1/* trees.c -- output deflated data using Huffman coding
2 * Copyright (C) 1995-2017 Jean-loup Gailly
3 * detect_data_type() function provided freely by Cosmin Truta, 2006
4 * For conditions of distribution and use, see copyright notice in zlib.h
5 */
6
7/*
8 * ALGORITHM
9 *
10 * The "deflation" process uses several Huffman trees. The more
11 * common source values are represented by shorter bit sequences.
12 *
13 * Each code tree is stored in a compressed form which is itself
14 * a Huffman encoding of the lengths of all the code strings (in
15 * ascending order by source values). The actual code strings are
16 * reconstructed from the lengths in the inflate process, as described
17 * in the deflate specification.
18 *
19 * REFERENCES
20 *
21 * Deutsch, L.P.,"'Deflate' Compressed Data Format Specification".
22 * Available in ftp.uu.net:/pub/archiving/zip/doc/deflate-1.1.doc
23 *
24 * Storer, James A.
25 * Data Compression: Methods and Theory, pp. 49-50.
26 * Computer Science Press, 1988. ISBN 0-7167-8156-5.
27 *
28 * Sedgewick, R.
29 * Algorithms, p290.
30 * Addison-Wesley, 1983. ISBN 0-201-06672-6.
31 */
32
33/* @(#) $Id$ */
34
35#include "zbuild.h"
36#include "deflate.h"
37#include "trees_p.h"
38#include "trees.h"
39
40#ifdef ZLIB_DEBUG
41# include <ctype.h>
42#endif
43
44/* The lengths of the bit length codes are sent in order of decreasing
45 * probability, to avoid transmitting the lengths for unused bit length codes.
46 */
47
48/* ===========================================================================
49 * Local data. These are initialized only once.
50 */
51
52struct static_tree_desc_s {
53 const ct_data *static_tree; /* static tree or NULL */
54 const int *extra_bits; /* extra bits for each code or NULL */
55 int extra_base; /* base index for extra_bits */
56 int elems; /* max number of elements in the tree */
57 unsigned int max_length; /* max bit length for the codes */
58};
59
60static const static_tree_desc static_l_desc =
61{static_ltree, extra_lbits, LITERALS+1, L_CODES, MAX_BITS};
62
63static const static_tree_desc static_d_desc =
64{static_dtree, extra_dbits, 0, D_CODES, MAX_BITS};
65
66static const static_tree_desc static_bl_desc =
67{(const ct_data *)0, extra_blbits, 0, BL_CODES, MAX_BL_BITS};
68
69/* ===========================================================================
70 * Local (static) routines in this file.
71 */
72
73static void init_block (deflate_state *s);
74static void pqdownheap (deflate_state *s, ct_data *tree, int k);
75static void gen_bitlen (deflate_state *s, tree_desc *desc);
76static void build_tree (deflate_state *s, tree_desc *desc);
77static void scan_tree (deflate_state *s, ct_data *tree, int max_code);
78static void send_tree (deflate_state *s, ct_data *tree, int max_code);
79static int build_bl_tree (deflate_state *s);
80static void send_all_trees (deflate_state *s, int lcodes, int dcodes, int blcodes);
81static void compress_block (deflate_state *s, const ct_data *ltree, const ct_data *dtree);
82static int detect_data_type (deflate_state *s);
83static void bi_flush (deflate_state *s);
84
85/* ===========================================================================
86 * Initialize the tree data structures for a new zlib stream.
87 */
88void ZLIB_INTERNAL zng_tr_init(deflate_state *s) {
89 s->l_desc.dyn_tree = s->dyn_ltree;
90 s->l_desc.stat_desc = &static_l_desc;
91
92 s->d_desc.dyn_tree = s->dyn_dtree;
93 s->d_desc.stat_desc = &static_d_desc;
94
95 s->bl_desc.dyn_tree = s->bl_tree;
96 s->bl_desc.stat_desc = &static_bl_desc;
97
98 s->bi_buf = 0;
99 s->bi_valid = 0;
100#ifdef ZLIB_DEBUG
101 s->compressed_len = 0L;
102 s->bits_sent = 0L;
103#endif
104
105 /* Initialize the first block of the first file: */
106 init_block(s);
107}
108
109/* ===========================================================================
110 * Initialize a new block.
111 */
112static void init_block(deflate_state *s) {
113 int n; /* iterates over tree elements */
114
115 /* Initialize the trees. */
116 for (n = 0; n < L_CODES; n++)
117 s->dyn_ltree[n].Freq = 0;
118 for (n = 0; n < D_CODES; n++)
119 s->dyn_dtree[n].Freq = 0;
120 for (n = 0; n < BL_CODES; n++)
121 s->bl_tree[n].Freq = 0;
122
123 s->dyn_ltree[END_BLOCK].Freq = 1;
124 s->opt_len = s->static_len = 0L;
125 s->sym_next = s->matches = 0;
126}
127
128#define SMALLEST 1
129/* Index within the heap array of least frequent node in the Huffman tree */
130
131
132/* ===========================================================================
133 * Remove the smallest element from the heap and recreate the heap with
134 * one less element. Updates heap and heap_len.
135 */
136#define pqremove(s, tree, top) \
137{\
138 top = s->heap[SMALLEST]; \
139 s->heap[SMALLEST] = s->heap[s->heap_len--]; \
140 pqdownheap(s, tree, SMALLEST); \
141}
142
143/* ===========================================================================
144 * Compares to subtrees, using the tree depth as tie breaker when
145 * the subtrees have equal frequency. This minimizes the worst case length.
146 */
147#define smaller(tree, n, m, depth) \
148 (tree[n].Freq < tree[m].Freq || \
149 (tree[n].Freq == tree[m].Freq && depth[n] <= depth[m]))
150
151/* ===========================================================================
152 * Restore the heap property by moving down the tree starting at node k,
153 * exchanging a node with the smallest of its two sons if necessary, stopping
154 * when the heap property is re-established (each father smaller than its
155 * two sons).
156 */
157static void pqdownheap(deflate_state *s, ct_data *tree, int k) {
158 /* tree: the tree to restore */
159 /* k: node to move down */
160 int v = s->heap[k];
161 int j = k << 1; /* left son of k */
162 while (j <= s->heap_len) {
163 /* Set j to the smallest of the two sons: */
164 if (j < s->heap_len && smaller(tree, s->heap[j+1], s->heap[j], s->depth)) {
165 j++;
166 }
167 /* Exit if v is smaller than both sons */
168 if (smaller(tree, v, s->heap[j], s->depth))
169 break;
170
171 /* Exchange v with the smallest son */
172 s->heap[k] = s->heap[j];
173 k = j;
174
175 /* And continue down the tree, setting j to the left son of k */
176 j <<= 1;
177 }
178 s->heap[k] = v;
179}
180
181/* ===========================================================================
182 * Compute the optimal bit lengths for a tree and update the total bit length
183 * for the current block.
184 * IN assertion: the fields freq and dad are set, heap[heap_max] and
185 * above are the tree nodes sorted by increasing frequency.
186 * OUT assertions: the field len is set to the optimal bit length, the
187 * array bl_count contains the frequencies for each bit length.
188 * The length opt_len is updated; static_len is also updated if stree is
189 * not null.
190 */
191static void gen_bitlen(deflate_state *s, tree_desc *desc) {
192 /* desc: the tree descriptor */
193 ct_data *tree = desc->dyn_tree;
194 int max_code = desc->max_code;
195 const ct_data *stree = desc->stat_desc->static_tree;
196 const int *extra = desc->stat_desc->extra_bits;
197 int base = desc->stat_desc->extra_base;
198 unsigned int max_length = desc->stat_desc->max_length;
199 int h; /* heap index */
200 int n, m; /* iterate over the tree elements */
201 unsigned int bits; /* bit length */
202 int xbits; /* extra bits */
203 uint16_t f; /* frequency */
204 int overflow = 0; /* number of elements with bit length too large */
205
206 for (bits = 0; bits <= MAX_BITS; bits++)
207 s->bl_count[bits] = 0;
208
209 /* In a first pass, compute the optimal bit lengths (which may
210 * overflow in the case of the bit length tree).
211 */
212 tree[s->heap[s->heap_max]].Len = 0; /* root of the heap */
213
214 for (h = s->heap_max+1; h < HEAP_SIZE; h++) {
215 n = s->heap[h];
216 bits = tree[tree[n].Dad].Len + 1;
217 if (bits > max_length)
218 bits = max_length, overflow++;
219 tree[n].Len = (uint16_t)bits;
220 /* We overwrite tree[n].Dad which is no longer needed */
221
222 if (n > max_code) /* not a leaf node */
223 continue;
224
225 s->bl_count[bits]++;
226 xbits = 0;
227 if (n >= base)
228 xbits = extra[n-base];
229 f = tree[n].Freq;
230 s->opt_len += (unsigned long)f * (unsigned int)(bits + xbits);
231 if (stree)
232 s->static_len += (unsigned long)f * (unsigned int)(stree[n].Len + xbits);
233 }
234 if (overflow == 0)
235 return;
236
237 Tracev((stderr, "\nbit length overflow\n"));
238 /* This happens for example on obj2 and pic of the Calgary corpus */
239
240 /* Find the first bit length which could increase: */
241 do {
242 bits = max_length-1;
243 while (s->bl_count[bits] == 0)
244 bits--;
245 s->bl_count[bits]--; /* move one leaf down the tree */
246 s->bl_count[bits+1] += 2; /* move one overflow item as its brother */
247 s->bl_count[max_length]--;
248 /* The brother of the overflow item also moves one step up,
249 * but this does not affect bl_count[max_length]
250 */
251 overflow -= 2;
252 } while (overflow > 0);
253
254 /* Now recompute all bit lengths, scanning in increasing frequency.
255 * h is still equal to HEAP_SIZE. (It is simpler to reconstruct all
256 * lengths instead of fixing only the wrong ones. This idea is taken
257 * from 'ar' written by Haruhiko Okumura.)
258 */
259 for (bits = max_length; bits != 0; bits--) {
260 n = s->bl_count[bits];
261 while (n != 0) {
262 m = s->heap[--h];
263 if (m > max_code)
264 continue;
265 if (tree[m].Len != bits) {
266 Tracev((stderr, "code %d bits %d->%u\n", m, tree[m].Len, bits));
267 s->opt_len += (unsigned long)(bits * tree[m].Freq);
268 s->opt_len -= (unsigned long)(tree[m].Len * tree[m].Freq);
269 tree[m].Len = (uint16_t)bits;
270 }
271 n--;
272 }
273 }
274}
275
276/* ===========================================================================
277 * Generate the codes for a given tree and bit counts (which need not be
278 * optimal).
279 * IN assertion: the array bl_count contains the bit length statistics for
280 * the given tree and the field len is set for all tree elements.
281 * OUT assertion: the field code is set for all tree elements of non
282 * zero code length.
283 */
284ZLIB_INTERNAL void gen_codes(ct_data *tree, int max_code, uint16_t *bl_count) {
285 /* tree: the tree to decorate */
286 /* max_code: largest code with non zero frequency */
287 /* bl_count: number of codes at each bit length */
288 uint16_t next_code[MAX_BITS+1]; /* next code value for each bit length */
289 unsigned int code = 0; /* running code value */
290 int bits; /* bit index */
291 int n; /* code index */
292
293 /* The distribution counts are first used to generate the code values
294 * without bit reversal.
295 */
296 for (bits = 1; bits <= MAX_BITS; bits++) {
297 code = (code + bl_count[bits-1]) << 1;
298 next_code[bits] = (uint16_t)code;
299 }
300 /* Check that the bit counts in bl_count are consistent. The last code
301 * must be all ones.
302 */
303 Assert(code + bl_count[MAX_BITS]-1 == (1 << MAX_BITS)-1, "inconsistent bit counts");
304 Tracev((stderr, "\ngen_codes: max_code %d ", max_code));
305
306 for (n = 0; n <= max_code; n++) {
307 int len = tree[n].Len;
308 if (len == 0)
309 continue;
310 /* Now reverse the bits */
311 tree[n].Code = (uint16_t)bi_reverse(next_code[len]++, len);
312
313 Tracecv(tree != static_ltree, (stderr, "\nn %3d %c l %2d c %4x (%x) ",
314 n, (isgraph(n) ? n : ' '), len, tree[n].Code, next_code[len]-1));
315 }
316}
317
318/* ===========================================================================
319 * Construct one Huffman tree and assigns the code bit strings and lengths.
320 * Update the total bit length for the current block.
321 * IN assertion: the field freq is set for all tree elements.
322 * OUT assertions: the fields len and code are set to the optimal bit length
323 * and corresponding code. The length opt_len is updated; static_len is
324 * also updated if stree is not null. The field max_code is set.
325 */
326static void build_tree(deflate_state *s, tree_desc *desc) {
327 /* desc: the tree descriptor */
328 ct_data *tree = desc->dyn_tree;
329 const ct_data *stree = desc->stat_desc->static_tree;
330 int elems = desc->stat_desc->elems;
331 int n, m; /* iterate over heap elements */
332 int max_code = -1; /* largest code with non zero frequency */
333 int node; /* new node being created */
334
335 /* Construct the initial heap, with least frequent element in
336 * heap[SMALLEST]. The sons of heap[n] are heap[2*n] and heap[2*n+1].
337 * heap[0] is not used.
338 */
339 s->heap_len = 0, s->heap_max = HEAP_SIZE;
340
341 for (n = 0; n < elems; n++) {
342 if (tree[n].Freq != 0) {
343 s->heap[++(s->heap_len)] = max_code = n;
344 s->depth[n] = 0;
345 } else {
346 tree[n].Len = 0;
347 }
348 }
349
350 /* The pkzip format requires that at least one distance code exists,
351 * and that at least one bit should be sent even if there is only one
352 * possible code. So to avoid special checks later on we force at least
353 * two codes of non zero frequency.
354 */
355 while (s->heap_len < 2) {
356 node = s->heap[++(s->heap_len)] = (max_code < 2 ? ++max_code : 0);
357 tree[node].Freq = 1;
358 s->depth[node] = 0;
359 s->opt_len--;
360 if (stree)
361 s->static_len -= stree[node].Len;
362 /* node is 0 or 1 so it does not have extra bits */
363 }
364 desc->max_code = max_code;
365
366 /* The elements heap[heap_len/2+1 .. heap_len] are leaves of the tree,
367 * establish sub-heaps of increasing lengths:
368 */
369 for (n = s->heap_len/2; n >= 1; n--)
370 pqdownheap(s, tree, n);
371
372 /* Construct the Huffman tree by repeatedly combining the least two
373 * frequent nodes.
374 */
375 node = elems; /* next internal node of the tree */
376 do {
377 pqremove(s, tree, n); /* n = node of least frequency */
378 m = s->heap[SMALLEST]; /* m = node of next least frequency */
379
380 s->heap[--(s->heap_max)] = n; /* keep the nodes sorted by frequency */
381 s->heap[--(s->heap_max)] = m;
382
383 /* Create a new node father of n and m */
384 tree[node].Freq = tree[n].Freq + tree[m].Freq;
385 s->depth[node] = (unsigned char)((s->depth[n] >= s->depth[m] ?
386 s->depth[n] : s->depth[m]) + 1);
387 tree[n].Dad = tree[m].Dad = (uint16_t)node;
388#ifdef DUMP_BL_TREE
389 if (tree == s->bl_tree) {
390 fprintf(stderr, "\nnode %d(%d), sons %d(%d) %d(%d)",
391 node, tree[node].Freq, n, tree[n].Freq, m, tree[m].Freq);
392 }
393#endif
394 /* and insert the new node in the heap */
395 s->heap[SMALLEST] = node++;
396 pqdownheap(s, tree, SMALLEST);
397 } while (s->heap_len >= 2);
398
399 s->heap[--(s->heap_max)] = s->heap[SMALLEST];
400
401 /* At this point, the fields freq and dad are set. We can now
402 * generate the bit lengths.
403 */
404 gen_bitlen(s, (tree_desc *)desc);
405
406 /* The field len is now set, we can generate the bit codes */
407 gen_codes((ct_data *)tree, max_code, s->bl_count);
408}
409
410/* ===========================================================================
411 * Scan a literal or distance tree to determine the frequencies of the codes
412 * in the bit length tree.
413 */
414static void scan_tree(deflate_state *s, ct_data *tree, int max_code) {
415 /* tree: the tree to be scanned */
416 /* max_code: and its largest code of non zero frequency */
417 int n; /* iterates over all tree elements */
418 int prevlen = -1; /* last emitted length */
419 int curlen; /* length of current code */
420 int nextlen = tree[0].Len; /* length of next code */
421 int count = 0; /* repeat count of the current code */
422 int max_count = 7; /* max repeat count */
423 int min_count = 4; /* min repeat count */
424
425 if (nextlen == 0)
426 max_count = 138, min_count = 3;
427
428 tree[max_code+1].Len = (uint16_t)0xffff; /* guard */
429
430 for (n = 0; n <= max_code; n++) {
431 curlen = nextlen;
432 nextlen = tree[n+1].Len;
433 if (++count < max_count && curlen == nextlen) {
434 continue;
435 } else if (count < min_count) {
436 s->bl_tree[curlen].Freq += count;
437 } else if (curlen != 0) {
438 if (curlen != prevlen)
439 s->bl_tree[curlen].Freq++;
440 s->bl_tree[REP_3_6].Freq++;
441 } else if (count <= 10) {
442 s->bl_tree[REPZ_3_10].Freq++;
443 } else {
444 s->bl_tree[REPZ_11_138].Freq++;
445 }
446 count = 0;
447 prevlen = curlen;
448 if (nextlen == 0) {
449 max_count = 138, min_count = 3;
450 } else if (curlen == nextlen) {
451 max_count = 6, min_count = 3;
452 } else {
453 max_count = 7, min_count = 4;
454 }
455 }
456}
457
458/* ===========================================================================
459 * Send a literal or distance tree in compressed form, using the codes in
460 * bl_tree.
461 */
462static void send_tree(deflate_state *s, ct_data *tree, int max_code) {
463 /* tree: the tree to be scanned */
464 /* max_code and its largest code of non zero frequency */
465 int n; /* iterates over all tree elements */
466 int prevlen = -1; /* last emitted length */
467 int curlen; /* length of current code */
468 int nextlen = tree[0].Len; /* length of next code */
469 int count = 0; /* repeat count of the current code */
470 int max_count = 7; /* max repeat count */
471 int min_count = 4; /* min repeat count */
472
473 /* tree[max_code+1].Len = -1; */ /* guard already set */
474 if (nextlen == 0)
475 max_count = 138, min_count = 3;
476
477 // Temp local variables
478 int filled = s->bi_valid;
479 uint16_t bit_buf = s->bi_buf;
480
481 for (n = 0; n <= max_code; n++) {
482 curlen = nextlen;
483 nextlen = tree[n+1].Len;
484 if (++count < max_count && curlen == nextlen) {
485 continue;
486 } else if (count < min_count) {
487 do {
488 send_code(s, curlen, s->bl_tree, bit_buf, filled);
489 } while (--count != 0);
490
491 } else if (curlen != 0) {
492 if (curlen != prevlen) {
493 send_code(s, curlen, s->bl_tree, bit_buf, filled);
494 count--;
495 }
496 Assert(count >= 3 && count <= 6, " 3_6?");
497 send_code(s, REP_3_6, s->bl_tree, bit_buf, filled);
498 send_bits(s, count-3, 2, bit_buf, filled);
499
500 } else if (count <= 10) {
501 send_code(s, REPZ_3_10, s->bl_tree, bit_buf, filled);
502 send_bits(s, count-3, 3, bit_buf, filled);
503
504 } else {
505 send_code(s, REPZ_11_138, s->bl_tree, bit_buf, filled);
506 send_bits(s, count-11, 7, bit_buf, filled);
507 }
508 count = 0;
509 prevlen = curlen;
510 if (nextlen == 0) {
511 max_count = 138, min_count = 3;
512 } else if (curlen == nextlen) {
513 max_count = 6, min_count = 3;
514 } else {
515 max_count = 7, min_count = 4;
516 }
517 }
518
519 // Store back temp variables
520 s->bi_buf = bit_buf;
521 s->bi_valid = filled;
522}
523
524/* ===========================================================================
525 * Construct the Huffman tree for the bit lengths and return the index in
526 * bl_order of the last bit length code to send.
527 */
528static int build_bl_tree(deflate_state *s) {
529 int max_blindex; /* index of last bit length code of non zero freq */
530
531 /* Determine the bit length frequencies for literal and distance trees */
532 scan_tree(s, (ct_data *)s->dyn_ltree, s->l_desc.max_code);
533 scan_tree(s, (ct_data *)s->dyn_dtree, s->d_desc.max_code);
534
535 /* Build the bit length tree: */
536 build_tree(s, (tree_desc *)(&(s->bl_desc)));
537 /* opt_len now includes the length of the tree representations, except
538 * the lengths of the bit lengths codes and the 5+5+4 bits for the counts.
539 */
540
541 /* Determine the number of bit length codes to send. The pkzip format
542 * requires that at least 4 bit length codes be sent. (appnote.txt says
543 * 3 but the actual value used is 4.)
544 */
545 for (max_blindex = BL_CODES-1; max_blindex >= 3; max_blindex--) {
546 if (s->bl_tree[bl_order[max_blindex]].Len != 0)
547 break;
548 }
549 /* Update opt_len to include the bit length tree and counts */
550 s->opt_len += 3*((unsigned long)max_blindex+1) + 5+5+4;
551 Tracev((stderr, "\ndyn trees: dyn %lu, stat %lu", s->opt_len, s->static_len));
552
553 return max_blindex;
554}
555
556/* ===========================================================================
557 * Send the header for a block using dynamic Huffman trees: the counts, the
558 * lengths of the bit length codes, the literal tree and the distance tree.
559 * IN assertion: lcodes >= 257, dcodes >= 1, blcodes >= 4.
560 */
561static void send_all_trees(deflate_state *s, int lcodes, int dcodes, int blcodes) {
562 int rank; /* index in bl_order */
563
564 Assert(lcodes >= 257 && dcodes >= 1 && blcodes >= 4, "not enough codes");
565 Assert(lcodes <= L_CODES && dcodes <= D_CODES && blcodes <= BL_CODES, "too many codes");
566
567 // Temp local variables
568 int filled = s->bi_valid;
569 uint16_t bit_buf = s->bi_buf;
570
571 Tracev((stderr, "\nbl counts: "));
572 send_bits(s, lcodes-257, 5, bit_buf, filled); /* not +255 as stated in appnote.txt */
573 send_bits(s, dcodes-1, 5, bit_buf, filled);
574 send_bits(s, blcodes-4, 4, bit_buf, filled); /* not -3 as stated in appnote.txt */
575 for (rank = 0; rank < blcodes; rank++) {
576 Tracev((stderr, "\nbl code %2u ", bl_order[rank]));
577 send_bits(s, s->bl_tree[bl_order[rank]].Len, 3, bit_buf, filled);
578 }
579 Tracev((stderr, "\nbl tree: sent %lu", s->bits_sent));
580
581 // Store back temp variables
582 s->bi_buf = bit_buf;
583 s->bi_valid = filled;
584
585 send_tree(s, (ct_data *)s->dyn_ltree, lcodes-1); /* literal tree */
586 Tracev((stderr, "\nlit tree: sent %lu", s->bits_sent));
587
588 send_tree(s, (ct_data *)s->dyn_dtree, dcodes-1); /* distance tree */
589 Tracev((stderr, "\ndist tree: sent %lu", s->bits_sent));
590}
591
592/* ===========================================================================
593 * Send a stored block
594 */
595void ZLIB_INTERNAL zng_tr_stored_block(deflate_state *s, char *buf, unsigned long stored_len, int last) {
596 /* buf: input block */
597 /* stored_len: length of input block */
598 /* last: one if this is the last block for a file */
599 send_bits(s, (STORED_BLOCK << 1)+last, 3, s->bi_buf, s->bi_valid); /* send block type */
600 bi_windup(s); /* align on byte boundary */
601 put_short(s, (uint16_t)stored_len);
602 put_short(s, (uint16_t)~stored_len);
603 if (stored_len)
604 memcpy(s->pending_buf + s->pending, (unsigned char *)buf, stored_len);
605 s->pending += stored_len;
606#ifdef ZLIB_DEBUG
607 s->compressed_len = (s->compressed_len + 3 + 7) & (unsigned long)~7L;
608 s->compressed_len += (stored_len + 4) << 3;
609 s->bits_sent += 2*16;
610 s->bits_sent += stored_len<<3;
611#endif
612}
613
614/* ===========================================================================
615 * Flush the bits in the bit buffer to pending output (leaves at most 7 bits)
616 */
617void ZLIB_INTERNAL zng_tr_flush_bits(deflate_state *s) {
618 bi_flush(s);
619}
620
621/* ===========================================================================
622 * Send one empty static block to give enough lookahead for inflate.
623 * This takes 10 bits, of which 7 may remain in the bit buffer.
624 */
625void ZLIB_INTERNAL zng_tr_align(deflate_state *s) {
626 send_bits(s, STATIC_TREES << 1, 3, s->bi_buf, s->bi_valid);
627 send_code(s, END_BLOCK, static_ltree, s->bi_buf, s->bi_valid);
628#ifdef ZLIB_DEBUG
629 s->compressed_len += 10L; /* 3 for block type, 7 for EOB */
630#endif
631 bi_flush(s);
632}
633
634/* ===========================================================================
635 * Determine the best encoding for the current block: dynamic trees, static
636 * trees or store, and write out the encoded block.
637 */
638void ZLIB_INTERNAL zng_tr_flush_block(deflate_state *s, char *buf, unsigned long stored_len, int last) {
639 /* buf: input block, or NULL if too old */
640 /* stored_len: length of input block */
641 /* last: one if this is the last block for a file */
642 unsigned long opt_lenb, static_lenb; /* opt_len and static_len in bytes */
643 int max_blindex = 0; /* index of last bit length code of non zero freq */
644
645 /* Build the Huffman trees unless a stored block is forced */
646 if (s->level > 0) {
647 /* Check if the file is binary or text */
648 if (s->strm->data_type == Z_UNKNOWN)
649 s->strm->data_type = detect_data_type(s);
650
651 /* Construct the literal and distance trees */
652 build_tree(s, (tree_desc *)(&(s->l_desc)));
653 Tracev((stderr, "\nlit data: dyn %lu, stat %lu", s->opt_len, s->static_len));
654
655 build_tree(s, (tree_desc *)(&(s->d_desc)));
656 Tracev((stderr, "\ndist data: dyn %lu, stat %lu", s->opt_len, s->static_len));
657 /* At this point, opt_len and static_len are the total bit lengths of
658 * the compressed block data, excluding the tree representations.
659 */
660
661 /* Build the bit length tree for the above two trees, and get the index
662 * in bl_order of the last bit length code to send.
663 */
664 max_blindex = build_bl_tree(s);
665
666 /* Determine the best encoding. Compute the block lengths in bytes. */
667 opt_lenb = (s->opt_len+3+7) >> 3;
668 static_lenb = (s->static_len+3+7) >> 3;
669
670 Tracev((stderr, "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u ",
671 opt_lenb, s->opt_len, static_lenb, s->static_len, stored_len,
672 s->sym_next / 3));
673
674 if (static_lenb <= opt_lenb)
675 opt_lenb = static_lenb;
676
677 } else {
678 Assert(buf != NULL, "lost buf");
679 opt_lenb = static_lenb = stored_len + 5; /* force a stored block */
680 }
681
682#ifdef FORCE_STORED
683 if (buf != NULL) { /* force stored block */
684#else
685 if (stored_len+4 <= opt_lenb && buf != NULL) {
686 /* 4: two words for the lengths */
687#endif
688 /* The test buf != NULL is only necessary if LIT_BUFSIZE > WSIZE.
689 * Otherwise we can't have processed more than WSIZE input bytes since
690 * the last block flush, because compression would have been
691 * successful. If LIT_BUFSIZE <= WSIZE, it is never too late to
692 * transform a block into a stored block.
693 */
694 zng_tr_stored_block(s, buf, stored_len, last);
695
696#ifdef FORCE_STATIC
697 } else if (static_lenb >= 0) { /* force static trees */
698#else
699 } else if (s->strategy == Z_FIXED || static_lenb == opt_lenb) {
700#endif
701 send_bits(s, (STATIC_TREES << 1)+last, 3, s->bi_buf, s->bi_valid);
702 compress_block(s, (const ct_data *)static_ltree, (const ct_data *)static_dtree);
703#ifdef ZLIB_DEBUG
704 s->compressed_len += 3 + s->static_len;
705#endif
706 } else {
707 send_bits(s, (DYN_TREES << 1)+last, 3, s->bi_buf, s->bi_valid);
708 send_all_trees(s, s->l_desc.max_code+1, s->d_desc.max_code+1, max_blindex+1);
709 compress_block(s, (const ct_data *)s->dyn_ltree, (const ct_data *)s->dyn_dtree);
710#ifdef ZLIB_DEBUG
711 s->compressed_len += 3 + s->opt_len;
712#endif
713 }
714 Assert(s->compressed_len == s->bits_sent, "bad compressed size");
715 /* The above check is made mod 2^32, for files larger than 512 MB
716 * and unsigned long implemented on 32 bits.
717 */
718 init_block(s);
719
720 if (last) {
721 bi_windup(s);
722#ifdef ZLIB_DEBUG
723 s->compressed_len += 7; /* align on byte boundary */
724#endif
725 }
726 Tracev((stderr, "\ncomprlen %lu(%lu) ", s->compressed_len>>3, s->compressed_len-7*last));
727}
728
729/* ===========================================================================
730 * Save the match info and tally the frequency counts. Return true if
731 * the current block must be flushed.
732 */
733int ZLIB_INTERNAL zng_tr_tally(deflate_state *s, unsigned dist, unsigned lc) {
734 /* dist: distance of matched string */
735 /* lc: match length-MIN_MATCH or unmatched char (if dist==0) */
736 s->sym_buf[s->sym_next++] = dist;
737 s->sym_buf[s->sym_next++] = dist >> 8;
738 s->sym_buf[s->sym_next++] = lc;
739 if (dist == 0) {
740 /* lc is the unmatched char */
741 s->dyn_ltree[lc].Freq++;
742 } else {
743 s->matches++;
744 /* Here, lc is the match length - MIN_MATCH */
745 dist--; /* dist = match distance - 1 */
746 Assert((uint16_t)dist < (uint16_t)MAX_DIST(s) &&
747 (uint16_t)lc <= (uint16_t)(MAX_MATCH-MIN_MATCH) &&
748 (uint16_t)d_code(dist) < (uint16_t)D_CODES, "zng_tr_tally: bad match");
749
750 s->dyn_ltree[zng_length_code[lc]+LITERALS+1].Freq++;
751 s->dyn_dtree[d_code(dist)].Freq++;
752 }
753 return (s->sym_next == s->sym_end);
754}
755
756/* ===========================================================================
757 * Send the block data compressed using the given Huffman trees
758 */
759static void compress_block(deflate_state *s, const ct_data *ltree, const ct_data *dtree) {
760 /* ltree: literal tree */
761 /* dtree: distance tree */
762 unsigned dist; /* distance of matched string */
763 int lc; /* match length or unmatched char (if dist == 0) */
764 unsigned sx = 0; /* running index in sym_buf */
765 int code; /* the code to send */
766 int extra; /* number of extra bits to send */
767
768 // Temp local variables
769 int filled = s->bi_valid;
770 uint16_t bit_buf = s->bi_buf;
771
772 if (s->sym_next != 0) {
773 do {
774 dist = s->sym_buf[sx++] & 0xff;
775 dist += (unsigned)(s->sym_buf[sx++] & 0xff) << 8;
776 lc = s->sym_buf[sx++];
777 if (dist == 0) {
778 send_code(s, lc, ltree, bit_buf, filled) /* send a literal byte */
779 Tracecv(isgraph(lc), (stderr, " '%c' ", lc));
780 } else {
781 /* Here, lc is the match length - MIN_MATCH */
782 code = zng_length_code[lc];
783 send_code(s, code+LITERALS+1, ltree, bit_buf, filled); /* send the length code */
784 extra = extra_lbits[code];
785 if (extra != 0) {
786 lc -= base_length[code];
787 send_bits(s, lc, extra, bit_buf, filled); /* send the extra length bits */
788 }
789 dist--; /* dist is now the match distance - 1 */
790 code = d_code(dist);
791 Assert(code < D_CODES, "bad d_code");
792
793 send_code(s, code, dtree, bit_buf, filled); /* send the distance code */
794 extra = extra_dbits[code];
795 if (extra != 0) {
796 dist -= (unsigned int)base_dist[code];
797 send_bits(s, dist, extra, bit_buf, filled); /* send the extra distance bits */
798 }
799 } /* literal or match pair ? */
800
801 /* Check that the overlay between pending_buf and sym_buf is ok: */
802 Assert(s->pending < s->lit_bufsize + sx, "pendingBuf overflow");
803 } while (sx < s->sym_next);
804 }
805
806 send_code(s, END_BLOCK, ltree, bit_buf, filled);
807
808 // Store back temp variables
809 s->bi_buf = bit_buf;
810 s->bi_valid = filled;
811}
812
813/* ===========================================================================
814 * Check if the data type is TEXT or BINARY, using the following algorithm:
815 * - TEXT if the two conditions below are satisfied:
816 * a) There are no non-portable control characters belonging to the
817 * "black list" (0..6, 14..25, 28..31).
818 * b) There is at least one printable character belonging to the
819 * "white list" (9 {TAB}, 10 {LF}, 13 {CR}, 32..255).
820 * - BINARY otherwise.
821 * - The following partially-portable control characters form a
822 * "gray list" that is ignored in this detection algorithm:
823 * (7 {BEL}, 8 {BS}, 11 {VT}, 12 {FF}, 26 {SUB}, 27 {ESC}).
824 * IN assertion: the fields Freq of dyn_ltree are set.
825 */
826static int detect_data_type(deflate_state *s) {
827 /* black_mask is the bit mask of black-listed bytes
828 * set bits 0..6, 14..25, and 28..31
829 * 0xf3ffc07f = binary 11110011111111111100000001111111
830 */
831 unsigned long black_mask = 0xf3ffc07fUL;
832 int n;
833
834 /* Check for non-textual ("black-listed") bytes. */
835 for (n = 0; n <= 31; n++, black_mask >>= 1)
836 if ((black_mask & 1) && (s->dyn_ltree[n].Freq != 0))
837 return Z_BINARY;
838
839 /* Check for textual ("white-listed") bytes. */
840 if (s->dyn_ltree[9].Freq != 0 || s->dyn_ltree[10].Freq != 0 || s->dyn_ltree[13].Freq != 0)
841 return Z_TEXT;
842 for (n = 32; n < LITERALS; n++)
843 if (s->dyn_ltree[n].Freq != 0)
844 return Z_TEXT;
845
846 /* There are no "black-listed" or "white-listed" bytes:
847 * this stream either is empty or has tolerated ("gray-listed") bytes only.
848 */
849 return Z_BINARY;
850}
851
852/* ===========================================================================
853 * Reverse the first len bits of a code, using straightforward code (a faster
854 * method would use a table)
855 * IN assertion: 1 <= len <= 15
856 */
857ZLIB_INTERNAL unsigned bi_reverse(unsigned code, int len) {
858 /* code: the value to invert */
859 /* len: its bit length */
860 register unsigned res = 0;
861 do {
862 res |= code & 1;
863 code >>= 1, res <<= 1;
864 } while (--len > 0);
865 return res >> 1;
866}
867
868/* ===========================================================================
869 * Flush the bit buffer, keeping at most 7 bits in it.
870 */
871static void bi_flush(deflate_state *s) {
872 if (s->bi_valid == 16) {
873 put_short(s, s->bi_buf);
874 s->bi_buf = 0;
875 s->bi_valid = 0;
876 } else if (s->bi_valid >= 8) {
877 put_byte(s, (unsigned char)s->bi_buf);
878 s->bi_buf >>= 8;
879 s->bi_valid -= 8;
880 }
881}
882
883/* ===========================================================================
884 * Flush the bit buffer and align the output on a byte boundary
885 */
886ZLIB_INTERNAL void bi_windup(deflate_state *s) {
887 if (s->bi_valid > 8) {
888 put_short(s, s->bi_buf);
889 } else if (s->bi_valid > 0) {
890 put_byte(s, (unsigned char)s->bi_buf);
891 }
892 s->bi_buf = 0;
893 s->bi_valid = 0;
894#ifdef ZLIB_DEBUG
895 s->bits_sent = (s->bits_sent+7) & ~7;
896#endif
897}
898