1 | n/a | /* inftrees.c -- generate Huffman trees for efficient decoding |
---|
2 | n/a | * Copyright (C) 1995-2017 Mark Adler |
---|
3 | n/a | * For conditions of distribution and use, see copyright notice in zlib.h |
---|
4 | n/a | */ |
---|
5 | n/a | |
---|
6 | n/a | #include "zutil.h" |
---|
7 | n/a | #include "inftrees.h" |
---|
8 | n/a | |
---|
9 | n/a | #define MAXBITS 15 |
---|
10 | n/a | |
---|
11 | n/a | const char inflate_copyright[] = |
---|
12 | n/a | " inflate 1.2.11 Copyright 1995-2017 Mark Adler "; |
---|
13 | n/a | /* |
---|
14 | n/a | If you use the zlib library in a product, an acknowledgment is welcome |
---|
15 | n/a | in the documentation of your product. If for some reason you cannot |
---|
16 | n/a | include such an acknowledgment, I would appreciate that you keep this |
---|
17 | n/a | copyright string in the executable of your product. |
---|
18 | n/a | */ |
---|
19 | n/a | |
---|
20 | n/a | /* |
---|
21 | n/a | Build a set of tables to decode the provided canonical Huffman code. |
---|
22 | n/a | The code lengths are lens[0..codes-1]. The result starts at *table, |
---|
23 | n/a | whose indices are 0..2^bits-1. work is a writable array of at least |
---|
24 | n/a | lens shorts, which is used as a work area. type is the type of code |
---|
25 | n/a | to be generated, CODES, LENS, or DISTS. On return, zero is success, |
---|
26 | n/a | -1 is an invalid code, and +1 means that ENOUGH isn't enough. table |
---|
27 | n/a | on return points to the next available entry's address. bits is the |
---|
28 | n/a | requested root table index bits, and on return it is the actual root |
---|
29 | n/a | table index bits. It will differ if the request is greater than the |
---|
30 | n/a | longest code or if it is less than the shortest code. |
---|
31 | n/a | */ |
---|
32 | n/a | int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work) |
---|
33 | n/a | codetype type; |
---|
34 | n/a | unsigned short FAR *lens; |
---|
35 | n/a | unsigned codes; |
---|
36 | n/a | code FAR * FAR *table; |
---|
37 | n/a | unsigned FAR *bits; |
---|
38 | n/a | unsigned short FAR *work; |
---|
39 | n/a | { |
---|
40 | n/a | unsigned len; /* a code's length in bits */ |
---|
41 | n/a | unsigned sym; /* index of code symbols */ |
---|
42 | n/a | unsigned min, max; /* minimum and maximum code lengths */ |
---|
43 | n/a | unsigned root; /* number of index bits for root table */ |
---|
44 | n/a | unsigned curr; /* number of index bits for current table */ |
---|
45 | n/a | unsigned drop; /* code bits to drop for sub-table */ |
---|
46 | n/a | int left; /* number of prefix codes available */ |
---|
47 | n/a | unsigned used; /* code entries in table used */ |
---|
48 | n/a | unsigned huff; /* Huffman code */ |
---|
49 | n/a | unsigned incr; /* for incrementing code, index */ |
---|
50 | n/a | unsigned fill; /* index for replicating entries */ |
---|
51 | n/a | unsigned low; /* low bits for current root entry */ |
---|
52 | n/a | unsigned mask; /* mask for low root bits */ |
---|
53 | n/a | code here; /* table entry for duplication */ |
---|
54 | n/a | code FAR *next; /* next available space in table */ |
---|
55 | n/a | const unsigned short FAR *base; /* base value table to use */ |
---|
56 | n/a | const unsigned short FAR *extra; /* extra bits table to use */ |
---|
57 | n/a | unsigned match; /* use base and extra for symbol >= match */ |
---|
58 | n/a | unsigned short count[MAXBITS+1]; /* number of codes of each length */ |
---|
59 | n/a | unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ |
---|
60 | n/a | static const unsigned short lbase[31] = { /* Length codes 257..285 base */ |
---|
61 | n/a | 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, |
---|
62 | n/a | 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; |
---|
63 | n/a | static const unsigned short lext[31] = { /* Length codes 257..285 extra */ |
---|
64 | n/a | 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, |
---|
65 | n/a | 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 77, 202}; |
---|
66 | n/a | static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ |
---|
67 | n/a | 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, |
---|
68 | n/a | 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, |
---|
69 | n/a | 8193, 12289, 16385, 24577, 0, 0}; |
---|
70 | n/a | static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ |
---|
71 | n/a | 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, |
---|
72 | n/a | 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, |
---|
73 | n/a | 28, 28, 29, 29, 64, 64}; |
---|
74 | n/a | |
---|
75 | n/a | /* |
---|
76 | n/a | Process a set of code lengths to create a canonical Huffman code. The |
---|
77 | n/a | code lengths are lens[0..codes-1]. Each length corresponds to the |
---|
78 | n/a | symbols 0..codes-1. The Huffman code is generated by first sorting the |
---|
79 | n/a | symbols by length from short to long, and retaining the symbol order |
---|
80 | n/a | for codes with equal lengths. Then the code starts with all zero bits |
---|
81 | n/a | for the first code of the shortest length, and the codes are integer |
---|
82 | n/a | increments for the same length, and zeros are appended as the length |
---|
83 | n/a | increases. For the deflate format, these bits are stored backwards |
---|
84 | n/a | from their more natural integer increment ordering, and so when the |
---|
85 | n/a | decoding tables are built in the large loop below, the integer codes |
---|
86 | n/a | are incremented backwards. |
---|
87 | n/a | |
---|
88 | n/a | This routine assumes, but does not check, that all of the entries in |
---|
89 | n/a | lens[] are in the range 0..MAXBITS. The caller must assure this. |
---|
90 | n/a | 1..MAXBITS is interpreted as that code length. zero means that that |
---|
91 | n/a | symbol does not occur in this code. |
---|
92 | n/a | |
---|
93 | n/a | The codes are sorted by computing a count of codes for each length, |
---|
94 | n/a | creating from that a table of starting indices for each length in the |
---|
95 | n/a | sorted table, and then entering the symbols in order in the sorted |
---|
96 | n/a | table. The sorted table is work[], with that space being provided by |
---|
97 | n/a | the caller. |
---|
98 | n/a | |
---|
99 | n/a | The length counts are used for other purposes as well, i.e. finding |
---|
100 | n/a | the minimum and maximum length codes, determining if there are any |
---|
101 | n/a | codes at all, checking for a valid set of lengths, and looking ahead |
---|
102 | n/a | at length counts to determine sub-table sizes when building the |
---|
103 | n/a | decoding tables. |
---|
104 | n/a | */ |
---|
105 | n/a | |
---|
106 | n/a | /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ |
---|
107 | n/a | for (len = 0; len <= MAXBITS; len++) |
---|
108 | n/a | count[len] = 0; |
---|
109 | n/a | for (sym = 0; sym < codes; sym++) |
---|
110 | n/a | count[lens[sym]]++; |
---|
111 | n/a | |
---|
112 | n/a | /* bound code lengths, force root to be within code lengths */ |
---|
113 | n/a | root = *bits; |
---|
114 | n/a | for (max = MAXBITS; max >= 1; max--) |
---|
115 | n/a | if (count[max] != 0) break; |
---|
116 | n/a | if (root > max) root = max; |
---|
117 | n/a | if (max == 0) { /* no symbols to code at all */ |
---|
118 | n/a | here.op = (unsigned char)64; /* invalid code marker */ |
---|
119 | n/a | here.bits = (unsigned char)1; |
---|
120 | n/a | here.val = (unsigned short)0; |
---|
121 | n/a | *(*table)++ = here; /* make a table to force an error */ |
---|
122 | n/a | *(*table)++ = here; |
---|
123 | n/a | *bits = 1; |
---|
124 | n/a | return 0; /* no symbols, but wait for decoding to report error */ |
---|
125 | n/a | } |
---|
126 | n/a | for (min = 1; min < max; min++) |
---|
127 | n/a | if (count[min] != 0) break; |
---|
128 | n/a | if (root < min) root = min; |
---|
129 | n/a | |
---|
130 | n/a | /* check for an over-subscribed or incomplete set of lengths */ |
---|
131 | n/a | left = 1; |
---|
132 | n/a | for (len = 1; len <= MAXBITS; len++) { |
---|
133 | n/a | left <<= 1; |
---|
134 | n/a | left -= count[len]; |
---|
135 | n/a | if (left < 0) return -1; /* over-subscribed */ |
---|
136 | n/a | } |
---|
137 | n/a | if (left > 0 && (type == CODES || max != 1)) |
---|
138 | n/a | return -1; /* incomplete set */ |
---|
139 | n/a | |
---|
140 | n/a | /* generate offsets into symbol table for each length for sorting */ |
---|
141 | n/a | offs[1] = 0; |
---|
142 | n/a | for (len = 1; len < MAXBITS; len++) |
---|
143 | n/a | offs[len + 1] = offs[len] + count[len]; |
---|
144 | n/a | |
---|
145 | n/a | /* sort symbols by length, by symbol order within each length */ |
---|
146 | n/a | for (sym = 0; sym < codes; sym++) |
---|
147 | n/a | if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; |
---|
148 | n/a | |
---|
149 | n/a | /* |
---|
150 | n/a | Create and fill in decoding tables. In this loop, the table being |
---|
151 | n/a | filled is at next and has curr index bits. The code being used is huff |
---|
152 | n/a | with length len. That code is converted to an index by dropping drop |
---|
153 | n/a | bits off of the bottom. For codes where len is less than drop + curr, |
---|
154 | n/a | those top drop + curr - len bits are incremented through all values to |
---|
155 | n/a | fill the table with replicated entries. |
---|
156 | n/a | |
---|
157 | n/a | root is the number of index bits for the root table. When len exceeds |
---|
158 | n/a | root, sub-tables are created pointed to by the root entry with an index |
---|
159 | n/a | of the low root bits of huff. This is saved in low to check for when a |
---|
160 | n/a | new sub-table should be started. drop is zero when the root table is |
---|
161 | n/a | being filled, and drop is root when sub-tables are being filled. |
---|
162 | n/a | |
---|
163 | n/a | When a new sub-table is needed, it is necessary to look ahead in the |
---|
164 | n/a | code lengths to determine what size sub-table is needed. The length |
---|
165 | n/a | counts are used for this, and so count[] is decremented as codes are |
---|
166 | n/a | entered in the tables. |
---|
167 | n/a | |
---|
168 | n/a | used keeps track of how many table entries have been allocated from the |
---|
169 | n/a | provided *table space. It is checked for LENS and DIST tables against |
---|
170 | n/a | the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in |
---|
171 | n/a | the initial root table size constants. See the comments in inftrees.h |
---|
172 | n/a | for more information. |
---|
173 | n/a | |
---|
174 | n/a | sym increments through all symbols, and the loop terminates when |
---|
175 | n/a | all codes of length max, i.e. all codes, have been processed. This |
---|
176 | n/a | routine permits incomplete codes, so another loop after this one fills |
---|
177 | n/a | in the rest of the decoding tables with invalid code markers. |
---|
178 | n/a | */ |
---|
179 | n/a | |
---|
180 | n/a | /* set up for code type */ |
---|
181 | n/a | switch (type) { |
---|
182 | n/a | case CODES: |
---|
183 | n/a | base = extra = work; /* dummy value--not used */ |
---|
184 | n/a | match = 20; |
---|
185 | n/a | break; |
---|
186 | n/a | case LENS: |
---|
187 | n/a | base = lbase; |
---|
188 | n/a | extra = lext; |
---|
189 | n/a | match = 257; |
---|
190 | n/a | break; |
---|
191 | n/a | default: /* DISTS */ |
---|
192 | n/a | base = dbase; |
---|
193 | n/a | extra = dext; |
---|
194 | n/a | match = 0; |
---|
195 | n/a | } |
---|
196 | n/a | |
---|
197 | n/a | /* initialize state for loop */ |
---|
198 | n/a | huff = 0; /* starting code */ |
---|
199 | n/a | sym = 0; /* starting code symbol */ |
---|
200 | n/a | len = min; /* starting code length */ |
---|
201 | n/a | next = *table; /* current table to fill in */ |
---|
202 | n/a | curr = root; /* current table index bits */ |
---|
203 | n/a | drop = 0; /* current bits to drop from code for index */ |
---|
204 | n/a | low = (unsigned)(-1); /* trigger new sub-table when len > root */ |
---|
205 | n/a | used = 1U << root; /* use root table entries */ |
---|
206 | n/a | mask = used - 1; /* mask for comparing low */ |
---|
207 | n/a | |
---|
208 | n/a | /* check available table space */ |
---|
209 | n/a | if ((type == LENS && used > ENOUGH_LENS) || |
---|
210 | n/a | (type == DISTS && used > ENOUGH_DISTS)) |
---|
211 | n/a | return 1; |
---|
212 | n/a | |
---|
213 | n/a | /* process all codes and make table entries */ |
---|
214 | n/a | for (;;) { |
---|
215 | n/a | /* create table entry */ |
---|
216 | n/a | here.bits = (unsigned char)(len - drop); |
---|
217 | n/a | if (work[sym] + 1U < match) { |
---|
218 | n/a | here.op = (unsigned char)0; |
---|
219 | n/a | here.val = work[sym]; |
---|
220 | n/a | } |
---|
221 | n/a | else if (work[sym] >= match) { |
---|
222 | n/a | here.op = (unsigned char)(extra[work[sym] - match]); |
---|
223 | n/a | here.val = base[work[sym] - match]; |
---|
224 | n/a | } |
---|
225 | n/a | else { |
---|
226 | n/a | here.op = (unsigned char)(32 + 64); /* end of block */ |
---|
227 | n/a | here.val = 0; |
---|
228 | n/a | } |
---|
229 | n/a | |
---|
230 | n/a | /* replicate for those indices with low len bits equal to huff */ |
---|
231 | n/a | incr = 1U << (len - drop); |
---|
232 | n/a | fill = 1U << curr; |
---|
233 | n/a | min = fill; /* save offset to next table */ |
---|
234 | n/a | do { |
---|
235 | n/a | fill -= incr; |
---|
236 | n/a | next[(huff >> drop) + fill] = here; |
---|
237 | n/a | } while (fill != 0); |
---|
238 | n/a | |
---|
239 | n/a | /* backwards increment the len-bit code huff */ |
---|
240 | n/a | incr = 1U << (len - 1); |
---|
241 | n/a | while (huff & incr) |
---|
242 | n/a | incr >>= 1; |
---|
243 | n/a | if (incr != 0) { |
---|
244 | n/a | huff &= incr - 1; |
---|
245 | n/a | huff += incr; |
---|
246 | n/a | } |
---|
247 | n/a | else |
---|
248 | n/a | huff = 0; |
---|
249 | n/a | |
---|
250 | n/a | /* go to next symbol, update count, len */ |
---|
251 | n/a | sym++; |
---|
252 | n/a | if (--(count[len]) == 0) { |
---|
253 | n/a | if (len == max) break; |
---|
254 | n/a | len = lens[work[sym]]; |
---|
255 | n/a | } |
---|
256 | n/a | |
---|
257 | n/a | /* create new sub-table if needed */ |
---|
258 | n/a | if (len > root && (huff & mask) != low) { |
---|
259 | n/a | /* if first time, transition to sub-tables */ |
---|
260 | n/a | if (drop == 0) |
---|
261 | n/a | drop = root; |
---|
262 | n/a | |
---|
263 | n/a | /* increment past last table */ |
---|
264 | n/a | next += min; /* here min is 1 << curr */ |
---|
265 | n/a | |
---|
266 | n/a | /* determine length of next table */ |
---|
267 | n/a | curr = len - drop; |
---|
268 | n/a | left = (int)(1 << curr); |
---|
269 | n/a | while (curr + drop < max) { |
---|
270 | n/a | left -= count[curr + drop]; |
---|
271 | n/a | if (left <= 0) break; |
---|
272 | n/a | curr++; |
---|
273 | n/a | left <<= 1; |
---|
274 | n/a | } |
---|
275 | n/a | |
---|
276 | n/a | /* check for enough space */ |
---|
277 | n/a | used += 1U << curr; |
---|
278 | n/a | if ((type == LENS && used > ENOUGH_LENS) || |
---|
279 | n/a | (type == DISTS && used > ENOUGH_DISTS)) |
---|
280 | n/a | return 1; |
---|
281 | n/a | |
---|
282 | n/a | /* point entry in root table to sub-table */ |
---|
283 | n/a | low = huff & mask; |
---|
284 | n/a | (*table)[low].op = (unsigned char)curr; |
---|
285 | n/a | (*table)[low].bits = (unsigned char)root; |
---|
286 | n/a | (*table)[low].val = (unsigned short)(next - *table); |
---|
287 | n/a | } |
---|
288 | n/a | } |
---|
289 | n/a | |
---|
290 | n/a | /* fill in remaining table entry if code is incomplete (guaranteed to have |
---|
291 | n/a | at most one remaining entry, since if the code is incomplete, the |
---|
292 | n/a | maximum code length that was allowed to get this far is one bit) */ |
---|
293 | n/a | if (huff != 0) { |
---|
294 | n/a | here.op = (unsigned char)64; /* invalid code marker */ |
---|
295 | n/a | here.bits = (unsigned char)(len - drop); |
---|
296 | n/a | here.val = (unsigned short)0; |
---|
297 | n/a | next[huff] = here; |
---|
298 | n/a | } |
---|
299 | n/a | |
---|
300 | n/a | /* set return parameters */ |
---|
301 | n/a | *table += used; |
---|
302 | n/a | *bits = root; |
---|
303 | n/a | return 0; |
---|
304 | n/a | } |
---|