1 /* vi: set sw=4 ts=4: */
3 * Gzip implementation for busybox
5 * Based on GNU gzip Copyright (C) 1992-1993 Jean-loup Gailly.
7 * Originally adjusted for busybox by Charles P. Wright <cpw@unix.asb.com>
8 * "this is a stripped down version of gzip I put into busybox, it does
9 * only standard in to standard out with -9 compression. It also requires
10 * the zcat module for some important functions."
12 * Adjusted further by Erik Andersen <andersen@codepoet.org> to support
13 * files as well as stdin/stdout, and to generally behave itself wrt
14 * command line handling.
16 * Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
19 /* TODO: full support for -v for DESKTOP
20 /usr/bin/gzip -v a bogus aa
21 a: 85.1% -- replaced with a.gz
22 gzip: bogus: No such file or directory
23 aa: 85.1% -- replaced with aa.gz
31 /* ===========================================================================
34 /* Diagnostic functions */
36 # define Assert(cond,msg) {if(!(cond)) bb_error_msg(msg);}
37 # define Trace(x) fprintf x
38 # define Tracev(x) {if (verbose) fprintf x ;}
39 # define Tracevv(x) {if (verbose > 1) fprintf x ;}
40 # define Tracec(c,x) {if (verbose && (c)) fprintf x ;}
41 # define Tracecv(c,x) {if (verbose > 1 && (c)) fprintf x ;}
43 # define Assert(cond,msg)
52 /* ===========================================================================
56 /* Compression methods (see algorithm.doc) */
57 /* Only STORED and DEFLATED are supported by this BusyBox module */
59 /* methods 4 to 7 reserved */
64 # define INBUFSIZ 0x2000 /* input buffer size */
66 # define INBUFSIZ 0x8000 /* input buffer size */
70 #define INBUF_EXTRA 64 /* required by unlzw() */
74 # define OUTBUFSIZ 8192 /* output buffer size */
76 # define OUTBUFSIZ 16384 /* output buffer size */
79 #define OUTBUF_EXTRA 2048 /* required by unlzw() */
83 # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */
85 # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */
90 #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */
91 #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
92 #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
93 #define ORIG_NAME 0x08 /* bit 3 set: original file name present */
94 #define COMMENT 0x10 /* bit 4 set: file comment present */
95 #define RESERVED 0xC0 /* bit 6,7: reserved */
97 /* internal file attribute */
98 #define UNKNOWN 0xffff
103 # define WSIZE 0x8000 /* window size--must be a power of two, and */
104 #endif /* at least 32K for zip's deflate method */
107 #define MAX_MATCH 258
108 /* The minimum and maximum match lengths */
110 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
111 /* Minimum amount of lookahead, except at the end of the input file.
112 * See deflate.c for comments about the MIN_MATCH+1.
115 #define MAX_DIST (WSIZE-MIN_LOOKAHEAD)
116 /* In order to simplify the code, particularly on 16 bit machines, match
117 * distances are limited to MAX_DIST instead of WSIZE.
121 # define MAX_PATH_LEN 1024 /* max pathname length */
124 #define seekable() 0 /* force sequential output */
125 #define translate_eol 0 /* no option -a yet */
130 #define INIT_BITS 9 /* Initial number of bits per code */
132 #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */
133 /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free.
134 * It's a pity that old uncompress does not check bit 0x20. That makes
135 * extension of the format actually undesirable because old compress
136 * would just crash on the new format instead of giving a meaningful
137 * error message. It does check the number of bits, but it's more
138 * helpful to say "unsupported format, get a new version" than
139 * "can only handle 16 bits".
143 # define MAX_SUFFIX MAX_EXT_CHARS
145 # define MAX_SUFFIX 30
149 /* ===========================================================================
150 * Compile with MEDIUM_MEM to reduce the memory requirements or
151 * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the
152 * entire input file can be held in memory (not possible on 16 bit systems).
153 * Warning: defining these symbols affects HASH_BITS (see below) and thus
154 * affects the compression ratio. The compressed output
155 * is still correct, and might even be smaller in some cases.
159 # define HASH_BITS 13 /* Number of bits used to hash strings */
162 # define HASH_BITS 14
165 # define HASH_BITS 15
166 /* For portability to 16 bit machines, do not use values above 15. */
169 /* To save space (see unlzw.c), we overlay prev+head with tab_prefix and
170 * window with tab_suffix. Check that we can do this:
172 #if (WSIZE<<1) > (1<<BITS)
173 # error cannot overlay window with tab_suffix and prev with tab_prefix0
175 #if HASH_BITS > BITS-1
176 # error cannot overlay head with tab_prefix1
178 #define HASH_SIZE (unsigned)(1<<HASH_BITS)
179 #define HASH_MASK (HASH_SIZE-1)
180 #define WMASK (WSIZE-1)
181 /* HASH_SIZE and WSIZE must be powers of two */
183 # define TOO_FAR 4096
185 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
188 /* ===========================================================================
190 typedef unsigned char uch;
191 typedef unsigned short ush;
192 typedef unsigned long ulg;
195 /* ===========================================================================
196 * Local data used by the "longest match" routines.
199 typedef unsigned IPos;
201 /* A Pos is an index in the character window. We use short instead of int to
202 * save space in the various tables. IPos is used only for parameter passing.
205 #define DECLARE(type, array, size)\
207 #define ALLOC(type, array, size) { \
208 array = xzalloc((size_t)(((size)+1L)/2) * 2*sizeof(type)); \
211 #define FREE(array) { \
216 /* DECLARE(uch, window, 2L*WSIZE); */
217 /* Sliding window. Input bytes are read into the second half of the window,
218 * and move to the first half later to keep a dictionary of at least WSIZE
219 * bytes. With this organization, matches are limited to a distance of
220 * WSIZE-MAX_MATCH bytes, but this ensures that IO is always
221 * performed with a length multiple of the block size. Also, it limits
222 * the window size to 64K, which is quite useful on MSDOS.
223 * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would
224 * be less efficient).
227 /* DECLARE(Pos, prev, WSIZE); */
228 /* Link to older string with same hash index. To limit the size of this
229 * array to 64K, this link is maintained only for the last 32K strings.
230 * An index in this array is thus a window index modulo 32K.
233 /* DECLARE(Pos, head, 1<<HASH_BITS); */
234 /* Heads of the hash chains or 0. */
236 static long block_start;
238 /* window position at the beginning of the current output block. Gets
239 * negative when the window is moved backwards.
242 static unsigned ins_h; /* hash index of string to be inserted */
244 #define H_SHIFT ((HASH_BITS+MIN_MATCH-1)/MIN_MATCH)
245 /* Number of bits by which ins_h and del_h must be shifted at each
246 * input step. It must be such that after MIN_MATCH steps, the oldest
247 * byte no longer takes part in the hash key, that is:
248 * H_SHIFT * MIN_MATCH >= HASH_BITS
251 static unsigned int prev_length;
253 /* Length of the best match at previous step. Matches not greater than this
254 * are discarded. This is used in the lazy match evaluation.
257 static unsigned strstart; /* start of string to insert */
258 static unsigned match_start; /* start of matching string */
259 static int eofile; /* flag set at end of input file */
260 static unsigned lookahead; /* number of valid bytes ahead in window */
263 WINDOW_SIZE = 2 * WSIZE,
264 /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the
265 * input file length plus MIN_LOOKAHEAD.
268 max_chain_length = 4096,
269 /* To speed up deflation, hash chains are never searched beyond this length.
270 * A higher limit improves compression ratio but degrades the speed.
273 max_lazy_match = 258,
274 /* Attempt to find a better match only when the current match is strictly
275 * smaller than this value. This mechanism is used only for compression
279 max_insert_length = max_lazy_match,
280 /* Insert new strings in the hash table only if the match length
281 * is not greater than this length. This saves time but degrades compression.
282 * max_insert_length is used only for compression levels <= 3.
286 /* Use a faster search when the previous match is longer than this */
288 /* Values for max_lazy_match, good_match and max_chain_length, depending on
289 * the desired pack level (0..9). The values given below have been tuned to
290 * exclude worst case performance for pathological files. Better values may be
291 * found for specific files.
294 nice_match = 258 /* Stop searching when current match exceeds this */
295 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
296 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
302 /* ===========================================================================
303 * Prototypes for local functions.
305 static void fill_window(void);
307 static int longest_match(IPos cur_match);
310 static void check_match(IPos start, IPos match, int length);
315 static void lm_init(ush * flags);
316 static ulg deflate(void);
319 static void ct_init(ush * attr, int *methodp);
320 static int ct_tally(int dist, int lc);
321 static ulg flush_block(char *buf, ulg stored_len, int eof);
324 static void bi_init(int zipfile);
325 static void send_bits(int value, int length);
326 static unsigned bi_reverse(unsigned value, int length);
327 static void bi_windup(void);
328 static void copy_block(char *buf, unsigned len, int header);
332 /* To save memory for 16 bit systems, some arrays are overlaid between
333 * the various modules:
334 * deflate: prev+head window d_buf l_buf outbuf
335 * unlzw: tab_prefix tab_suffix stack inbuf outbuf
336 * For compression, input is done in window[]. For decompression, output
337 * is done in window except for unlzw.
340 #define tab_suffix window
341 #define tab_prefix prev /* hash link (see deflate.c) */
342 #define head (prev+WSIZE) /* hash head (see deflate.c) */
344 DECLARE(uch, inbuf, INBUFSIZ + INBUF_EXTRA);
345 DECLARE(uch, outbuf, OUTBUFSIZ + OUTBUF_EXTRA);
346 DECLARE(ush, d_buf, DIST_BUFSIZE);
347 DECLARE(uch, window, 2L * WSIZE);
348 DECLARE(ush, tab_prefix, 1L << BITS);
350 static long isize; /* number of input bytes */
352 static int foreground; /* set if program run in foreground */
353 static int method = DEFLATED; /* compression method */
354 static int exit_code; /* program exit code */
355 static long time_stamp; /* original time stamp (modification time) */
356 static char z_suffix[MAX_SUFFIX + 1]; /* default suffix (can be set with --suffix) */
358 static int ifd; /* input file descriptor */
359 static int ofd; /* output file descriptor */
361 static unsigned insize; /* valid bytes in inbuf */
363 static unsigned outcnt; /* bytes in output buffer */
365 static uint32_t *crc_32_tab;
368 /* ===========================================================================
369 * Local data used by the "bit string" routines.
372 static int zfile; /* output gzip file */
374 static unsigned short bi_buf;
376 /* Output buffer. bits are inserted starting at the bottom (least significant
381 #define BUF_SIZE (8 * sizeof(bi_buf))
382 /* Number of bits used within bi_buf. (bi_buf might be implemented on
383 * more than 16 bits on some systems.)
388 /* Current input function. Set to mem_read for in-memory compression */
391 static ulg bits_sent; /* bit length of the compressed data */
395 /* ===========================================================================
396 * Write the output buffer outbuf[0..outcnt-1] and update bytes_out.
397 * (used for the compressed data only)
399 static void flush_outbuf(void)
404 xwrite(ofd, (char *) outbuf, outcnt);
409 /* ===========================================================================
411 /* put_8bit is used for the compressed output */
412 #define put_8bit(c) \
414 outbuf[outcnt++] = (c); \
415 if (outcnt == OUTBUFSIZ) flush_outbuf(); \
418 /* Output a 16 bit value, lsb first */
419 static void put_16bit(ush w)
421 if (outcnt < OUTBUFSIZ - 2) {
422 outbuf[outcnt++] = w;
423 outbuf[outcnt++] = w >> 8;
430 static void put_32bit(ulg n)
436 /* put_header_byte is used for the compressed output
437 * - for the initial 4 bytes that can't overflow the buffer.
439 #define put_header_byte(c) \
441 outbuf[outcnt++] = (c); \
445 /* ===========================================================================
446 * Clear input and output buffers
448 static void clear_bufs(void)
458 /* ===========================================================================
459 * Run a set of bytes through the crc shift register. If s is a NULL
460 * pointer, then initialize the crc shift register contents instead.
461 * Return the current crc in either case.
463 static uint32_t crc; /* shift register contents */
464 static uint32_t updcrc(uch * s, unsigned n)
468 c = crc_32_tab[(uch)(c ^ *s++)] ^ (c >> 8);
476 /* ===========================================================================
477 * Read a new buffer from the current input file, perform end-of-line
478 * translation, and update the crc and input file size.
479 * IN assertion: size >= 2 (for end-of-line translation)
481 static unsigned file_read(void *buf, unsigned size)
485 Assert(insize == 0, "inbuf not empty");
487 len = safe_read(ifd, buf, size);
488 if (len == (unsigned)(-1) || len == 0)
497 /* ===========================================================================
498 * Initialize the bit string routines.
500 static void bi_init(int zipfile)
511 /* ===========================================================================
512 * Send a value on a given number of bits.
513 * IN assertion: length <= 16 and value fits in length bits.
515 static void send_bits(int value, int length)
518 Tracev((stderr, " l %2d v %4x ", length, value));
519 Assert(length > 0 && length <= 15, "invalid length");
522 /* If not enough room in bi_buf, use (valid) bits from bi_buf and
523 * (16 - bi_valid) bits from value, leaving (width - (16-bi_valid))
524 * unused bits in value.
526 if (bi_valid > (int) BUF_SIZE - length) {
527 bi_buf |= (value << bi_valid);
529 bi_buf = (ush) value >> (BUF_SIZE - bi_valid);
530 bi_valid += length - BUF_SIZE;
532 bi_buf |= value << bi_valid;
538 /* ===========================================================================
539 * Reverse the first len bits of a code, using straightforward code (a faster
540 * method would use a table)
541 * IN assertion: 1 <= len <= 15
543 static unsigned bi_reverse(unsigned code, int len)
549 code >>= 1, res <<= 1;
555 /* ===========================================================================
556 * Write out any remaining bits in an incomplete byte.
558 static void bi_windup(void)
562 } else if (bi_valid > 0) {
568 bits_sent = (bits_sent + 7) & ~7;
573 /* ===========================================================================
574 * Copy a stored block to the zip file, storing first the length and its
575 * one's complement if requested.
577 static void copy_block(char *buf, unsigned len, int header)
579 bi_windup(); /* align on byte boundary */
589 bits_sent += (ulg) len << 3;
597 /* ===========================================================================
598 * Update a hash value with the given input byte
599 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
600 * input characters, so that a running hash key can be computed from the
601 * previous key instead of complete recalculation each time.
603 #define UPDATE_HASH(h, c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK)
606 /* ===========================================================================
607 * Initialize the "longest match" routines for a new file
609 static void lm_init(ush * flags)
613 /* Initialize the hash table. */
614 memset(head, 0, HASH_SIZE * sizeof(*head));
615 /* prev will be initialized on the fly */
617 /*speed options for the general purpose bit flag */
618 *flags |= 2; /* FAST 4, SLOW 2 */
619 /* ??? reduce max_chain_length for binary files */
624 lookahead = file_read(window,
625 sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE);
627 if (lookahead == 0 || lookahead == (unsigned) -1) {
633 /* Make sure that we always have enough lookahead. This is important
634 * if input comes from a device such as a tty.
636 while (lookahead < MIN_LOOKAHEAD && !eofile)
640 for (j = 0; j < MIN_MATCH - 1; j++)
641 UPDATE_HASH(ins_h, window[j]);
642 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
643 * not important since only literal bytes will be emitted.
647 /* ===========================================================================
648 * Set match_start to the longest match starting at the given string and
649 * return its length. Matches shorter or equal to prev_length are discarded,
650 * in which case the result is equal to prev_length and match_start is
652 * IN assertions: cur_match is the head of the hash chain for the current
653 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
656 /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or
657 * match.s. The code is functionally equivalent, so you can use the C version
660 static int longest_match(IPos cur_match)
662 unsigned chain_length = max_chain_length; /* max hash chain length */
663 uch *scan = window + strstart; /* current string */
664 uch *match; /* matched string */
665 int len; /* length of current match */
666 int best_len = prev_length; /* best match length so far */
667 IPos limit = strstart > (IPos) MAX_DIST ? strstart - (IPos) MAX_DIST : 0;
668 /* Stop when cur_match becomes <= limit. To simplify the code,
669 * we prevent matches with the string of window index 0.
672 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
673 * It is easy to get rid of this optimization if necessary.
675 #if HASH_BITS < 8 || MAX_MATCH != 258
676 # error Code too clever
678 uch *strend = window + strstart + MAX_MATCH;
679 uch scan_end1 = scan[best_len - 1];
680 uch scan_end = scan[best_len];
682 /* Do not waste too much time if we already have a good match: */
683 if (prev_length >= good_match) {
686 Assert(strstart <= WINDOW_SIZE - MIN_LOOKAHEAD, "insufficient lookahead");
689 Assert(cur_match < strstart, "no future");
690 match = window + cur_match;
692 /* Skip to next match if the match length cannot increase
693 * or if the match length is less than 2:
695 if (match[best_len] != scan_end ||
696 match[best_len - 1] != scan_end1 ||
697 *match != *scan || *++match != scan[1])
700 /* The check at best_len-1 can be removed because it will be made
701 * again later. (This heuristic is not always a win.)
702 * It is not necessary to compare scan[2] and match[2] since they
703 * are always equal when the other bytes match, given that
704 * the hash keys are equal and that HASH_BITS >= 8.
708 /* We check for insufficient lookahead only every 8th comparison;
709 * the 256th check will be made at strstart+258.
712 } while (*++scan == *++match && *++scan == *++match &&
713 *++scan == *++match && *++scan == *++match &&
714 *++scan == *++match && *++scan == *++match &&
715 *++scan == *++match && *++scan == *++match && scan < strend);
717 len = MAX_MATCH - (int) (strend - scan);
718 scan = strend - MAX_MATCH;
720 if (len > best_len) {
721 match_start = cur_match;
723 if (len >= nice_match)
725 scan_end1 = scan[best_len - 1];
726 scan_end = scan[best_len];
728 } while ((cur_match = prev[cur_match & WMASK]) > limit
729 && --chain_length != 0);
736 /* ===========================================================================
737 * Check that the match at match_start is indeed a match.
739 static void check_match(IPos start, IPos match, int length)
741 /* check that the match is indeed a match */
742 if (memcmp(window + match, window + start, length) != 0) {
743 bb_error_msg(" start %d, match %d, length %d", start, match, length);
744 bb_error_msg("invalid match");
747 bb_error_msg("\\[%d,%d]", start - match, length);
749 putc(window[start++], stderr);
750 } while (--length != 0);
754 # define check_match(start, match, length) ((void)0)
758 /* ===========================================================================
759 * Fill the window when the lookahead becomes insufficient.
760 * Updates strstart and lookahead, and sets eofile if end of input file.
761 * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0
762 * OUT assertions: at least one byte has been read, or eofile is set;
763 * file reads are performed for at least two bytes (required for the
764 * translate_eol option).
766 static void fill_window(void)
769 unsigned more = WINDOW_SIZE - lookahead - strstart;
770 /* Amount of free space at the end of the window. */
772 /* If the window is almost full and there is insufficient lookahead,
773 * move the upper half to the lower one to make room in the upper half.
775 if (more == (unsigned) -1) {
776 /* Very unlikely, but possible on 16 bit machine if strstart == 0
777 * and lookahead == 1 (input done one byte at time)
780 } else if (strstart >= WSIZE + MAX_DIST) {
781 /* By the IN assertion, the window is not empty so we can't confuse
782 * more == 0 with more == 64K on a 16 bit machine.
784 Assert(WINDOW_SIZE == 2 * WSIZE, "no sliding with BIG_MEM");
786 memcpy(window, window + WSIZE, WSIZE);
787 match_start -= WSIZE;
788 strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
790 block_start -= WSIZE;
792 for (n = 0; n < HASH_SIZE; n++) {
794 head[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
796 for (n = 0; n < WSIZE; n++) {
798 prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0);
799 /* If n is not on any hash chain, prev[n] is garbage but
800 * its value will never be used.
805 /* At this point, more >= 2 */
807 n = file_read(window + strstart + lookahead, more);
808 if (n == 0 || n == (unsigned) -1) {
817 /* ===========================================================================
818 * Same as above, but achieves better compression. We use a lazy
819 * evaluation for matches: a match is finally adopted only if there is
820 * no better match at the next window position.
822 * Processes a new input file and return its compressed length. Sets
823 * the compressed length, crc, deflate flags and internal file
827 /* Flush the current block, with given end-of-file flag.
828 * IN assertion: strstart is set to the end of the current match. */
829 #define FLUSH_BLOCK(eof) \
832 ? (char*)&window[(unsigned)block_start] \
834 (long)strstart - block_start, \
838 /* Insert string s in the dictionary and set match_head to the previous head
839 * of the hash chain (the most recent string with same hash key). Return
840 * the previous length of the hash chain.
841 * IN assertion: all calls to to INSERT_STRING are made with consecutive
842 * input characters and the first MIN_MATCH bytes of s are valid
843 * (except for the last MIN_MATCH-1 bytes of the input file). */
844 #define INSERT_STRING(s, match_head) \
846 UPDATE_HASH(ins_h, window[(s) + MIN_MATCH-1]); \
847 prev[(s) & WMASK] = match_head = head[ins_h]; \
851 static ulg deflate(void)
853 IPos hash_head; /* head of hash chain */
854 IPos prev_match; /* previous match */
855 int flush; /* set if current block must be flushed */
856 int match_available = 0; /* set if previous match exists */
857 unsigned match_length = MIN_MATCH - 1; /* length of best match */
859 /* Process the input block. */
860 while (lookahead != 0) {
861 /* Insert the string window[strstart .. strstart+2] in the
862 * dictionary, and set hash_head to the head of the hash chain:
864 INSERT_STRING(strstart, hash_head);
866 /* Find the longest match, discarding those <= prev_length.
868 prev_length = match_length, prev_match = match_start;
869 match_length = MIN_MATCH - 1;
871 if (hash_head != 0 && prev_length < max_lazy_match
872 && strstart - hash_head <= MAX_DIST
874 /* To simplify the code, we prevent matches with the string
875 * of window index 0 (in particular we have to avoid a match
876 * of the string with itself at the start of the input file).
878 match_length = longest_match(hash_head);
879 /* longest_match() sets match_start */
880 if (match_length > lookahead)
881 match_length = lookahead;
883 /* Ignore a length 3 match if it is too distant: */
884 if (match_length == MIN_MATCH && strstart - match_start > TOO_FAR) {
885 /* If prev_match is also MIN_MATCH, match_start is garbage
886 * but we will ignore the current match anyway.
891 /* If there was a match at the previous step and the current
892 * match is not better, output the previous match:
894 if (prev_length >= MIN_MATCH && match_length <= prev_length) {
895 check_match(strstart - 1, prev_match, prev_length);
896 flush = ct_tally(strstart - 1 - prev_match, prev_length - MIN_MATCH);
898 /* Insert in hash table all strings up to the end of the match.
899 * strstart-1 and strstart are already inserted.
901 lookahead -= prev_length - 1;
905 INSERT_STRING(strstart, hash_head);
906 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
907 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
908 * these bytes are garbage, but it does not matter since the
909 * next lookahead bytes will always be emitted as literals.
911 } while (--prev_length != 0);
913 match_length = MIN_MATCH - 1;
917 block_start = strstart;
919 } else if (match_available) {
920 /* If there was no match at the previous position, output a
921 * single literal. If there was a match but the current match
922 * is longer, truncate the previous match to a single literal.
924 Tracevv((stderr, "%c", window[strstart - 1]));
925 if (ct_tally(0, window[strstart - 1])) {
927 block_start = strstart;
932 /* There is no previous match to compare with, wait for
933 * the next step to decide.
939 Assert(strstart <= isize && lookahead <= isize, "a bit too far");
941 /* Make sure that we always have enough lookahead, except
942 * at the end of the input file. We need MAX_MATCH bytes
943 * for the next match, plus MIN_MATCH bytes to insert the
944 * string following the next match.
946 while (lookahead < MIN_LOOKAHEAD && !eofile)
950 ct_tally(0, window[strstart - 1]);
952 return FLUSH_BLOCK(1); /* eof */
954 /* trees.c -- output deflated data using Huffman coding
955 * Copyright (C) 1992-1993 Jean-loup Gailly
956 * This is free software; you can redistribute it and/or modify it under the
957 * terms of the GNU General Public License, see the file COPYING.
963 * Encode various sets of source values using variable-length
968 * The PKZIP "deflation" process uses several Huffman trees. The more
969 * common source values are represented by shorter bit sequences.
971 * Each code tree is stored in the ZIP file in a compressed form
972 * which is itself a Huffman encoding of the lengths of
973 * all the code strings (in ascending order by source values).
974 * The actual code strings are reconstructed from the lengths in
975 * the UNZIP process, as described in the "application note"
976 * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program.
981 * Data Compression: Techniques and Applications, pp. 53-55.
982 * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7.
985 * Data Compression: Methods and Theory, pp. 49-50.
986 * Computer Science Press, 1988. ISBN 0-7167-8156-5.
990 * Addison-Wesley, 1983. ISBN 0-201-06672-6.
994 * void ct_init(ush *attr, int *methodp)
995 * Allocate the match buffer, initialize the various tables and save
996 * the location of the internal file attribute (ascii/binary) and
997 * method (DEFLATE/STORE)
999 * void ct_tally(int dist, int lc);
1000 * Save the match info and tally the frequency counts.
1002 * long flush_block (char *buf, ulg stored_len, int eof)
1003 * Determine the best encoding for the current block: dynamic trees,
1004 * static trees or store, and output the encoded block to the zip
1005 * file. Returns the total compressed length for the file so far.
1009 /* ===========================================================================
1014 /* All codes must not exceed MAX_BITS bits */
1016 #define MAX_BL_BITS 7
1017 /* Bit length codes must not exceed MAX_BL_BITS bits */
1019 #define LENGTH_CODES 29
1020 /* number of length codes, not counting the special END_BLOCK code */
1022 #define LITERALS 256
1023 /* number of literal bytes 0..255 */
1025 #define END_BLOCK 256
1026 /* end of block literal code */
1028 #define L_CODES (LITERALS+1+LENGTH_CODES)
1029 /* number of Literal or Length codes, including the END_BLOCK code */
1032 /* number of distance codes */
1035 /* number of codes used to transfer the bit lengths */
1037 typedef uch extra_bits_t;
1039 /* extra bits for each length code */
1040 static const extra_bits_t extra_lbits[LENGTH_CODES]
1041 = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4,
1045 /* extra bits for each distance code */
1046 static const extra_bits_t extra_dbits[D_CODES]
1047 = { 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,
1048 10, 10, 11, 11, 12, 12, 13, 13
1051 /* extra bits for each bit length code */
1052 static const extra_bits_t extra_blbits[BL_CODES]
1053 = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 };
1055 #define STORED_BLOCK 0
1056 #define STATIC_TREES 1
1058 /* The three kinds of block type */
1062 # define LIT_BUFSIZE 0x2000
1065 # define LIT_BUFSIZE 0x4000
1067 # define LIT_BUFSIZE 0x8000
1071 #ifndef DIST_BUFSIZE
1072 # define DIST_BUFSIZE LIT_BUFSIZE
1074 /* Sizes of match buffers for literals/lengths and distances. There are
1075 * 4 reasons for limiting LIT_BUFSIZE to 64K:
1076 * - frequencies can be kept in 16 bit counters
1077 * - if compression is not successful for the first block, all input data is
1078 * still in the window so we can still emit a stored block even when input
1079 * comes from standard input. (This can also be done for all blocks if
1080 * LIT_BUFSIZE is not greater than 32K.)
1081 * - if compression is not successful for a file smaller than 64K, we can
1082 * even emit a stored file instead of a stored block (saving 5 bytes).
1083 * - creating new Huffman trees less frequently may not provide fast
1084 * adaptation to changes in the input data statistics. (Take for
1085 * example a binary file with poorly compressible code followed by
1086 * a highly compressible string table.) Smaller buffer sizes give
1087 * fast adaptation but have of course the overhead of transmitting trees
1089 * - I can't count above 4
1090 * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save
1091 * memory at the expense of compression). Some optimizations would be possible
1092 * if we rely on DIST_BUFSIZE == LIT_BUFSIZE.
1094 #if LIT_BUFSIZE > INBUFSIZ
1095 #error cannot overlay l_buf and inbuf
1098 /* repeat previous bit length 3-6 times (2 bits of repeat count) */
1099 #define REPZ_3_10 17
1100 /* repeat a zero length 3-10 times (3 bits of repeat count) */
1101 #define REPZ_11_138 18
1102 /* repeat a zero length 11-138 times (7 bits of repeat count) */
1104 /* ===========================================================================
1108 /* Data structure describing a single value and its code string. */
1109 typedef struct ct_data {
1111 ush freq; /* frequency count */
1112 ush code; /* bit string */
1115 ush dad; /* father node in Huffman tree */
1116 ush len; /* length of bit string */
1120 #define Freq fc.freq
1121 #define Code fc.code
1125 #define HEAP_SIZE (2*L_CODES+1)
1126 /* maximum heap size */
1128 static ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */
1129 static ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */
1131 static ct_data static_ltree[L_CODES + 2];
1133 /* The static literal tree. Since the bit lengths are imposed, there is no
1134 * need for the L_CODES extra codes used during heap construction. However
1135 * The codes 286 and 287 are needed to build a canonical tree (see ct_init
1139 static ct_data static_dtree[D_CODES];
1141 /* The static distance tree. (Actually a trivial tree since all codes use
1145 static ct_data bl_tree[2 * BL_CODES + 1];
1147 /* Huffman tree for the bit lengths */
1149 typedef struct tree_desc {
1150 ct_data *dyn_tree; /* the dynamic tree */
1151 ct_data *static_tree; /* corresponding static tree or NULL */
1152 const extra_bits_t *extra_bits; /* extra bits for each code or NULL */
1153 int extra_base; /* base index for extra_bits */
1154 int elems; /* max number of elements in the tree */
1155 int max_length; /* max bit length for the codes */
1156 int max_code; /* largest code with non zero frequency */
1159 static tree_desc l_desc = {
1160 dyn_ltree, static_ltree, extra_lbits,
1161 LITERALS + 1, L_CODES, MAX_BITS, 0
1164 static tree_desc d_desc = {
1165 dyn_dtree, static_dtree, extra_dbits, 0, D_CODES, MAX_BITS, 0
1168 static tree_desc bl_desc = {
1169 bl_tree, NULL, extra_blbits, 0, BL_CODES, MAX_BL_BITS, 0
1173 static ush bl_count[MAX_BITS + 1];
1175 /* number of codes at each bit length for an optimal tree */
1177 static const uch bl_order[BL_CODES] = {
1178 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15
1181 /* The lengths of the bit length codes are sent in order of decreasing
1182 * probability, to avoid transmitting the lengths for unused bit length codes.
1185 static int heap[2 * L_CODES + 1]; /* heap used to build the Huffman trees */
1186 static int heap_len; /* number of elements in the heap */
1187 static int heap_max; /* element of largest frequency */
1189 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
1190 * The same heap array is used to build all trees.
1193 static uch depth[2 * L_CODES + 1];
1195 /* Depth of each subtree used as tie breaker for trees of equal frequency */
1197 static uch length_code[MAX_MATCH - MIN_MATCH + 1];
1199 /* length code for each normalized match length (0 == MIN_MATCH) */
1201 static uch dist_code[512];
1203 /* distance codes. The first 256 values correspond to the distances
1204 * 3 .. 258, the last 256 values correspond to the top 8 bits of
1205 * the 15 bit distances.
1208 static int base_length[LENGTH_CODES];
1210 /* First normalized length for each code (0 = MIN_MATCH) */
1212 static int base_dist[D_CODES];
1214 /* First normalized distance for each code (0 = distance of 1) */
1217 /* DECLARE(uch, l_buf, LIT_BUFSIZE); buffer for literals or lengths */
1219 /* DECLARE(ush, d_buf, DIST_BUFSIZE); buffer for distances */
1221 static uch flag_buf[(LIT_BUFSIZE / 8)];
1223 /* flag_buf is a bit array distinguishing literals from lengths in
1224 * l_buf, thus indicating the presence or absence of a distance.
1227 static unsigned last_lit; /* running index in l_buf */
1228 static unsigned last_dist; /* running index in d_buf */
1229 static unsigned last_flags; /* running index in flag_buf */
1230 static uch flags; /* current flags not yet saved in flag_buf */
1231 static uch flag_bit; /* current bit used in flags */
1233 /* bits are filled in flags starting at bit 0 (least significant).
1234 * Note: these flags are overkill in the current code since we don't
1235 * take advantage of DIST_BUFSIZE == LIT_BUFSIZE.
1238 static ulg opt_len; /* bit length of current block with optimal trees */
1239 static ulg static_len; /* bit length of current block with static trees */
1241 static ulg compressed_len; /* total bit length of compressed file */
1244 static ush *file_type; /* pointer to UNKNOWN, BINARY or ASCII */
1245 static int *file_method; /* pointer to DEFLATE or STORE */
1247 /* ===========================================================================
1248 * Local (static) routines in this file.
1251 static void init_block(void);
1252 static void pqdownheap(ct_data * tree, int k);
1253 static void gen_bitlen(tree_desc * desc);
1254 static void gen_codes(ct_data * tree, int max_code);
1255 static void build_tree(tree_desc * desc);
1256 static void scan_tree(ct_data * tree, int max_code);
1257 static void send_tree(ct_data * tree, int max_code);
1258 static int build_bl_tree(void);
1259 static void send_all_trees(int lcodes, int dcodes, int blcodes);
1260 static void compress_block(ct_data * ltree, ct_data * dtree);
1261 static void set_file_type(void);
1265 /* Send a code of the given tree. c and tree must not have side effects */
1266 # define send_code(c, tree) send_bits(tree[c].Code, tree[c].Len)
1268 # define send_code(c, tree) \
1270 if (verbose > 1) bb_error_msg("\ncd %3d ",(c)); \
1271 send_bits(tree[c].Code, tree[c].Len); \
1275 #define d_code(dist) \
1276 ((dist) < 256 ? dist_code[dist] : dist_code[256 + ((dist)>>7)])
1277 /* Mapping from a distance to a distance code. dist is the distance - 1 and
1278 * must not have side effects. dist_code[256] and dist_code[257] are never
1282 /* the arguments must not have side effects */
1285 /* ===========================================================================
1286 * Allocate the match buffer, initialize the various tables and save the
1287 * location of the internal file attribute (ascii/binary) and method
1290 static void ct_init(ush * attr, int *methodp)
1292 int n; /* iterates over tree elements */
1293 int bits; /* bit counter */
1294 int length; /* length value */
1295 int code; /* code value */
1296 int dist; /* distance index */
1299 file_method = methodp;
1300 compressed_len = 0L;
1302 if (static_dtree[0].Len != 0)
1303 return; /* ct_init already called */
1305 /* Initialize the mapping length (0..255) -> length code (0..28) */
1307 for (code = 0; code < LENGTH_CODES - 1; code++) {
1308 base_length[code] = length;
1309 for (n = 0; n < (1 << extra_lbits[code]); n++) {
1310 length_code[length++] = (uch) code;
1313 Assert(length == 256, "ct_init: length != 256");
1314 /* Note that the length 255 (match length 258) can be represented
1315 * in two different ways: code 284 + 5 bits or code 285, so we
1316 * overwrite length_code[255] to use the best encoding:
1318 length_code[length - 1] = (uch) code;
1320 /* Initialize the mapping dist (0..32K) -> dist code (0..29) */
1322 for (code = 0; code < 16; code++) {
1323 base_dist[code] = dist;
1324 for (n = 0; n < (1 << extra_dbits[code]); n++) {
1325 dist_code[dist++] = code;
1328 Assert(dist == 256, "ct_init: dist != 256");
1329 dist >>= 7; /* from now on, all distances are divided by 128 */
1330 for (; code < D_CODES; code++) {
1331 base_dist[code] = dist << 7;
1332 for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) {
1333 dist_code[256 + dist++] = (uch) code;
1336 Assert(dist == 256, "ct_init: 256+dist != 512");
1338 /* Construct the codes of the static literal tree */
1339 for (bits = 0; bits <= MAX_BITS; bits++)
1344 static_ltree[n++].Len = 8;
1348 static_ltree[n++].Len = 9;
1352 static_ltree[n++].Len = 7;
1356 static_ltree[n++].Len = 8;
1359 /* Codes 286 and 287 do not exist, but we must include them in the
1360 * tree construction to get a canonical Huffman tree (longest code
1363 gen_codes((ct_data *) static_ltree, L_CODES + 1);
1365 /* The static distance tree is trivial: */
1366 for (n = 0; n < D_CODES; n++) {
1367 static_dtree[n].Len = 5;
1368 static_dtree[n].Code = bi_reverse(n, 5);
1371 /* Initialize the first block of the first file: */
1376 /* ===========================================================================
1377 * Initialize a new block.
1379 static void init_block(void)
1381 int n; /* iterates over tree elements */
1383 /* Initialize the trees. */
1384 for (n = 0; n < L_CODES; n++)
1385 dyn_ltree[n].Freq = 0;
1386 for (n = 0; n < D_CODES; n++)
1387 dyn_dtree[n].Freq = 0;
1388 for (n = 0; n < BL_CODES; n++)
1389 bl_tree[n].Freq = 0;
1391 dyn_ltree[END_BLOCK].Freq = 1;
1392 opt_len = static_len = 0L;
1393 last_lit = last_dist = last_flags = 0;
1399 /* ===========================================================================
1400 * Remove the smallest element from the heap and recreate the heap with
1401 * one less element. Updates heap and heap_len.
1405 /* Index within the heap array of least frequent node in the Huffman tree */
1407 #define pqremove(tree, top) \
1409 top = heap[SMALLEST]; \
1410 heap[SMALLEST] = heap[heap_len--]; \
1411 pqdownheap(tree, SMALLEST); \
1415 /* ===========================================================================
1416 * Restore the heap property by moving down the tree starting at node k,
1417 * exchanging a node with the smallest of its two sons if necessary, stopping
1418 * when the heap property is re-established (each father smaller than its
1422 /* Compares to subtrees, using the tree depth as tie breaker when
1423 * the subtrees have equal frequency. This minimizes the worst case length.
1425 #define smaller(tree, n, m) \
1426 (tree[n].Freq < tree[m].Freq \
1427 || (tree[n].Freq == tree[m].Freq && depth[n] <= depth[m]))
1429 static void pqdownheap(ct_data * tree, int k)
1432 int j = k << 1; /* left son of k */
1434 while (j <= heap_len) {
1435 /* Set j to the smallest of the two sons: */
1436 if (j < heap_len && smaller(tree, heap[j + 1], heap[j]))
1439 /* Exit if v is smaller than both sons */
1440 if (smaller(tree, v, heap[j]))
1443 /* Exchange v with the smallest son */
1447 /* And continue down the tree, setting j to the left son of k */
1454 /* ===========================================================================
1455 * Compute the optimal bit lengths for a tree and update the total bit length
1456 * for the current block.
1457 * IN assertion: the fields freq and dad are set, heap[heap_max] and
1458 * above are the tree nodes sorted by increasing frequency.
1459 * OUT assertions: the field len is set to the optimal bit length, the
1460 * array bl_count contains the frequencies for each bit length.
1461 * The length opt_len is updated; static_len is also updated if stree is
1464 static void gen_bitlen(tree_desc * desc)
1466 ct_data *tree = desc->dyn_tree;
1467 const extra_bits_t *extra = desc->extra_bits;
1468 int base = desc->extra_base;
1469 int max_code = desc->max_code;
1470 int max_length = desc->max_length;
1471 ct_data *stree = desc->static_tree;
1472 int h; /* heap index */
1473 int n, m; /* iterate over the tree elements */
1474 int bits; /* bit length */
1475 int xbits; /* extra bits */
1476 ush f; /* frequency */
1477 int overflow = 0; /* number of elements with bit length too large */
1479 for (bits = 0; bits <= MAX_BITS; bits++)
1482 /* In a first pass, compute the optimal bit lengths (which may
1483 * overflow in the case of the bit length tree).
1485 tree[heap[heap_max]].Len = 0; /* root of the heap */
1487 for (h = heap_max + 1; h < HEAP_SIZE; h++) {
1489 bits = tree[tree[n].Dad].Len + 1;
1490 if (bits > max_length) {
1494 tree[n].Len = (ush) bits;
1495 /* We overwrite tree[n].Dad which is no longer needed */
1498 continue; /* not a leaf node */
1503 xbits = extra[n - base];
1505 opt_len += (ulg) f *(bits + xbits);
1508 static_len += (ulg) f *(stree[n].Len + xbits);
1513 Trace((stderr, "\nbit length overflow\n"));
1514 /* This happens for example on obj2 and pic of the Calgary corpus */
1516 /* Find the first bit length which could increase: */
1518 bits = max_length - 1;
1519 while (bl_count[bits] == 0)
1521 bl_count[bits]--; /* move one leaf down the tree */
1522 bl_count[bits + 1] += 2; /* move one overflow item as its brother */
1523 bl_count[max_length]--;
1524 /* The brother of the overflow item also moves one step up,
1525 * but this does not affect bl_count[max_length]
1528 } while (overflow > 0);
1530 /* Now recompute all bit lengths, scanning in increasing frequency.
1531 * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all
1532 * lengths instead of fixing only the wrong ones. This idea is taken
1533 * from 'ar' written by Haruhiko Okumura.)
1535 for (bits = max_length; bits != 0; bits--) {
1541 if (tree[m].Len != (unsigned) bits) {
1542 Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len, bits));
1543 opt_len += ((long) bits - (long) tree[m].Len) * (long) tree[m].Freq;
1544 tree[m].Len = (ush) bits;
1551 /* ===========================================================================
1552 * Generate the codes for a given tree and bit counts (which need not be
1554 * IN assertion: the array bl_count contains the bit length statistics for
1555 * the given tree and the field len is set for all tree elements.
1556 * OUT assertion: the field code is set for all tree elements of non
1559 static void gen_codes(ct_data * tree, int max_code)
1561 ush next_code[MAX_BITS + 1]; /* next code value for each bit length */
1562 ush code = 0; /* running code value */
1563 int bits; /* bit index */
1564 int n; /* code index */
1566 /* The distribution counts are first used to generate the code values
1567 * without bit reversal.
1569 for (bits = 1; bits <= MAX_BITS; bits++) {
1570 next_code[bits] = code = (code + bl_count[bits - 1]) << 1;
1572 /* Check that the bit counts in bl_count are consistent. The last code
1575 Assert(code + bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1,
1576 "inconsistent bit counts");
1577 Tracev((stderr, "\ngen_codes: max_code %d ", max_code));
1579 for (n = 0; n <= max_code; n++) {
1580 int len = tree[n].Len;
1584 /* Now reverse the bits */
1585 tree[n].Code = bi_reverse(next_code[len]++, len);
1587 Tracec(tree != static_ltree,
1588 (stderr, "\nn %3d %c l %2d c %4x (%x) ", n,
1589 (isgraph(n) ? n : ' '), len, tree[n].Code,
1590 next_code[len] - 1));
1594 /* ===========================================================================
1595 * Construct one Huffman tree and assigns the code bit strings and lengths.
1596 * Update the total bit length for the current block.
1597 * IN assertion: the field freq is set for all tree elements.
1598 * OUT assertions: the fields len and code are set to the optimal bit length
1599 * and corresponding code. The length opt_len is updated; static_len is
1600 * also updated if stree is not null. The field max_code is set.
1602 static void build_tree(tree_desc * desc)
1604 ct_data *tree = desc->dyn_tree;
1605 ct_data *stree = desc->static_tree;
1606 int elems = desc->elems;
1607 int n, m; /* iterate over heap elements */
1608 int max_code = -1; /* largest code with non zero frequency */
1609 int node = elems; /* next internal node of the tree */
1611 /* Construct the initial heap, with least frequent element in
1612 * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1].
1613 * heap[0] is not used.
1615 heap_len = 0, heap_max = HEAP_SIZE;
1617 for (n = 0; n < elems; n++) {
1618 if (tree[n].Freq != 0) {
1619 heap[++heap_len] = max_code = n;
1626 /* The pkzip format requires that at least one distance code exists,
1627 * and that at least one bit should be sent even if there is only one
1628 * possible code. So to avoid special checks later on we force at least
1629 * two codes of non zero frequency.
1631 while (heap_len < 2) {
1632 int new = heap[++heap_len] = (max_code < 2 ? ++max_code : 0);
1638 static_len -= stree[new].Len;
1639 /* new is 0 or 1 so it does not have extra bits */
1641 desc->max_code = max_code;
1643 /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree,
1644 * establish sub-heaps of increasing lengths:
1646 for (n = heap_len / 2; n >= 1; n--)
1647 pqdownheap(tree, n);
1649 /* Construct the Huffman tree by repeatedly combining the least two
1653 pqremove(tree, n); /* n = node of least frequency */
1654 m = heap[SMALLEST]; /* m = node of next least frequency */
1656 heap[--heap_max] = n; /* keep the nodes sorted by frequency */
1657 heap[--heap_max] = m;
1659 /* Create a new node father of n and m */
1660 tree[node].Freq = tree[n].Freq + tree[m].Freq;
1661 depth[node] = (uch) (MAX(depth[n], depth[m]) + 1);
1662 tree[n].Dad = tree[m].Dad = (ush) node;
1664 if (tree == bl_tree) {
1665 bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)",
1666 node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq);
1669 /* and insert the new node in the heap */
1670 heap[SMALLEST] = node++;
1671 pqdownheap(tree, SMALLEST);
1673 } while (heap_len >= 2);
1675 heap[--heap_max] = heap[SMALLEST];
1677 /* At this point, the fields freq and dad are set. We can now
1678 * generate the bit lengths.
1680 gen_bitlen((tree_desc *) desc);
1682 /* The field len is now set, we can generate the bit codes */
1683 gen_codes((ct_data *) tree, max_code);
1686 /* ===========================================================================
1687 * Scan a literal or distance tree to determine the frequencies of the codes
1688 * in the bit length tree. Updates opt_len to take into account the repeat
1689 * counts. (The contribution of the bit length codes will be added later
1690 * during the construction of bl_tree.)
1692 static void scan_tree(ct_data * tree, int max_code)
1694 int n; /* iterates over all tree elements */
1695 int prevlen = -1; /* last emitted length */
1696 int curlen; /* length of current code */
1697 int nextlen = tree[0].Len; /* length of next code */
1698 int count = 0; /* repeat count of the current code */
1699 int max_count = 7; /* max repeat count */
1700 int min_count = 4; /* min repeat count */
1706 tree[max_code + 1].Len = (ush) 0xffff; /* guard */
1708 for (n = 0; n <= max_code; n++) {
1710 nextlen = tree[n + 1].Len;
1711 if (++count < max_count && curlen == nextlen) {
1713 } else if (count < min_count) {
1714 bl_tree[curlen].Freq += count;
1715 } else if (curlen != 0) {
1716 if (curlen != prevlen)
1717 bl_tree[curlen].Freq++;
1718 bl_tree[REP_3_6].Freq++;
1719 } else if (count <= 10) {
1720 bl_tree[REPZ_3_10].Freq++;
1722 bl_tree[REPZ_11_138].Freq++;
1729 } else if (curlen == nextlen) {
1739 /* ===========================================================================
1740 * Send a literal or distance tree in compressed form, using the codes in
1743 static void send_tree(ct_data * tree, int max_code)
1745 int n; /* iterates over all tree elements */
1746 int prevlen = -1; /* last emitted length */
1747 int curlen; /* length of current code */
1748 int nextlen = tree[0].Len; /* length of next code */
1749 int count = 0; /* repeat count of the current code */
1750 int max_count = 7; /* max repeat count */
1751 int min_count = 4; /* min repeat count */
1753 /* tree[max_code+1].Len = -1; *//* guard already set */
1755 max_count = 138, min_count = 3;
1757 for (n = 0; n <= max_code; n++) {
1759 nextlen = tree[n + 1].Len;
1760 if (++count < max_count && curlen == nextlen) {
1762 } else if (count < min_count) {
1764 send_code(curlen, bl_tree);
1766 } else if (curlen != 0) {
1767 if (curlen != prevlen) {
1768 send_code(curlen, bl_tree);
1771 Assert(count >= 3 && count <= 6, " 3_6?");
1772 send_code(REP_3_6, bl_tree);
1773 send_bits(count - 3, 2);
1774 } else if (count <= 10) {
1775 send_code(REPZ_3_10, bl_tree);
1776 send_bits(count - 3, 3);
1778 send_code(REPZ_11_138, bl_tree);
1779 send_bits(count - 11, 7);
1786 } else if (curlen == nextlen) {
1796 /* ===========================================================================
1797 * Construct the Huffman tree for the bit lengths and return the index in
1798 * bl_order of the last bit length code to send.
1800 static int build_bl_tree(void)
1802 int max_blindex; /* index of last bit length code of non zero freq */
1804 /* Determine the bit length frequencies for literal and distance trees */
1805 scan_tree((ct_data *) dyn_ltree, l_desc.max_code);
1806 scan_tree((ct_data *) dyn_dtree, d_desc.max_code);
1808 /* Build the bit length tree: */
1809 build_tree((tree_desc *) (&bl_desc));
1810 /* opt_len now includes the length of the tree representations, except
1811 * the lengths of the bit lengths codes and the 5+5+4 bits for the counts.
1814 /* Determine the number of bit length codes to send. The pkzip format
1815 * requires that at least 4 bit length codes be sent. (appnote.txt says
1816 * 3 but the actual value used is 4.)
1818 for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) {
1819 if (bl_tree[bl_order[max_blindex]].Len != 0)
1822 /* Update opt_len to include the bit length tree and counts */
1823 opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4;
1824 Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", opt_len, static_len));
1829 /* ===========================================================================
1830 * Send the header for a block using dynamic Huffman trees: the counts, the
1831 * lengths of the bit length codes, the literal tree and the distance tree.
1832 * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4.
1834 static void send_all_trees(int lcodes, int dcodes, int blcodes)
1836 int rank; /* index in bl_order */
1838 Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes");
1839 Assert(lcodes <= L_CODES && dcodes <= D_CODES
1840 && blcodes <= BL_CODES, "too many codes");
1841 Tracev((stderr, "\nbl counts: "));
1842 send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */
1843 send_bits(dcodes - 1, 5);
1844 send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */
1845 for (rank = 0; rank < blcodes; rank++) {
1846 Tracev((stderr, "\nbl code %2d ", bl_order[rank]));
1847 send_bits(bl_tree[bl_order[rank]].Len, 3);
1849 Tracev((stderr, "\nbl tree: sent %ld", bits_sent));
1851 send_tree((ct_data *) dyn_ltree, lcodes - 1); /* send the literal tree */
1852 Tracev((stderr, "\nlit tree: sent %ld", bits_sent));
1854 send_tree((ct_data *) dyn_dtree, dcodes - 1); /* send the distance tree */
1855 Tracev((stderr, "\ndist tree: sent %ld", bits_sent));
1858 /* ===========================================================================
1859 * Determine the best encoding for the current block: dynamic trees, static
1860 * trees or store, and output the encoded block to the zip file. This function
1861 * returns the total compressed length for the file so far.
1863 static ulg flush_block(char *buf, ulg stored_len, int eof)
1865 ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
1866 int max_blindex; /* index of last bit length code of non zero freq */
1868 flag_buf[last_flags] = flags; /* Save the flags for the last 8 items */
1870 /* Check if the file is ascii or binary */
1871 if (*file_type == (ush) UNKNOWN)
1874 /* Construct the literal and distance trees */
1875 build_tree((tree_desc *) (&l_desc));
1876 Tracev((stderr, "\nlit data: dyn %ld, stat %ld", opt_len, static_len));
1878 build_tree((tree_desc *) (&d_desc));
1879 Tracev((stderr, "\ndist data: dyn %ld, stat %ld", opt_len, static_len));
1880 /* At this point, opt_len and static_len are the total bit lengths of
1881 * the compressed block data, excluding the tree representations.
1884 /* Build the bit length tree for the above two trees, and get the index
1885 * in bl_order of the last bit length code to send.
1887 max_blindex = build_bl_tree();
1889 /* Determine the best encoding. Compute first the block length in bytes */
1890 opt_lenb = (opt_len + 3 + 7) >> 3;
1891 static_lenb = (static_len + 3 + 7) >> 3;
1894 "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ",
1895 opt_lenb, opt_len, static_lenb, static_len, stored_len,
1896 last_lit, last_dist));
1898 if (static_lenb <= opt_lenb)
1899 opt_lenb = static_lenb;
1901 /* If compression failed and this is the first and last block,
1902 * and if the zip file can be seeked (to rewrite the local header),
1903 * the whole file is transformed into a stored file:
1905 if (stored_len <= opt_lenb && eof && compressed_len == 0L && seekable()) {
1906 /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */
1908 bb_error_msg("block vanished");
1910 copy_block(buf, (unsigned) stored_len, 0); /* without header */
1911 compressed_len = stored_len << 3;
1912 *file_method = STORED;
1914 } else if (stored_len + 4 <= opt_lenb && buf != (char *) 0) {
1915 /* 4: two words for the lengths */
1916 /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE.
1917 * Otherwise we can't have processed more than WSIZE input bytes since
1918 * the last block flush, because compression would have been
1919 * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to
1920 * transform a block into a stored block.
1922 send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */
1923 compressed_len = (compressed_len + 3 + 7) & ~7L;
1924 compressed_len += (stored_len + 4) << 3;
1926 copy_block(buf, (unsigned) stored_len, 1); /* with header */
1928 } else if (static_lenb == opt_lenb) {
1929 send_bits((STATIC_TREES << 1) + eof, 3);
1930 compress_block((ct_data *) static_ltree, (ct_data *) static_dtree);
1931 compressed_len += 3 + static_len;
1933 send_bits((DYN_TREES << 1) + eof, 3);
1934 send_all_trees(l_desc.max_code + 1, d_desc.max_code + 1,
1936 compress_block((ct_data *) dyn_ltree, (ct_data *) dyn_dtree);
1937 compressed_len += 3 + opt_len;
1939 Assert(compressed_len == bits_sent, "bad compressed size");
1944 compressed_len += 7; /* align on byte boundary */
1946 Tracev((stderr, "\ncomprlen %lu(%lu) ", compressed_len >> 3,
1947 compressed_len - 7 * eof));
1949 return compressed_len >> 3;
1952 /* ===========================================================================
1953 * Save the match info and tally the frequency counts. Return true if
1954 * the current block must be flushed.
1956 static int ct_tally(int dist, int lc)
1958 l_buf[last_lit++] = lc;
1960 /* lc is the unmatched char */
1961 dyn_ltree[lc].Freq++;
1963 /* Here, lc is the match length - MIN_MATCH */
1964 dist--; /* dist = match distance - 1 */
1965 Assert((ush) dist < (ush) MAX_DIST
1966 && (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH)
1967 && (ush) d_code(dist) < (ush) D_CODES, "ct_tally: bad match"
1970 dyn_ltree[length_code[lc] + LITERALS + 1].Freq++;
1971 dyn_dtree[d_code(dist)].Freq++;
1973 d_buf[last_dist++] = dist;
1978 /* Output the flags if they fill a byte: */
1979 if ((last_lit & 7) == 0) {
1980 flag_buf[last_flags++] = flags;
1981 flags = 0, flag_bit = 1;
1983 /* Try to guess if it is profitable to stop the current block here */
1984 if ((last_lit & 0xfff) == 0) {
1985 /* Compute an upper bound for the compressed length */
1986 ulg out_length = last_lit * 8L;
1987 ulg in_length = (ulg) strstart - block_start;
1990 for (dcode = 0; dcode < D_CODES; dcode++) {
1991 out_length += dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]);
1995 "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ",
1996 last_lit, last_dist, in_length, out_length,
1997 100L - out_length * 100L / in_length));
1998 if (last_dist < last_lit / 2 && out_length < in_length / 2)
2001 return (last_lit == LIT_BUFSIZE - 1 || last_dist == DIST_BUFSIZE);
2002 /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K
2003 * on 16 bit machines and because stored blocks are restricted to
2008 /* ===========================================================================
2009 * Send the block data compressed using the given Huffman trees
2011 static void compress_block(ct_data * ltree, ct_data * dtree)
2013 unsigned dist; /* distance of matched string */
2014 int lc; /* match length or unmatched char (if dist == 0) */
2015 unsigned lx = 0; /* running index in l_buf */
2016 unsigned dx = 0; /* running index in d_buf */
2017 unsigned fx = 0; /* running index in flag_buf */
2018 uch flag = 0; /* current flags */
2019 unsigned code; /* the code to send */
2020 int extra; /* number of extra bits to send */
2022 if (last_lit != 0) {
2025 flag = flag_buf[fx++];
2027 if ((flag & 1) == 0) {
2028 send_code(lc, ltree); /* send a literal byte */
2029 Tracecv(isgraph(lc), (stderr, " '%c' ", lc));
2031 /* Here, lc is the match length - MIN_MATCH */
2032 code = length_code[lc];
2033 send_code(code + LITERALS + 1, ltree); /* send the length code */
2034 extra = extra_lbits[code];
2036 lc -= base_length[code];
2037 send_bits(lc, extra); /* send the extra length bits */
2040 /* Here, dist is the match distance - 1 */
2041 code = d_code(dist);
2042 Assert(code < D_CODES, "bad d_code");
2044 send_code(code, dtree); /* send the distance code */
2045 extra = extra_dbits[code];
2047 dist -= base_dist[code];
2048 send_bits(dist, extra); /* send the extra distance bits */
2050 } /* literal or match pair ? */
2052 } while (lx < last_lit);
2055 send_code(END_BLOCK, ltree);
2058 /* ===========================================================================
2059 * Set the file type to ASCII or BINARY, using a crude approximation:
2060 * binary if more than 20% of the bytes are <= 6 or >= 128, ascii otherwise.
2061 * IN assertion: the fields freq of dyn_ltree are set and the total of all
2062 * frequencies does not exceed 64K (to fit in an int on 16 bit machines).
2064 static void set_file_type(void)
2067 unsigned ascii_freq = 0;
2068 unsigned bin_freq = 0;
2071 bin_freq += dyn_ltree[n++].Freq;
2073 ascii_freq += dyn_ltree[n++].Freq;
2074 while (n < LITERALS)
2075 bin_freq += dyn_ltree[n++].Freq;
2076 *file_type = (bin_freq > (ascii_freq >> 2)) ? BINARY : ASCII;
2077 if (*file_type == BINARY && translate_eol) {
2078 bb_error_msg("-l used on binary file");
2082 /* ===========================================================================
2083 * Deflate in to out.
2084 * IN assertions: the input and output buffers are cleared.
2085 * The variables time_stamp and save_orig_name are initialized.
2087 static int zip(int in, int out)
2089 uch my_flags = 0; /* general purpose bit flags */
2090 ush attr = 0; /* ascii/binary flag */
2091 ush deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */
2097 /* Write the header to the gzip file. See algorithm.doc for the format */
2100 put_header_byte(0x1f); /* magic header for gzip files, 1F 8B */
2101 put_header_byte(0x8b);
2103 put_header_byte(DEFLATED); /* compression method */
2105 put_header_byte(my_flags); /* general flags */
2106 put_32bit(time_stamp);
2108 /* Write deflated file to zip file */
2112 ct_init(&attr, &method);
2113 lm_init(&deflate_flags);
2115 put_8bit(deflate_flags); /* extra flags */
2116 put_8bit(3); /* OS identifier = 3 (Unix) */
2120 /* Write the crc and uncompressed size */
2129 /* ======================================================================== */
2130 static void abort_gzip(int ATTRIBUTE_UNUSED ignored)
2135 int gzip_main(int argc, char **argv)
2147 struct stat statBuf;
2150 opt = getopt32(argc, argv, "cf123456789qv" USE_GUNZIP("d"));
2151 //if (opt & 0x1) // -c
2152 //if (opt & 0x2) // -f
2153 /* Ignore 1-9 (compression level) options */
2154 //if (opt & 0x4) // -1
2155 //if (opt & 0x8) // -2
2156 //if (opt & 0x10) // -3
2157 //if (opt & 0x20) // -4
2158 //if (opt & 0x40) // -5
2159 //if (opt & 0x80) // -6
2160 //if (opt & 0x100) // -7
2161 //if (opt & 0x200) // -8
2162 //if (opt & 0x400) // -9
2163 //if (opt & 0x800) // -q
2164 //if (opt & 0x1000) // -v
2165 #if ENABLE_GUNZIP /* gunzip_main may not be visible... */
2166 if (opt & 0x2000) { // -d
2167 /* FIXME: getopt32 should not depend on optind */
2169 return gunzip_main(argc, argv);
2173 foreground = signal(SIGINT, SIG_IGN) != SIG_IGN;
2175 signal(SIGINT, abort_gzip);
2178 if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
2179 signal(SIGTERM, abort_gzip);
2183 if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
2184 signal(SIGHUP, abort_gzip);
2188 strncpy(z_suffix, ".gz", sizeof(z_suffix) - 1);
2190 /* Allocate all global buffers (for DYN_ALLOC option) */
2191 ALLOC(uch, inbuf, INBUFSIZ + INBUF_EXTRA);
2192 ALLOC(uch, outbuf, OUTBUFSIZ + OUTBUF_EXTRA);
2193 ALLOC(ush, d_buf, DIST_BUFSIZE);
2194 ALLOC(uch, window, 2L * WSIZE);
2195 ALLOC(ush, tab_prefix, 1L << BITS);
2197 /* Initialise the CRC32 table */
2198 crc_32_tab = crc32_filltable(0);
2202 if (optind == argc) {
2204 zip(STDIN_FILENO, STDOUT_FILENO);
2208 for (i = optind; i < argc; i++) {
2212 if (LONE_DASH(argv[i])) {
2214 inFileNum = STDIN_FILENO;
2215 outFileNum = STDOUT_FILENO;
2217 inFileNum = xopen(argv[i], O_RDONLY);
2218 if (fstat(inFileNum, &statBuf) < 0)
2219 bb_perror_msg_and_die("%s", argv[i]);
2220 time_stamp = statBuf.st_ctime;
2222 if (!(opt & OPT_tostdout)) {
2223 path = xasprintf("%s.gz", argv[i]);
2225 /* Open output file */
2226 #if defined(__GLIBC__) && __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 1 && defined(O_NOFOLLOW)
2227 outFileNum = open(path, O_RDWR | O_CREAT | O_EXCL | O_NOFOLLOW);
2229 outFileNum = open(path, O_RDWR | O_CREAT | O_EXCL);
2231 if (outFileNum < 0) {
2232 bb_perror_msg("%s", path);
2237 /* Set permissions on the file */
2238 fchmod(outFileNum, statBuf.st_mode);
2240 outFileNum = STDOUT_FILENO;
2243 if (path == NULL && isatty(outFileNum) && !(opt & OPT_force)) {
2244 bb_error_msg("compressed data not written "
2245 "to a terminal. Use -f to force compression.");
2250 result = zip(inFileNum, outFileNum);
2256 /* Delete the original file */
2258 delFileName = argv[i];
2262 if (unlink(delFileName) < 0)
2263 bb_perror_msg("%s", delFileName);