Contents of /trunk/mkinitrd-magellan/busybox/archival/gzip.c
Parent Directory | Revision Log
Revision 532 -
(show annotations)
(download)
Sat Sep 1 22:45:15 2007 UTC (16 years, 8 months ago) by niro
File MIME type: text/plain
File size: 67240 byte(s)
Sat Sep 1 22:45:15 2007 UTC (16 years, 8 months ago) by niro
File MIME type: text/plain
File size: 67240 byte(s)
-import if magellan mkinitrd; it is a fork of redhats mkinitrd-5.0.8 with all magellan patches and features; deprecates magellan-src/mkinitrd
1 | /* vi: set sw=4 ts=4: */ |
2 | /* |
3 | * Gzip implementation for busybox |
4 | * |
5 | * Based on GNU gzip Copyright (C) 1992-1993 Jean-loup Gailly. |
6 | * |
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." |
11 | * |
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. |
15 | * |
16 | * Licensed under GPLv2 or later, see file LICENSE in this tarball for details. |
17 | */ |
18 | |
19 | /* big objects in bss: |
20 | * 00000020 b bl_count |
21 | * 00000074 b base_length |
22 | * 00000078 b base_dist |
23 | * 00000078 b static_dtree |
24 | * 0000009c b bl_tree |
25 | * 000000f4 b dyn_dtree |
26 | * 00000100 b length_code |
27 | * 00000200 b dist_code |
28 | * 0000023d b depth |
29 | * 00000400 b flag_buf |
30 | * 0000047a b heap |
31 | * 00000480 b static_ltree |
32 | * 000008f4 b dyn_ltree |
33 | */ |
34 | |
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 |
40 | */ |
41 | |
42 | #include "busybox.h" |
43 | |
44 | |
45 | /* =========================================================================== |
46 | */ |
47 | //#define DEBUG 1 |
48 | /* Diagnostic functions */ |
49 | #ifdef DEBUG |
50 | # define Assert(cond,msg) {if(!(cond)) bb_error_msg(msg);} |
51 | # define Trace(x) fprintf x |
52 | # define Tracev(x) {if (verbose) fprintf x ;} |
53 | # define Tracevv(x) {if (verbose > 1) fprintf x ;} |
54 | # define Tracec(c,x) {if (verbose && (c)) fprintf x ;} |
55 | # define Tracecv(c,x) {if (verbose > 1 && (c)) fprintf x ;} |
56 | #else |
57 | # define Assert(cond,msg) |
58 | # define Trace(x) |
59 | # define Tracev(x) |
60 | # define Tracevv(x) |
61 | # define Tracec(c,x) |
62 | # define Tracecv(c,x) |
63 | #endif |
64 | |
65 | |
66 | /* =========================================================================== |
67 | */ |
68 | #define SMALL_MEM |
69 | |
70 | /* Compression methods (see algorithm.doc) */ |
71 | /* Only STORED and DEFLATED are supported by this BusyBox module */ |
72 | #define STORED 0 |
73 | /* methods 4 to 7 reserved */ |
74 | #define DEFLATED 8 |
75 | |
76 | #ifndef INBUFSIZ |
77 | # ifdef SMALL_MEM |
78 | # define INBUFSIZ 0x2000 /* input buffer size */ |
79 | # else |
80 | # define INBUFSIZ 0x8000 /* input buffer size */ |
81 | # endif |
82 | #endif |
83 | |
84 | #ifndef OUTBUFSIZ |
85 | # ifdef SMALL_MEM |
86 | # define OUTBUFSIZ 8192 /* output buffer size */ |
87 | # else |
88 | # define OUTBUFSIZ 16384 /* output buffer size */ |
89 | # endif |
90 | #endif |
91 | |
92 | #ifndef DIST_BUFSIZE |
93 | # ifdef SMALL_MEM |
94 | # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */ |
95 | # else |
96 | # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */ |
97 | # endif |
98 | #endif |
99 | |
100 | /* gzip flag byte */ |
101 | #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */ |
102 | #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */ |
103 | #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */ |
104 | #define ORIG_NAME 0x08 /* bit 3 set: original file name present */ |
105 | #define COMMENT 0x10 /* bit 4 set: file comment present */ |
106 | #define RESERVED 0xC0 /* bit 6,7: reserved */ |
107 | |
108 | /* internal file attribute */ |
109 | #define UNKNOWN 0xffff |
110 | #define BINARY 0 |
111 | #define ASCII 1 |
112 | |
113 | #ifndef WSIZE |
114 | # define WSIZE 0x8000 /* window size--must be a power of two, and */ |
115 | #endif /* at least 32K for zip's deflate method */ |
116 | |
117 | #define MIN_MATCH 3 |
118 | #define MAX_MATCH 258 |
119 | /* The minimum and maximum match lengths */ |
120 | |
121 | #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) |
122 | /* Minimum amount of lookahead, except at the end of the input file. |
123 | * See deflate.c for comments about the MIN_MATCH+1. |
124 | */ |
125 | |
126 | #define MAX_DIST (WSIZE-MIN_LOOKAHEAD) |
127 | /* In order to simplify the code, particularly on 16 bit machines, match |
128 | * distances are limited to MAX_DIST instead of WSIZE. |
129 | */ |
130 | |
131 | #ifndef MAX_PATH_LEN |
132 | # define MAX_PATH_LEN 1024 /* max pathname length */ |
133 | #endif |
134 | |
135 | #define seekable() 0 /* force sequential output */ |
136 | #define translate_eol 0 /* no option -a yet */ |
137 | |
138 | #ifndef BITS |
139 | # define BITS 16 |
140 | #endif |
141 | #define INIT_BITS 9 /* Initial number of bits per code */ |
142 | |
143 | #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */ |
144 | /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free. |
145 | * It's a pity that old uncompress does not check bit 0x20. That makes |
146 | * extension of the format actually undesirable because old compress |
147 | * would just crash on the new format instead of giving a meaningful |
148 | * error message. It does check the number of bits, but it's more |
149 | * helpful to say "unsupported format, get a new version" than |
150 | * "can only handle 16 bits". |
151 | */ |
152 | |
153 | #ifdef MAX_EXT_CHARS |
154 | # define MAX_SUFFIX MAX_EXT_CHARS |
155 | #else |
156 | # define MAX_SUFFIX 30 |
157 | #endif |
158 | |
159 | |
160 | /* =========================================================================== |
161 | * Compile with MEDIUM_MEM to reduce the memory requirements or |
162 | * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the |
163 | * entire input file can be held in memory (not possible on 16 bit systems). |
164 | * Warning: defining these symbols affects HASH_BITS (see below) and thus |
165 | * affects the compression ratio. The compressed output |
166 | * is still correct, and might even be smaller in some cases. |
167 | */ |
168 | |
169 | #ifdef SMALL_MEM |
170 | # define HASH_BITS 13 /* Number of bits used to hash strings */ |
171 | #endif |
172 | #ifdef MEDIUM_MEM |
173 | # define HASH_BITS 14 |
174 | #endif |
175 | #ifndef HASH_BITS |
176 | # define HASH_BITS 15 |
177 | /* For portability to 16 bit machines, do not use values above 15. */ |
178 | #endif |
179 | |
180 | #define HASH_SIZE (unsigned)(1<<HASH_BITS) |
181 | #define HASH_MASK (HASH_SIZE-1) |
182 | #define WMASK (WSIZE-1) |
183 | /* HASH_SIZE and WSIZE must be powers of two */ |
184 | #ifndef TOO_FAR |
185 | # define TOO_FAR 4096 |
186 | #endif |
187 | /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ |
188 | |
189 | |
190 | /* =========================================================================== |
191 | * These types are not really 'char', 'short' and 'long' |
192 | */ |
193 | typedef uint8_t uch; |
194 | typedef uint16_t ush; |
195 | typedef uint32_t ulg; |
196 | typedef int32_t lng; |
197 | |
198 | |
199 | /* =========================================================================== |
200 | */ |
201 | typedef ush Pos; |
202 | typedef unsigned IPos; |
203 | |
204 | /* A Pos is an index in the character window. We use short instead of int to |
205 | * save space in the various tables. IPos is used only for parameter passing. |
206 | */ |
207 | |
208 | static lng block_start; |
209 | |
210 | /* window position at the beginning of the current output block. Gets |
211 | * negative when the window is moved backwards. |
212 | */ |
213 | |
214 | static unsigned ins_h; /* hash index of string to be inserted */ |
215 | |
216 | #define H_SHIFT ((HASH_BITS+MIN_MATCH-1) / MIN_MATCH) |
217 | /* Number of bits by which ins_h and del_h must be shifted at each |
218 | * input step. It must be such that after MIN_MATCH steps, the oldest |
219 | * byte no longer takes part in the hash key, that is: |
220 | * H_SHIFT * MIN_MATCH >= HASH_BITS |
221 | */ |
222 | |
223 | static unsigned int prev_length; |
224 | |
225 | /* Length of the best match at previous step. Matches not greater than this |
226 | * are discarded. This is used in the lazy match evaluation. |
227 | */ |
228 | |
229 | static unsigned strstart; /* start of string to insert */ |
230 | static unsigned match_start; /* start of matching string */ |
231 | static int eofile; /* flag set at end of input file */ |
232 | static unsigned lookahead; /* number of valid bytes ahead in window */ |
233 | |
234 | enum { |
235 | WINDOW_SIZE = 2 * WSIZE, |
236 | /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the |
237 | * input file length plus MIN_LOOKAHEAD. |
238 | */ |
239 | |
240 | max_chain_length = 4096, |
241 | /* To speed up deflation, hash chains are never searched beyond this length. |
242 | * A higher limit improves compression ratio but degrades the speed. |
243 | */ |
244 | |
245 | max_lazy_match = 258, |
246 | /* Attempt to find a better match only when the current match is strictly |
247 | * smaller than this value. This mechanism is used only for compression |
248 | * levels >= 4. |
249 | */ |
250 | |
251 | max_insert_length = max_lazy_match, |
252 | /* Insert new strings in the hash table only if the match length |
253 | * is not greater than this length. This saves time but degrades compression. |
254 | * max_insert_length is used only for compression levels <= 3. |
255 | */ |
256 | |
257 | good_match = 32, |
258 | /* Use a faster search when the previous match is longer than this */ |
259 | |
260 | /* Values for max_lazy_match, good_match and max_chain_length, depending on |
261 | * the desired pack level (0..9). The values given below have been tuned to |
262 | * exclude worst case performance for pathological files. Better values may be |
263 | * found for specific files. |
264 | */ |
265 | |
266 | nice_match = 258, /* Stop searching when current match exceeds this */ |
267 | /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 |
268 | * For deflate_fast() (levels <= 3) good is ignored and lazy has a different |
269 | * meaning. |
270 | */ |
271 | }; |
272 | |
273 | |
274 | /* =========================================================================== |
275 | */ |
276 | #define DECLARE(type, array, size) \ |
277 | static type * array |
278 | |
279 | #define ALLOC(type, array, size) \ |
280 | { \ |
281 | array = xzalloc((size_t)(((size)+1L)/2) * 2*sizeof(type)); \ |
282 | } |
283 | |
284 | #define FREE(array) \ |
285 | { \ |
286 | free(array); \ |
287 | array = NULL; \ |
288 | } |
289 | |
290 | /* global buffers */ |
291 | |
292 | /* buffer for literals or lengths */ |
293 | /* DECLARE(uch, l_buf, LIT_BUFSIZE); */ |
294 | DECLARE(uch, l_buf, INBUFSIZ); |
295 | |
296 | DECLARE(ush, d_buf, DIST_BUFSIZE); |
297 | DECLARE(uch, outbuf, OUTBUFSIZ); |
298 | |
299 | /* Sliding window. Input bytes are read into the second half of the window, |
300 | * and move to the first half later to keep a dictionary of at least WSIZE |
301 | * bytes. With this organization, matches are limited to a distance of |
302 | * WSIZE-MAX_MATCH bytes, but this ensures that IO is always |
303 | * performed with a length multiple of the block size. Also, it limits |
304 | * the window size to 64K, which is quite useful on MSDOS. |
305 | * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would |
306 | * be less efficient). |
307 | */ |
308 | DECLARE(uch, window, 2L * WSIZE); |
309 | |
310 | /* Link to older string with same hash index. To limit the size of this |
311 | * array to 64K, this link is maintained only for the last 32K strings. |
312 | * An index in this array is thus a window index modulo 32K. |
313 | */ |
314 | /* DECLARE(Pos, prev, WSIZE); */ |
315 | DECLARE(ush, prev, 1L << BITS); |
316 | |
317 | /* Heads of the hash chains or 0. */ |
318 | /* DECLARE(Pos, head, 1<<HASH_BITS); */ |
319 | #define head (prev+WSIZE) /* hash head (see deflate.c) */ |
320 | |
321 | |
322 | /* number of input bytes */ |
323 | static ulg isize; /* only 32 bits stored in .gz file */ |
324 | |
325 | static int foreground; /* set if program run in foreground */ |
326 | static int method = DEFLATED; /* compression method */ |
327 | static int exit_code; /* program exit code */ |
328 | |
329 | /* original time stamp (modification time) */ |
330 | static ulg time_stamp; /* only 32 bits stored in .gz file */ |
331 | |
332 | static int ifd; /* input file descriptor */ |
333 | static int ofd; /* output file descriptor */ |
334 | #ifdef DEBUG |
335 | static unsigned insize; /* valid bytes in l_buf */ |
336 | #endif |
337 | static unsigned outcnt; /* bytes in output buffer */ |
338 | |
339 | static uint32_t *crc_32_tab; |
340 | |
341 | |
342 | /* =========================================================================== |
343 | * Local data used by the "bit string" routines. |
344 | */ |
345 | |
346 | //// static int zfile; /* output gzip file */ |
347 | |
348 | static unsigned short bi_buf; |
349 | |
350 | /* Output buffer. bits are inserted starting at the bottom (least significant |
351 | * bits). |
352 | */ |
353 | |
354 | #undef BUF_SIZE |
355 | #define BUF_SIZE (8 * sizeof(bi_buf)) |
356 | /* Number of bits used within bi_buf. (bi_buf might be implemented on |
357 | * more than 16 bits on some systems.) |
358 | */ |
359 | |
360 | static int bi_valid; |
361 | |
362 | /* Current input function. Set to mem_read for in-memory compression */ |
363 | |
364 | #ifdef DEBUG |
365 | static ulg bits_sent; /* bit length of the compressed data */ |
366 | #endif |
367 | |
368 | |
369 | /* =========================================================================== |
370 | * Write the output buffer outbuf[0..outcnt-1] and update bytes_out. |
371 | * (used for the compressed data only) |
372 | */ |
373 | static void flush_outbuf(void) |
374 | { |
375 | if (outcnt == 0) |
376 | return; |
377 | |
378 | xwrite(ofd, (char *) outbuf, outcnt); |
379 | outcnt = 0; |
380 | } |
381 | |
382 | |
383 | /* =========================================================================== |
384 | */ |
385 | /* put_8bit is used for the compressed output */ |
386 | #define put_8bit(c) \ |
387 | { \ |
388 | outbuf[outcnt++] = (c); \ |
389 | if (outcnt == OUTBUFSIZ) flush_outbuf(); \ |
390 | } |
391 | |
392 | /* Output a 16 bit value, lsb first */ |
393 | static void put_16bit(ush w) |
394 | { |
395 | if (outcnt < OUTBUFSIZ - 2) { |
396 | outbuf[outcnt++] = w; |
397 | outbuf[outcnt++] = w >> 8; |
398 | } else { |
399 | put_8bit(w); |
400 | put_8bit(w >> 8); |
401 | } |
402 | } |
403 | |
404 | static void put_32bit(ulg n) |
405 | { |
406 | put_16bit(n); |
407 | put_16bit(n >> 16); |
408 | } |
409 | |
410 | /* =========================================================================== |
411 | * Clear input and output buffers |
412 | */ |
413 | static void clear_bufs(void) |
414 | { |
415 | outcnt = 0; |
416 | #ifdef DEBUG |
417 | insize = 0; |
418 | #endif |
419 | isize = 0; |
420 | } |
421 | |
422 | |
423 | /* =========================================================================== |
424 | * Run a set of bytes through the crc shift register. If s is a NULL |
425 | * pointer, then initialize the crc shift register contents instead. |
426 | * Return the current crc in either case. |
427 | */ |
428 | static uint32_t crc; /* shift register contents */ |
429 | static uint32_t updcrc(uch * s, unsigned n) |
430 | { |
431 | uint32_t c = crc; |
432 | while (n) { |
433 | c = crc_32_tab[(uch)(c ^ *s++)] ^ (c >> 8); |
434 | n--; |
435 | } |
436 | crc = c; |
437 | return c; |
438 | } |
439 | |
440 | |
441 | /* =========================================================================== |
442 | * Read a new buffer from the current input file, perform end-of-line |
443 | * translation, and update the crc and input file size. |
444 | * IN assertion: size >= 2 (for end-of-line translation) |
445 | */ |
446 | static unsigned file_read(void *buf, unsigned size) |
447 | { |
448 | unsigned len; |
449 | |
450 | Assert(insize == 0, "l_buf not empty"); |
451 | |
452 | len = safe_read(ifd, buf, size); |
453 | if (len == (unsigned)(-1) || len == 0) |
454 | return len; |
455 | |
456 | updcrc(buf, len); |
457 | isize += len; |
458 | return len; |
459 | } |
460 | |
461 | |
462 | /* =========================================================================== |
463 | * Send a value on a given number of bits. |
464 | * IN assertion: length <= 16 and value fits in length bits. |
465 | */ |
466 | static void send_bits(int value, int length) |
467 | { |
468 | #ifdef DEBUG |
469 | Tracev((stderr, " l %2d v %4x ", length, value)); |
470 | Assert(length > 0 && length <= 15, "invalid length"); |
471 | bits_sent += length; |
472 | #endif |
473 | /* If not enough room in bi_buf, use (valid) bits from bi_buf and |
474 | * (16 - bi_valid) bits from value, leaving (width - (16-bi_valid)) |
475 | * unused bits in value. |
476 | */ |
477 | if (bi_valid > (int) BUF_SIZE - length) { |
478 | bi_buf |= (value << bi_valid); |
479 | put_16bit(bi_buf); |
480 | bi_buf = (ush) value >> (BUF_SIZE - bi_valid); |
481 | bi_valid += length - BUF_SIZE; |
482 | } else { |
483 | bi_buf |= value << bi_valid; |
484 | bi_valid += length; |
485 | } |
486 | } |
487 | |
488 | |
489 | /* =========================================================================== |
490 | * Reverse the first len bits of a code, using straightforward code (a faster |
491 | * method would use a table) |
492 | * IN assertion: 1 <= len <= 15 |
493 | */ |
494 | static unsigned bi_reverse(unsigned code, int len) |
495 | { |
496 | unsigned res = 0; |
497 | |
498 | while (1) { |
499 | res |= code & 1; |
500 | if (--len <= 0) return res; |
501 | code >>= 1; |
502 | res <<= 1; |
503 | } |
504 | } |
505 | |
506 | |
507 | /* =========================================================================== |
508 | * Write out any remaining bits in an incomplete byte. |
509 | */ |
510 | static void bi_windup(void) |
511 | { |
512 | if (bi_valid > 8) { |
513 | put_16bit(bi_buf); |
514 | } else if (bi_valid > 0) { |
515 | put_8bit(bi_buf); |
516 | } |
517 | bi_buf = 0; |
518 | bi_valid = 0; |
519 | #ifdef DEBUG |
520 | bits_sent = (bits_sent + 7) & ~7; |
521 | #endif |
522 | } |
523 | |
524 | |
525 | /* =========================================================================== |
526 | * Copy a stored block to the zip file, storing first the length and its |
527 | * one's complement if requested. |
528 | */ |
529 | static void copy_block(char *buf, unsigned len, int header) |
530 | { |
531 | bi_windup(); /* align on byte boundary */ |
532 | |
533 | if (header) { |
534 | put_16bit(len); |
535 | put_16bit(~len); |
536 | #ifdef DEBUG |
537 | bits_sent += 2 * 16; |
538 | #endif |
539 | } |
540 | #ifdef DEBUG |
541 | bits_sent += (ulg) len << 3; |
542 | #endif |
543 | while (len--) { |
544 | put_8bit(*buf++); |
545 | } |
546 | } |
547 | |
548 | |
549 | /* =========================================================================== |
550 | * Fill the window when the lookahead becomes insufficient. |
551 | * Updates strstart and lookahead, and sets eofile if end of input file. |
552 | * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0 |
553 | * OUT assertions: at least one byte has been read, or eofile is set; |
554 | * file reads are performed for at least two bytes (required for the |
555 | * translate_eol option). |
556 | */ |
557 | static void fill_window(void) |
558 | { |
559 | unsigned n, m; |
560 | unsigned more = WINDOW_SIZE - lookahead - strstart; |
561 | /* Amount of free space at the end of the window. */ |
562 | |
563 | /* If the window is almost full and there is insufficient lookahead, |
564 | * move the upper half to the lower one to make room in the upper half. |
565 | */ |
566 | if (more == (unsigned) -1) { |
567 | /* Very unlikely, but possible on 16 bit machine if strstart == 0 |
568 | * and lookahead == 1 (input done one byte at time) |
569 | */ |
570 | more--; |
571 | } else if (strstart >= WSIZE + MAX_DIST) { |
572 | /* By the IN assertion, the window is not empty so we can't confuse |
573 | * more == 0 with more == 64K on a 16 bit machine. |
574 | */ |
575 | Assert(WINDOW_SIZE == 2 * WSIZE, "no sliding with BIG_MEM"); |
576 | |
577 | memcpy(window, window + WSIZE, WSIZE); |
578 | match_start -= WSIZE; |
579 | strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */ |
580 | |
581 | block_start -= WSIZE; |
582 | |
583 | for (n = 0; n < HASH_SIZE; n++) { |
584 | m = head[n]; |
585 | head[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0); |
586 | } |
587 | for (n = 0; n < WSIZE; n++) { |
588 | m = prev[n]; |
589 | prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0); |
590 | /* If n is not on any hash chain, prev[n] is garbage but |
591 | * its value will never be used. |
592 | */ |
593 | } |
594 | more += WSIZE; |
595 | } |
596 | /* At this point, more >= 2 */ |
597 | if (!eofile) { |
598 | n = file_read(window + strstart + lookahead, more); |
599 | if (n == 0 || n == (unsigned) -1) { |
600 | eofile = 1; |
601 | } else { |
602 | lookahead += n; |
603 | } |
604 | } |
605 | } |
606 | |
607 | |
608 | /* =========================================================================== |
609 | * Set match_start to the longest match starting at the given string and |
610 | * return its length. Matches shorter or equal to prev_length are discarded, |
611 | * in which case the result is equal to prev_length and match_start is |
612 | * garbage. |
613 | * IN assertions: cur_match is the head of the hash chain for the current |
614 | * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 |
615 | */ |
616 | |
617 | /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or |
618 | * match.s. The code is functionally equivalent, so you can use the C version |
619 | * if desired. |
620 | */ |
621 | static int longest_match(IPos cur_match) |
622 | { |
623 | unsigned chain_length = max_chain_length; /* max hash chain length */ |
624 | uch *scan = window + strstart; /* current string */ |
625 | uch *match; /* matched string */ |
626 | int len; /* length of current match */ |
627 | int best_len = prev_length; /* best match length so far */ |
628 | IPos limit = strstart > (IPos) MAX_DIST ? strstart - (IPos) MAX_DIST : 0; |
629 | /* Stop when cur_match becomes <= limit. To simplify the code, |
630 | * we prevent matches with the string of window index 0. |
631 | */ |
632 | |
633 | /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. |
634 | * It is easy to get rid of this optimization if necessary. |
635 | */ |
636 | #if HASH_BITS < 8 || MAX_MATCH != 258 |
637 | # error Code too clever |
638 | #endif |
639 | uch *strend = window + strstart + MAX_MATCH; |
640 | uch scan_end1 = scan[best_len - 1]; |
641 | uch scan_end = scan[best_len]; |
642 | |
643 | /* Do not waste too much time if we already have a good match: */ |
644 | if (prev_length >= good_match) { |
645 | chain_length >>= 2; |
646 | } |
647 | Assert(strstart <= WINDOW_SIZE - MIN_LOOKAHEAD, "insufficient lookahead"); |
648 | |
649 | do { |
650 | Assert(cur_match < strstart, "no future"); |
651 | match = window + cur_match; |
652 | |
653 | /* Skip to next match if the match length cannot increase |
654 | * or if the match length is less than 2: |
655 | */ |
656 | if (match[best_len] != scan_end || |
657 | match[best_len - 1] != scan_end1 || |
658 | *match != *scan || *++match != scan[1]) |
659 | continue; |
660 | |
661 | /* The check at best_len-1 can be removed because it will be made |
662 | * again later. (This heuristic is not always a win.) |
663 | * It is not necessary to compare scan[2] and match[2] since they |
664 | * are always equal when the other bytes match, given that |
665 | * the hash keys are equal and that HASH_BITS >= 8. |
666 | */ |
667 | scan += 2, match++; |
668 | |
669 | /* We check for insufficient lookahead only every 8th comparison; |
670 | * the 256th check will be made at strstart+258. |
671 | */ |
672 | do { |
673 | } while (*++scan == *++match && *++scan == *++match && |
674 | *++scan == *++match && *++scan == *++match && |
675 | *++scan == *++match && *++scan == *++match && |
676 | *++scan == *++match && *++scan == *++match && scan < strend); |
677 | |
678 | len = MAX_MATCH - (int) (strend - scan); |
679 | scan = strend - MAX_MATCH; |
680 | |
681 | if (len > best_len) { |
682 | match_start = cur_match; |
683 | best_len = len; |
684 | if (len >= nice_match) |
685 | break; |
686 | scan_end1 = scan[best_len - 1]; |
687 | scan_end = scan[best_len]; |
688 | } |
689 | } while ((cur_match = prev[cur_match & WMASK]) > limit |
690 | && --chain_length != 0); |
691 | |
692 | return best_len; |
693 | } |
694 | |
695 | |
696 | #ifdef DEBUG |
697 | /* =========================================================================== |
698 | * Check that the match at match_start is indeed a match. |
699 | */ |
700 | static void check_match(IPos start, IPos match, int length) |
701 | { |
702 | /* check that the match is indeed a match */ |
703 | if (memcmp(window + match, window + start, length) != 0) { |
704 | bb_error_msg(" start %d, match %d, length %d", start, match, length); |
705 | bb_error_msg("invalid match"); |
706 | } |
707 | if (verbose > 1) { |
708 | bb_error_msg("\\[%d,%d]", start - match, length); |
709 | do { |
710 | putc(window[start++], stderr); |
711 | } while (--length != 0); |
712 | } |
713 | } |
714 | #else |
715 | # define check_match(start, match, length) ((void)0) |
716 | #endif |
717 | |
718 | |
719 | /* trees.c -- output deflated data using Huffman coding |
720 | * Copyright (C) 1992-1993 Jean-loup Gailly |
721 | * This is free software; you can redistribute it and/or modify it under the |
722 | * terms of the GNU General Public License, see the file COPYING. |
723 | */ |
724 | |
725 | /* PURPOSE |
726 | * Encode various sets of source values using variable-length |
727 | * binary code trees. |
728 | * |
729 | * DISCUSSION |
730 | * The PKZIP "deflation" process uses several Huffman trees. The more |
731 | * common source values are represented by shorter bit sequences. |
732 | * |
733 | * Each code tree is stored in the ZIP file in a compressed form |
734 | * which is itself a Huffman encoding of the lengths of |
735 | * all the code strings (in ascending order by source values). |
736 | * The actual code strings are reconstructed from the lengths in |
737 | * the UNZIP process, as described in the "application note" |
738 | * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program. |
739 | * |
740 | * REFERENCES |
741 | * Lynch, Thomas J. |
742 | * Data Compression: Techniques and Applications, pp. 53-55. |
743 | * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7. |
744 | * |
745 | * Storer, James A. |
746 | * Data Compression: Methods and Theory, pp. 49-50. |
747 | * Computer Science Press, 1988. ISBN 0-7167-8156-5. |
748 | * |
749 | * Sedgewick, R. |
750 | * Algorithms, p290. |
751 | * Addison-Wesley, 1983. ISBN 0-201-06672-6. |
752 | * |
753 | * INTERFACE |
754 | * void ct_init(ush *attr, int *methodp) |
755 | * Allocate the match buffer, initialize the various tables and save |
756 | * the location of the internal file attribute (ascii/binary) and |
757 | * method (DEFLATE/STORE) |
758 | * |
759 | * void ct_tally(int dist, int lc); |
760 | * Save the match info and tally the frequency counts. |
761 | * |
762 | * ulg flush_block(char *buf, ulg stored_len, int eof) |
763 | * Determine the best encoding for the current block: dynamic trees, |
764 | * static trees or store, and output the encoded block to the zip |
765 | * file. Returns the total compressed length for the file so far. |
766 | */ |
767 | |
768 | #define MAX_BITS 15 |
769 | /* All codes must not exceed MAX_BITS bits */ |
770 | |
771 | #define MAX_BL_BITS 7 |
772 | /* Bit length codes must not exceed MAX_BL_BITS bits */ |
773 | |
774 | #define LENGTH_CODES 29 |
775 | /* number of length codes, not counting the special END_BLOCK code */ |
776 | |
777 | #define LITERALS 256 |
778 | /* number of literal bytes 0..255 */ |
779 | |
780 | #define END_BLOCK 256 |
781 | /* end of block literal code */ |
782 | |
783 | #define L_CODES (LITERALS+1+LENGTH_CODES) |
784 | /* number of Literal or Length codes, including the END_BLOCK code */ |
785 | |
786 | #define D_CODES 30 |
787 | /* number of distance codes */ |
788 | |
789 | #define BL_CODES 19 |
790 | /* number of codes used to transfer the bit lengths */ |
791 | |
792 | typedef uch extra_bits_t; |
793 | |
794 | /* extra bits for each length code */ |
795 | static const extra_bits_t extra_lbits[LENGTH_CODES]= { |
796 | 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, |
797 | 4, 4, 5, 5, 5, 5, 0 |
798 | }; |
799 | |
800 | /* extra bits for each distance code */ |
801 | static const extra_bits_t extra_dbits[D_CODES] = { |
802 | 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, |
803 | 10, 10, 11, 11, 12, 12, 13, 13 |
804 | }; |
805 | |
806 | /* extra bits for each bit length code */ |
807 | static const extra_bits_t extra_blbits[BL_CODES] = { |
808 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 }; |
809 | |
810 | #define STORED_BLOCK 0 |
811 | #define STATIC_TREES 1 |
812 | #define DYN_TREES 2 |
813 | /* The three kinds of block type */ |
814 | |
815 | #ifndef LIT_BUFSIZE |
816 | # ifdef SMALL_MEM |
817 | # define LIT_BUFSIZE 0x2000 |
818 | # else |
819 | # ifdef MEDIUM_MEM |
820 | # define LIT_BUFSIZE 0x4000 |
821 | # else |
822 | # define LIT_BUFSIZE 0x8000 |
823 | # endif |
824 | # endif |
825 | #endif |
826 | #ifndef DIST_BUFSIZE |
827 | # define DIST_BUFSIZE LIT_BUFSIZE |
828 | #endif |
829 | /* Sizes of match buffers for literals/lengths and distances. There are |
830 | * 4 reasons for limiting LIT_BUFSIZE to 64K: |
831 | * - frequencies can be kept in 16 bit counters |
832 | * - if compression is not successful for the first block, all input data is |
833 | * still in the window so we can still emit a stored block even when input |
834 | * comes from standard input. (This can also be done for all blocks if |
835 | * LIT_BUFSIZE is not greater than 32K.) |
836 | * - if compression is not successful for a file smaller than 64K, we can |
837 | * even emit a stored file instead of a stored block (saving 5 bytes). |
838 | * - creating new Huffman trees less frequently may not provide fast |
839 | * adaptation to changes in the input data statistics. (Take for |
840 | * example a binary file with poorly compressible code followed by |
841 | * a highly compressible string table.) Smaller buffer sizes give |
842 | * fast adaptation but have of course the overhead of transmitting trees |
843 | * more frequently. |
844 | * - I can't count above 4 |
845 | * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save |
846 | * memory at the expense of compression). Some optimizations would be possible |
847 | * if we rely on DIST_BUFSIZE == LIT_BUFSIZE. |
848 | */ |
849 | #define REP_3_6 16 |
850 | /* repeat previous bit length 3-6 times (2 bits of repeat count) */ |
851 | #define REPZ_3_10 17 |
852 | /* repeat a zero length 3-10 times (3 bits of repeat count) */ |
853 | #define REPZ_11_138 18 |
854 | /* repeat a zero length 11-138 times (7 bits of repeat count) */ |
855 | |
856 | /* =========================================================================== |
857 | */ |
858 | /* Data structure describing a single value and its code string. */ |
859 | typedef struct ct_data { |
860 | union { |
861 | ush freq; /* frequency count */ |
862 | ush code; /* bit string */ |
863 | } fc; |
864 | union { |
865 | ush dad; /* father node in Huffman tree */ |
866 | ush len; /* length of bit string */ |
867 | } dl; |
868 | } ct_data; |
869 | |
870 | #define Freq fc.freq |
871 | #define Code fc.code |
872 | #define Dad dl.dad |
873 | #define Len dl.len |
874 | |
875 | #define HEAP_SIZE (2*L_CODES + 1) |
876 | /* maximum heap size */ |
877 | |
878 | ////static int heap[HEAP_SIZE]; /* heap used to build the Huffman trees */ |
879 | ////let's try this |
880 | static ush heap[HEAP_SIZE]; /* heap used to build the Huffman trees */ |
881 | static int heap_len; /* number of elements in the heap */ |
882 | static int heap_max; /* element of largest frequency */ |
883 | |
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. |
886 | */ |
887 | |
888 | static ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */ |
889 | static ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */ |
890 | |
891 | static ct_data static_ltree[L_CODES + 2]; |
892 | |
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 |
896 | * below). |
897 | */ |
898 | |
899 | static ct_data static_dtree[D_CODES]; |
900 | |
901 | /* The static distance tree. (Actually a trivial tree since all codes use |
902 | * 5 bits.) |
903 | */ |
904 | |
905 | static ct_data bl_tree[2 * BL_CODES + 1]; |
906 | |
907 | /* Huffman tree for the bit lengths */ |
908 | |
909 | typedef struct tree_desc { |
910 | ct_data *dyn_tree; /* the dynamic tree */ |
911 | ct_data *static_tree; /* corresponding static tree or NULL */ |
912 | const extra_bits_t *extra_bits; /* extra bits for each code or NULL */ |
913 | int extra_base; /* base index for extra_bits */ |
914 | int elems; /* max number of elements in the tree */ |
915 | int max_length; /* max bit length for the codes */ |
916 | int max_code; /* largest code with non zero frequency */ |
917 | } tree_desc; |
918 | |
919 | static tree_desc l_desc = { |
920 | dyn_ltree, static_ltree, extra_lbits, |
921 | LITERALS + 1, L_CODES, MAX_BITS, 0 |
922 | }; |
923 | |
924 | static tree_desc d_desc = { |
925 | dyn_dtree, static_dtree, extra_dbits, 0, D_CODES, MAX_BITS, 0 |
926 | }; |
927 | |
928 | static tree_desc bl_desc = { |
929 | bl_tree, NULL, extra_blbits, 0, BL_CODES, MAX_BL_BITS, 0 |
930 | }; |
931 | |
932 | |
933 | static ush bl_count[MAX_BITS + 1]; |
934 | |
935 | /* number of codes at each bit length for an optimal tree */ |
936 | |
937 | static const uch bl_order[BL_CODES] = { |
938 | 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 |
939 | }; |
940 | |
941 | /* The lengths of the bit length codes are sent in order of decreasing |
942 | * probability, to avoid transmitting the lengths for unused bit length codes. |
943 | */ |
944 | |
945 | static uch depth[2 * L_CODES + 1]; |
946 | |
947 | /* Depth of each subtree used as tie breaker for trees of equal frequency */ |
948 | |
949 | static uch length_code[MAX_MATCH - MIN_MATCH + 1]; |
950 | |
951 | /* length code for each normalized match length (0 == MIN_MATCH) */ |
952 | |
953 | static uch dist_code[512]; |
954 | |
955 | /* distance codes. The first 256 values correspond to the distances |
956 | * 3 .. 258, the last 256 values correspond to the top 8 bits of |
957 | * the 15 bit distances. |
958 | */ |
959 | |
960 | static int base_length[LENGTH_CODES]; |
961 | |
962 | /* First normalized length for each code (0 = MIN_MATCH) */ |
963 | |
964 | static int base_dist[D_CODES]; |
965 | |
966 | /* First normalized distance for each code (0 = distance of 1) */ |
967 | |
968 | static uch flag_buf[LIT_BUFSIZE / 8]; |
969 | |
970 | /* flag_buf is a bit array distinguishing literals from lengths in |
971 | * l_buf, thus indicating the presence or absence of a distance. |
972 | */ |
973 | |
974 | static unsigned last_lit; /* running index in l_buf */ |
975 | static unsigned last_dist; /* running index in d_buf */ |
976 | static unsigned last_flags; /* running index in flag_buf */ |
977 | static uch flags; /* current flags not yet saved in flag_buf */ |
978 | static uch flag_bit; /* current bit used in flags */ |
979 | |
980 | /* bits are filled in flags starting at bit 0 (least significant). |
981 | * Note: these flags are overkill in the current code since we don't |
982 | * take advantage of DIST_BUFSIZE == LIT_BUFSIZE. |
983 | */ |
984 | |
985 | static ulg opt_len; /* bit length of current block with optimal trees */ |
986 | static ulg static_len; /* bit length of current block with static trees */ |
987 | |
988 | static ulg compressed_len; /* total bit length of compressed file */ |
989 | |
990 | static ush *file_type; /* pointer to UNKNOWN, BINARY or ASCII */ |
991 | static int *file_method; /* pointer to DEFLATE or STORE */ |
992 | |
993 | /* =========================================================================== |
994 | */ |
995 | static void gen_codes(ct_data * tree, int max_code); |
996 | static void build_tree(tree_desc * desc); |
997 | static void scan_tree(ct_data * tree, int max_code); |
998 | static void send_tree(ct_data * tree, int max_code); |
999 | static int build_bl_tree(void); |
1000 | static void send_all_trees(int lcodes, int dcodes, int blcodes); |
1001 | static void compress_block(ct_data * ltree, ct_data * dtree); |
1002 | |
1003 | |
1004 | #ifndef DEBUG |
1005 | /* Send a code of the given tree. c and tree must not have side effects */ |
1006 | # define SEND_CODE(c, tree) send_bits(tree[c].Code, tree[c].Len) |
1007 | #else |
1008 | # define SEND_CODE(c, tree) \ |
1009 | { \ |
1010 | if (verbose > 1) bb_error_msg("\ncd %3d ",(c)); \ |
1011 | send_bits(tree[c].Code, tree[c].Len); \ |
1012 | } |
1013 | #endif |
1014 | |
1015 | #define D_CODE(dist) \ |
1016 | ((dist) < 256 ? dist_code[dist] : dist_code[256 + ((dist)>>7)]) |
1017 | /* Mapping from a distance to a distance code. dist is the distance - 1 and |
1018 | * must not have side effects. dist_code[256] and dist_code[257] are never |
1019 | * used. |
1020 | * The arguments must not have side effects. |
1021 | */ |
1022 | |
1023 | |
1024 | /* =========================================================================== |
1025 | * Initialize a new block. |
1026 | */ |
1027 | static void init_block(void) |
1028 | { |
1029 | int n; /* iterates over tree elements */ |
1030 | |
1031 | /* Initialize the trees. */ |
1032 | for (n = 0; n < L_CODES; n++) |
1033 | dyn_ltree[n].Freq = 0; |
1034 | for (n = 0; n < D_CODES; n++) |
1035 | dyn_dtree[n].Freq = 0; |
1036 | for (n = 0; n < BL_CODES; n++) |
1037 | bl_tree[n].Freq = 0; |
1038 | |
1039 | dyn_ltree[END_BLOCK].Freq = 1; |
1040 | opt_len = static_len = 0; |
1041 | last_lit = last_dist = last_flags = 0; |
1042 | flags = 0; |
1043 | flag_bit = 1; |
1044 | } |
1045 | |
1046 | |
1047 | /* =========================================================================== |
1048 | * Restore the heap property by moving down the tree starting at node k, |
1049 | * exchanging a node with the smallest of its two sons if necessary, stopping |
1050 | * when the heap property is re-established (each father smaller than its |
1051 | * two sons). |
1052 | */ |
1053 | |
1054 | /* Compares to subtrees, using the tree depth as tie breaker when |
1055 | * the subtrees have equal frequency. This minimizes the worst case length. */ |
1056 | #define SMALLER(tree, n, m) \ |
1057 | (tree[n].Freq < tree[m].Freq \ |
1058 | || (tree[n].Freq == tree[m].Freq && depth[n] <= depth[m])) |
1059 | |
1060 | static void pqdownheap(ct_data * tree, int k) |
1061 | { |
1062 | int v = heap[k]; |
1063 | int j = k << 1; /* left son of k */ |
1064 | |
1065 | while (j <= heap_len) { |
1066 | /* Set j to the smallest of the two sons: */ |
1067 | if (j < heap_len && SMALLER(tree, heap[j + 1], heap[j])) |
1068 | j++; |
1069 | |
1070 | /* Exit if v is smaller than both sons */ |
1071 | if (SMALLER(tree, v, heap[j])) |
1072 | break; |
1073 | |
1074 | /* Exchange v with the smallest son */ |
1075 | heap[k] = heap[j]; |
1076 | k = j; |
1077 | |
1078 | /* And continue down the tree, setting j to the left son of k */ |
1079 | j <<= 1; |
1080 | } |
1081 | heap[k] = v; |
1082 | } |
1083 | |
1084 | |
1085 | /* =========================================================================== |
1086 | * Compute the optimal bit lengths for a tree and update the total bit length |
1087 | * for the current block. |
1088 | * IN assertion: the fields freq and dad are set, heap[heap_max] and |
1089 | * above are the tree nodes sorted by increasing frequency. |
1090 | * OUT assertions: the field len is set to the optimal bit length, the |
1091 | * array bl_count contains the frequencies for each bit length. |
1092 | * The length opt_len is updated; static_len is also updated if stree is |
1093 | * not null. |
1094 | */ |
1095 | static void gen_bitlen(tree_desc * desc) |
1096 | { |
1097 | ct_data *tree = desc->dyn_tree; |
1098 | const extra_bits_t *extra = desc->extra_bits; |
1099 | int base = desc->extra_base; |
1100 | int max_code = desc->max_code; |
1101 | int max_length = desc->max_length; |
1102 | ct_data *stree = desc->static_tree; |
1103 | int h; /* heap index */ |
1104 | int n, m; /* iterate over the tree elements */ |
1105 | int bits; /* bit length */ |
1106 | int xbits; /* extra bits */ |
1107 | ush f; /* frequency */ |
1108 | int overflow = 0; /* number of elements with bit length too large */ |
1109 | |
1110 | for (bits = 0; bits <= MAX_BITS; bits++) |
1111 | bl_count[bits] = 0; |
1112 | |
1113 | /* In a first pass, compute the optimal bit lengths (which may |
1114 | * overflow in the case of the bit length tree). |
1115 | */ |
1116 | tree[heap[heap_max]].Len = 0; /* root of the heap */ |
1117 | |
1118 | for (h = heap_max + 1; h < HEAP_SIZE; h++) { |
1119 | n = heap[h]; |
1120 | bits = tree[tree[n].Dad].Len + 1; |
1121 | if (bits > max_length) { |
1122 | bits = max_length; |
1123 | overflow++; |
1124 | } |
1125 | tree[n].Len = (ush) bits; |
1126 | /* We overwrite tree[n].Dad which is no longer needed */ |
1127 | |
1128 | if (n > max_code) |
1129 | continue; /* not a leaf node */ |
1130 | |
1131 | bl_count[bits]++; |
1132 | xbits = 0; |
1133 | if (n >= base) |
1134 | xbits = extra[n - base]; |
1135 | f = tree[n].Freq; |
1136 | opt_len += (ulg) f *(bits + xbits); |
1137 | |
1138 | if (stree) |
1139 | static_len += (ulg) f * (stree[n].Len + xbits); |
1140 | } |
1141 | if (overflow == 0) |
1142 | return; |
1143 | |
1144 | Trace((stderr, "\nbit length overflow\n")); |
1145 | /* This happens for example on obj2 and pic of the Calgary corpus */ |
1146 | |
1147 | /* Find the first bit length which could increase: */ |
1148 | do { |
1149 | bits = max_length - 1; |
1150 | while (bl_count[bits] == 0) |
1151 | bits--; |
1152 | bl_count[bits]--; /* move one leaf down the tree */ |
1153 | bl_count[bits + 1] += 2; /* move one overflow item as its brother */ |
1154 | bl_count[max_length]--; |
1155 | /* The brother of the overflow item also moves one step up, |
1156 | * but this does not affect bl_count[max_length] |
1157 | */ |
1158 | overflow -= 2; |
1159 | } while (overflow > 0); |
1160 | |
1161 | /* Now recompute all bit lengths, scanning in increasing frequency. |
1162 | * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all |
1163 | * lengths instead of fixing only the wrong ones. This idea is taken |
1164 | * from 'ar' written by Haruhiko Okumura.) |
1165 | */ |
1166 | for (bits = max_length; bits != 0; bits--) { |
1167 | n = bl_count[bits]; |
1168 | while (n != 0) { |
1169 | m = heap[--h]; |
1170 | if (m > max_code) |
1171 | continue; |
1172 | if (tree[m].Len != (unsigned) bits) { |
1173 | Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len, bits)); |
1174 | opt_len += ((int32_t) bits - tree[m].Len) * tree[m].Freq; |
1175 | tree[m].Len = bits; |
1176 | } |
1177 | n--; |
1178 | } |
1179 | } |
1180 | } |
1181 | |
1182 | |
1183 | /* =========================================================================== |
1184 | * Generate the codes for a given tree and bit counts (which need not be |
1185 | * optimal). |
1186 | * IN assertion: the array bl_count contains the bit length statistics for |
1187 | * the given tree and the field len is set for all tree elements. |
1188 | * OUT assertion: the field code is set for all tree elements of non |
1189 | * zero code length. |
1190 | */ |
1191 | static void gen_codes(ct_data * tree, int max_code) |
1192 | { |
1193 | ush next_code[MAX_BITS + 1]; /* next code value for each bit length */ |
1194 | ush code = 0; /* running code value */ |
1195 | int bits; /* bit index */ |
1196 | int n; /* code index */ |
1197 | |
1198 | /* The distribution counts are first used to generate the code values |
1199 | * without bit reversal. |
1200 | */ |
1201 | for (bits = 1; bits <= MAX_BITS; bits++) { |
1202 | next_code[bits] = code = (code + bl_count[bits - 1]) << 1; |
1203 | } |
1204 | /* Check that the bit counts in bl_count are consistent. The last code |
1205 | * must be all ones. |
1206 | */ |
1207 | Assert(code + bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1, |
1208 | "inconsistent bit counts"); |
1209 | Tracev((stderr, "\ngen_codes: max_code %d ", max_code)); |
1210 | |
1211 | for (n = 0; n <= max_code; n++) { |
1212 | int len = tree[n].Len; |
1213 | |
1214 | if (len == 0) |
1215 | continue; |
1216 | /* Now reverse the bits */ |
1217 | tree[n].Code = bi_reverse(next_code[len]++, len); |
1218 | |
1219 | Tracec(tree != static_ltree, |
1220 | (stderr, "\nn %3d %c l %2d c %4x (%x) ", n, |
1221 | (isgraph(n) ? n : ' '), len, tree[n].Code, |
1222 | next_code[len] - 1)); |
1223 | } |
1224 | } |
1225 | |
1226 | |
1227 | /* =========================================================================== |
1228 | * Construct one Huffman tree and assigns the code bit strings and lengths. |
1229 | * Update the total bit length for the current block. |
1230 | * IN assertion: the field freq is set for all tree elements. |
1231 | * OUT assertions: the fields len and code are set to the optimal bit length |
1232 | * and corresponding code. The length opt_len is updated; static_len is |
1233 | * also updated if stree is not null. The field max_code is set. |
1234 | */ |
1235 | |
1236 | /* Remove the smallest element from the heap and recreate the heap with |
1237 | * one less element. Updates heap and heap_len. */ |
1238 | |
1239 | #define SMALLEST 1 |
1240 | /* Index within the heap array of least frequent node in the Huffman tree */ |
1241 | |
1242 | #define PQREMOVE(tree, top) \ |
1243 | { \ |
1244 | top = heap[SMALLEST]; \ |
1245 | heap[SMALLEST] = heap[heap_len--]; \ |
1246 | pqdownheap(tree, SMALLEST); \ |
1247 | } |
1248 | |
1249 | static void build_tree(tree_desc * desc) |
1250 | { |
1251 | ct_data *tree = desc->dyn_tree; |
1252 | ct_data *stree = desc->static_tree; |
1253 | int elems = desc->elems; |
1254 | int n, m; /* iterate over heap elements */ |
1255 | int max_code = -1; /* largest code with non zero frequency */ |
1256 | int node = elems; /* next internal node of the tree */ |
1257 | |
1258 | /* Construct the initial heap, with least frequent element in |
1259 | * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1]. |
1260 | * heap[0] is not used. |
1261 | */ |
1262 | heap_len = 0, heap_max = HEAP_SIZE; |
1263 | |
1264 | for (n = 0; n < elems; n++) { |
1265 | if (tree[n].Freq != 0) { |
1266 | heap[++heap_len] = max_code = n; |
1267 | depth[n] = 0; |
1268 | } else { |
1269 | tree[n].Len = 0; |
1270 | } |
1271 | } |
1272 | |
1273 | /* The pkzip format requires that at least one distance code exists, |
1274 | * and that at least one bit should be sent even if there is only one |
1275 | * possible code. So to avoid special checks later on we force at least |
1276 | * two codes of non zero frequency. |
1277 | */ |
1278 | while (heap_len < 2) { |
1279 | int new = heap[++heap_len] = (max_code < 2 ? ++max_code : 0); |
1280 | |
1281 | tree[new].Freq = 1; |
1282 | depth[new] = 0; |
1283 | opt_len--; |
1284 | if (stree) |
1285 | static_len -= stree[new].Len; |
1286 | /* new is 0 or 1 so it does not have extra bits */ |
1287 | } |
1288 | desc->max_code = max_code; |
1289 | |
1290 | /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree, |
1291 | * establish sub-heaps of increasing lengths: |
1292 | */ |
1293 | for (n = heap_len / 2; n >= 1; n--) |
1294 | pqdownheap(tree, n); |
1295 | |
1296 | /* Construct the Huffman tree by repeatedly combining the least two |
1297 | * frequent nodes. |
1298 | */ |
1299 | do { |
1300 | PQREMOVE(tree, n); /* n = node of least frequency */ |
1301 | m = heap[SMALLEST]; /* m = node of next least frequency */ |
1302 | |
1303 | heap[--heap_max] = n; /* keep the nodes sorted by frequency */ |
1304 | heap[--heap_max] = m; |
1305 | |
1306 | /* Create a new node father of n and m */ |
1307 | tree[node].Freq = tree[n].Freq + tree[m].Freq; |
1308 | depth[node] = MAX(depth[n], depth[m]) + 1; |
1309 | tree[n].Dad = tree[m].Dad = (ush) node; |
1310 | #ifdef DUMP_BL_TREE |
1311 | if (tree == bl_tree) { |
1312 | bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)", |
1313 | node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq); |
1314 | } |
1315 | #endif |
1316 | /* and insert the new node in the heap */ |
1317 | heap[SMALLEST] = node++; |
1318 | pqdownheap(tree, SMALLEST); |
1319 | |
1320 | } while (heap_len >= 2); |
1321 | |
1322 | heap[--heap_max] = heap[SMALLEST]; |
1323 | |
1324 | /* At this point, the fields freq and dad are set. We can now |
1325 | * generate the bit lengths. |
1326 | */ |
1327 | gen_bitlen((tree_desc *) desc); |
1328 | |
1329 | /* The field len is now set, we can generate the bit codes */ |
1330 | gen_codes((ct_data *) tree, max_code); |
1331 | } |
1332 | |
1333 | |
1334 | /* =========================================================================== |
1335 | * Scan a literal or distance tree to determine the frequencies of the codes |
1336 | * in the bit length tree. Updates opt_len to take into account the repeat |
1337 | * counts. (The contribution of the bit length codes will be added later |
1338 | * during the construction of bl_tree.) |
1339 | */ |
1340 | static void scan_tree(ct_data * tree, int max_code) |
1341 | { |
1342 | int n; /* iterates over all tree elements */ |
1343 | int prevlen = -1; /* last emitted length */ |
1344 | int curlen; /* length of current code */ |
1345 | int nextlen = tree[0].Len; /* length of next code */ |
1346 | int count = 0; /* repeat count of the current code */ |
1347 | int max_count = 7; /* max repeat count */ |
1348 | int min_count = 4; /* min repeat count */ |
1349 | |
1350 | if (nextlen == 0) { |
1351 | max_count = 138; |
1352 | min_count = 3; |
1353 | } |
1354 | tree[max_code + 1].Len = 0xffff; /* guard */ |
1355 | |
1356 | for (n = 0; n <= max_code; n++) { |
1357 | curlen = nextlen; |
1358 | nextlen = tree[n + 1].Len; |
1359 | if (++count < max_count && curlen == nextlen) |
1360 | continue; |
1361 | |
1362 | if (count < min_count) { |
1363 | bl_tree[curlen].Freq += count; |
1364 | } else if (curlen != 0) { |
1365 | if (curlen != prevlen) |
1366 | bl_tree[curlen].Freq++; |
1367 | bl_tree[REP_3_6].Freq++; |
1368 | } else if (count <= 10) { |
1369 | bl_tree[REPZ_3_10].Freq++; |
1370 | } else { |
1371 | bl_tree[REPZ_11_138].Freq++; |
1372 | } |
1373 | count = 0; |
1374 | prevlen = curlen; |
1375 | |
1376 | max_count = 7; |
1377 | min_count = 4; |
1378 | if (nextlen == 0) { |
1379 | max_count = 138; |
1380 | min_count = 3; |
1381 | } else if (curlen == nextlen) { |
1382 | max_count = 6; |
1383 | min_count = 3; |
1384 | } |
1385 | } |
1386 | } |
1387 | |
1388 | |
1389 | /* =========================================================================== |
1390 | * Send a literal or distance tree in compressed form, using the codes in |
1391 | * bl_tree. |
1392 | */ |
1393 | static void send_tree(ct_data * tree, int max_code) |
1394 | { |
1395 | int n; /* iterates over all tree elements */ |
1396 | int prevlen = -1; /* last emitted length */ |
1397 | int curlen; /* length of current code */ |
1398 | int nextlen = tree[0].Len; /* length of next code */ |
1399 | int count = 0; /* repeat count of the current code */ |
1400 | int max_count = 7; /* max repeat count */ |
1401 | int min_count = 4; /* min repeat count */ |
1402 | |
1403 | /* tree[max_code+1].Len = -1; *//* guard already set */ |
1404 | if (nextlen == 0) |
1405 | max_count = 138, min_count = 3; |
1406 | |
1407 | for (n = 0; n <= max_code; n++) { |
1408 | curlen = nextlen; |
1409 | nextlen = tree[n + 1].Len; |
1410 | if (++count < max_count && curlen == nextlen) { |
1411 | continue; |
1412 | } else if (count < min_count) { |
1413 | do { |
1414 | SEND_CODE(curlen, bl_tree); |
1415 | } while (--count); |
1416 | } else if (curlen != 0) { |
1417 | if (curlen != prevlen) { |
1418 | SEND_CODE(curlen, bl_tree); |
1419 | count--; |
1420 | } |
1421 | Assert(count >= 3 && count <= 6, " 3_6?"); |
1422 | SEND_CODE(REP_3_6, bl_tree); |
1423 | send_bits(count - 3, 2); |
1424 | } else if (count <= 10) { |
1425 | SEND_CODE(REPZ_3_10, bl_tree); |
1426 | send_bits(count - 3, 3); |
1427 | } else { |
1428 | SEND_CODE(REPZ_11_138, bl_tree); |
1429 | send_bits(count - 11, 7); |
1430 | } |
1431 | count = 0; |
1432 | prevlen = curlen; |
1433 | if (nextlen == 0) { |
1434 | max_count = 138; |
1435 | min_count = 3; |
1436 | } else if (curlen == nextlen) { |
1437 | max_count = 6; |
1438 | min_count = 3; |
1439 | } else { |
1440 | max_count = 7; |
1441 | min_count = 4; |
1442 | } |
1443 | } |
1444 | } |
1445 | |
1446 | |
1447 | /* =========================================================================== |
1448 | * Construct the Huffman tree for the bit lengths and return the index in |
1449 | * bl_order of the last bit length code to send. |
1450 | */ |
1451 | static int build_bl_tree(void) |
1452 | { |
1453 | int max_blindex; /* index of last bit length code of non zero freq */ |
1454 | |
1455 | /* Determine the bit length frequencies for literal and distance trees */ |
1456 | scan_tree((ct_data *) dyn_ltree, l_desc.max_code); |
1457 | scan_tree((ct_data *) dyn_dtree, d_desc.max_code); |
1458 | |
1459 | /* Build the bit length tree: */ |
1460 | build_tree((tree_desc *) &bl_desc); |
1461 | /* opt_len now includes the length of the tree representations, except |
1462 | * the lengths of the bit lengths codes and the 5+5+4 bits for the counts. |
1463 | */ |
1464 | |
1465 | /* Determine the number of bit length codes to send. The pkzip format |
1466 | * requires that at least 4 bit length codes be sent. (appnote.txt says |
1467 | * 3 but the actual value used is 4.) |
1468 | */ |
1469 | for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) { |
1470 | if (bl_tree[bl_order[max_blindex]].Len != 0) |
1471 | break; |
1472 | } |
1473 | /* Update opt_len to include the bit length tree and counts */ |
1474 | opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4; |
1475 | Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", opt_len, static_len)); |
1476 | |
1477 | return max_blindex; |
1478 | } |
1479 | |
1480 | |
1481 | /* =========================================================================== |
1482 | * Send the header for a block using dynamic Huffman trees: the counts, the |
1483 | * lengths of the bit length codes, the literal tree and the distance tree. |
1484 | * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4. |
1485 | */ |
1486 | static void send_all_trees(int lcodes, int dcodes, int blcodes) |
1487 | { |
1488 | int rank; /* index in bl_order */ |
1489 | |
1490 | Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes"); |
1491 | Assert(lcodes <= L_CODES && dcodes <= D_CODES |
1492 | && blcodes <= BL_CODES, "too many codes"); |
1493 | Tracev((stderr, "\nbl counts: ")); |
1494 | send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */ |
1495 | send_bits(dcodes - 1, 5); |
1496 | send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */ |
1497 | for (rank = 0; rank < blcodes; rank++) { |
1498 | Tracev((stderr, "\nbl code %2d ", bl_order[rank])); |
1499 | send_bits(bl_tree[bl_order[rank]].Len, 3); |
1500 | } |
1501 | Tracev((stderr, "\nbl tree: sent %ld", bits_sent)); |
1502 | |
1503 | send_tree((ct_data *) dyn_ltree, lcodes - 1); /* send the literal tree */ |
1504 | Tracev((stderr, "\nlit tree: sent %ld", bits_sent)); |
1505 | |
1506 | send_tree((ct_data *) dyn_dtree, dcodes - 1); /* send the distance tree */ |
1507 | Tracev((stderr, "\ndist tree: sent %ld", bits_sent)); |
1508 | } |
1509 | |
1510 | |
1511 | /* =========================================================================== |
1512 | * Set the file type to ASCII or BINARY, using a crude approximation: |
1513 | * binary if more than 20% of the bytes are <= 6 or >= 128, ascii otherwise. |
1514 | * IN assertion: the fields freq of dyn_ltree are set and the total of all |
1515 | * frequencies does not exceed 64K (to fit in an int on 16 bit machines). |
1516 | */ |
1517 | static void set_file_type(void) |
1518 | { |
1519 | int n = 0; |
1520 | unsigned ascii_freq = 0; |
1521 | unsigned bin_freq = 0; |
1522 | |
1523 | while (n < 7) |
1524 | bin_freq += dyn_ltree[n++].Freq; |
1525 | while (n < 128) |
1526 | ascii_freq += dyn_ltree[n++].Freq; |
1527 | while (n < LITERALS) |
1528 | bin_freq += dyn_ltree[n++].Freq; |
1529 | *file_type = (bin_freq > (ascii_freq >> 2)) ? BINARY : ASCII; |
1530 | if (*file_type == BINARY && translate_eol) { |
1531 | bb_error_msg("-l used on binary file"); |
1532 | } |
1533 | } |
1534 | |
1535 | |
1536 | /* =========================================================================== |
1537 | * Save the match info and tally the frequency counts. Return true if |
1538 | * the current block must be flushed. |
1539 | */ |
1540 | static int ct_tally(int dist, int lc) |
1541 | { |
1542 | l_buf[last_lit++] = lc; |
1543 | if (dist == 0) { |
1544 | /* lc is the unmatched char */ |
1545 | dyn_ltree[lc].Freq++; |
1546 | } else { |
1547 | /* Here, lc is the match length - MIN_MATCH */ |
1548 | dist--; /* dist = match distance - 1 */ |
1549 | Assert((ush) dist < (ush) MAX_DIST |
1550 | && (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH) |
1551 | && (ush) D_CODE(dist) < (ush) D_CODES, "ct_tally: bad match" |
1552 | ); |
1553 | |
1554 | dyn_ltree[length_code[lc] + LITERALS + 1].Freq++; |
1555 | dyn_dtree[D_CODE(dist)].Freq++; |
1556 | |
1557 | d_buf[last_dist++] = dist; |
1558 | flags |= flag_bit; |
1559 | } |
1560 | flag_bit <<= 1; |
1561 | |
1562 | /* Output the flags if they fill a byte: */ |
1563 | if ((last_lit & 7) == 0) { |
1564 | flag_buf[last_flags++] = flags; |
1565 | flags = 0, flag_bit = 1; |
1566 | } |
1567 | /* Try to guess if it is profitable to stop the current block here */ |
1568 | if ((last_lit & 0xfff) == 0) { |
1569 | /* Compute an upper bound for the compressed length */ |
1570 | ulg out_length = last_lit * 8L; |
1571 | ulg in_length = (ulg) strstart - block_start; |
1572 | int dcode; |
1573 | |
1574 | for (dcode = 0; dcode < D_CODES; dcode++) { |
1575 | out_length += dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]); |
1576 | } |
1577 | out_length >>= 3; |
1578 | Trace((stderr, |
1579 | "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ", |
1580 | last_lit, last_dist, in_length, out_length, |
1581 | 100L - out_length * 100L / in_length)); |
1582 | if (last_dist < last_lit / 2 && out_length < in_length / 2) |
1583 | return 1; |
1584 | } |
1585 | return (last_lit == LIT_BUFSIZE - 1 || last_dist == DIST_BUFSIZE); |
1586 | /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K |
1587 | * on 16 bit machines and because stored blocks are restricted to |
1588 | * 64K-1 bytes. |
1589 | */ |
1590 | } |
1591 | |
1592 | /* =========================================================================== |
1593 | * Send the block data compressed using the given Huffman trees |
1594 | */ |
1595 | static void compress_block(ct_data * ltree, ct_data * dtree) |
1596 | { |
1597 | unsigned dist; /* distance of matched string */ |
1598 | int lc; /* match length or unmatched char (if dist == 0) */ |
1599 | unsigned lx = 0; /* running index in l_buf */ |
1600 | unsigned dx = 0; /* running index in d_buf */ |
1601 | unsigned fx = 0; /* running index in flag_buf */ |
1602 | uch flag = 0; /* current flags */ |
1603 | unsigned code; /* the code to send */ |
1604 | int extra; /* number of extra bits to send */ |
1605 | |
1606 | if (last_lit != 0) do { |
1607 | if ((lx & 7) == 0) |
1608 | flag = flag_buf[fx++]; |
1609 | lc = l_buf[lx++]; |
1610 | if ((flag & 1) == 0) { |
1611 | SEND_CODE(lc, ltree); /* send a literal byte */ |
1612 | Tracecv(isgraph(lc), (stderr, " '%c' ", lc)); |
1613 | } else { |
1614 | /* Here, lc is the match length - MIN_MATCH */ |
1615 | code = length_code[lc]; |
1616 | SEND_CODE(code + LITERALS + 1, ltree); /* send the length code */ |
1617 | extra = extra_lbits[code]; |
1618 | if (extra != 0) { |
1619 | lc -= base_length[code]; |
1620 | send_bits(lc, extra); /* send the extra length bits */ |
1621 | } |
1622 | dist = d_buf[dx++]; |
1623 | /* Here, dist is the match distance - 1 */ |
1624 | code = D_CODE(dist); |
1625 | Assert(code < D_CODES, "bad d_code"); |
1626 | |
1627 | SEND_CODE(code, dtree); /* send the distance code */ |
1628 | extra = extra_dbits[code]; |
1629 | if (extra != 0) { |
1630 | dist -= base_dist[code]; |
1631 | send_bits(dist, extra); /* send the extra distance bits */ |
1632 | } |
1633 | } /* literal or match pair ? */ |
1634 | flag >>= 1; |
1635 | } while (lx < last_lit); |
1636 | |
1637 | SEND_CODE(END_BLOCK, ltree); |
1638 | } |
1639 | |
1640 | |
1641 | /* =========================================================================== |
1642 | * Determine the best encoding for the current block: dynamic trees, static |
1643 | * trees or store, and output the encoded block to the zip file. This function |
1644 | * returns the total compressed length for the file so far. |
1645 | */ |
1646 | static ulg flush_block(char *buf, ulg stored_len, int eof) |
1647 | { |
1648 | ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */ |
1649 | int max_blindex; /* index of last bit length code of non zero freq */ |
1650 | |
1651 | flag_buf[last_flags] = flags; /* Save the flags for the last 8 items */ |
1652 | |
1653 | /* Check if the file is ascii or binary */ |
1654 | if (*file_type == (ush) UNKNOWN) |
1655 | set_file_type(); |
1656 | |
1657 | /* Construct the literal and distance trees */ |
1658 | build_tree((tree_desc *) &l_desc); |
1659 | Tracev((stderr, "\nlit data: dyn %ld, stat %ld", opt_len, static_len)); |
1660 | |
1661 | build_tree((tree_desc *) &d_desc); |
1662 | Tracev((stderr, "\ndist data: dyn %ld, stat %ld", opt_len, static_len)); |
1663 | /* At this point, opt_len and static_len are the total bit lengths of |
1664 | * the compressed block data, excluding the tree representations. |
1665 | */ |
1666 | |
1667 | /* Build the bit length tree for the above two trees, and get the index |
1668 | * in bl_order of the last bit length code to send. |
1669 | */ |
1670 | max_blindex = build_bl_tree(); |
1671 | |
1672 | /* Determine the best encoding. Compute first the block length in bytes */ |
1673 | opt_lenb = (opt_len + 3 + 7) >> 3; |
1674 | static_lenb = (static_len + 3 + 7) >> 3; |
1675 | |
1676 | Trace((stderr, |
1677 | "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ", |
1678 | opt_lenb, opt_len, static_lenb, static_len, stored_len, |
1679 | last_lit, last_dist)); |
1680 | |
1681 | if (static_lenb <= opt_lenb) |
1682 | opt_lenb = static_lenb; |
1683 | |
1684 | /* If compression failed and this is the first and last block, |
1685 | * and if the zip file can be seeked (to rewrite the local header), |
1686 | * the whole file is transformed into a stored file: |
1687 | */ |
1688 | if (stored_len <= opt_lenb && eof && compressed_len == 0L && seekable()) { |
1689 | /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */ |
1690 | if (buf == NULL) |
1691 | bb_error_msg("block vanished"); |
1692 | |
1693 | copy_block(buf, (unsigned) stored_len, 0); /* without header */ |
1694 | compressed_len = stored_len << 3; |
1695 | *file_method = STORED; |
1696 | |
1697 | } else if (stored_len + 4 <= opt_lenb && buf != NULL) { |
1698 | /* 4: two words for the lengths */ |
1699 | /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE. |
1700 | * Otherwise we can't have processed more than WSIZE input bytes since |
1701 | * the last block flush, because compression would have been |
1702 | * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to |
1703 | * transform a block into a stored block. |
1704 | */ |
1705 | send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */ |
1706 | compressed_len = (compressed_len + 3 + 7) & ~7L; |
1707 | compressed_len += (stored_len + 4) << 3; |
1708 | |
1709 | copy_block(buf, (unsigned) stored_len, 1); /* with header */ |
1710 | |
1711 | } else if (static_lenb == opt_lenb) { |
1712 | send_bits((STATIC_TREES << 1) + eof, 3); |
1713 | compress_block((ct_data *) static_ltree, (ct_data *) static_dtree); |
1714 | compressed_len += 3 + static_len; |
1715 | } else { |
1716 | send_bits((DYN_TREES << 1) + eof, 3); |
1717 | send_all_trees(l_desc.max_code + 1, d_desc.max_code + 1, |
1718 | max_blindex + 1); |
1719 | compress_block((ct_data *) dyn_ltree, (ct_data *) dyn_dtree); |
1720 | compressed_len += 3 + opt_len; |
1721 | } |
1722 | Assert(compressed_len == bits_sent, "bad compressed size"); |
1723 | init_block(); |
1724 | |
1725 | if (eof) { |
1726 | bi_windup(); |
1727 | compressed_len += 7; /* align on byte boundary */ |
1728 | } |
1729 | Tracev((stderr, "\ncomprlen %lu(%lu) ", compressed_len >> 3, |
1730 | compressed_len - 7 * eof)); |
1731 | |
1732 | return compressed_len >> 3; |
1733 | } |
1734 | |
1735 | |
1736 | /* =========================================================================== |
1737 | * Update a hash value with the given input byte |
1738 | * IN assertion: all calls to to UPDATE_HASH are made with consecutive |
1739 | * input characters, so that a running hash key can be computed from the |
1740 | * previous key instead of complete recalculation each time. |
1741 | */ |
1742 | #define UPDATE_HASH(h, c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK) |
1743 | |
1744 | |
1745 | /* =========================================================================== |
1746 | * Same as above, but achieves better compression. We use a lazy |
1747 | * evaluation for matches: a match is finally adopted only if there is |
1748 | * no better match at the next window position. |
1749 | * |
1750 | * Processes a new input file and return its compressed length. Sets |
1751 | * the compressed length, crc, deflate flags and internal file |
1752 | * attributes. |
1753 | */ |
1754 | |
1755 | /* Flush the current block, with given end-of-file flag. |
1756 | * IN assertion: strstart is set to the end of the current match. */ |
1757 | #define FLUSH_BLOCK(eof) \ |
1758 | flush_block( \ |
1759 | block_start >= 0L \ |
1760 | ? (char*)&window[(unsigned)block_start] \ |
1761 | : (char*)NULL, \ |
1762 | (ulg)strstart - block_start, \ |
1763 | (eof) \ |
1764 | ) |
1765 | |
1766 | /* Insert string s in the dictionary and set match_head to the previous head |
1767 | * of the hash chain (the most recent string with same hash key). Return |
1768 | * the previous length of the hash chain. |
1769 | * IN assertion: all calls to to INSERT_STRING are made with consecutive |
1770 | * input characters and the first MIN_MATCH bytes of s are valid |
1771 | * (except for the last MIN_MATCH-1 bytes of the input file). */ |
1772 | #define INSERT_STRING(s, match_head) \ |
1773 | { \ |
1774 | UPDATE_HASH(ins_h, window[(s) + MIN_MATCH-1]); \ |
1775 | prev[(s) & WMASK] = match_head = head[ins_h]; \ |
1776 | head[ins_h] = (s); \ |
1777 | } |
1778 | |
1779 | static ulg deflate(void) |
1780 | { |
1781 | IPos hash_head; /* head of hash chain */ |
1782 | IPos prev_match; /* previous match */ |
1783 | int flush; /* set if current block must be flushed */ |
1784 | int match_available = 0; /* set if previous match exists */ |
1785 | unsigned match_length = MIN_MATCH - 1; /* length of best match */ |
1786 | |
1787 | /* Process the input block. */ |
1788 | while (lookahead != 0) { |
1789 | /* Insert the string window[strstart .. strstart+2] in the |
1790 | * dictionary, and set hash_head to the head of the hash chain: |
1791 | */ |
1792 | INSERT_STRING(strstart, hash_head); |
1793 | |
1794 | /* Find the longest match, discarding those <= prev_length. |
1795 | */ |
1796 | prev_length = match_length, prev_match = match_start; |
1797 | match_length = MIN_MATCH - 1; |
1798 | |
1799 | if (hash_head != 0 && prev_length < max_lazy_match |
1800 | && strstart - hash_head <= MAX_DIST |
1801 | ) { |
1802 | /* To simplify the code, we prevent matches with the string |
1803 | * of window index 0 (in particular we have to avoid a match |
1804 | * of the string with itself at the start of the input file). |
1805 | */ |
1806 | match_length = longest_match(hash_head); |
1807 | /* longest_match() sets match_start */ |
1808 | if (match_length > lookahead) |
1809 | match_length = lookahead; |
1810 | |
1811 | /* Ignore a length 3 match if it is too distant: */ |
1812 | if (match_length == MIN_MATCH && strstart - match_start > TOO_FAR) { |
1813 | /* If prev_match is also MIN_MATCH, match_start is garbage |
1814 | * but we will ignore the current match anyway. |
1815 | */ |
1816 | match_length--; |
1817 | } |
1818 | } |
1819 | /* If there was a match at the previous step and the current |
1820 | * match is not better, output the previous match: |
1821 | */ |
1822 | if (prev_length >= MIN_MATCH && match_length <= prev_length) { |
1823 | check_match(strstart - 1, prev_match, prev_length); |
1824 | flush = ct_tally(strstart - 1 - prev_match, prev_length - MIN_MATCH); |
1825 | |
1826 | /* Insert in hash table all strings up to the end of the match. |
1827 | * strstart-1 and strstart are already inserted. |
1828 | */ |
1829 | lookahead -= prev_length - 1; |
1830 | prev_length -= 2; |
1831 | do { |
1832 | strstart++; |
1833 | INSERT_STRING(strstart, hash_head); |
1834 | /* strstart never exceeds WSIZE-MAX_MATCH, so there are |
1835 | * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH |
1836 | * these bytes are garbage, but it does not matter since the |
1837 | * next lookahead bytes will always be emitted as literals. |
1838 | */ |
1839 | } while (--prev_length != 0); |
1840 | match_available = 0; |
1841 | match_length = MIN_MATCH - 1; |
1842 | strstart++; |
1843 | if (flush) { |
1844 | FLUSH_BLOCK(0); |
1845 | block_start = strstart; |
1846 | } |
1847 | } else if (match_available) { |
1848 | /* If there was no match at the previous position, output a |
1849 | * single literal. If there was a match but the current match |
1850 | * is longer, truncate the previous match to a single literal. |
1851 | */ |
1852 | Tracevv((stderr, "%c", window[strstart - 1])); |
1853 | if (ct_tally(0, window[strstart - 1])) { |
1854 | FLUSH_BLOCK(0); |
1855 | block_start = strstart; |
1856 | } |
1857 | strstart++; |
1858 | lookahead--; |
1859 | } else { |
1860 | /* There is no previous match to compare with, wait for |
1861 | * the next step to decide. |
1862 | */ |
1863 | match_available = 1; |
1864 | strstart++; |
1865 | lookahead--; |
1866 | } |
1867 | Assert(strstart <= isize && lookahead <= isize, "a bit too far"); |
1868 | |
1869 | /* Make sure that we always have enough lookahead, except |
1870 | * at the end of the input file. We need MAX_MATCH bytes |
1871 | * for the next match, plus MIN_MATCH bytes to insert the |
1872 | * string following the next match. |
1873 | */ |
1874 | while (lookahead < MIN_LOOKAHEAD && !eofile) |
1875 | fill_window(); |
1876 | } |
1877 | if (match_available) |
1878 | ct_tally(0, window[strstart - 1]); |
1879 | |
1880 | return FLUSH_BLOCK(1); /* eof */ |
1881 | } |
1882 | |
1883 | |
1884 | /* =========================================================================== |
1885 | * Initialize the bit string routines. |
1886 | */ |
1887 | static void bi_init(void) //// int zipfile) |
1888 | { |
1889 | //// zfile = zipfile; |
1890 | bi_buf = 0; |
1891 | bi_valid = 0; |
1892 | #ifdef DEBUG |
1893 | bits_sent = 0L; |
1894 | #endif |
1895 | } |
1896 | |
1897 | |
1898 | /* =========================================================================== |
1899 | * Initialize the "longest match" routines for a new file |
1900 | */ |
1901 | static void lm_init(ush * flagsp) |
1902 | { |
1903 | unsigned j; |
1904 | |
1905 | /* Initialize the hash table. */ |
1906 | memset(head, 0, HASH_SIZE * sizeof(*head)); |
1907 | /* prev will be initialized on the fly */ |
1908 | |
1909 | /* speed options for the general purpose bit flag */ |
1910 | *flagsp |= 2; /* FAST 4, SLOW 2 */ |
1911 | /* ??? reduce max_chain_length for binary files */ |
1912 | |
1913 | strstart = 0; |
1914 | block_start = 0L; |
1915 | |
1916 | lookahead = file_read(window, |
1917 | sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE); |
1918 | |
1919 | if (lookahead == 0 || lookahead == (unsigned) -1) { |
1920 | eofile = 1; |
1921 | lookahead = 0; |
1922 | return; |
1923 | } |
1924 | eofile = 0; |
1925 | /* Make sure that we always have enough lookahead. This is important |
1926 | * if input comes from a device such as a tty. |
1927 | */ |
1928 | while (lookahead < MIN_LOOKAHEAD && !eofile) |
1929 | fill_window(); |
1930 | |
1931 | ins_h = 0; |
1932 | for (j = 0; j < MIN_MATCH - 1; j++) |
1933 | UPDATE_HASH(ins_h, window[j]); |
1934 | /* If lookahead < MIN_MATCH, ins_h is garbage, but this is |
1935 | * not important since only literal bytes will be emitted. |
1936 | */ |
1937 | } |
1938 | |
1939 | |
1940 | /* =========================================================================== |
1941 | * Allocate the match buffer, initialize the various tables and save the |
1942 | * location of the internal file attribute (ascii/binary) and method |
1943 | * (DEFLATE/STORE). |
1944 | * One callsite in zip() |
1945 | */ |
1946 | static void ct_init(ush * attr, int *methodp) |
1947 | { |
1948 | int n; /* iterates over tree elements */ |
1949 | int length; /* length value */ |
1950 | int code; /* code value */ |
1951 | int dist; /* distance index */ |
1952 | |
1953 | file_type = attr; |
1954 | file_method = methodp; |
1955 | compressed_len = 0L; |
1956 | |
1957 | #ifdef NOT_NEEDED |
1958 | if (static_dtree[0].Len != 0) |
1959 | return; /* ct_init already called */ |
1960 | #endif |
1961 | |
1962 | /* Initialize the mapping length (0..255) -> length code (0..28) */ |
1963 | length = 0; |
1964 | for (code = 0; code < LENGTH_CODES - 1; code++) { |
1965 | base_length[code] = length; |
1966 | for (n = 0; n < (1 << extra_lbits[code]); n++) { |
1967 | length_code[length++] = code; |
1968 | } |
1969 | } |
1970 | Assert(length == 256, "ct_init: length != 256"); |
1971 | /* Note that the length 255 (match length 258) can be represented |
1972 | * in two different ways: code 284 + 5 bits or code 285, so we |
1973 | * overwrite length_code[255] to use the best encoding: |
1974 | */ |
1975 | length_code[length - 1] = code; |
1976 | |
1977 | /* Initialize the mapping dist (0..32K) -> dist code (0..29) */ |
1978 | dist = 0; |
1979 | for (code = 0; code < 16; code++) { |
1980 | base_dist[code] = dist; |
1981 | for (n = 0; n < (1 << extra_dbits[code]); n++) { |
1982 | dist_code[dist++] = code; |
1983 | } |
1984 | } |
1985 | Assert(dist == 256, "ct_init: dist != 256"); |
1986 | dist >>= 7; /* from now on, all distances are divided by 128 */ |
1987 | for (; code < D_CODES; code++) { |
1988 | base_dist[code] = dist << 7; |
1989 | for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) { |
1990 | dist_code[256 + dist++] = code; |
1991 | } |
1992 | } |
1993 | Assert(dist == 256, "ct_init: 256+dist != 512"); |
1994 | |
1995 | /* Construct the codes of the static literal tree */ |
1996 | /* already zeroed - it's in bss |
1997 | for (n = 0; n <= MAX_BITS; n++) |
1998 | bl_count[n] = 0; */ |
1999 | |
2000 | n = 0; |
2001 | while (n <= 143) { |
2002 | static_ltree[n++].Len = 8; |
2003 | bl_count[8]++; |
2004 | } |
2005 | while (n <= 255) { |
2006 | static_ltree[n++].Len = 9; |
2007 | bl_count[9]++; |
2008 | } |
2009 | while (n <= 279) { |
2010 | static_ltree[n++].Len = 7; |
2011 | bl_count[7]++; |
2012 | } |
2013 | while (n <= 287) { |
2014 | static_ltree[n++].Len = 8; |
2015 | bl_count[8]++; |
2016 | } |
2017 | /* Codes 286 and 287 do not exist, but we must include them in the |
2018 | * tree construction to get a canonical Huffman tree (longest code |
2019 | * all ones) |
2020 | */ |
2021 | gen_codes((ct_data *) static_ltree, L_CODES + 1); |
2022 | |
2023 | /* The static distance tree is trivial: */ |
2024 | for (n = 0; n < D_CODES; n++) { |
2025 | static_dtree[n].Len = 5; |
2026 | static_dtree[n].Code = bi_reverse(n, 5); |
2027 | } |
2028 | |
2029 | /* Initialize the first block of the first file: */ |
2030 | init_block(); |
2031 | } |
2032 | |
2033 | |
2034 | /* =========================================================================== |
2035 | * Deflate in to out. |
2036 | * IN assertions: the input and output buffers are cleared. |
2037 | * The variables time_stamp and save_orig_name are initialized. |
2038 | */ |
2039 | |
2040 | /* put_header_byte is used for the compressed output |
2041 | * - for the initial 4 bytes that can't overflow the buffer. */ |
2042 | #define put_header_byte(c) outbuf[outcnt++] = (c) |
2043 | |
2044 | static void zip(int in, int out) |
2045 | { |
2046 | uch my_flags = 0; /* general purpose bit flags */ |
2047 | ush attr = 0; /* ascii/binary flag */ |
2048 | ush deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */ |
2049 | |
2050 | ifd = in; |
2051 | ofd = out; |
2052 | outcnt = 0; |
2053 | |
2054 | /* Write the header to the gzip file. See algorithm.doc for the format */ |
2055 | |
2056 | method = DEFLATED; |
2057 | put_header_byte(0x1f); /* magic header for gzip files, 1F 8B */ |
2058 | put_header_byte(0x8b); |
2059 | put_header_byte(DEFLATED); /* compression method */ |
2060 | put_header_byte(my_flags); /* general flags */ |
2061 | put_32bit(time_stamp); |
2062 | |
2063 | /* Write deflated file to zip file */ |
2064 | crc = ~0; |
2065 | |
2066 | bi_init(); //// (out); |
2067 | ct_init(&attr, &method); |
2068 | lm_init(&deflate_flags); |
2069 | |
2070 | put_8bit(deflate_flags); /* extra flags */ |
2071 | put_8bit(3); /* OS identifier = 3 (Unix) */ |
2072 | |
2073 | deflate(); |
2074 | |
2075 | /* Write the crc and uncompressed size */ |
2076 | put_32bit(~crc); |
2077 | put_32bit(isize); |
2078 | |
2079 | flush_outbuf(); |
2080 | } |
2081 | |
2082 | |
2083 | /* ======================================================================== */ |
2084 | static void abort_gzip(int ATTRIBUTE_UNUSED ignored) |
2085 | { |
2086 | exit(1); |
2087 | } |
2088 | |
2089 | int gzip_main(int argc, char **argv) |
2090 | { |
2091 | enum { |
2092 | OPT_tostdout = 0x1, |
2093 | OPT_force = 0x2, |
2094 | }; |
2095 | |
2096 | unsigned opt; |
2097 | int inFileNum; |
2098 | int outFileNum; |
2099 | int i; |
2100 | struct stat statBuf; |
2101 | |
2102 | opt = getopt32(argc, argv, "cf123456789qv" USE_GUNZIP("d")); |
2103 | //if (opt & 0x1) // -c |
2104 | //if (opt & 0x2) // -f |
2105 | /* Ignore 1-9 (compression level) options */ |
2106 | //if (opt & 0x4) // -1 |
2107 | //if (opt & 0x8) // -2 |
2108 | //if (opt & 0x10) // -3 |
2109 | //if (opt & 0x20) // -4 |
2110 | //if (opt & 0x40) // -5 |
2111 | //if (opt & 0x80) // -6 |
2112 | //if (opt & 0x100) // -7 |
2113 | //if (opt & 0x200) // -8 |
2114 | //if (opt & 0x400) // -9 |
2115 | //if (opt & 0x800) // -q |
2116 | //if (opt & 0x1000) // -v |
2117 | #if ENABLE_GUNZIP /* gunzip_main may not be visible... */ |
2118 | if (opt & 0x2000) { // -d |
2119 | /* FIXME: getopt32 should not depend on optind */ |
2120 | optind = 1; |
2121 | return gunzip_main(argc, argv); |
2122 | } |
2123 | #endif |
2124 | |
2125 | foreground = signal(SIGINT, SIG_IGN) != SIG_IGN; |
2126 | if (foreground) { |
2127 | signal(SIGINT, abort_gzip); |
2128 | } |
2129 | #ifdef SIGTERM |
2130 | if (signal(SIGTERM, SIG_IGN) != SIG_IGN) { |
2131 | signal(SIGTERM, abort_gzip); |
2132 | } |
2133 | #endif |
2134 | #ifdef SIGHUP |
2135 | if (signal(SIGHUP, SIG_IGN) != SIG_IGN) { |
2136 | signal(SIGHUP, abort_gzip); |
2137 | } |
2138 | #endif |
2139 | |
2140 | /* Allocate all global buffers (for DYN_ALLOC option) */ |
2141 | ALLOC(uch, l_buf, INBUFSIZ); |
2142 | ALLOC(uch, outbuf, OUTBUFSIZ); |
2143 | ALLOC(ush, d_buf, DIST_BUFSIZE); |
2144 | ALLOC(uch, window, 2L * WSIZE); |
2145 | ALLOC(ush, prev, 1L << BITS); |
2146 | |
2147 | /* Initialise the CRC32 table */ |
2148 | crc_32_tab = crc32_filltable(0); |
2149 | |
2150 | clear_bufs(); |
2151 | |
2152 | if (optind == argc) { |
2153 | time_stamp = 0; |
2154 | zip(STDIN_FILENO, STDOUT_FILENO); |
2155 | return exit_code; |
2156 | } |
2157 | |
2158 | for (i = optind; i < argc; i++) { |
2159 | char *path = NULL; |
2160 | |
2161 | clear_bufs(); |
2162 | if (LONE_DASH(argv[i])) { |
2163 | time_stamp = 0; |
2164 | inFileNum = STDIN_FILENO; |
2165 | outFileNum = STDOUT_FILENO; |
2166 | } else { |
2167 | inFileNum = xopen(argv[i], O_RDONLY); |
2168 | if (fstat(inFileNum, &statBuf) < 0) |
2169 | bb_perror_msg_and_die("%s", argv[i]); |
2170 | time_stamp = statBuf.st_ctime; |
2171 | |
2172 | if (!(opt & OPT_tostdout)) { |
2173 | path = xasprintf("%s.gz", argv[i]); |
2174 | |
2175 | /* Open output file */ |
2176 | #if defined(__GLIBC__) && __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 1 && defined(O_NOFOLLOW) |
2177 | outFileNum = open(path, O_RDWR | O_CREAT | O_EXCL | O_NOFOLLOW); |
2178 | #else |
2179 | outFileNum = open(path, O_RDWR | O_CREAT | O_EXCL); |
2180 | #endif |
2181 | if (outFileNum < 0) { |
2182 | bb_perror_msg("%s", path); |
2183 | free(path); |
2184 | continue; |
2185 | } |
2186 | |
2187 | /* Set permissions on the file */ |
2188 | fchmod(outFileNum, statBuf.st_mode); |
2189 | } else |
2190 | outFileNum = STDOUT_FILENO; |
2191 | } |
2192 | |
2193 | if (path == NULL && isatty(outFileNum) && !(opt & OPT_force)) { |
2194 | bb_error_msg("compressed data not written " |
2195 | "to a terminal. Use -f to force compression."); |
2196 | free(path); |
2197 | continue; |
2198 | } |
2199 | |
2200 | zip(inFileNum, outFileNum); |
2201 | |
2202 | if (path != NULL) { |
2203 | char *delFileName; |
2204 | |
2205 | close(inFileNum); |
2206 | close(outFileNum); |
2207 | |
2208 | /* Delete the original file */ |
2209 | // Pity we don't propagate zip failures to this place... |
2210 | //if (zip_is_ok) |
2211 | delFileName = argv[i]; |
2212 | //else |
2213 | // delFileName = path; |
2214 | if (unlink(delFileName) < 0) |
2215 | bb_perror_msg("%s", delFileName); |
2216 | } |
2217 | |
2218 | free(path); |
2219 | } |
2220 | |
2221 | return exit_code; |
2222 | } |