Contents of /tags/mkinitrd-6_2_1/busybox/archival/gzip.c
Parent Directory | Revision Log
Revision 999 -
(show annotations)
(download)
Sun May 30 12:16:23 2010 UTC (14 years, 4 months ago) by niro
File MIME type: text/plain
File size: 65853 byte(s)
Sun May 30 12:16:23 2010 UTC (14 years, 4 months ago) by niro
File MIME type: text/plain
File size: 65853 byte(s)
tagged 'mkinitrd-6_2_1'
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 "libbb.h" |
43 | #include "unarchive.h" |
44 | |
45 | |
46 | /* =========================================================================== |
47 | */ |
48 | //#define DEBUG 1 |
49 | /* Diagnostic functions */ |
50 | #ifdef DEBUG |
51 | # define Assert(cond,msg) { if (!(cond)) bb_error_msg(msg); } |
52 | # define Trace(x) fprintf x |
53 | # define Tracev(x) {if (verbose) fprintf x; } |
54 | # define Tracevv(x) {if (verbose > 1) fprintf x; } |
55 | # define Tracec(c,x) {if (verbose && (c)) fprintf x; } |
56 | # define Tracecv(c,x) {if (verbose > 1 && (c)) fprintf x; } |
57 | #else |
58 | # define Assert(cond,msg) |
59 | # define Trace(x) |
60 | # define Tracev(x) |
61 | # define Tracevv(x) |
62 | # define Tracec(c,x) |
63 | # define Tracecv(c,x) |
64 | #endif |
65 | |
66 | |
67 | /* =========================================================================== |
68 | */ |
69 | #define SMALL_MEM |
70 | |
71 | #ifndef INBUFSIZ |
72 | # ifdef SMALL_MEM |
73 | # define INBUFSIZ 0x2000 /* input buffer size */ |
74 | # else |
75 | # define INBUFSIZ 0x8000 /* input buffer size */ |
76 | # endif |
77 | #endif |
78 | |
79 | #ifndef OUTBUFSIZ |
80 | # ifdef SMALL_MEM |
81 | # define OUTBUFSIZ 8192 /* output buffer size */ |
82 | # else |
83 | # define OUTBUFSIZ 16384 /* output buffer size */ |
84 | # endif |
85 | #endif |
86 | |
87 | #ifndef DIST_BUFSIZE |
88 | # ifdef SMALL_MEM |
89 | # define DIST_BUFSIZE 0x2000 /* buffer for distances, see trees.c */ |
90 | # else |
91 | # define DIST_BUFSIZE 0x8000 /* buffer for distances, see trees.c */ |
92 | # endif |
93 | #endif |
94 | |
95 | /* gzip flag byte */ |
96 | #define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */ |
97 | #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */ |
98 | #define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */ |
99 | #define ORIG_NAME 0x08 /* bit 3 set: original file name present */ |
100 | #define COMMENT 0x10 /* bit 4 set: file comment present */ |
101 | #define RESERVED 0xC0 /* bit 6,7: reserved */ |
102 | |
103 | /* internal file attribute */ |
104 | #define UNKNOWN 0xffff |
105 | #define BINARY 0 |
106 | #define ASCII 1 |
107 | |
108 | #ifndef WSIZE |
109 | # define WSIZE 0x8000 /* window size--must be a power of two, and */ |
110 | #endif /* at least 32K for zip's deflate method */ |
111 | |
112 | #define MIN_MATCH 3 |
113 | #define MAX_MATCH 258 |
114 | /* The minimum and maximum match lengths */ |
115 | |
116 | #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) |
117 | /* Minimum amount of lookahead, except at the end of the input file. |
118 | * See deflate.c for comments about the MIN_MATCH+1. |
119 | */ |
120 | |
121 | #define MAX_DIST (WSIZE-MIN_LOOKAHEAD) |
122 | /* In order to simplify the code, particularly on 16 bit machines, match |
123 | * distances are limited to MAX_DIST instead of WSIZE. |
124 | */ |
125 | |
126 | #ifndef MAX_PATH_LEN |
127 | # define MAX_PATH_LEN 1024 /* max pathname length */ |
128 | #endif |
129 | |
130 | #define seekable() 0 /* force sequential output */ |
131 | #define translate_eol 0 /* no option -a yet */ |
132 | |
133 | #ifndef BITS |
134 | # define BITS 16 |
135 | #endif |
136 | #define INIT_BITS 9 /* Initial number of bits per code */ |
137 | |
138 | #define BIT_MASK 0x1f /* Mask for 'number of compression bits' */ |
139 | /* Mask 0x20 is reserved to mean a fourth header byte, and 0x40 is free. |
140 | * It's a pity that old uncompress does not check bit 0x20. That makes |
141 | * extension of the format actually undesirable because old compress |
142 | * would just crash on the new format instead of giving a meaningful |
143 | * error message. It does check the number of bits, but it's more |
144 | * helpful to say "unsupported format, get a new version" than |
145 | * "can only handle 16 bits". |
146 | */ |
147 | |
148 | #ifdef MAX_EXT_CHARS |
149 | # define MAX_SUFFIX MAX_EXT_CHARS |
150 | #else |
151 | # define MAX_SUFFIX 30 |
152 | #endif |
153 | |
154 | |
155 | /* =========================================================================== |
156 | * Compile with MEDIUM_MEM to reduce the memory requirements or |
157 | * with SMALL_MEM to use as little memory as possible. Use BIG_MEM if the |
158 | * entire input file can be held in memory (not possible on 16 bit systems). |
159 | * Warning: defining these symbols affects HASH_BITS (see below) and thus |
160 | * affects the compression ratio. The compressed output |
161 | * is still correct, and might even be smaller in some cases. |
162 | */ |
163 | |
164 | #ifdef SMALL_MEM |
165 | # define HASH_BITS 13 /* Number of bits used to hash strings */ |
166 | #endif |
167 | #ifdef MEDIUM_MEM |
168 | # define HASH_BITS 14 |
169 | #endif |
170 | #ifndef HASH_BITS |
171 | # define HASH_BITS 15 |
172 | /* For portability to 16 bit machines, do not use values above 15. */ |
173 | #endif |
174 | |
175 | #define HASH_SIZE (unsigned)(1<<HASH_BITS) |
176 | #define HASH_MASK (HASH_SIZE-1) |
177 | #define WMASK (WSIZE-1) |
178 | /* HASH_SIZE and WSIZE must be powers of two */ |
179 | #ifndef TOO_FAR |
180 | # define TOO_FAR 4096 |
181 | #endif |
182 | /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ |
183 | |
184 | |
185 | /* =========================================================================== |
186 | * These types are not really 'char', 'short' and 'long' |
187 | */ |
188 | typedef uint8_t uch; |
189 | typedef uint16_t ush; |
190 | typedef uint32_t ulg; |
191 | typedef int32_t lng; |
192 | |
193 | typedef ush Pos; |
194 | typedef unsigned IPos; |
195 | /* A Pos is an index in the character window. We use short instead of int to |
196 | * save space in the various tables. IPos is used only for parameter passing. |
197 | */ |
198 | |
199 | enum { |
200 | WINDOW_SIZE = 2 * WSIZE, |
201 | /* window size, 2*WSIZE except for MMAP or BIG_MEM, where it is the |
202 | * input file length plus MIN_LOOKAHEAD. |
203 | */ |
204 | |
205 | max_chain_length = 4096, |
206 | /* To speed up deflation, hash chains are never searched beyond this length. |
207 | * A higher limit improves compression ratio but degrades the speed. |
208 | */ |
209 | |
210 | max_lazy_match = 258, |
211 | /* Attempt to find a better match only when the current match is strictly |
212 | * smaller than this value. This mechanism is used only for compression |
213 | * levels >= 4. |
214 | */ |
215 | |
216 | max_insert_length = max_lazy_match, |
217 | /* Insert new strings in the hash table only if the match length |
218 | * is not greater than this length. This saves time but degrades compression. |
219 | * max_insert_length is used only for compression levels <= 3. |
220 | */ |
221 | |
222 | good_match = 32, |
223 | /* Use a faster search when the previous match is longer than this */ |
224 | |
225 | /* Values for max_lazy_match, good_match and max_chain_length, depending on |
226 | * the desired pack level (0..9). The values given below have been tuned to |
227 | * exclude worst case performance for pathological files. Better values may be |
228 | * found for specific files. |
229 | */ |
230 | |
231 | nice_match = 258, /* Stop searching when current match exceeds this */ |
232 | /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 |
233 | * For deflate_fast() (levels <= 3) good is ignored and lazy has a different |
234 | * meaning. |
235 | */ |
236 | }; |
237 | |
238 | |
239 | struct globals { |
240 | |
241 | lng block_start; |
242 | |
243 | /* window position at the beginning of the current output block. Gets |
244 | * negative when the window is moved backwards. |
245 | */ |
246 | unsigned ins_h; /* hash index of string to be inserted */ |
247 | |
248 | #define H_SHIFT ((HASH_BITS+MIN_MATCH-1) / MIN_MATCH) |
249 | /* Number of bits by which ins_h and del_h must be shifted at each |
250 | * input step. It must be such that after MIN_MATCH steps, the oldest |
251 | * byte no longer takes part in the hash key, that is: |
252 | * H_SHIFT * MIN_MATCH >= HASH_BITS |
253 | */ |
254 | |
255 | unsigned prev_length; |
256 | |
257 | /* Length of the best match at previous step. Matches not greater than this |
258 | * are discarded. This is used in the lazy match evaluation. |
259 | */ |
260 | |
261 | unsigned strstart; /* start of string to insert */ |
262 | unsigned match_start; /* start of matching string */ |
263 | unsigned lookahead; /* number of valid bytes ahead in window */ |
264 | |
265 | /* =========================================================================== |
266 | */ |
267 | #define DECLARE(type, array, size) \ |
268 | type * array |
269 | #define ALLOC(type, array, size) \ |
270 | array = xzalloc((size_t)(((size)+1L)/2) * 2*sizeof(type)) |
271 | #define FREE(array) \ |
272 | do { free(array); array = NULL; } while (0) |
273 | |
274 | /* global buffers */ |
275 | |
276 | /* buffer for literals or lengths */ |
277 | /* DECLARE(uch, l_buf, LIT_BUFSIZE); */ |
278 | DECLARE(uch, l_buf, INBUFSIZ); |
279 | |
280 | DECLARE(ush, d_buf, DIST_BUFSIZE); |
281 | DECLARE(uch, outbuf, OUTBUFSIZ); |
282 | |
283 | /* Sliding window. Input bytes are read into the second half of the window, |
284 | * and move to the first half later to keep a dictionary of at least WSIZE |
285 | * bytes. With this organization, matches are limited to a distance of |
286 | * WSIZE-MAX_MATCH bytes, but this ensures that IO is always |
287 | * performed with a length multiple of the block size. Also, it limits |
288 | * the window size to 64K, which is quite useful on MSDOS. |
289 | * To do: limit the window size to WSIZE+BSZ if SMALL_MEM (the code would |
290 | * be less efficient). |
291 | */ |
292 | DECLARE(uch, window, 2L * WSIZE); |
293 | |
294 | /* Link to older string with same hash index. To limit the size of this |
295 | * array to 64K, this link is maintained only for the last 32K strings. |
296 | * An index in this array is thus a window index modulo 32K. |
297 | */ |
298 | /* DECLARE(Pos, prev, WSIZE); */ |
299 | DECLARE(ush, prev, 1L << BITS); |
300 | |
301 | /* Heads of the hash chains or 0. */ |
302 | /* DECLARE(Pos, head, 1<<HASH_BITS); */ |
303 | #define head (G1.prev + WSIZE) /* hash head (see deflate.c) */ |
304 | |
305 | /* number of input bytes */ |
306 | ulg isize; /* only 32 bits stored in .gz file */ |
307 | |
308 | /* bbox always use stdin/stdout */ |
309 | #define ifd STDIN_FILENO /* input file descriptor */ |
310 | #define ofd STDOUT_FILENO /* output file descriptor */ |
311 | |
312 | #ifdef DEBUG |
313 | unsigned insize; /* valid bytes in l_buf */ |
314 | #endif |
315 | unsigned outcnt; /* bytes in output buffer */ |
316 | |
317 | smallint eofile; /* flag set at end of input file */ |
318 | |
319 | /* =========================================================================== |
320 | * Local data used by the "bit string" routines. |
321 | */ |
322 | |
323 | unsigned short bi_buf; |
324 | |
325 | /* Output buffer. bits are inserted starting at the bottom (least significant |
326 | * bits). |
327 | */ |
328 | |
329 | #undef BUF_SIZE |
330 | #define BUF_SIZE (8 * sizeof(G1.bi_buf)) |
331 | /* Number of bits used within bi_buf. (bi_buf might be implemented on |
332 | * more than 16 bits on some systems.) |
333 | */ |
334 | |
335 | int bi_valid; |
336 | |
337 | /* Current input function. Set to mem_read for in-memory compression */ |
338 | |
339 | #ifdef DEBUG |
340 | ulg bits_sent; /* bit length of the compressed data */ |
341 | #endif |
342 | |
343 | uint32_t *crc_32_tab; |
344 | uint32_t crc; /* shift register contents */ |
345 | }; |
346 | |
347 | #define G1 (*(ptr_to_globals - 1)) |
348 | |
349 | |
350 | /* =========================================================================== |
351 | * Write the output buffer outbuf[0..outcnt-1] and update bytes_out. |
352 | * (used for the compressed data only) |
353 | */ |
354 | static void flush_outbuf(void) |
355 | { |
356 | if (G1.outcnt == 0) |
357 | return; |
358 | |
359 | xwrite(ofd, (char *) G1.outbuf, G1.outcnt); |
360 | G1.outcnt = 0; |
361 | } |
362 | |
363 | |
364 | /* =========================================================================== |
365 | */ |
366 | /* put_8bit is used for the compressed output */ |
367 | #define put_8bit(c) \ |
368 | do { \ |
369 | G1.outbuf[G1.outcnt++] = (c); \ |
370 | if (G1.outcnt == OUTBUFSIZ) flush_outbuf(); \ |
371 | } while (0) |
372 | |
373 | /* Output a 16 bit value, lsb first */ |
374 | static void put_16bit(ush w) |
375 | { |
376 | if (G1.outcnt < OUTBUFSIZ - 2) { |
377 | G1.outbuf[G1.outcnt++] = w; |
378 | G1.outbuf[G1.outcnt++] = w >> 8; |
379 | } else { |
380 | put_8bit(w); |
381 | put_8bit(w >> 8); |
382 | } |
383 | } |
384 | |
385 | static void put_32bit(ulg n) |
386 | { |
387 | put_16bit(n); |
388 | put_16bit(n >> 16); |
389 | } |
390 | |
391 | /* =========================================================================== |
392 | * Run a set of bytes through the crc shift register. If s is a NULL |
393 | * pointer, then initialize the crc shift register contents instead. |
394 | * Return the current crc in either case. |
395 | */ |
396 | static uint32_t updcrc(uch * s, unsigned n) |
397 | { |
398 | uint32_t c = G1.crc; |
399 | while (n) { |
400 | c = G1.crc_32_tab[(uch)(c ^ *s++)] ^ (c >> 8); |
401 | n--; |
402 | } |
403 | G1.crc = c; |
404 | return c; |
405 | } |
406 | |
407 | |
408 | /* =========================================================================== |
409 | * Read a new buffer from the current input file, perform end-of-line |
410 | * translation, and update the crc and input file size. |
411 | * IN assertion: size >= 2 (for end-of-line translation) |
412 | */ |
413 | static unsigned file_read(void *buf, unsigned size) |
414 | { |
415 | unsigned len; |
416 | |
417 | Assert(G1.insize == 0, "l_buf not empty"); |
418 | |
419 | len = safe_read(ifd, buf, size); |
420 | if (len == (unsigned)(-1) || len == 0) |
421 | return len; |
422 | |
423 | updcrc(buf, len); |
424 | G1.isize += len; |
425 | return len; |
426 | } |
427 | |
428 | |
429 | /* =========================================================================== |
430 | * Send a value on a given number of bits. |
431 | * IN assertion: length <= 16 and value fits in length bits. |
432 | */ |
433 | static void send_bits(int value, int length) |
434 | { |
435 | #ifdef DEBUG |
436 | Tracev((stderr, " l %2d v %4x ", length, value)); |
437 | Assert(length > 0 && length <= 15, "invalid length"); |
438 | G1.bits_sent += length; |
439 | #endif |
440 | /* If not enough room in bi_buf, use (valid) bits from bi_buf and |
441 | * (16 - bi_valid) bits from value, leaving (width - (16-bi_valid)) |
442 | * unused bits in value. |
443 | */ |
444 | if (G1.bi_valid > (int) BUF_SIZE - length) { |
445 | G1.bi_buf |= (value << G1.bi_valid); |
446 | put_16bit(G1.bi_buf); |
447 | G1.bi_buf = (ush) value >> (BUF_SIZE - G1.bi_valid); |
448 | G1.bi_valid += length - BUF_SIZE; |
449 | } else { |
450 | G1.bi_buf |= value << G1.bi_valid; |
451 | G1.bi_valid += length; |
452 | } |
453 | } |
454 | |
455 | |
456 | /* =========================================================================== |
457 | * Reverse the first len bits of a code, using straightforward code (a faster |
458 | * method would use a table) |
459 | * IN assertion: 1 <= len <= 15 |
460 | */ |
461 | static unsigned bi_reverse(unsigned code, int len) |
462 | { |
463 | unsigned res = 0; |
464 | |
465 | while (1) { |
466 | res |= code & 1; |
467 | if (--len <= 0) return res; |
468 | code >>= 1; |
469 | res <<= 1; |
470 | } |
471 | } |
472 | |
473 | |
474 | /* =========================================================================== |
475 | * Write out any remaining bits in an incomplete byte. |
476 | */ |
477 | static void bi_windup(void) |
478 | { |
479 | if (G1.bi_valid > 8) { |
480 | put_16bit(G1.bi_buf); |
481 | } else if (G1.bi_valid > 0) { |
482 | put_8bit(G1.bi_buf); |
483 | } |
484 | G1.bi_buf = 0; |
485 | G1.bi_valid = 0; |
486 | #ifdef DEBUG |
487 | G1.bits_sent = (G1.bits_sent + 7) & ~7; |
488 | #endif |
489 | } |
490 | |
491 | |
492 | /* =========================================================================== |
493 | * Copy a stored block to the zip file, storing first the length and its |
494 | * one's complement if requested. |
495 | */ |
496 | static void copy_block(char *buf, unsigned len, int header) |
497 | { |
498 | bi_windup(); /* align on byte boundary */ |
499 | |
500 | if (header) { |
501 | put_16bit(len); |
502 | put_16bit(~len); |
503 | #ifdef DEBUG |
504 | G1.bits_sent += 2 * 16; |
505 | #endif |
506 | } |
507 | #ifdef DEBUG |
508 | G1.bits_sent += (ulg) len << 3; |
509 | #endif |
510 | while (len--) { |
511 | put_8bit(*buf++); |
512 | } |
513 | } |
514 | |
515 | |
516 | /* =========================================================================== |
517 | * Fill the window when the lookahead becomes insufficient. |
518 | * Updates strstart and lookahead, and sets eofile if end of input file. |
519 | * IN assertion: lookahead < MIN_LOOKAHEAD && strstart + lookahead > 0 |
520 | * OUT assertions: at least one byte has been read, or eofile is set; |
521 | * file reads are performed for at least two bytes (required for the |
522 | * translate_eol option). |
523 | */ |
524 | static void fill_window(void) |
525 | { |
526 | unsigned n, m; |
527 | unsigned more = WINDOW_SIZE - G1.lookahead - G1.strstart; |
528 | /* Amount of free space at the end of the window. */ |
529 | |
530 | /* If the window is almost full and there is insufficient lookahead, |
531 | * move the upper half to the lower one to make room in the upper half. |
532 | */ |
533 | if (more == (unsigned) -1) { |
534 | /* Very unlikely, but possible on 16 bit machine if strstart == 0 |
535 | * and lookahead == 1 (input done one byte at time) |
536 | */ |
537 | more--; |
538 | } else if (G1.strstart >= WSIZE + MAX_DIST) { |
539 | /* By the IN assertion, the window is not empty so we can't confuse |
540 | * more == 0 with more == 64K on a 16 bit machine. |
541 | */ |
542 | Assert(WINDOW_SIZE == 2 * WSIZE, "no sliding with BIG_MEM"); |
543 | |
544 | memcpy(G1.window, G1.window + WSIZE, WSIZE); |
545 | G1.match_start -= WSIZE; |
546 | G1.strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */ |
547 | |
548 | G1.block_start -= WSIZE; |
549 | |
550 | for (n = 0; n < HASH_SIZE; n++) { |
551 | m = head[n]; |
552 | head[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0); |
553 | } |
554 | for (n = 0; n < WSIZE; n++) { |
555 | m = G1.prev[n]; |
556 | G1.prev[n] = (Pos) (m >= WSIZE ? m - WSIZE : 0); |
557 | /* If n is not on any hash chain, prev[n] is garbage but |
558 | * its value will never be used. |
559 | */ |
560 | } |
561 | more += WSIZE; |
562 | } |
563 | /* At this point, more >= 2 */ |
564 | if (!G1.eofile) { |
565 | n = file_read(G1.window + G1.strstart + G1.lookahead, more); |
566 | if (n == 0 || n == (unsigned) -1) { |
567 | G1.eofile = 1; |
568 | } else { |
569 | G1.lookahead += n; |
570 | } |
571 | } |
572 | } |
573 | |
574 | |
575 | /* =========================================================================== |
576 | * Set match_start to the longest match starting at the given string and |
577 | * return its length. Matches shorter or equal to prev_length are discarded, |
578 | * in which case the result is equal to prev_length and match_start is |
579 | * garbage. |
580 | * IN assertions: cur_match is the head of the hash chain for the current |
581 | * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 |
582 | */ |
583 | |
584 | /* For MSDOS, OS/2 and 386 Unix, an optimized version is in match.asm or |
585 | * match.s. The code is functionally equivalent, so you can use the C version |
586 | * if desired. |
587 | */ |
588 | static int longest_match(IPos cur_match) |
589 | { |
590 | unsigned chain_length = max_chain_length; /* max hash chain length */ |
591 | uch *scan = G1.window + G1.strstart; /* current string */ |
592 | uch *match; /* matched string */ |
593 | int len; /* length of current match */ |
594 | int best_len = G1.prev_length; /* best match length so far */ |
595 | IPos limit = G1.strstart > (IPos) MAX_DIST ? G1.strstart - (IPos) MAX_DIST : 0; |
596 | /* Stop when cur_match becomes <= limit. To simplify the code, |
597 | * we prevent matches with the string of window index 0. |
598 | */ |
599 | |
600 | /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. |
601 | * It is easy to get rid of this optimization if necessary. |
602 | */ |
603 | #if HASH_BITS < 8 || MAX_MATCH != 258 |
604 | # error Code too clever |
605 | #endif |
606 | uch *strend = G1.window + G1.strstart + MAX_MATCH; |
607 | uch scan_end1 = scan[best_len - 1]; |
608 | uch scan_end = scan[best_len]; |
609 | |
610 | /* Do not waste too much time if we already have a good match: */ |
611 | if (G1.prev_length >= good_match) { |
612 | chain_length >>= 2; |
613 | } |
614 | Assert(G1.strstart <= WINDOW_SIZE - MIN_LOOKAHEAD, "insufficient lookahead"); |
615 | |
616 | do { |
617 | Assert(cur_match < G1.strstart, "no future"); |
618 | match = G1.window + cur_match; |
619 | |
620 | /* Skip to next match if the match length cannot increase |
621 | * or if the match length is less than 2: |
622 | */ |
623 | if (match[best_len] != scan_end || |
624 | match[best_len - 1] != scan_end1 || |
625 | *match != *scan || *++match != scan[1]) |
626 | continue; |
627 | |
628 | /* The check at best_len-1 can be removed because it will be made |
629 | * again later. (This heuristic is not always a win.) |
630 | * It is not necessary to compare scan[2] and match[2] since they |
631 | * are always equal when the other bytes match, given that |
632 | * the hash keys are equal and that HASH_BITS >= 8. |
633 | */ |
634 | scan += 2, match++; |
635 | |
636 | /* We check for insufficient lookahead only every 8th comparison; |
637 | * the 256th check will be made at strstart+258. |
638 | */ |
639 | do { |
640 | } while (*++scan == *++match && *++scan == *++match && |
641 | *++scan == *++match && *++scan == *++match && |
642 | *++scan == *++match && *++scan == *++match && |
643 | *++scan == *++match && *++scan == *++match && scan < strend); |
644 | |
645 | len = MAX_MATCH - (int) (strend - scan); |
646 | scan = strend - MAX_MATCH; |
647 | |
648 | if (len > best_len) { |
649 | G1.match_start = cur_match; |
650 | best_len = len; |
651 | if (len >= nice_match) |
652 | break; |
653 | scan_end1 = scan[best_len - 1]; |
654 | scan_end = scan[best_len]; |
655 | } |
656 | } while ((cur_match = G1.prev[cur_match & WMASK]) > limit |
657 | && --chain_length != 0); |
658 | |
659 | return best_len; |
660 | } |
661 | |
662 | |
663 | #ifdef DEBUG |
664 | /* =========================================================================== |
665 | * Check that the match at match_start is indeed a match. |
666 | */ |
667 | static void check_match(IPos start, IPos match, int length) |
668 | { |
669 | /* check that the match is indeed a match */ |
670 | if (memcmp(G1.window + match, G1.window + start, length) != 0) { |
671 | bb_error_msg(" start %d, match %d, length %d", start, match, length); |
672 | bb_error_msg("invalid match"); |
673 | } |
674 | if (verbose > 1) { |
675 | bb_error_msg("\\[%d,%d]", start - match, length); |
676 | do { |
677 | fputc(G1.window[start++], stderr); |
678 | } while (--length != 0); |
679 | } |
680 | } |
681 | #else |
682 | # define check_match(start, match, length) ((void)0) |
683 | #endif |
684 | |
685 | |
686 | /* trees.c -- output deflated data using Huffman coding |
687 | * Copyright (C) 1992-1993 Jean-loup Gailly |
688 | * This is free software; you can redistribute it and/or modify it under the |
689 | * terms of the GNU General Public License, see the file COPYING. |
690 | */ |
691 | |
692 | /* PURPOSE |
693 | * Encode various sets of source values using variable-length |
694 | * binary code trees. |
695 | * |
696 | * DISCUSSION |
697 | * The PKZIP "deflation" process uses several Huffman trees. The more |
698 | * common source values are represented by shorter bit sequences. |
699 | * |
700 | * Each code tree is stored in the ZIP file in a compressed form |
701 | * which is itself a Huffman encoding of the lengths of |
702 | * all the code strings (in ascending order by source values). |
703 | * The actual code strings are reconstructed from the lengths in |
704 | * the UNZIP process, as described in the "application note" |
705 | * (APPNOTE.TXT) distributed as part of PKWARE's PKZIP program. |
706 | * |
707 | * REFERENCES |
708 | * Lynch, Thomas J. |
709 | * Data Compression: Techniques and Applications, pp. 53-55. |
710 | * Lifetime Learning Publications, 1985. ISBN 0-534-03418-7. |
711 | * |
712 | * Storer, James A. |
713 | * Data Compression: Methods and Theory, pp. 49-50. |
714 | * Computer Science Press, 1988. ISBN 0-7167-8156-5. |
715 | * |
716 | * Sedgewick, R. |
717 | * Algorithms, p290. |
718 | * Addison-Wesley, 1983. ISBN 0-201-06672-6. |
719 | * |
720 | * INTERFACE |
721 | * void ct_init() |
722 | * Allocate the match buffer, initialize the various tables [and save |
723 | * the location of the internal file attribute (ascii/binary) and |
724 | * method (DEFLATE/STORE) -- deleted in bbox] |
725 | * |
726 | * void ct_tally(int dist, int lc); |
727 | * Save the match info and tally the frequency counts. |
728 | * |
729 | * ulg flush_block(char *buf, ulg stored_len, int eof) |
730 | * Determine the best encoding for the current block: dynamic trees, |
731 | * static trees or store, and output the encoded block to the zip |
732 | * file. Returns the total compressed length for the file so far. |
733 | */ |
734 | |
735 | #define MAX_BITS 15 |
736 | /* All codes must not exceed MAX_BITS bits */ |
737 | |
738 | #define MAX_BL_BITS 7 |
739 | /* Bit length codes must not exceed MAX_BL_BITS bits */ |
740 | |
741 | #define LENGTH_CODES 29 |
742 | /* number of length codes, not counting the special END_BLOCK code */ |
743 | |
744 | #define LITERALS 256 |
745 | /* number of literal bytes 0..255 */ |
746 | |
747 | #define END_BLOCK 256 |
748 | /* end of block literal code */ |
749 | |
750 | #define L_CODES (LITERALS+1+LENGTH_CODES) |
751 | /* number of Literal or Length codes, including the END_BLOCK code */ |
752 | |
753 | #define D_CODES 30 |
754 | /* number of distance codes */ |
755 | |
756 | #define BL_CODES 19 |
757 | /* number of codes used to transfer the bit lengths */ |
758 | |
759 | /* extra bits for each length code */ |
760 | static const uint8_t extra_lbits[LENGTH_CODES] ALIGN1 = { |
761 | 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, |
762 | 4, 4, 5, 5, 5, 5, 0 |
763 | }; |
764 | |
765 | /* extra bits for each distance code */ |
766 | static const uint8_t extra_dbits[D_CODES] ALIGN1 = { |
767 | 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, |
768 | 10, 10, 11, 11, 12, 12, 13, 13 |
769 | }; |
770 | |
771 | /* extra bits for each bit length code */ |
772 | static const uint8_t extra_blbits[BL_CODES] ALIGN1 = { |
773 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 7 }; |
774 | |
775 | /* number of codes at each bit length for an optimal tree */ |
776 | static const uint8_t bl_order[BL_CODES] ALIGN1 = { |
777 | 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 }; |
778 | |
779 | #define STORED_BLOCK 0 |
780 | #define STATIC_TREES 1 |
781 | #define DYN_TREES 2 |
782 | /* The three kinds of block type */ |
783 | |
784 | #ifndef LIT_BUFSIZE |
785 | # ifdef SMALL_MEM |
786 | # define LIT_BUFSIZE 0x2000 |
787 | # else |
788 | # ifdef MEDIUM_MEM |
789 | # define LIT_BUFSIZE 0x4000 |
790 | # else |
791 | # define LIT_BUFSIZE 0x8000 |
792 | # endif |
793 | # endif |
794 | #endif |
795 | #ifndef DIST_BUFSIZE |
796 | # define DIST_BUFSIZE LIT_BUFSIZE |
797 | #endif |
798 | /* Sizes of match buffers for literals/lengths and distances. There are |
799 | * 4 reasons for limiting LIT_BUFSIZE to 64K: |
800 | * - frequencies can be kept in 16 bit counters |
801 | * - if compression is not successful for the first block, all input data is |
802 | * still in the window so we can still emit a stored block even when input |
803 | * comes from standard input. (This can also be done for all blocks if |
804 | * LIT_BUFSIZE is not greater than 32K.) |
805 | * - if compression is not successful for a file smaller than 64K, we can |
806 | * even emit a stored file instead of a stored block (saving 5 bytes). |
807 | * - creating new Huffman trees less frequently may not provide fast |
808 | * adaptation to changes in the input data statistics. (Take for |
809 | * example a binary file with poorly compressible code followed by |
810 | * a highly compressible string table.) Smaller buffer sizes give |
811 | * fast adaptation but have of course the overhead of transmitting trees |
812 | * more frequently. |
813 | * - I can't count above 4 |
814 | * The current code is general and allows DIST_BUFSIZE < LIT_BUFSIZE (to save |
815 | * memory at the expense of compression). Some optimizations would be possible |
816 | * if we rely on DIST_BUFSIZE == LIT_BUFSIZE. |
817 | */ |
818 | #define REP_3_6 16 |
819 | /* repeat previous bit length 3-6 times (2 bits of repeat count) */ |
820 | #define REPZ_3_10 17 |
821 | /* repeat a zero length 3-10 times (3 bits of repeat count) */ |
822 | #define REPZ_11_138 18 |
823 | /* repeat a zero length 11-138 times (7 bits of repeat count) */ |
824 | |
825 | /* =========================================================================== |
826 | */ |
827 | /* Data structure describing a single value and its code string. */ |
828 | typedef struct ct_data { |
829 | union { |
830 | ush freq; /* frequency count */ |
831 | ush code; /* bit string */ |
832 | } fc; |
833 | union { |
834 | ush dad; /* father node in Huffman tree */ |
835 | ush len; /* length of bit string */ |
836 | } dl; |
837 | } ct_data; |
838 | |
839 | #define Freq fc.freq |
840 | #define Code fc.code |
841 | #define Dad dl.dad |
842 | #define Len dl.len |
843 | |
844 | #define HEAP_SIZE (2*L_CODES + 1) |
845 | /* maximum heap size */ |
846 | |
847 | typedef struct tree_desc { |
848 | ct_data *dyn_tree; /* the dynamic tree */ |
849 | ct_data *static_tree; /* corresponding static tree or NULL */ |
850 | const uint8_t *extra_bits; /* extra bits for each code or NULL */ |
851 | int extra_base; /* base index for extra_bits */ |
852 | int elems; /* max number of elements in the tree */ |
853 | int max_length; /* max bit length for the codes */ |
854 | int max_code; /* largest code with non zero frequency */ |
855 | } tree_desc; |
856 | |
857 | struct globals2 { |
858 | |
859 | ush heap[HEAP_SIZE]; /* heap used to build the Huffman trees */ |
860 | int heap_len; /* number of elements in the heap */ |
861 | int heap_max; /* element of largest frequency */ |
862 | |
863 | /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. |
864 | * The same heap array is used to build all trees. |
865 | */ |
866 | |
867 | ct_data dyn_ltree[HEAP_SIZE]; /* literal and length tree */ |
868 | ct_data dyn_dtree[2 * D_CODES + 1]; /* distance tree */ |
869 | |
870 | ct_data static_ltree[L_CODES + 2]; |
871 | |
872 | /* The static literal tree. Since the bit lengths are imposed, there is no |
873 | * need for the L_CODES extra codes used during heap construction. However |
874 | * The codes 286 and 287 are needed to build a canonical tree (see ct_init |
875 | * below). |
876 | */ |
877 | |
878 | ct_data static_dtree[D_CODES]; |
879 | |
880 | /* The static distance tree. (Actually a trivial tree since all codes use |
881 | * 5 bits.) |
882 | */ |
883 | |
884 | ct_data bl_tree[2 * BL_CODES + 1]; |
885 | |
886 | /* Huffman tree for the bit lengths */ |
887 | |
888 | tree_desc l_desc; |
889 | tree_desc d_desc; |
890 | tree_desc bl_desc; |
891 | |
892 | ush bl_count[MAX_BITS + 1]; |
893 | |
894 | /* The lengths of the bit length codes are sent in order of decreasing |
895 | * probability, to avoid transmitting the lengths for unused bit length codes. |
896 | */ |
897 | |
898 | uch depth[2 * L_CODES + 1]; |
899 | |
900 | /* Depth of each subtree used as tie breaker for trees of equal frequency */ |
901 | |
902 | uch length_code[MAX_MATCH - MIN_MATCH + 1]; |
903 | |
904 | /* length code for each normalized match length (0 == MIN_MATCH) */ |
905 | |
906 | uch dist_code[512]; |
907 | |
908 | /* distance codes. The first 256 values correspond to the distances |
909 | * 3 .. 258, the last 256 values correspond to the top 8 bits of |
910 | * the 15 bit distances. |
911 | */ |
912 | |
913 | int base_length[LENGTH_CODES]; |
914 | |
915 | /* First normalized length for each code (0 = MIN_MATCH) */ |
916 | |
917 | int base_dist[D_CODES]; |
918 | |
919 | /* First normalized distance for each code (0 = distance of 1) */ |
920 | |
921 | uch flag_buf[LIT_BUFSIZE / 8]; |
922 | |
923 | /* flag_buf is a bit array distinguishing literals from lengths in |
924 | * l_buf, thus indicating the presence or absence of a distance. |
925 | */ |
926 | |
927 | unsigned last_lit; /* running index in l_buf */ |
928 | unsigned last_dist; /* running index in d_buf */ |
929 | unsigned last_flags; /* running index in flag_buf */ |
930 | uch flags; /* current flags not yet saved in flag_buf */ |
931 | uch flag_bit; /* current bit used in flags */ |
932 | |
933 | /* bits are filled in flags starting at bit 0 (least significant). |
934 | * Note: these flags are overkill in the current code since we don't |
935 | * take advantage of DIST_BUFSIZE == LIT_BUFSIZE. |
936 | */ |
937 | |
938 | ulg opt_len; /* bit length of current block with optimal trees */ |
939 | ulg static_len; /* bit length of current block with static trees */ |
940 | |
941 | ulg compressed_len; /* total bit length of compressed file */ |
942 | }; |
943 | |
944 | #define G2ptr ((struct globals2*)(ptr_to_globals)) |
945 | #define G2 (*G2ptr) |
946 | |
947 | |
948 | /* =========================================================================== |
949 | */ |
950 | static void gen_codes(ct_data * tree, int max_code); |
951 | static void build_tree(tree_desc * desc); |
952 | static void scan_tree(ct_data * tree, int max_code); |
953 | static void send_tree(ct_data * tree, int max_code); |
954 | static int build_bl_tree(void); |
955 | static void send_all_trees(int lcodes, int dcodes, int blcodes); |
956 | static void compress_block(ct_data * ltree, ct_data * dtree); |
957 | |
958 | |
959 | #ifndef DEBUG |
960 | /* Send a code of the given tree. c and tree must not have side effects */ |
961 | # define SEND_CODE(c, tree) send_bits(tree[c].Code, tree[c].Len) |
962 | #else |
963 | # define SEND_CODE(c, tree) \ |
964 | { \ |
965 | if (verbose > 1) bb_error_msg("\ncd %3d ",(c)); \ |
966 | send_bits(tree[c].Code, tree[c].Len); \ |
967 | } |
968 | #endif |
969 | |
970 | #define D_CODE(dist) \ |
971 | ((dist) < 256 ? G2.dist_code[dist] : G2.dist_code[256 + ((dist)>>7)]) |
972 | /* Mapping from a distance to a distance code. dist is the distance - 1 and |
973 | * must not have side effects. dist_code[256] and dist_code[257] are never |
974 | * used. |
975 | * The arguments must not have side effects. |
976 | */ |
977 | |
978 | |
979 | /* =========================================================================== |
980 | * Initialize a new block. |
981 | */ |
982 | static void init_block(void) |
983 | { |
984 | int n; /* iterates over tree elements */ |
985 | |
986 | /* Initialize the trees. */ |
987 | for (n = 0; n < L_CODES; n++) |
988 | G2.dyn_ltree[n].Freq = 0; |
989 | for (n = 0; n < D_CODES; n++) |
990 | G2.dyn_dtree[n].Freq = 0; |
991 | for (n = 0; n < BL_CODES; n++) |
992 | G2.bl_tree[n].Freq = 0; |
993 | |
994 | G2.dyn_ltree[END_BLOCK].Freq = 1; |
995 | G2.opt_len = G2.static_len = 0; |
996 | G2.last_lit = G2.last_dist = G2.last_flags = 0; |
997 | G2.flags = 0; |
998 | G2.flag_bit = 1; |
999 | } |
1000 | |
1001 | |
1002 | /* =========================================================================== |
1003 | * Restore the heap property by moving down the tree starting at node k, |
1004 | * exchanging a node with the smallest of its two sons if necessary, stopping |
1005 | * when the heap property is re-established (each father smaller than its |
1006 | * two sons). |
1007 | */ |
1008 | |
1009 | /* Compares to subtrees, using the tree depth as tie breaker when |
1010 | * the subtrees have equal frequency. This minimizes the worst case length. */ |
1011 | #define SMALLER(tree, n, m) \ |
1012 | (tree[n].Freq < tree[m].Freq \ |
1013 | || (tree[n].Freq == tree[m].Freq && G2.depth[n] <= G2.depth[m])) |
1014 | |
1015 | static void pqdownheap(ct_data * tree, int k) |
1016 | { |
1017 | int v = G2.heap[k]; |
1018 | int j = k << 1; /* left son of k */ |
1019 | |
1020 | while (j <= G2.heap_len) { |
1021 | /* Set j to the smallest of the two sons: */ |
1022 | if (j < G2.heap_len && SMALLER(tree, G2.heap[j + 1], G2.heap[j])) |
1023 | j++; |
1024 | |
1025 | /* Exit if v is smaller than both sons */ |
1026 | if (SMALLER(tree, v, G2.heap[j])) |
1027 | break; |
1028 | |
1029 | /* Exchange v with the smallest son */ |
1030 | G2.heap[k] = G2.heap[j]; |
1031 | k = j; |
1032 | |
1033 | /* And continue down the tree, setting j to the left son of k */ |
1034 | j <<= 1; |
1035 | } |
1036 | G2.heap[k] = v; |
1037 | } |
1038 | |
1039 | |
1040 | /* =========================================================================== |
1041 | * Compute the optimal bit lengths for a tree and update the total bit length |
1042 | * for the current block. |
1043 | * IN assertion: the fields freq and dad are set, heap[heap_max] and |
1044 | * above are the tree nodes sorted by increasing frequency. |
1045 | * OUT assertions: the field len is set to the optimal bit length, the |
1046 | * array bl_count contains the frequencies for each bit length. |
1047 | * The length opt_len is updated; static_len is also updated if stree is |
1048 | * not null. |
1049 | */ |
1050 | static void gen_bitlen(tree_desc * desc) |
1051 | { |
1052 | ct_data *tree = desc->dyn_tree; |
1053 | const uint8_t *extra = desc->extra_bits; |
1054 | int base = desc->extra_base; |
1055 | int max_code = desc->max_code; |
1056 | int max_length = desc->max_length; |
1057 | ct_data *stree = desc->static_tree; |
1058 | int h; /* heap index */ |
1059 | int n, m; /* iterate over the tree elements */ |
1060 | int bits; /* bit length */ |
1061 | int xbits; /* extra bits */ |
1062 | ush f; /* frequency */ |
1063 | int overflow = 0; /* number of elements with bit length too large */ |
1064 | |
1065 | for (bits = 0; bits <= MAX_BITS; bits++) |
1066 | G2.bl_count[bits] = 0; |
1067 | |
1068 | /* In a first pass, compute the optimal bit lengths (which may |
1069 | * overflow in the case of the bit length tree). |
1070 | */ |
1071 | tree[G2.heap[G2.heap_max]].Len = 0; /* root of the heap */ |
1072 | |
1073 | for (h = G2.heap_max + 1; h < HEAP_SIZE; h++) { |
1074 | n = G2.heap[h]; |
1075 | bits = tree[tree[n].Dad].Len + 1; |
1076 | if (bits > max_length) { |
1077 | bits = max_length; |
1078 | overflow++; |
1079 | } |
1080 | tree[n].Len = (ush) bits; |
1081 | /* We overwrite tree[n].Dad which is no longer needed */ |
1082 | |
1083 | if (n > max_code) |
1084 | continue; /* not a leaf node */ |
1085 | |
1086 | G2.bl_count[bits]++; |
1087 | xbits = 0; |
1088 | if (n >= base) |
1089 | xbits = extra[n - base]; |
1090 | f = tree[n].Freq; |
1091 | G2.opt_len += (ulg) f *(bits + xbits); |
1092 | |
1093 | if (stree) |
1094 | G2.static_len += (ulg) f * (stree[n].Len + xbits); |
1095 | } |
1096 | if (overflow == 0) |
1097 | return; |
1098 | |
1099 | Trace((stderr, "\nbit length overflow\n")); |
1100 | /* This happens for example on obj2 and pic of the Calgary corpus */ |
1101 | |
1102 | /* Find the first bit length which could increase: */ |
1103 | do { |
1104 | bits = max_length - 1; |
1105 | while (G2.bl_count[bits] == 0) |
1106 | bits--; |
1107 | G2.bl_count[bits]--; /* move one leaf down the tree */ |
1108 | G2.bl_count[bits + 1] += 2; /* move one overflow item as its brother */ |
1109 | G2.bl_count[max_length]--; |
1110 | /* The brother of the overflow item also moves one step up, |
1111 | * but this does not affect bl_count[max_length] |
1112 | */ |
1113 | overflow -= 2; |
1114 | } while (overflow > 0); |
1115 | |
1116 | /* Now recompute all bit lengths, scanning in increasing frequency. |
1117 | * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all |
1118 | * lengths instead of fixing only the wrong ones. This idea is taken |
1119 | * from 'ar' written by Haruhiko Okumura.) |
1120 | */ |
1121 | for (bits = max_length; bits != 0; bits--) { |
1122 | n = G2.bl_count[bits]; |
1123 | while (n != 0) { |
1124 | m = G2.heap[--h]; |
1125 | if (m > max_code) |
1126 | continue; |
1127 | if (tree[m].Len != (unsigned) bits) { |
1128 | Trace((stderr, "code %d bits %d->%d\n", m, tree[m].Len, bits)); |
1129 | G2.opt_len += ((int32_t) bits - tree[m].Len) * tree[m].Freq; |
1130 | tree[m].Len = bits; |
1131 | } |
1132 | n--; |
1133 | } |
1134 | } |
1135 | } |
1136 | |
1137 | |
1138 | /* =========================================================================== |
1139 | * Generate the codes for a given tree and bit counts (which need not be |
1140 | * optimal). |
1141 | * IN assertion: the array bl_count contains the bit length statistics for |
1142 | * the given tree and the field len is set for all tree elements. |
1143 | * OUT assertion: the field code is set for all tree elements of non |
1144 | * zero code length. |
1145 | */ |
1146 | static void gen_codes(ct_data * tree, int max_code) |
1147 | { |
1148 | ush next_code[MAX_BITS + 1]; /* next code value for each bit length */ |
1149 | ush code = 0; /* running code value */ |
1150 | int bits; /* bit index */ |
1151 | int n; /* code index */ |
1152 | |
1153 | /* The distribution counts are first used to generate the code values |
1154 | * without bit reversal. |
1155 | */ |
1156 | for (bits = 1; bits <= MAX_BITS; bits++) { |
1157 | next_code[bits] = code = (code + G2.bl_count[bits - 1]) << 1; |
1158 | } |
1159 | /* Check that the bit counts in bl_count are consistent. The last code |
1160 | * must be all ones. |
1161 | */ |
1162 | Assert(code + G2.bl_count[MAX_BITS] - 1 == (1 << MAX_BITS) - 1, |
1163 | "inconsistent bit counts"); |
1164 | Tracev((stderr, "\ngen_codes: max_code %d ", max_code)); |
1165 | |
1166 | for (n = 0; n <= max_code; n++) { |
1167 | int len = tree[n].Len; |
1168 | |
1169 | if (len == 0) |
1170 | continue; |
1171 | /* Now reverse the bits */ |
1172 | tree[n].Code = bi_reverse(next_code[len]++, len); |
1173 | |
1174 | Tracec(tree != G2.static_ltree, |
1175 | (stderr, "\nn %3d %c l %2d c %4x (%x) ", n, |
1176 | (n > ' ' ? n : ' '), len, tree[n].Code, |
1177 | next_code[len] - 1)); |
1178 | } |
1179 | } |
1180 | |
1181 | |
1182 | /* =========================================================================== |
1183 | * Construct one Huffman tree and assigns the code bit strings and lengths. |
1184 | * Update the total bit length for the current block. |
1185 | * IN assertion: the field freq is set for all tree elements. |
1186 | * OUT assertions: the fields len and code are set to the optimal bit length |
1187 | * and corresponding code. The length opt_len is updated; static_len is |
1188 | * also updated if stree is not null. The field max_code is set. |
1189 | */ |
1190 | |
1191 | /* Remove the smallest element from the heap and recreate the heap with |
1192 | * one less element. Updates heap and heap_len. */ |
1193 | |
1194 | #define SMALLEST 1 |
1195 | /* Index within the heap array of least frequent node in the Huffman tree */ |
1196 | |
1197 | #define PQREMOVE(tree, top) \ |
1198 | do { \ |
1199 | top = G2.heap[SMALLEST]; \ |
1200 | G2.heap[SMALLEST] = G2.heap[G2.heap_len--]; \ |
1201 | pqdownheap(tree, SMALLEST); \ |
1202 | } while (0) |
1203 | |
1204 | static void build_tree(tree_desc * desc) |
1205 | { |
1206 | ct_data *tree = desc->dyn_tree; |
1207 | ct_data *stree = desc->static_tree; |
1208 | int elems = desc->elems; |
1209 | int n, m; /* iterate over heap elements */ |
1210 | int max_code = -1; /* largest code with non zero frequency */ |
1211 | int node = elems; /* next internal node of the tree */ |
1212 | |
1213 | /* Construct the initial heap, with least frequent element in |
1214 | * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1]. |
1215 | * heap[0] is not used. |
1216 | */ |
1217 | G2.heap_len = 0; |
1218 | G2.heap_max = HEAP_SIZE; |
1219 | |
1220 | for (n = 0; n < elems; n++) { |
1221 | if (tree[n].Freq != 0) { |
1222 | G2.heap[++G2.heap_len] = max_code = n; |
1223 | G2.depth[n] = 0; |
1224 | } else { |
1225 | tree[n].Len = 0; |
1226 | } |
1227 | } |
1228 | |
1229 | /* The pkzip format requires that at least one distance code exists, |
1230 | * and that at least one bit should be sent even if there is only one |
1231 | * possible code. So to avoid special checks later on we force at least |
1232 | * two codes of non zero frequency. |
1233 | */ |
1234 | while (G2.heap_len < 2) { |
1235 | int new = G2.heap[++G2.heap_len] = (max_code < 2 ? ++max_code : 0); |
1236 | |
1237 | tree[new].Freq = 1; |
1238 | G2.depth[new] = 0; |
1239 | G2.opt_len--; |
1240 | if (stree) |
1241 | G2.static_len -= stree[new].Len; |
1242 | /* new is 0 or 1 so it does not have extra bits */ |
1243 | } |
1244 | desc->max_code = max_code; |
1245 | |
1246 | /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree, |
1247 | * establish sub-heaps of increasing lengths: |
1248 | */ |
1249 | for (n = G2.heap_len / 2; n >= 1; n--) |
1250 | pqdownheap(tree, n); |
1251 | |
1252 | /* Construct the Huffman tree by repeatedly combining the least two |
1253 | * frequent nodes. |
1254 | */ |
1255 | do { |
1256 | PQREMOVE(tree, n); /* n = node of least frequency */ |
1257 | m = G2.heap[SMALLEST]; /* m = node of next least frequency */ |
1258 | |
1259 | G2.heap[--G2.heap_max] = n; /* keep the nodes sorted by frequency */ |
1260 | G2.heap[--G2.heap_max] = m; |
1261 | |
1262 | /* Create a new node father of n and m */ |
1263 | tree[node].Freq = tree[n].Freq + tree[m].Freq; |
1264 | G2.depth[node] = MAX(G2.depth[n], G2.depth[m]) + 1; |
1265 | tree[n].Dad = tree[m].Dad = (ush) node; |
1266 | #ifdef DUMP_BL_TREE |
1267 | if (tree == G2.bl_tree) { |
1268 | bb_error_msg("\nnode %d(%d), sons %d(%d) %d(%d)", |
1269 | node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq); |
1270 | } |
1271 | #endif |
1272 | /* and insert the new node in the heap */ |
1273 | G2.heap[SMALLEST] = node++; |
1274 | pqdownheap(tree, SMALLEST); |
1275 | |
1276 | } while (G2.heap_len >= 2); |
1277 | |
1278 | G2.heap[--G2.heap_max] = G2.heap[SMALLEST]; |
1279 | |
1280 | /* At this point, the fields freq and dad are set. We can now |
1281 | * generate the bit lengths. |
1282 | */ |
1283 | gen_bitlen((tree_desc *) desc); |
1284 | |
1285 | /* The field len is now set, we can generate the bit codes */ |
1286 | gen_codes((ct_data *) tree, max_code); |
1287 | } |
1288 | |
1289 | |
1290 | /* =========================================================================== |
1291 | * Scan a literal or distance tree to determine the frequencies of the codes |
1292 | * in the bit length tree. Updates opt_len to take into account the repeat |
1293 | * counts. (The contribution of the bit length codes will be added later |
1294 | * during the construction of bl_tree.) |
1295 | */ |
1296 | static void scan_tree(ct_data * tree, int max_code) |
1297 | { |
1298 | int n; /* iterates over all tree elements */ |
1299 | int prevlen = -1; /* last emitted length */ |
1300 | int curlen; /* length of current code */ |
1301 | int nextlen = tree[0].Len; /* length of next code */ |
1302 | int count = 0; /* repeat count of the current code */ |
1303 | int max_count = 7; /* max repeat count */ |
1304 | int min_count = 4; /* min repeat count */ |
1305 | |
1306 | if (nextlen == 0) { |
1307 | max_count = 138; |
1308 | min_count = 3; |
1309 | } |
1310 | tree[max_code + 1].Len = 0xffff; /* guard */ |
1311 | |
1312 | for (n = 0; n <= max_code; n++) { |
1313 | curlen = nextlen; |
1314 | nextlen = tree[n + 1].Len; |
1315 | if (++count < max_count && curlen == nextlen) |
1316 | continue; |
1317 | |
1318 | if (count < min_count) { |
1319 | G2.bl_tree[curlen].Freq += count; |
1320 | } else if (curlen != 0) { |
1321 | if (curlen != prevlen) |
1322 | G2.bl_tree[curlen].Freq++; |
1323 | G2.bl_tree[REP_3_6].Freq++; |
1324 | } else if (count <= 10) { |
1325 | G2.bl_tree[REPZ_3_10].Freq++; |
1326 | } else { |
1327 | G2.bl_tree[REPZ_11_138].Freq++; |
1328 | } |
1329 | count = 0; |
1330 | prevlen = curlen; |
1331 | |
1332 | max_count = 7; |
1333 | min_count = 4; |
1334 | if (nextlen == 0) { |
1335 | max_count = 138; |
1336 | min_count = 3; |
1337 | } else if (curlen == nextlen) { |
1338 | max_count = 6; |
1339 | min_count = 3; |
1340 | } |
1341 | } |
1342 | } |
1343 | |
1344 | |
1345 | /* =========================================================================== |
1346 | * Send a literal or distance tree in compressed form, using the codes in |
1347 | * bl_tree. |
1348 | */ |
1349 | static void send_tree(ct_data * tree, int max_code) |
1350 | { |
1351 | int n; /* iterates over all tree elements */ |
1352 | int prevlen = -1; /* last emitted length */ |
1353 | int curlen; /* length of current code */ |
1354 | int nextlen = tree[0].Len; /* length of next code */ |
1355 | int count = 0; /* repeat count of the current code */ |
1356 | int max_count = 7; /* max repeat count */ |
1357 | int min_count = 4; /* min repeat count */ |
1358 | |
1359 | /* tree[max_code+1].Len = -1; *//* guard already set */ |
1360 | if (nextlen == 0) |
1361 | max_count = 138, min_count = 3; |
1362 | |
1363 | for (n = 0; n <= max_code; n++) { |
1364 | curlen = nextlen; |
1365 | nextlen = tree[n + 1].Len; |
1366 | if (++count < max_count && curlen == nextlen) { |
1367 | continue; |
1368 | } else if (count < min_count) { |
1369 | do { |
1370 | SEND_CODE(curlen, G2.bl_tree); |
1371 | } while (--count); |
1372 | } else if (curlen != 0) { |
1373 | if (curlen != prevlen) { |
1374 | SEND_CODE(curlen, G2.bl_tree); |
1375 | count--; |
1376 | } |
1377 | Assert(count >= 3 && count <= 6, " 3_6?"); |
1378 | SEND_CODE(REP_3_6, G2.bl_tree); |
1379 | send_bits(count - 3, 2); |
1380 | } else if (count <= 10) { |
1381 | SEND_CODE(REPZ_3_10, G2.bl_tree); |
1382 | send_bits(count - 3, 3); |
1383 | } else { |
1384 | SEND_CODE(REPZ_11_138, G2.bl_tree); |
1385 | send_bits(count - 11, 7); |
1386 | } |
1387 | count = 0; |
1388 | prevlen = curlen; |
1389 | if (nextlen == 0) { |
1390 | max_count = 138; |
1391 | min_count = 3; |
1392 | } else if (curlen == nextlen) { |
1393 | max_count = 6; |
1394 | min_count = 3; |
1395 | } else { |
1396 | max_count = 7; |
1397 | min_count = 4; |
1398 | } |
1399 | } |
1400 | } |
1401 | |
1402 | |
1403 | /* =========================================================================== |
1404 | * Construct the Huffman tree for the bit lengths and return the index in |
1405 | * bl_order of the last bit length code to send. |
1406 | */ |
1407 | static int build_bl_tree(void) |
1408 | { |
1409 | int max_blindex; /* index of last bit length code of non zero freq */ |
1410 | |
1411 | /* Determine the bit length frequencies for literal and distance trees */ |
1412 | scan_tree(G2.dyn_ltree, G2.l_desc.max_code); |
1413 | scan_tree(G2.dyn_dtree, G2.d_desc.max_code); |
1414 | |
1415 | /* Build the bit length tree: */ |
1416 | build_tree(&G2.bl_desc); |
1417 | /* opt_len now includes the length of the tree representations, except |
1418 | * the lengths of the bit lengths codes and the 5+5+4 bits for the counts. |
1419 | */ |
1420 | |
1421 | /* Determine the number of bit length codes to send. The pkzip format |
1422 | * requires that at least 4 bit length codes be sent. (appnote.txt says |
1423 | * 3 but the actual value used is 4.) |
1424 | */ |
1425 | for (max_blindex = BL_CODES - 1; max_blindex >= 3; max_blindex--) { |
1426 | if (G2.bl_tree[bl_order[max_blindex]].Len != 0) |
1427 | break; |
1428 | } |
1429 | /* Update opt_len to include the bit length tree and counts */ |
1430 | G2.opt_len += 3 * (max_blindex + 1) + 5 + 5 + 4; |
1431 | Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", G2.opt_len, G2.static_len)); |
1432 | |
1433 | return max_blindex; |
1434 | } |
1435 | |
1436 | |
1437 | /* =========================================================================== |
1438 | * Send the header for a block using dynamic Huffman trees: the counts, the |
1439 | * lengths of the bit length codes, the literal tree and the distance tree. |
1440 | * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4. |
1441 | */ |
1442 | static void send_all_trees(int lcodes, int dcodes, int blcodes) |
1443 | { |
1444 | int rank; /* index in bl_order */ |
1445 | |
1446 | Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes"); |
1447 | Assert(lcodes <= L_CODES && dcodes <= D_CODES |
1448 | && blcodes <= BL_CODES, "too many codes"); |
1449 | Tracev((stderr, "\nbl counts: ")); |
1450 | send_bits(lcodes - 257, 5); /* not +255 as stated in appnote.txt */ |
1451 | send_bits(dcodes - 1, 5); |
1452 | send_bits(blcodes - 4, 4); /* not -3 as stated in appnote.txt */ |
1453 | for (rank = 0; rank < blcodes; rank++) { |
1454 | Tracev((stderr, "\nbl code %2d ", bl_order[rank])); |
1455 | send_bits(G2.bl_tree[bl_order[rank]].Len, 3); |
1456 | } |
1457 | Tracev((stderr, "\nbl tree: sent %ld", G1.bits_sent)); |
1458 | |
1459 | send_tree((ct_data *) G2.dyn_ltree, lcodes - 1); /* send the literal tree */ |
1460 | Tracev((stderr, "\nlit tree: sent %ld", G1.bits_sent)); |
1461 | |
1462 | send_tree((ct_data *) G2.dyn_dtree, dcodes - 1); /* send the distance tree */ |
1463 | Tracev((stderr, "\ndist tree: sent %ld", G1.bits_sent)); |
1464 | } |
1465 | |
1466 | |
1467 | /* =========================================================================== |
1468 | * Save the match info and tally the frequency counts. Return true if |
1469 | * the current block must be flushed. |
1470 | */ |
1471 | static int ct_tally(int dist, int lc) |
1472 | { |
1473 | G1.l_buf[G2.last_lit++] = lc; |
1474 | if (dist == 0) { |
1475 | /* lc is the unmatched char */ |
1476 | G2.dyn_ltree[lc].Freq++; |
1477 | } else { |
1478 | /* Here, lc is the match length - MIN_MATCH */ |
1479 | dist--; /* dist = match distance - 1 */ |
1480 | Assert((ush) dist < (ush) MAX_DIST |
1481 | && (ush) lc <= (ush) (MAX_MATCH - MIN_MATCH) |
1482 | && (ush) D_CODE(dist) < (ush) D_CODES, "ct_tally: bad match" |
1483 | ); |
1484 | |
1485 | G2.dyn_ltree[G2.length_code[lc] + LITERALS + 1].Freq++; |
1486 | G2.dyn_dtree[D_CODE(dist)].Freq++; |
1487 | |
1488 | G1.d_buf[G2.last_dist++] = dist; |
1489 | G2.flags |= G2.flag_bit; |
1490 | } |
1491 | G2.flag_bit <<= 1; |
1492 | |
1493 | /* Output the flags if they fill a byte: */ |
1494 | if ((G2.last_lit & 7) == 0) { |
1495 | G2.flag_buf[G2.last_flags++] = G2.flags; |
1496 | G2.flags = 0; |
1497 | G2.flag_bit = 1; |
1498 | } |
1499 | /* Try to guess if it is profitable to stop the current block here */ |
1500 | if ((G2.last_lit & 0xfff) == 0) { |
1501 | /* Compute an upper bound for the compressed length */ |
1502 | ulg out_length = G2.last_lit * 8L; |
1503 | ulg in_length = (ulg) G1.strstart - G1.block_start; |
1504 | int dcode; |
1505 | |
1506 | for (dcode = 0; dcode < D_CODES; dcode++) { |
1507 | out_length += G2.dyn_dtree[dcode].Freq * (5L + extra_dbits[dcode]); |
1508 | } |
1509 | out_length >>= 3; |
1510 | Trace((stderr, |
1511 | "\nlast_lit %u, last_dist %u, in %ld, out ~%ld(%ld%%) ", |
1512 | G2.last_lit, G2.last_dist, in_length, out_length, |
1513 | 100L - out_length * 100L / in_length)); |
1514 | if (G2.last_dist < G2.last_lit / 2 && out_length < in_length / 2) |
1515 | return 1; |
1516 | } |
1517 | return (G2.last_lit == LIT_BUFSIZE - 1 || G2.last_dist == DIST_BUFSIZE); |
1518 | /* We avoid equality with LIT_BUFSIZE because of wraparound at 64K |
1519 | * on 16 bit machines and because stored blocks are restricted to |
1520 | * 64K-1 bytes. |
1521 | */ |
1522 | } |
1523 | |
1524 | /* =========================================================================== |
1525 | * Send the block data compressed using the given Huffman trees |
1526 | */ |
1527 | static void compress_block(ct_data * ltree, ct_data * dtree) |
1528 | { |
1529 | unsigned dist; /* distance of matched string */ |
1530 | int lc; /* match length or unmatched char (if dist == 0) */ |
1531 | unsigned lx = 0; /* running index in l_buf */ |
1532 | unsigned dx = 0; /* running index in d_buf */ |
1533 | unsigned fx = 0; /* running index in flag_buf */ |
1534 | uch flag = 0; /* current flags */ |
1535 | unsigned code; /* the code to send */ |
1536 | int extra; /* number of extra bits to send */ |
1537 | |
1538 | if (G2.last_lit != 0) do { |
1539 | if ((lx & 7) == 0) |
1540 | flag = G2.flag_buf[fx++]; |
1541 | lc = G1.l_buf[lx++]; |
1542 | if ((flag & 1) == 0) { |
1543 | SEND_CODE(lc, ltree); /* send a literal byte */ |
1544 | Tracecv(lc > ' ', (stderr, " '%c' ", lc)); |
1545 | } else { |
1546 | /* Here, lc is the match length - MIN_MATCH */ |
1547 | code = G2.length_code[lc]; |
1548 | SEND_CODE(code + LITERALS + 1, ltree); /* send the length code */ |
1549 | extra = extra_lbits[code]; |
1550 | if (extra != 0) { |
1551 | lc -= G2.base_length[code]; |
1552 | send_bits(lc, extra); /* send the extra length bits */ |
1553 | } |
1554 | dist = G1.d_buf[dx++]; |
1555 | /* Here, dist is the match distance - 1 */ |
1556 | code = D_CODE(dist); |
1557 | Assert(code < D_CODES, "bad d_code"); |
1558 | |
1559 | SEND_CODE(code, dtree); /* send the distance code */ |
1560 | extra = extra_dbits[code]; |
1561 | if (extra != 0) { |
1562 | dist -= G2.base_dist[code]; |
1563 | send_bits(dist, extra); /* send the extra distance bits */ |
1564 | } |
1565 | } /* literal or match pair ? */ |
1566 | flag >>= 1; |
1567 | } while (lx < G2.last_lit); |
1568 | |
1569 | SEND_CODE(END_BLOCK, ltree); |
1570 | } |
1571 | |
1572 | |
1573 | /* =========================================================================== |
1574 | * Determine the best encoding for the current block: dynamic trees, static |
1575 | * trees or store, and output the encoded block to the zip file. This function |
1576 | * returns the total compressed length for the file so far. |
1577 | */ |
1578 | static ulg flush_block(char *buf, ulg stored_len, int eof) |
1579 | { |
1580 | ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */ |
1581 | int max_blindex; /* index of last bit length code of non zero freq */ |
1582 | |
1583 | G2.flag_buf[G2.last_flags] = G2.flags; /* Save the flags for the last 8 items */ |
1584 | |
1585 | /* Construct the literal and distance trees */ |
1586 | build_tree(&G2.l_desc); |
1587 | Tracev((stderr, "\nlit data: dyn %ld, stat %ld", G2.opt_len, G2.static_len)); |
1588 | |
1589 | build_tree(&G2.d_desc); |
1590 | Tracev((stderr, "\ndist data: dyn %ld, stat %ld", G2.opt_len, G2.static_len)); |
1591 | /* At this point, opt_len and static_len are the total bit lengths of |
1592 | * the compressed block data, excluding the tree representations. |
1593 | */ |
1594 | |
1595 | /* Build the bit length tree for the above two trees, and get the index |
1596 | * in bl_order of the last bit length code to send. |
1597 | */ |
1598 | max_blindex = build_bl_tree(); |
1599 | |
1600 | /* Determine the best encoding. Compute first the block length in bytes */ |
1601 | opt_lenb = (G2.opt_len + 3 + 7) >> 3; |
1602 | static_lenb = (G2.static_len + 3 + 7) >> 3; |
1603 | |
1604 | Trace((stderr, |
1605 | "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u dist %u ", |
1606 | opt_lenb, G2.opt_len, static_lenb, G2.static_len, stored_len, |
1607 | G2.last_lit, G2.last_dist)); |
1608 | |
1609 | if (static_lenb <= opt_lenb) |
1610 | opt_lenb = static_lenb; |
1611 | |
1612 | /* If compression failed and this is the first and last block, |
1613 | * and if the zip file can be seeked (to rewrite the local header), |
1614 | * the whole file is transformed into a stored file: |
1615 | */ |
1616 | if (stored_len <= opt_lenb && eof && G2.compressed_len == 0L && seekable()) { |
1617 | /* Since LIT_BUFSIZE <= 2*WSIZE, the input data must be there: */ |
1618 | if (buf == NULL) |
1619 | bb_error_msg("block vanished"); |
1620 | |
1621 | copy_block(buf, (unsigned) stored_len, 0); /* without header */ |
1622 | G2.compressed_len = stored_len << 3; |
1623 | |
1624 | } else if (stored_len + 4 <= opt_lenb && buf != NULL) { |
1625 | /* 4: two words for the lengths */ |
1626 | /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE. |
1627 | * Otherwise we can't have processed more than WSIZE input bytes since |
1628 | * the last block flush, because compression would have been |
1629 | * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to |
1630 | * transform a block into a stored block. |
1631 | */ |
1632 | send_bits((STORED_BLOCK << 1) + eof, 3); /* send block type */ |
1633 | G2.compressed_len = (G2.compressed_len + 3 + 7) & ~7L; |
1634 | G2.compressed_len += (stored_len + 4) << 3; |
1635 | |
1636 | copy_block(buf, (unsigned) stored_len, 1); /* with header */ |
1637 | |
1638 | } else if (static_lenb == opt_lenb) { |
1639 | send_bits((STATIC_TREES << 1) + eof, 3); |
1640 | compress_block((ct_data *) G2.static_ltree, (ct_data *) G2.static_dtree); |
1641 | G2.compressed_len += 3 + G2.static_len; |
1642 | } else { |
1643 | send_bits((DYN_TREES << 1) + eof, 3); |
1644 | send_all_trees(G2.l_desc.max_code + 1, G2.d_desc.max_code + 1, |
1645 | max_blindex + 1); |
1646 | compress_block((ct_data *) G2.dyn_ltree, (ct_data *) G2.dyn_dtree); |
1647 | G2.compressed_len += 3 + G2.opt_len; |
1648 | } |
1649 | Assert(G2.compressed_len == G1.bits_sent, "bad compressed size"); |
1650 | init_block(); |
1651 | |
1652 | if (eof) { |
1653 | bi_windup(); |
1654 | G2.compressed_len += 7; /* align on byte boundary */ |
1655 | } |
1656 | Tracev((stderr, "\ncomprlen %lu(%lu) ", G2.compressed_len >> 3, |
1657 | G2.compressed_len - 7 * eof)); |
1658 | |
1659 | return G2.compressed_len >> 3; |
1660 | } |
1661 | |
1662 | |
1663 | /* =========================================================================== |
1664 | * Update a hash value with the given input byte |
1665 | * IN assertion: all calls to to UPDATE_HASH are made with consecutive |
1666 | * input characters, so that a running hash key can be computed from the |
1667 | * previous key instead of complete recalculation each time. |
1668 | */ |
1669 | #define UPDATE_HASH(h, c) (h = (((h)<<H_SHIFT) ^ (c)) & HASH_MASK) |
1670 | |
1671 | |
1672 | /* =========================================================================== |
1673 | * Same as above, but achieves better compression. We use a lazy |
1674 | * evaluation for matches: a match is finally adopted only if there is |
1675 | * no better match at the next window position. |
1676 | * |
1677 | * Processes a new input file and return its compressed length. Sets |
1678 | * the compressed length, crc, deflate flags and internal file |
1679 | * attributes. |
1680 | */ |
1681 | |
1682 | /* Flush the current block, with given end-of-file flag. |
1683 | * IN assertion: strstart is set to the end of the current match. */ |
1684 | #define FLUSH_BLOCK(eof) \ |
1685 | flush_block( \ |
1686 | G1.block_start >= 0L \ |
1687 | ? (char*)&G1.window[(unsigned)G1.block_start] \ |
1688 | : (char*)NULL, \ |
1689 | (ulg)G1.strstart - G1.block_start, \ |
1690 | (eof) \ |
1691 | ) |
1692 | |
1693 | /* Insert string s in the dictionary and set match_head to the previous head |
1694 | * of the hash chain (the most recent string with same hash key). Return |
1695 | * the previous length of the hash chain. |
1696 | * IN assertion: all calls to to INSERT_STRING are made with consecutive |
1697 | * input characters and the first MIN_MATCH bytes of s are valid |
1698 | * (except for the last MIN_MATCH-1 bytes of the input file). */ |
1699 | #define INSERT_STRING(s, match_head) \ |
1700 | do { \ |
1701 | UPDATE_HASH(G1.ins_h, G1.window[(s) + MIN_MATCH-1]); \ |
1702 | G1.prev[(s) & WMASK] = match_head = head[G1.ins_h]; \ |
1703 | head[G1.ins_h] = (s); \ |
1704 | } while (0) |
1705 | |
1706 | static ulg deflate(void) |
1707 | { |
1708 | IPos hash_head; /* head of hash chain */ |
1709 | IPos prev_match; /* previous match */ |
1710 | int flush; /* set if current block must be flushed */ |
1711 | int match_available = 0; /* set if previous match exists */ |
1712 | unsigned match_length = MIN_MATCH - 1; /* length of best match */ |
1713 | |
1714 | /* Process the input block. */ |
1715 | while (G1.lookahead != 0) { |
1716 | /* Insert the string window[strstart .. strstart+2] in the |
1717 | * dictionary, and set hash_head to the head of the hash chain: |
1718 | */ |
1719 | INSERT_STRING(G1.strstart, hash_head); |
1720 | |
1721 | /* Find the longest match, discarding those <= prev_length. |
1722 | */ |
1723 | G1.prev_length = match_length; |
1724 | prev_match = G1.match_start; |
1725 | match_length = MIN_MATCH - 1; |
1726 | |
1727 | if (hash_head != 0 && G1.prev_length < max_lazy_match |
1728 | && G1.strstart - hash_head <= MAX_DIST |
1729 | ) { |
1730 | /* To simplify the code, we prevent matches with the string |
1731 | * of window index 0 (in particular we have to avoid a match |
1732 | * of the string with itself at the start of the input file). |
1733 | */ |
1734 | match_length = longest_match(hash_head); |
1735 | /* longest_match() sets match_start */ |
1736 | if (match_length > G1.lookahead) |
1737 | match_length = G1.lookahead; |
1738 | |
1739 | /* Ignore a length 3 match if it is too distant: */ |
1740 | if (match_length == MIN_MATCH && G1.strstart - G1.match_start > TOO_FAR) { |
1741 | /* If prev_match is also MIN_MATCH, G1.match_start is garbage |
1742 | * but we will ignore the current match anyway. |
1743 | */ |
1744 | match_length--; |
1745 | } |
1746 | } |
1747 | /* If there was a match at the previous step and the current |
1748 | * match is not better, output the previous match: |
1749 | */ |
1750 | if (G1.prev_length >= MIN_MATCH && match_length <= G1.prev_length) { |
1751 | check_match(G1.strstart - 1, prev_match, G1.prev_length); |
1752 | flush = ct_tally(G1.strstart - 1 - prev_match, G1.prev_length - MIN_MATCH); |
1753 | |
1754 | /* Insert in hash table all strings up to the end of the match. |
1755 | * strstart-1 and strstart are already inserted. |
1756 | */ |
1757 | G1.lookahead -= G1.prev_length - 1; |
1758 | G1.prev_length -= 2; |
1759 | do { |
1760 | G1.strstart++; |
1761 | INSERT_STRING(G1.strstart, hash_head); |
1762 | /* strstart never exceeds WSIZE-MAX_MATCH, so there are |
1763 | * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH |
1764 | * these bytes are garbage, but it does not matter since the |
1765 | * next lookahead bytes will always be emitted as literals. |
1766 | */ |
1767 | } while (--G1.prev_length != 0); |
1768 | match_available = 0; |
1769 | match_length = MIN_MATCH - 1; |
1770 | G1.strstart++; |
1771 | if (flush) { |
1772 | FLUSH_BLOCK(0); |
1773 | G1.block_start = G1.strstart; |
1774 | } |
1775 | } else if (match_available) { |
1776 | /* If there was no match at the previous position, output a |
1777 | * single literal. If there was a match but the current match |
1778 | * is longer, truncate the previous match to a single literal. |
1779 | */ |
1780 | Tracevv((stderr, "%c", G1.window[G1.strstart - 1])); |
1781 | if (ct_tally(0, G1.window[G1.strstart - 1])) { |
1782 | FLUSH_BLOCK(0); |
1783 | G1.block_start = G1.strstart; |
1784 | } |
1785 | G1.strstart++; |
1786 | G1.lookahead--; |
1787 | } else { |
1788 | /* There is no previous match to compare with, wait for |
1789 | * the next step to decide. |
1790 | */ |
1791 | match_available = 1; |
1792 | G1.strstart++; |
1793 | G1.lookahead--; |
1794 | } |
1795 | Assert(G1.strstart <= G1.isize && lookahead <= G1.isize, "a bit too far"); |
1796 | |
1797 | /* Make sure that we always have enough lookahead, except |
1798 | * at the end of the input file. We need MAX_MATCH bytes |
1799 | * for the next match, plus MIN_MATCH bytes to insert the |
1800 | * string following the next match. |
1801 | */ |
1802 | while (G1.lookahead < MIN_LOOKAHEAD && !G1.eofile) |
1803 | fill_window(); |
1804 | } |
1805 | if (match_available) |
1806 | ct_tally(0, G1.window[G1.strstart - 1]); |
1807 | |
1808 | return FLUSH_BLOCK(1); /* eof */ |
1809 | } |
1810 | |
1811 | |
1812 | /* =========================================================================== |
1813 | * Initialize the bit string routines. |
1814 | */ |
1815 | static void bi_init(void) |
1816 | { |
1817 | G1.bi_buf = 0; |
1818 | G1.bi_valid = 0; |
1819 | #ifdef DEBUG |
1820 | G1.bits_sent = 0L; |
1821 | #endif |
1822 | } |
1823 | |
1824 | |
1825 | /* =========================================================================== |
1826 | * Initialize the "longest match" routines for a new file |
1827 | */ |
1828 | static void lm_init(ush * flagsp) |
1829 | { |
1830 | unsigned j; |
1831 | |
1832 | /* Initialize the hash table. */ |
1833 | memset(head, 0, HASH_SIZE * sizeof(*head)); |
1834 | /* prev will be initialized on the fly */ |
1835 | |
1836 | /* speed options for the general purpose bit flag */ |
1837 | *flagsp |= 2; /* FAST 4, SLOW 2 */ |
1838 | /* ??? reduce max_chain_length for binary files */ |
1839 | |
1840 | G1.strstart = 0; |
1841 | G1.block_start = 0L; |
1842 | |
1843 | G1.lookahead = file_read(G1.window, |
1844 | sizeof(int) <= 2 ? (unsigned) WSIZE : 2 * WSIZE); |
1845 | |
1846 | if (G1.lookahead == 0 || G1.lookahead == (unsigned) -1) { |
1847 | G1.eofile = 1; |
1848 | G1.lookahead = 0; |
1849 | return; |
1850 | } |
1851 | G1.eofile = 0; |
1852 | /* Make sure that we always have enough lookahead. This is important |
1853 | * if input comes from a device such as a tty. |
1854 | */ |
1855 | while (G1.lookahead < MIN_LOOKAHEAD && !G1.eofile) |
1856 | fill_window(); |
1857 | |
1858 | G1.ins_h = 0; |
1859 | for (j = 0; j < MIN_MATCH - 1; j++) |
1860 | UPDATE_HASH(G1.ins_h, G1.window[j]); |
1861 | /* If lookahead < MIN_MATCH, ins_h is garbage, but this is |
1862 | * not important since only literal bytes will be emitted. |
1863 | */ |
1864 | } |
1865 | |
1866 | |
1867 | /* =========================================================================== |
1868 | * Allocate the match buffer, initialize the various tables and save the |
1869 | * location of the internal file attribute (ascii/binary) and method |
1870 | * (DEFLATE/STORE). |
1871 | * One callsite in zip() |
1872 | */ |
1873 | static void ct_init(void) |
1874 | { |
1875 | int n; /* iterates over tree elements */ |
1876 | int length; /* length value */ |
1877 | int code; /* code value */ |
1878 | int dist; /* distance index */ |
1879 | |
1880 | G2.compressed_len = 0L; |
1881 | |
1882 | #ifdef NOT_NEEDED |
1883 | if (G2.static_dtree[0].Len != 0) |
1884 | return; /* ct_init already called */ |
1885 | #endif |
1886 | |
1887 | /* Initialize the mapping length (0..255) -> length code (0..28) */ |
1888 | length = 0; |
1889 | for (code = 0; code < LENGTH_CODES - 1; code++) { |
1890 | G2.base_length[code] = length; |
1891 | for (n = 0; n < (1 << extra_lbits[code]); n++) { |
1892 | G2.length_code[length++] = code; |
1893 | } |
1894 | } |
1895 | Assert(length == 256, "ct_init: length != 256"); |
1896 | /* Note that the length 255 (match length 258) can be represented |
1897 | * in two different ways: code 284 + 5 bits or code 285, so we |
1898 | * overwrite length_code[255] to use the best encoding: |
1899 | */ |
1900 | G2.length_code[length - 1] = code; |
1901 | |
1902 | /* Initialize the mapping dist (0..32K) -> dist code (0..29) */ |
1903 | dist = 0; |
1904 | for (code = 0; code < 16; code++) { |
1905 | G2.base_dist[code] = dist; |
1906 | for (n = 0; n < (1 << extra_dbits[code]); n++) { |
1907 | G2.dist_code[dist++] = code; |
1908 | } |
1909 | } |
1910 | Assert(dist == 256, "ct_init: dist != 256"); |
1911 | dist >>= 7; /* from now on, all distances are divided by 128 */ |
1912 | for (; code < D_CODES; code++) { |
1913 | G2.base_dist[code] = dist << 7; |
1914 | for (n = 0; n < (1 << (extra_dbits[code] - 7)); n++) { |
1915 | G2.dist_code[256 + dist++] = code; |
1916 | } |
1917 | } |
1918 | Assert(dist == 256, "ct_init: 256+dist != 512"); |
1919 | |
1920 | /* Construct the codes of the static literal tree */ |
1921 | /* already zeroed - it's in bss |
1922 | for (n = 0; n <= MAX_BITS; n++) |
1923 | G2.bl_count[n] = 0; */ |
1924 | |
1925 | n = 0; |
1926 | while (n <= 143) { |
1927 | G2.static_ltree[n++].Len = 8; |
1928 | G2.bl_count[8]++; |
1929 | } |
1930 | while (n <= 255) { |
1931 | G2.static_ltree[n++].Len = 9; |
1932 | G2.bl_count[9]++; |
1933 | } |
1934 | while (n <= 279) { |
1935 | G2.static_ltree[n++].Len = 7; |
1936 | G2.bl_count[7]++; |
1937 | } |
1938 | while (n <= 287) { |
1939 | G2.static_ltree[n++].Len = 8; |
1940 | G2.bl_count[8]++; |
1941 | } |
1942 | /* Codes 286 and 287 do not exist, but we must include them in the |
1943 | * tree construction to get a canonical Huffman tree (longest code |
1944 | * all ones) |
1945 | */ |
1946 | gen_codes((ct_data *) G2.static_ltree, L_CODES + 1); |
1947 | |
1948 | /* The static distance tree is trivial: */ |
1949 | for (n = 0; n < D_CODES; n++) { |
1950 | G2.static_dtree[n].Len = 5; |
1951 | G2.static_dtree[n].Code = bi_reverse(n, 5); |
1952 | } |
1953 | |
1954 | /* Initialize the first block of the first file: */ |
1955 | init_block(); |
1956 | } |
1957 | |
1958 | |
1959 | /* =========================================================================== |
1960 | * Deflate in to out. |
1961 | * IN assertions: the input and output buffers are cleared. |
1962 | */ |
1963 | |
1964 | static void zip(ulg time_stamp) |
1965 | { |
1966 | ush deflate_flags = 0; /* pkzip -es, -en or -ex equivalent */ |
1967 | |
1968 | G1.outcnt = 0; |
1969 | |
1970 | /* Write the header to the gzip file. See algorithm.doc for the format */ |
1971 | /* magic header for gzip files: 1F 8B */ |
1972 | /* compression method: 8 (DEFLATED) */ |
1973 | /* general flags: 0 */ |
1974 | put_32bit(0x00088b1f); |
1975 | put_32bit(time_stamp); |
1976 | |
1977 | /* Write deflated file to zip file */ |
1978 | G1.crc = ~0; |
1979 | |
1980 | bi_init(); |
1981 | ct_init(); |
1982 | lm_init(&deflate_flags); |
1983 | |
1984 | put_8bit(deflate_flags); /* extra flags */ |
1985 | put_8bit(3); /* OS identifier = 3 (Unix) */ |
1986 | |
1987 | deflate(); |
1988 | |
1989 | /* Write the crc and uncompressed size */ |
1990 | put_32bit(~G1.crc); |
1991 | put_32bit(G1.isize); |
1992 | |
1993 | flush_outbuf(); |
1994 | } |
1995 | |
1996 | |
1997 | /* ======================================================================== */ |
1998 | static |
1999 | char* make_new_name_gzip(char *filename) |
2000 | { |
2001 | return xasprintf("%s.gz", filename); |
2002 | } |
2003 | |
2004 | static |
2005 | IF_DESKTOP(long long) int pack_gzip(unpack_info_t *info UNUSED_PARAM) |
2006 | { |
2007 | struct stat s; |
2008 | |
2009 | /* Clear input and output buffers */ |
2010 | G1.outcnt = 0; |
2011 | #ifdef DEBUG |
2012 | G1.insize = 0; |
2013 | #endif |
2014 | G1.isize = 0; |
2015 | |
2016 | /* Reinit G2.xxx */ |
2017 | memset(&G2, 0, sizeof(G2)); |
2018 | G2.l_desc.dyn_tree = G2.dyn_ltree; |
2019 | G2.l_desc.static_tree = G2.static_ltree; |
2020 | G2.l_desc.extra_bits = extra_lbits; |
2021 | G2.l_desc.extra_base = LITERALS + 1; |
2022 | G2.l_desc.elems = L_CODES; |
2023 | G2.l_desc.max_length = MAX_BITS; |
2024 | //G2.l_desc.max_code = 0; |
2025 | G2.d_desc.dyn_tree = G2.dyn_dtree; |
2026 | G2.d_desc.static_tree = G2.static_dtree; |
2027 | G2.d_desc.extra_bits = extra_dbits; |
2028 | //G2.d_desc.extra_base = 0; |
2029 | G2.d_desc.elems = D_CODES; |
2030 | G2.d_desc.max_length = MAX_BITS; |
2031 | //G2.d_desc.max_code = 0; |
2032 | G2.bl_desc.dyn_tree = G2.bl_tree; |
2033 | //G2.bl_desc.static_tree = NULL; |
2034 | G2.bl_desc.extra_bits = extra_blbits, |
2035 | //G2.bl_desc.extra_base = 0; |
2036 | G2.bl_desc.elems = BL_CODES; |
2037 | G2.bl_desc.max_length = MAX_BL_BITS; |
2038 | //G2.bl_desc.max_code = 0; |
2039 | |
2040 | s.st_ctime = 0; |
2041 | fstat(STDIN_FILENO, &s); |
2042 | zip(s.st_ctime); |
2043 | return 0; |
2044 | } |
2045 | |
2046 | #if ENABLE_FEATURE_GZIP_LONG_OPTIONS |
2047 | static const char gzip_longopts[] ALIGN1 = |
2048 | "stdout\0" No_argument "c" |
2049 | "to-stdout\0" No_argument "c" |
2050 | "force\0" No_argument "f" |
2051 | "verbose\0" No_argument "v" |
2052 | #if ENABLE_GUNZIP |
2053 | "decompress\0" No_argument "d" |
2054 | "uncompress\0" No_argument "d" |
2055 | "test\0" No_argument "t" |
2056 | #endif |
2057 | "quiet\0" No_argument "q" |
2058 | "fast\0" No_argument "1" |
2059 | "best\0" No_argument "9" |
2060 | ; |
2061 | #endif |
2062 | |
2063 | /* |
2064 | * Linux kernel build uses gzip -d -n. We accept and ignore it. |
2065 | * Man page says: |
2066 | * -n --no-name |
2067 | * gzip: do not save the original file name and time stamp. |
2068 | * (The original name is always saved if the name had to be truncated.) |
2069 | * gunzip: do not restore the original file name/time even if present |
2070 | * (remove only the gzip suffix from the compressed file name). |
2071 | * This option is the default when decompressing. |
2072 | * -N --name |
2073 | * gzip: always save the original file name and time stamp (this is the default) |
2074 | * gunzip: restore the original file name and time stamp if present. |
2075 | */ |
2076 | |
2077 | int gzip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; |
2078 | #if ENABLE_GUNZIP |
2079 | int gzip_main(int argc, char **argv) |
2080 | #else |
2081 | int gzip_main(int argc UNUSED_PARAM, char **argv) |
2082 | #endif |
2083 | { |
2084 | unsigned opt; |
2085 | |
2086 | #if ENABLE_FEATURE_GZIP_LONG_OPTIONS |
2087 | applet_long_options = gzip_longopts; |
2088 | #endif |
2089 | /* Must match bbunzip's constants OPT_STDOUT, OPT_FORCE! */ |
2090 | opt = getopt32(argv, "cfv" IF_GUNZIP("dt") "q123456789n"); |
2091 | #if ENABLE_GUNZIP /* gunzip_main may not be visible... */ |
2092 | if (opt & 0x18) // -d and/or -t |
2093 | return gunzip_main(argc, argv); |
2094 | #endif |
2095 | option_mask32 &= 0x7; /* ignore -q, -0..9 */ |
2096 | //if (opt & 0x1) // -c |
2097 | //if (opt & 0x2) // -f |
2098 | //if (opt & 0x4) // -v |
2099 | argv += optind; |
2100 | |
2101 | SET_PTR_TO_GLOBALS((char *)xzalloc(sizeof(struct globals)+sizeof(struct globals2)) |
2102 | + sizeof(struct globals)); |
2103 | |
2104 | /* Allocate all global buffers (for DYN_ALLOC option) */ |
2105 | ALLOC(uch, G1.l_buf, INBUFSIZ); |
2106 | ALLOC(uch, G1.outbuf, OUTBUFSIZ); |
2107 | ALLOC(ush, G1.d_buf, DIST_BUFSIZE); |
2108 | ALLOC(uch, G1.window, 2L * WSIZE); |
2109 | ALLOC(ush, G1.prev, 1L << BITS); |
2110 | |
2111 | /* Initialise the CRC32 table */ |
2112 | G1.crc_32_tab = crc32_filltable(NULL, 0); |
2113 | |
2114 | return bbunpack(argv, make_new_name_gzip, pack_gzip); |
2115 | } |