1 /* inftrees.c -- generate Huffman trees for efficient decoding
2 * Copyright (C) 1995-2024 Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
5
6 #include "zutil.h"
7 #include "inftrees.h"
8
9 #define MAXBITS 15
10
11 /*
12 If you use the zlib library in a product, an acknowledgment is welcome
13 in the documentation of your product. If for some reason you cannot
14 include such an acknowledgment, I would appreciate that you keep this
15 copyright string in the executable of your product.
16 */
17
18 /*
19 Build a set of tables to decode the provided canonical Huffman code.
20 The code lengths are lens[0..codes-1]. The result starts at *table,
21 whose indices are 0..2^bits-1. work is a writable array of at least
22 lens shorts, which is used as a work area. type is the type of code
23 to be generated, CODES, LENS, or DISTS. On return, zero is success,
24 -1 is an invalid code, and +1 means that ENOUGH isn't enough. table
25 on return points to the next available entry's address. bits is the
26 requested root table index bits, and on return it is the actual root
27 table index bits. It will differ if the request is greater than the
28 longest code or if it is less than the shortest code.
29 */
inflate_table(codetype type,unsigned short FAR * lens,unsigned codes,code FAR * FAR * table,unsigned FAR * bits,unsigned short FAR * work)30 int ZLIB_INTERNAL inflate_table(codetype type, unsigned short FAR *lens,
31 unsigned codes, code FAR * FAR *table,
32 unsigned FAR *bits, unsigned short FAR *work) {
33 unsigned len; /* a code's length in bits */
34 unsigned sym; /* index of code symbols */
35 unsigned min, max; /* minimum and maximum code lengths */
36 unsigned root; /* number of index bits for root table */
37 unsigned curr; /* number of index bits for current table */
38 unsigned drop; /* code bits to drop for sub-table */
39 int left; /* number of prefix codes available */
40 unsigned used; /* code entries in table used */
41 unsigned huff; /* Huffman code */
42 unsigned incr; /* for incrementing code, index */
43 unsigned fill; /* index for replicating entries */
44 unsigned low; /* low bits for current root entry */
45 unsigned mask; /* mask for low root bits */
46 code here; /* table entry for duplication */
47 code FAR *next; /* next available space in table */
48 const unsigned short FAR *base; /* base value table to use */
49 const unsigned short FAR *extra; /* extra bits table to use */
50 unsigned match; /* use base and extra for symbol >= match */
51 unsigned short count[MAXBITS+1]; /* number of codes of each length */
52 unsigned short offs[MAXBITS+1]; /* offsets in table for each length */
53 static const unsigned short lbase[31] = { /* Length codes 257..285 base */
54 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
55 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
56 static const unsigned short lext[31] = { /* Length codes 257..285 extra */
57 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
58 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 73, 200};
59 static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
60 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
61 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
62 8193, 12289, 16385, 24577, 0, 0};
63 static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
64 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
65 23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
66 28, 28, 29, 29, 64, 64};
67
68 /*
69 Process a set of code lengths to create a canonical Huffman code. The
70 code lengths are lens[0..codes-1]. Each length corresponds to the
71 symbols 0..codes-1. The Huffman code is generated by first sorting the
72 symbols by length from short to long, and retaining the symbol order
73 for codes with equal lengths. Then the code starts with all zero bits
74 for the first code of the shortest length, and the codes are integer
75 increments for the same length, and zeros are appended as the length
76 increases. For the deflate format, these bits are stored backwards
77 from their more natural integer increment ordering, and so when the
78 decoding tables are built in the large loop below, the integer codes
79 are incremented backwards.
80
81 This routine assumes, but does not check, that all of the entries in
82 lens[] are in the range 0..MAXBITS. The caller must assure this.
83 1..MAXBITS is interpreted as that code length. zero means that that
84 symbol does not occur in this code.
85
86 The codes are sorted by computing a count of codes for each length,
87 creating from that a table of starting indices for each length in the
88 sorted table, and then entering the symbols in order in the sorted
89 table. The sorted table is work[], with that space being provided by
90 the caller.
91
92 The length counts are used for other purposes as well, i.e. finding
93 the minimum and maximum length codes, determining if there are any
94 codes at all, checking for a valid set of lengths, and looking ahead
95 at length counts to determine sub-table sizes when building the
96 decoding tables.
97 */
98
99 /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
100 for (len = 0; len <= MAXBITS; len++)
101 count[len] = 0;
102 for (sym = 0; sym < codes; sym++)
103 count[lens[sym]]++;
104
105 /* bound code lengths, force root to be within code lengths */
106 root = *bits;
107 for (max = MAXBITS; max >= 1; max--)
108 if (count[max] != 0) break;
109 if (root > max) root = max;
110 if (max == 0) { /* no symbols to code at all */
111 here.op = (unsigned char)64; /* invalid code marker */
112 here.bits = (unsigned char)1;
113 here.val = (unsigned short)0;
114 *(*table)++ = here; /* make a table to force an error */
115 *(*table)++ = here;
116 *bits = 1;
117 return 0; /* no symbols, but wait for decoding to report error */
118 }
119 for (min = 1; min < max; min++)
120 if (count[min] != 0) break;
121 if (root < min) root = min;
122
123 /* check for an over-subscribed or incomplete set of lengths */
124 left = 1;
125 for (len = 1; len <= MAXBITS; len++) {
126 left <<= 1;
127 left -= count[len];
128 if (left < 0) return -1; /* over-subscribed */
129 }
130 if (left > 0 && (type == CODES || max != 1))
131 return -1; /* incomplete set */
132
133 /* generate offsets into symbol table for each length for sorting */
134 offs[1] = 0;
135 for (len = 1; len < MAXBITS; len++)
136 offs[len + 1] = offs[len] + count[len];
137
138 /* sort symbols by length, by symbol order within each length */
139 for (sym = 0; sym < codes; sym++)
140 if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
141
142 /*
143 Create and fill in decoding tables. In this loop, the table being
144 filled is at next and has curr index bits. The code being used is huff
145 with length len. That code is converted to an index by dropping drop
146 bits off of the bottom. For codes where len is less than drop + curr,
147 those top drop + curr - len bits are incremented through all values to
148 fill the table with replicated entries.
149
150 root is the number of index bits for the root table. When len exceeds
151 root, sub-tables are created pointed to by the root entry with an index
152 of the low root bits of huff. This is saved in low to check for when a
153 new sub-table should be started. drop is zero when the root table is
154 being filled, and drop is root when sub-tables are being filled.
155
156 When a new sub-table is needed, it is necessary to look ahead in the
157 code lengths to determine what size sub-table is needed. The length
158 counts are used for this, and so count[] is decremented as codes are
159 entered in the tables.
160
161 used keeps track of how many table entries have been allocated from the
162 provided *table space. It is checked for LENS and DIST tables against
163 the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
164 the initial root table size constants. See the comments in inftrees.h
165 for more information.
166
167 sym increments through all symbols, and the loop terminates when
168 all codes of length max, i.e. all codes, have been processed. This
169 routine permits incomplete codes, so another loop after this one fills
170 in the rest of the decoding tables with invalid code markers.
171 */
172
173 /* set up for code type */
174 switch (type) {
175 case CODES:
176 base = extra = work; /* dummy value--not used */
177 match = 20;
178 break;
179 case LENS:
180 base = lbase;
181 extra = lext;
182 match = 257;
183 break;
184 default: /* DISTS */
185 base = dbase;
186 extra = dext;
187 match = 0;
188 }
189
190 /* initialize state for loop */
191 huff = 0; /* starting code */
192 sym = 0; /* starting code symbol */
193 len = min; /* starting code length */
194 next = *table; /* current table to fill in */
195 curr = root; /* current table index bits */
196 drop = 0; /* current bits to drop from code for index */
197 low = (unsigned)(-1); /* trigger new sub-table when len > root */
198 used = 1U << root; /* use root table entries */
199 mask = used - 1; /* mask for comparing low */
200
201 /* check available table space */
202 if ((type == LENS && used > ENOUGH_LENS) ||
203 (type == DISTS && used > ENOUGH_DISTS))
204 return 1;
205
206 /* process all codes and make table entries */
207 for (;;) {
208 /* create table entry */
209 here.bits = (unsigned char)(len - drop);
210 if (work[sym] + 1U < match) {
211 here.op = (unsigned char)0;
212 here.val = work[sym];
213 }
214 else if (work[sym] >= match) {
215 here.op = (unsigned char)(extra[work[sym] - match]);
216 here.val = base[work[sym] - match];
217 }
218 else {
219 here.op = (unsigned char)(32 + 64); /* end of block */
220 here.val = 0;
221 }
222
223 /* replicate for those indices with low len bits equal to huff */
224 incr = 1U << (len - drop);
225 fill = 1U << curr;
226 min = fill; /* save offset to next table */
227 do {
228 fill -= incr;
229 next[(huff >> drop) + fill] = here;
230 } while (fill != 0);
231
232 /* backwards increment the len-bit code huff */
233 incr = 1U << (len - 1);
234 while (huff & incr)
235 incr >>= 1;
236 if (incr != 0) {
237 huff &= incr - 1;
238 huff += incr;
239 }
240 else
241 huff = 0;
242
243 /* go to next symbol, update count, len */
244 sym++;
245 if (--(count[len]) == 0) {
246 if (len == max) break;
247 len = lens[work[sym]];
248 }
249
250 /* create new sub-table if needed */
251 if (len > root && (huff & mask) != low) {
252 /* if first time, transition to sub-tables */
253 if (drop == 0)
254 drop = root;
255
256 /* increment past last table */
257 next += min; /* here min is 1 << curr */
258
259 /* determine length of next table */
260 curr = len - drop;
261 left = (int)(1 << curr);
262 while (curr + drop < max) {
263 left -= count[curr + drop];
264 if (left <= 0) break;
265 curr++;
266 left <<= 1;
267 }
268
269 /* check for enough space */
270 used += 1U << curr;
271 if ((type == LENS && used > ENOUGH_LENS) ||
272 (type == DISTS && used > ENOUGH_DISTS))
273 return 1;
274
275 /* point entry in root table to sub-table */
276 low = huff & mask;
277 (*table)[low].op = (unsigned char)curr;
278 (*table)[low].bits = (unsigned char)root;
279 (*table)[low].val = (unsigned short)(next - *table);
280 }
281 }
282
283 /* fill in remaining table entry if code is incomplete (guaranteed to have
284 at most one remaining entry, since if the code is incomplete, the
285 maximum code length that was allowed to get this far is one bit) */
286 if (huff != 0) {
287 here.op = (unsigned char)64; /* invalid code marker */
288 here.bits = (unsigned char)(len - drop);
289 here.val = (unsigned short)0;
290 next[huff] = here;
291 }
292
293 /* set return parameters */
294 *table += used;
295 *bits = root;
296 return 0;
297 }
298