 /* opt_len now includes the length of the tree representations, except
1439 * the lengths of the bit lengths codes and the 5+5+4 bits for the counts.
1442 /* Determine the number of bit length codes to send. The pkzip format
1443 * requires that at least 4 bit length codes be sent. (appnote.txt says
1444 * 3 but the actual value used is 4.)
1446 for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) {
1447 if (G2.bl_tree[bl_order[max_blindex]].Len != 0)
1450 /* Update opt_len to include the bit length tree and counts */
1451 G2.opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4;
1452 Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", G2.opt_len, G2.static_len));
1458 /* ===========================================================================
1459 * Send the header for a block using dynamic Huffman trees: the counts, the
1460 * lengths of the bit length codes, the literal tree and the distance tree.
1461 * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4.
1463 static void send_all_trees(int lcodes, int dcodes, int blcodes)
1465 int rank; /* index in bl_order */
1467 Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes");
1468 Assert(lcodes <= L_CODES && dcodes <= D_CODES
1469 && blcodes <= BL_CODES, "too many codes");
1470 Tracev((stderr, "\nbl counts: "));
1471 send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */
1472 send_bits(dcodes - 1, 5);
1473 send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */
1474 for (rank = 0; rank < blcodes; rank++) {
1475 Tracev((stderr, "\nbl code %2d ", bl_order[rank]));
1476 send_bits(G2.bl_tree[bl_order[rank]].Len, 3);
1478 Tracev((stderr, "\nbl tree: sent %ld", G1.bits_sent));
1480 send_tree((ct_data *) G2.dyn_ltree, lcodes - 1); /* send the literal tree */
1481 Tracev((stderr, "\nlit tree: sent %ld", G1.bits_sent));
1483 send_tree((ct_data *) G2.dyn_dtree, dcodes - 1); /* send the distance tree */
1484 Tracev((stderr, "\ndist tree: sent %ld", G1.bits_sent));
1488 /* ===========================================================================
1489 * Save the match info and tally the frequency counts. Return true if
1490 * the current block must be flushed.
1492 static int ct_tally(int dist, int lc)
1494 G1.l_buf[G2.last_lit++] = lc;
1496 /* lc is the unmatched char */
1497 G2.dyn_ltree[lc].Freq++;
1499 /* Here, lc is the match length - MIN_MATCH */
1500 dist--; /* dist = match distance - 1 */
1501 Assert((ush) dist < (ush) MAX_DIST
1502 && (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH)
1503 && (ush) D_CODE(dist) < (ush) D_CODES, "ct_tally: bad match"
1506 G2.dyn_ltree[G2.length_code[lc] + LITERALS + 1].Freq++;
1507 G2.dyn_dtree[D_CODE(dist)].Freq++;
1509 G1.d_buf[G2.last_dist++] = dist;
1510 G2.flags |= G2.flag_bit;
1514 /* Output the flags if they fill a byte: */
1515 if ((G2.last_lit & 7) == 0) {
1516 G2.flag_buf[G2.last_flags++] = G2.flags;
1520 /* Try to guess if it is profitable to stop the current block here */
1521 if ((G2.last_lit & 0xfff) == 0) {
1522 /* Compute an upper bound for the compressed length */
1523 ulg out_length = G2.last_lit * 8L;
1524 ulg in_length = (ulg) G1.strstart - G1.block_start;
1527 for (dcode = 0; dcode < D_CODES; dcode++) {
1528 out_length += G2.dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]);
1532 "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ",
1533 G2.last_lit, G2.last_dist, in_length, out_length,
1534 100L - out_length * 100L / in_length));
1535 if (G2.last_dist < G2.last_lit / 2 && out_length < in_length / 2)
1538 return (G2.last_lit == LIT_BUFSIZE - 1 || G2.last_dist == DIST_BUFSIZE);
1539 /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K
1540 * on 16 bit machines and because stored blocks are restricted to
1545 /* ===========================================================================
1546 * Send the block data compressed using the given Huffman trees
1548 static void compress_block(ct_data * ltree, ct_data * dtree)
1550 unsigned dist; /* distance of matched string */
1551 int lc; /* match length or unmatched char (if dist == 0) */
1552 unsigned lx = 0; /* running index in l_buf */
1553 unsigned dx = 0; /* running index in d_buf */
1554 unsigned fx = 0; /* running index in flag_buf */
1555 uch flag = 0; /* current flags */
1556 unsigned code; /* the code to send */
1557 int extra; /* number of extra bits to send */
1559 if (G2.last_lit != 0) do {
1561 flag = G2.flag_buf[fx++];
1562 lc = G1.l_buf[lx++];
1563 if ((flag & 1) == 0) {
1564 SEND_CODE(lc, ltree); /* send a literal byte */
1565 Tracecv(lc > ' ', (stderr, " '%c' ", lc));
1567 /* Here, lc is the match length - MIN_MATCH */
1568 code = G2.length_code[lc];
1569 SEND_CODE(code + LITERALS + 1, ltree); /* send the length code */
1570 extra = extra_lbits[code];
1572 lc -= G2.base_length[code];
1573 send_bits(lc, extra); /* send the extra length bits */
1575 dist = G1.d_buf[dx++];
1576 /* Here, dist is the match distance - 1 */
1577 code = D_CODE(dist);
1578 Assert(code < D_CODES, "bad d_code");
1580 SEND_CODE(code, dtree); /* send the distance code */
1581 extra = extra_dbits[code];
1583 dist -= G2.base_dist[code];
1584 send_bits(dist, extra); /* send the extra distance bits */
1586 } /* literal or match pair ? */
1588 } while (lx < G2.last_lit);
1590 SEND_CODE(END_BLOCK, ltree);
1594 /* ===========================================================================
1595 * Determine the best encoding for the current block: dynamic trees, static
1596 * trees or store, and output the encoded block to the zip file. This function
1597 * returns the total compressed length for the file so far.
1599 static ulg flush_block(char *buf, ulg stored_len, int eof)
1601 ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
1602 int max_blindex; /* index of last bit length code of non zero freq */
1604 G2.flag_buf[G2.last_flags] = G2.flags; /* Save the flags for the last 8 items */
1606 /* Construct the literal and distance trees */
1607 build_tree(&G2.l_desc);
1608 Tracev((stderr, "\nlit data: dyn %ld, stat %ld", G2.opt_len, G2.static_len));
1610 build_tree(&G2.d_desc);
1611 Tracev((stderr, "\ndist data: dyn %ld, stat %ld", G2.opt_len, G2.static_len));
1612 /* At this point, opt_len and static_len are the total bit lengths of
1613 * the compressed block data, excluding the tree representations.
1616 /* Build the bit length tree for the above two trees, and get the index
1617 * in bl_order of the last bit length code to send.
1619 max_blindex = build_bl_tree();
1621 /* Determine the best encoding. Compute first the block length in bytes */
1622 opt_lenb = (G2.opt_len + 3 + 7) >> 3;
1623 static_lenb = (G2.static_len + 3 + 7) >> 3;
1626 "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ",
1627 opt_lenb, G2.opt_len, static_lenb, G2.static_len, stored_len,
1628 G2.last_lit, G2.last_dist));
1630 if (static_lenb <= opt_lenb)
1631 opt_lenb = static_lenb;
1633 /* If compression failed and this is the first and last block,
1634 * and if the zip file can be seeked (to rewrite the local header),
1635 * the whole file is transformed into a stored file:
1637 if (stored_len <= opt_lenb && eof && G2.compressed_len == 0L && seekable()) {
1638 /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */
1640 bb_error_msg("block vanished");
1642 copy_block(buf, (unsigned) stored_len, 0); /* without header */
1643 G2.compressed_len = stored_len << 3;
1645 } else if (stored_len + 4 <= opt_lenb && buf != NULL) {
1646 /* 4: two words for the lengths */
1647 /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE.
1648 * Otherwise we can't have processed more than WSIZE input bytes since
1649 * the last block flush, because compression would have been
1650 * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to
1651 * transform a block into a stored block.
1653 send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */
1654 G2.compressed_len = (G2.compressed_len + 3 + 7) & ~7L;
1655 G2.compressed_len += (stored_len + 4) << 3;
1657 copy_block(buf, (unsigned) stored_len, 1); /* with header */
1659 } else if (static_lenb == opt_lenb) {
1660 send_bits((STATIC_TREES << 1) + eof, 3);
1661 compress_block((ct_data *) G2.static_ltree, (ct_data *) G2.static_dtree);
1662 G2.compressed_len += 3 + G2.static_len;
1664 send_bits((DYN_TREES << 1) + eof, 3);
1665 send_all_trees(G2.l_desc.max_code + 1, G2.d_desc.max_code + 1,
1667 compress_block((ct_data *) G2.dyn_ltree, (ct_data *) G2.dyn_dtree);
1668 G2.compressed_len += 3 + G2.opt_len;
1670 Assert(G2.compressed_len == G1.bits_sent, "bad compressed size");
1675 G2.compressed_len += 7; /* align on byte boundary */
1677 Tracev((stderr, "\ncomprlen %lu(%lu) ", G2.compressed_len >> 3,
1678 G2.compressed_len - 7 * eof));
1680 return G2.compressed_len >> 3;
1684 /* ===========================================================================
1685 * Update a hash value with the given input byte
1686 * IN assertion: all calls to UPDATE_HASH are made with consecutive
1687 * input characters, so that a running hash key can be computed from the
1688 * previous key instead of complete recalculation each time.
1690 #define UPDATE_HASH(h, c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
1693 /* ===========================================================================
1694 * Same as above, but achieves better compression. We use a lazy
1695 * evaluation for matches: a match is finally adopted only if there is
1696 * no better match at the next window position.
1698 * Processes a new input file and return its compressed length. Sets
1699 * the compressed length, crc, deflate flags and internal file
1703 /* Flush the current block, with given end-of-file flag.
1704 * IN assertion: strstart is set to the end of the current match. */
1705 #define FLUSH_BLOCK(eof) \
1707 G1.block_start >= 0L \
1708 ? (char*)&G1.window[(unsigned)G1.block_start] \
1710 (ulg)G1.strstart - G1.block_start, \
1714 /* Insert string s in the dictionary and set match_head to the previous head
1715 * of the hash chain (the most recent string with same hash key). Return
1716 * the previous length of the hash chain.
1717 * IN assertion: all calls to INSERT_STRING are made with consecutive
1718 * input characters and the first MIN_MATCH bytes of s are valid
1719 * (except for the last MIN_MATCH-1 bytes of the input file). */
1720 #define INSERT_STRING(s, match_head) \
1722 UPDATE_HASH(G1.ins_h, G1.window[(s) + MIN_MATCH-1]); \
1723 G1.prev[(s) & WMASK] = match_head = head[G1.ins_h]; \
1724 head[G1.ins_h] = (s); \
1727 static ulg deflate(void)
1729 IPos hash_head; /* head of hash chain */
1730 IPos prev_match; /* previous match */
1731 int flush; /* set if current block must be flushed */
1732 int match_available = 0; /* set if previous match exists */
1733 unsigned match_length = MIN_MATCH - 1; /* length of best match */
1735 /* Process the input block. */
1736 while (G1.lookahead != 0) {
1737 /* Insert the string window[strstart .. strstart+2] in the
1738 * dictionary, and set hash_head to the head of the hash chain:
1740 INSERT_STRING(G1.strstart, hash_head);
1742 /* Find the longest match, discarding those <= prev_length.
1744 G1.prev_length = match_length;
1745 prev_match = G1.match_start;
1746 match_length = MIN_MATCH - 1;
1748 if (hash_head != 0 && G1.prev_length < max_lazy_match
1749 && G1.strstart - hash_head <= MAX_DIST
1751 /* To simplify the code, we prevent matches with the string
1752 * of window index 0 (in particular we have to avoid a match
1753 * of the string with itself at the start of the input file).
1755 match_length = longest_match(hash_head);
1756 /* longest_match() sets match_start */
1757 if (match_length > G1.lookahead)
1758 match_length = G1.lookahead;
1760 /* Ignore a length 3 match if it is too distant: */
1761 if (match_length == MIN_MATCH && G1.strstart - G1.match_start > TOO_FAR) {
1762 /* If prev_match is also MIN_MATCH, G1.match_start is garbage
1763 * but we will ignore the current match anyway.
1768 /* If there was a match at the previous step and the current
1769 * match is not better, output the previous match:
1771 if (G1.prev_length >= MIN_MATCH && match_length <= G1.prev_length) {
1772 check_match(G1.strstart - 1, prev_match, G1.prev_length);
1773 flush = ct_tally(G1.strstart - 1 - prev_match, G1.prev_length - MIN_MATCH);
1775 /* Insert in hash table all strings up to the end of the match.
1776 * strstart-1 and strstart are already inserted.
1778 G1.lookahead -= G1.prev_length - 1;
1779 G1.prev_length -= 2;
1782 INSERT_STRING(G1.strstart, hash_head);
1783 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1784 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
1785 * these bytes are garbage, but it does not matter since the
1786 * next lookahead bytes will always be emitted as literals.
1788 } while (--G1.prev_length != 0);
1789 match_available = 0;
1790 match_length = MIN_MATCH - 1;
1794 G1.block_start = G1.strstart;
1796 } else if (match_available) {
1797 /* If there was no match at the previous position, output a
1798 * single literal. If there was a match but the current match
1799 * is longer, truncate the previous match to a single literal.
1801 Tracevv((stderr, "%c", G1.window[G1.strstart - 1]));
1802 if (ct_tally(0, G1.window[G1.strstart - 1])) {
1804 G1.block_start = G1.strstart;
1809 /* There is no previous match to compare with, wait for
1810 * the next step to decide.
1812 match_available = 1;
1816 Assert(G1.strstart <= G1.isize && lookahead <= G1.isize, "a bit too far");
1818 /* Make sure that we always have enough lookahead, except
1819 * at the end of the input file. We need MAX_MATCH bytes
1820 * for the next match, plus MIN_MATCH bytes to insert the
1821 * string following the next match.
1823 while (G1.lookahead < MIN_LOOKAHEAD && !G1.eofile)
1826 if (match_available)
1827 ct_tally(0, G1.window[G1.strstart - 1]);
1829 return FLUSH_BLOCK(1); /* eof */
1833 /* ===========================================================================
1834 * Initialize the bit string routines.
1836 static void bi_init(void)
1846 /* ===========================================================================
1847 * Initialize the "longest match" routines for a new file
1849 static void lm_init(ush * flagsp)
1853 /* Initialize the hash table. */
1854 memset(head, 0, HASH_SIZE * sizeof(*head));
1855 /* prev will be initialized on the fly */
1857 /* speed options for the general purpose bit flag */
1858 *flagsp |= 2; /* FAST 4, SLOW 2 */
1859 /* ??? reduce max_chain_length for binary files */
1862 G1.block_start = 0L;
1864 G1.lookahead = file_read(G1.window,
1865 sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE);
1867 if (G1.lookahead == 0 || G1.lookahead == (unsigned) -1) {
1873 /* Make sure that we always have enough lookahead. This is important
1874 * if input comes from a device such as a tty.
1876 while (G1.lookahead < MIN_LOOKAHEAD && !G1.eofile)
1880 for (j = 0; j < MIN_MATCH - 1; j++)
1881 UPDATE_HASH(G1.ins_h, G1.window[j]);
1882 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
1883 * not important since only literal bytes will be emitted.
1888 /* ===========================================================================
1889 * Allocate the match buffer, initialize the various tables and save the
1890 * location of the internal file attribute (ascii/binary) and method
1892 * One callsite in zip()
1894 static void ct_init(void)
1896 int n; /* iterates over tree elements */
1897 int length; /* length value */
1898 int code; /* code value */
1899 int dist; /* distance index */
1901 G2.compressed_len = 0L;
1904 if (G2.static_dtree[0].Len != 0)
1905 return; /* ct_init already called */
1908 /* Initialize the mapping length (0..255) -> length code (0..28) */
1910 for (code = 0; code < LENGTH_CODES - 1; code++) {
1911 G2.base_length[code] = length;
1912 for (n = 0; n < (1 << extra_lbits[code]); n++) {
1913 G2.length_code[length++] = code;
1916 Assert(length == 256, "ct_init: length != 256");
1917 /* Note that the length 255 (match length 258) can be represented
1918 * in two different ways: code 284 + 5 bits or code 285, so we
1919 * overwrite length_code[255] to use the best encoding:
1921 G2.length_code[length - 1] = code;
1923 /* Initialize the mapping dist (0..32K) -> dist code (0..29) */
1925 for (code = 0; code < 16; code++) {
1926 G2.base_dist[code] = dist;
1927 for (n = 0; n < (1 << extra_dbits[code]); n++) {
1928 G2.dist_code[dist++] = code;
1931 Assert(dist == 256, "ct_init: dist != 256");
1932 dist >>= 7; /* from now on, all distances are divided by 128 */
1933 for (; code < D_CODES; code++) {
1934 G2.base_dist[code] = dist << 7;
1935 for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) {
1936 G2.dist_code[256 + dist++] = code;
1939 Assert(dist == 256, "ct_init: 256+dist != 512");
1941 /* Construct the codes of the static literal tree */
1942 /* already zeroed - it's in bss
1943 for (n = 0; n <= MAX_BITS; n++)
1944 G2.bl_count[n] = 0; */
1948 G2.static_ltree[n++].Len = 8;
1952 G2.static_ltree[n++].Len = 9;
1956 G2.static_ltree[n++].Len = 7;
1960 G2.static_ltree[n++].Len = 8;
1963 /* Codes 286 and 287 do not exist, but we must include them in the
1964 * tree construction to get a canonical Huffman tree (longest code
1967 gen_codes((ct_data *) G2.static_ltree, L_CODES + 1);
1969 /* The static distance tree is trivial: */
1970 for (n = 0; n < D_CODES; n++) {
1971 G2.static_dtree[n].Len = 5;
1972 G2.static_dtree[n].Code = bi_reverse(n, 5);
1975 /* Initialize the first block of the first file: */
1980 /* ===========================================================================
1981 * Deflate in to out.
1982 * IN assertions: the input and output buffers are cleared.
1985 static void zip(ulg time_stamp)
1987 ush deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */
1991 /* Write the header to the gzip file. See algorithm.doc for the format */
1992 /* magic header for gzip files: 1F 8B */
1993 /* compression method: 8 (DEFLATED) */
1994 /* general flags: 0 */
1995 put_32bit(0x00088b1f);
1996 put_32bit(time_stamp);
1998 /* Write deflated file to zip file */
2003 lm_init(&deflate_flags);
2005 put_8bit(deflate_flags); /* extra flags */
2006 put_8bit(3); /* OS identifier = 3 (Unix) */
2010 /* Write the crc and uncompressed size */
2012 put_32bit(G1.isize);
2018 /* ======================================================================== */
2020 IF_DESKTOP(long long) int FAST_FUNC pack_gzip(transformer_aux_data_t *aux UNUSED_PARAM)
2024 /* Clear input and output buffers */
2032 memset(&G2, 0, sizeof(G2));
2033 G2.l_desc.dyn_tree = G2.dyn_ltree;
2034 G2.l_desc.static_tree = G2.static_ltree;
2035 G2.l_desc.extra_bits = extra_lbits;
2036 G2.l_desc.extra_base = LITERALS + 1;
2037 G2.l_desc.elems = L_CODES;
2038 G2.l_desc.max_length = MAX_BITS;
2039 //G2.l_desc.max_code = 0;
2040 G2.d_desc.dyn_tree = G2.dyn_dtree;
2041 G2.d_desc.static_tree = G2.static_dtree;
2042 G2.d_desc.extra_bits = extra_dbits;
2043 //G2.d_desc.extra_base = 0;
2044 G2.d_desc.elems = D_CODES;
2045 G2.d_desc.max_length = MAX_BITS;
2046 //G2.d_desc.max_code = 0;
2047 G2.bl_desc.dyn_tree = G2.bl_tree;
2048 //G2.bl_desc.static_tree = NULL;
2049 G2.bl_desc.extra_bits = extra_blbits,
2050 //G2.bl_desc.extra_base = 0;
2051 G2.bl_desc.elems = BL_CODES;
2052 G2.bl_desc.max_length = MAX_BL_BITS;
2053 //G2.bl_desc.max_code = 0;
2056 fstat(STDIN_FILENO, &s);
2061 #if ENABLE_FEATURE_GZIP_LONG_OPTIONS
2062 static const char gzip_longopts[] ALIGN1 =
2063 "stdout\0" No_argument "c"
2064 "to-stdout\0" No_argument "c"
2065 "force\0" No_argument "f"
2066 "verbose\0" No_argument "v"
2068 "decompress\0" No_argument "d"
2069 "uncompress\0" No_argument "d"
2070 "test\0" No_argument "t"
2072 "quiet\0" No_argument "q"
2073 "fast\0" No_argument "1"
2074 "best\0" No_argument "9"
2079 * Linux kernel build uses gzip -d -n. We accept and ignore -n.
2082 * gzip: do not save the original file name and time stamp.
2083 * (The original name is always saved if the name had to be truncated.)
2084 * gunzip: do not restore the original file name/time even if present
2085 * (remove only the gzip suffix from the compressed file name).
2086 * This option is the default when decompressing.
2088 * gzip: always save the original file name and time stamp (this is the default)
2089 * gunzip: restore the original file name and time stamp if present.
2092 int gzip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
2094 int gzip_main(int argc, char **argv)
2096 int gzip_main(int argc UNUSED_PARAM, char **argv)
2101 #if ENABLE_FEATURE_GZIP_LONG_OPTIONS
2102 applet_long_options = gzip_longopts;
2104 /* Must match bbunzip's constants OPT_STDOUT, OPT_FORCE! */
2105 opt = getopt32(argv, "cfv" IF_GUNZIP("dt") "q123456789n");
2106 #if ENABLE_GUNZIP /* gunzip_main may not be visible... */
2107 if (opt & 0x18) // -d and/or -t
2108 return gunzip_main(argc, argv);
2110 option_mask32 &= 0x7; /* ignore -q, -0..9 */
2111 //if (opt & 0x1) // -c
2112 //if (opt & 0x2) // -f
2113 //if (opt & 0x4) // -v
2116 SET_PTR_TO_GLOBALS((char *)xzalloc(sizeof(struct globals)+sizeof(struct globals2))
2117 + sizeof(struct globals));
2119 /* Allocate all global buffers (for DYN_ALLOC option) */
2120 ALLOC(uch, G1.l_buf, INBUFSIZ);
2121 ALLOC(uch, G1.outbuf, OUTBUFSIZ);
2122 ALLOC(ush, G1.d_buf, DIST_BUFSIZE);
2123 ALLOC(uch, G1.window, 2L * WSIZE);
2124 ALLOC(ush, G1.prev, 1L << BITS);
2126 /* Initialize the CRC32 table */
2127 global_crc32_table = crc32_filltable(NULL, 0);
2129 return bbunpack(argv, pack_gzip, append_ext, "gz");