Thu, 22 Jan 2015 13:21:57 +0100
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
michael@0 | 1 | /* inftrees.c -- generate Huffman trees for efficient decoding |
michael@0 | 2 | * Copyright (C) 1995-2010 Mark Adler |
michael@0 | 3 | * For conditions of distribution and use, see copyright notice in zlib.h |
michael@0 | 4 | */ |
michael@0 | 5 | |
michael@0 | 6 | #include "zutil.h" |
michael@0 | 7 | #include "inftrees.h" |
michael@0 | 8 | |
michael@0 | 9 | #define MAXBITS 15 |
michael@0 | 10 | |
michael@0 | 11 | const char inflate_copyright[] = |
michael@0 | 12 | " inflate 1.2.5 Copyright 1995-2010 Mark Adler "; |
michael@0 | 13 | /* |
michael@0 | 14 | If you use the zlib library in a product, an acknowledgment is welcome |
michael@0 | 15 | in the documentation of your product. If for some reason you cannot |
michael@0 | 16 | include such an acknowledgment, I would appreciate that you keep this |
michael@0 | 17 | copyright string in the executable of your product. |
michael@0 | 18 | */ |
michael@0 | 19 | |
michael@0 | 20 | /* |
michael@0 | 21 | Build a set of tables to decode the provided canonical Huffman code. |
michael@0 | 22 | The code lengths are lens[0..codes-1]. The result starts at *table, |
michael@0 | 23 | whose indices are 0..2^bits-1. work is a writable array of at least |
michael@0 | 24 | lens shorts, which is used as a work area. type is the type of code |
michael@0 | 25 | to be generated, CODES, LENS, or DISTS. On return, zero is success, |
michael@0 | 26 | -1 is an invalid code, and +1 means that ENOUGH isn't enough. table |
michael@0 | 27 | on return points to the next available entry's address. bits is the |
michael@0 | 28 | requested root table index bits, and on return it is the actual root |
michael@0 | 29 | table index bits. It will differ if the request is greater than the |
michael@0 | 30 | longest code or if it is less than the shortest code. |
michael@0 | 31 | */ |
michael@0 | 32 | int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work) |
michael@0 | 33 | codetype type; |
michael@0 | 34 | unsigned short FAR *lens; |
michael@0 | 35 | unsigned codes; |
michael@0 | 36 | code FAR * FAR *table; |
michael@0 | 37 | unsigned FAR *bits; |
michael@0 | 38 | unsigned short FAR *work; |
michael@0 | 39 | { |
michael@0 | 40 | unsigned len; /* a code's length in bits */ |
michael@0 | 41 | unsigned sym; /* index of code symbols */ |
michael@0 | 42 | unsigned min, max; /* minimum and maximum code lengths */ |
michael@0 | 43 | unsigned root; /* number of index bits for root table */ |
michael@0 | 44 | unsigned curr; /* number of index bits for current table */ |
michael@0 | 45 | unsigned drop; /* code bits to drop for sub-table */ |
michael@0 | 46 | int left; /* number of prefix codes available */ |
michael@0 | 47 | unsigned used; /* code entries in table used */ |
michael@0 | 48 | unsigned huff; /* Huffman code */ |
michael@0 | 49 | unsigned incr; /* for incrementing code, index */ |
michael@0 | 50 | unsigned fill; /* index for replicating entries */ |
michael@0 | 51 | unsigned low; /* low bits for current root entry */ |
michael@0 | 52 | unsigned mask; /* mask for low root bits */ |
michael@0 | 53 | code here; /* table entry for duplication */ |
michael@0 | 54 | code FAR *next; /* next available space in table */ |
michael@0 | 55 | const unsigned short FAR *base; /* base value table to use */ |
michael@0 | 56 | const unsigned short FAR *extra; /* extra bits table to use */ |
michael@0 | 57 | int end; /* use base and extra for symbol > end */ |
michael@0 | 58 | unsigned short count[MAXBITS+1]; /* number of codes of each length */ |
michael@0 | 59 | unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ |
michael@0 | 60 | static const unsigned short lbase[31] = { /* Length codes 257..285 base */ |
michael@0 | 61 | 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, |
michael@0 | 62 | 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; |
michael@0 | 63 | static const unsigned short lext[31] = { /* Length codes 257..285 extra */ |
michael@0 | 64 | 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, |
michael@0 | 65 | 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 73, 195}; |
michael@0 | 66 | static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ |
michael@0 | 67 | 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, |
michael@0 | 68 | 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, |
michael@0 | 69 | 8193, 12289, 16385, 24577, 0, 0}; |
michael@0 | 70 | static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ |
michael@0 | 71 | 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, |
michael@0 | 72 | 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, |
michael@0 | 73 | 28, 28, 29, 29, 64, 64}; |
michael@0 | 74 | |
michael@0 | 75 | /* |
michael@0 | 76 | Process a set of code lengths to create a canonical Huffman code. The |
michael@0 | 77 | code lengths are lens[0..codes-1]. Each length corresponds to the |
michael@0 | 78 | symbols 0..codes-1. The Huffman code is generated by first sorting the |
michael@0 | 79 | symbols by length from short to long, and retaining the symbol order |
michael@0 | 80 | for codes with equal lengths. Then the code starts with all zero bits |
michael@0 | 81 | for the first code of the shortest length, and the codes are integer |
michael@0 | 82 | increments for the same length, and zeros are appended as the length |
michael@0 | 83 | increases. For the deflate format, these bits are stored backwards |
michael@0 | 84 | from their more natural integer increment ordering, and so when the |
michael@0 | 85 | decoding tables are built in the large loop below, the integer codes |
michael@0 | 86 | are incremented backwards. |
michael@0 | 87 | |
michael@0 | 88 | This routine assumes, but does not check, that all of the entries in |
michael@0 | 89 | lens[] are in the range 0..MAXBITS. The caller must assure this. |
michael@0 | 90 | 1..MAXBITS is interpreted as that code length. zero means that that |
michael@0 | 91 | symbol does not occur in this code. |
michael@0 | 92 | |
michael@0 | 93 | The codes are sorted by computing a count of codes for each length, |
michael@0 | 94 | creating from that a table of starting indices for each length in the |
michael@0 | 95 | sorted table, and then entering the symbols in order in the sorted |
michael@0 | 96 | table. The sorted table is work[], with that space being provided by |
michael@0 | 97 | the caller. |
michael@0 | 98 | |
michael@0 | 99 | The length counts are used for other purposes as well, i.e. finding |
michael@0 | 100 | the minimum and maximum length codes, determining if there are any |
michael@0 | 101 | codes at all, checking for a valid set of lengths, and looking ahead |
michael@0 | 102 | at length counts to determine sub-table sizes when building the |
michael@0 | 103 | decoding tables. |
michael@0 | 104 | */ |
michael@0 | 105 | |
michael@0 | 106 | /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ |
michael@0 | 107 | for (len = 0; len <= MAXBITS; len++) |
michael@0 | 108 | count[len] = 0; |
michael@0 | 109 | for (sym = 0; sym < codes; sym++) |
michael@0 | 110 | count[lens[sym]]++; |
michael@0 | 111 | |
michael@0 | 112 | /* bound code lengths, force root to be within code lengths */ |
michael@0 | 113 | root = *bits; |
michael@0 | 114 | for (max = MAXBITS; max >= 1; max--) |
michael@0 | 115 | if (count[max] != 0) break; |
michael@0 | 116 | if (root > max) root = max; |
michael@0 | 117 | if (max == 0) { /* no symbols to code at all */ |
michael@0 | 118 | here.op = (unsigned char)64; /* invalid code marker */ |
michael@0 | 119 | here.bits = (unsigned char)1; |
michael@0 | 120 | here.val = (unsigned short)0; |
michael@0 | 121 | *(*table)++ = here; /* make a table to force an error */ |
michael@0 | 122 | *(*table)++ = here; |
michael@0 | 123 | *bits = 1; |
michael@0 | 124 | return 0; /* no symbols, but wait for decoding to report error */ |
michael@0 | 125 | } |
michael@0 | 126 | for (min = 1; min < max; min++) |
michael@0 | 127 | if (count[min] != 0) break; |
michael@0 | 128 | if (root < min) root = min; |
michael@0 | 129 | |
michael@0 | 130 | /* check for an over-subscribed or incomplete set of lengths */ |
michael@0 | 131 | left = 1; |
michael@0 | 132 | for (len = 1; len <= MAXBITS; len++) { |
michael@0 | 133 | left <<= 1; |
michael@0 | 134 | left -= count[len]; |
michael@0 | 135 | if (left < 0) return -1; /* over-subscribed */ |
michael@0 | 136 | } |
michael@0 | 137 | if (left > 0 && (type == CODES || max != 1)) |
michael@0 | 138 | return -1; /* incomplete set */ |
michael@0 | 139 | |
michael@0 | 140 | /* generate offsets into symbol table for each length for sorting */ |
michael@0 | 141 | offs[1] = 0; |
michael@0 | 142 | for (len = 1; len < MAXBITS; len++) |
michael@0 | 143 | offs[len + 1] = offs[len] + count[len]; |
michael@0 | 144 | |
michael@0 | 145 | /* sort symbols by length, by symbol order within each length */ |
michael@0 | 146 | for (sym = 0; sym < codes; sym++) |
michael@0 | 147 | if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; |
michael@0 | 148 | |
michael@0 | 149 | /* |
michael@0 | 150 | Create and fill in decoding tables. In this loop, the table being |
michael@0 | 151 | filled is at next and has curr index bits. The code being used is huff |
michael@0 | 152 | with length len. That code is converted to an index by dropping drop |
michael@0 | 153 | bits off of the bottom. For codes where len is less than drop + curr, |
michael@0 | 154 | those top drop + curr - len bits are incremented through all values to |
michael@0 | 155 | fill the table with replicated entries. |
michael@0 | 156 | |
michael@0 | 157 | root is the number of index bits for the root table. When len exceeds |
michael@0 | 158 | root, sub-tables are created pointed to by the root entry with an index |
michael@0 | 159 | of the low root bits of huff. This is saved in low to check for when a |
michael@0 | 160 | new sub-table should be started. drop is zero when the root table is |
michael@0 | 161 | being filled, and drop is root when sub-tables are being filled. |
michael@0 | 162 | |
michael@0 | 163 | When a new sub-table is needed, it is necessary to look ahead in the |
michael@0 | 164 | code lengths to determine what size sub-table is needed. The length |
michael@0 | 165 | counts are used for this, and so count[] is decremented as codes are |
michael@0 | 166 | entered in the tables. |
michael@0 | 167 | |
michael@0 | 168 | used keeps track of how many table entries have been allocated from the |
michael@0 | 169 | provided *table space. It is checked for LENS and DIST tables against |
michael@0 | 170 | the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in |
michael@0 | 171 | the initial root table size constants. See the comments in inftrees.h |
michael@0 | 172 | for more information. |
michael@0 | 173 | |
michael@0 | 174 | sym increments through all symbols, and the loop terminates when |
michael@0 | 175 | all codes of length max, i.e. all codes, have been processed. This |
michael@0 | 176 | routine permits incomplete codes, so another loop after this one fills |
michael@0 | 177 | in the rest of the decoding tables with invalid code markers. |
michael@0 | 178 | */ |
michael@0 | 179 | |
michael@0 | 180 | /* set up for code type */ |
michael@0 | 181 | switch (type) { |
michael@0 | 182 | case CODES: |
michael@0 | 183 | base = extra = work; /* dummy value--not used */ |
michael@0 | 184 | end = 19; |
michael@0 | 185 | break; |
michael@0 | 186 | case LENS: |
michael@0 | 187 | base = lbase; |
michael@0 | 188 | base -= 257; |
michael@0 | 189 | extra = lext; |
michael@0 | 190 | extra -= 257; |
michael@0 | 191 | end = 256; |
michael@0 | 192 | break; |
michael@0 | 193 | default: /* DISTS */ |
michael@0 | 194 | base = dbase; |
michael@0 | 195 | extra = dext; |
michael@0 | 196 | end = -1; |
michael@0 | 197 | } |
michael@0 | 198 | |
michael@0 | 199 | /* initialize state for loop */ |
michael@0 | 200 | huff = 0; /* starting code */ |
michael@0 | 201 | sym = 0; /* starting code symbol */ |
michael@0 | 202 | len = min; /* starting code length */ |
michael@0 | 203 | next = *table; /* current table to fill in */ |
michael@0 | 204 | curr = root; /* current table index bits */ |
michael@0 | 205 | drop = 0; /* current bits to drop from code for index */ |
michael@0 | 206 | low = (unsigned)(-1); /* trigger new sub-table when len > root */ |
michael@0 | 207 | used = 1U << root; /* use root table entries */ |
michael@0 | 208 | mask = used - 1; /* mask for comparing low */ |
michael@0 | 209 | |
michael@0 | 210 | /* check available table space */ |
michael@0 | 211 | if ((type == LENS && used >= ENOUGH_LENS) || |
michael@0 | 212 | (type == DISTS && used >= ENOUGH_DISTS)) |
michael@0 | 213 | return 1; |
michael@0 | 214 | |
michael@0 | 215 | /* process all codes and make table entries */ |
michael@0 | 216 | for (;;) { |
michael@0 | 217 | /* create table entry */ |
michael@0 | 218 | here.bits = (unsigned char)(len - drop); |
michael@0 | 219 | if ((int)(work[sym]) < end) { |
michael@0 | 220 | here.op = (unsigned char)0; |
michael@0 | 221 | here.val = work[sym]; |
michael@0 | 222 | } |
michael@0 | 223 | else if ((int)(work[sym]) > end) { |
michael@0 | 224 | here.op = (unsigned char)(extra[work[sym]]); |
michael@0 | 225 | here.val = base[work[sym]]; |
michael@0 | 226 | } |
michael@0 | 227 | else { |
michael@0 | 228 | here.op = (unsigned char)(32 + 64); /* end of block */ |
michael@0 | 229 | here.val = 0; |
michael@0 | 230 | } |
michael@0 | 231 | |
michael@0 | 232 | /* replicate for those indices with low len bits equal to huff */ |
michael@0 | 233 | incr = 1U << (len - drop); |
michael@0 | 234 | fill = 1U << curr; |
michael@0 | 235 | min = fill; /* save offset to next table */ |
michael@0 | 236 | do { |
michael@0 | 237 | fill -= incr; |
michael@0 | 238 | next[(huff >> drop) + fill] = here; |
michael@0 | 239 | } while (fill != 0); |
michael@0 | 240 | |
michael@0 | 241 | /* backwards increment the len-bit code huff */ |
michael@0 | 242 | incr = 1U << (len - 1); |
michael@0 | 243 | while (huff & incr) |
michael@0 | 244 | incr >>= 1; |
michael@0 | 245 | if (incr != 0) { |
michael@0 | 246 | huff &= incr - 1; |
michael@0 | 247 | huff += incr; |
michael@0 | 248 | } |
michael@0 | 249 | else |
michael@0 | 250 | huff = 0; |
michael@0 | 251 | |
michael@0 | 252 | /* go to next symbol, update count, len */ |
michael@0 | 253 | sym++; |
michael@0 | 254 | if (--(count[len]) == 0) { |
michael@0 | 255 | if (len == max) break; |
michael@0 | 256 | len = lens[work[sym]]; |
michael@0 | 257 | } |
michael@0 | 258 | |
michael@0 | 259 | /* create new sub-table if needed */ |
michael@0 | 260 | if (len > root && (huff & mask) != low) { |
michael@0 | 261 | /* if first time, transition to sub-tables */ |
michael@0 | 262 | if (drop == 0) |
michael@0 | 263 | drop = root; |
michael@0 | 264 | |
michael@0 | 265 | /* increment past last table */ |
michael@0 | 266 | next += min; /* here min is 1 << curr */ |
michael@0 | 267 | |
michael@0 | 268 | /* determine length of next table */ |
michael@0 | 269 | curr = len - drop; |
michael@0 | 270 | left = (int)(1 << curr); |
michael@0 | 271 | while (curr + drop < max) { |
michael@0 | 272 | left -= count[curr + drop]; |
michael@0 | 273 | if (left <= 0) break; |
michael@0 | 274 | curr++; |
michael@0 | 275 | left <<= 1; |
michael@0 | 276 | } |
michael@0 | 277 | |
michael@0 | 278 | /* check for enough space */ |
michael@0 | 279 | used += 1U << curr; |
michael@0 | 280 | if ((type == LENS && used >= ENOUGH_LENS) || |
michael@0 | 281 | (type == DISTS && used >= ENOUGH_DISTS)) |
michael@0 | 282 | return 1; |
michael@0 | 283 | |
michael@0 | 284 | /* point entry in root table to sub-table */ |
michael@0 | 285 | low = huff & mask; |
michael@0 | 286 | (*table)[low].op = (unsigned char)curr; |
michael@0 | 287 | (*table)[low].bits = (unsigned char)root; |
michael@0 | 288 | (*table)[low].val = (unsigned short)(next - *table); |
michael@0 | 289 | } |
michael@0 | 290 | } |
michael@0 | 291 | |
michael@0 | 292 | /* |
michael@0 | 293 | Fill in rest of table for incomplete codes. This loop is similar to the |
michael@0 | 294 | loop above in incrementing huff for table indices. It is assumed that |
michael@0 | 295 | len is equal to curr + drop, so there is no loop needed to increment |
michael@0 | 296 | through high index bits. When the current sub-table is filled, the loop |
michael@0 | 297 | drops back to the root table to fill in any remaining entries there. |
michael@0 | 298 | */ |
michael@0 | 299 | here.op = (unsigned char)64; /* invalid code marker */ |
michael@0 | 300 | here.bits = (unsigned char)(len - drop); |
michael@0 | 301 | here.val = (unsigned short)0; |
michael@0 | 302 | while (huff != 0) { |
michael@0 | 303 | /* when done with sub-table, drop back to root table */ |
michael@0 | 304 | if (drop != 0 && (huff & mask) != low) { |
michael@0 | 305 | drop = 0; |
michael@0 | 306 | len = root; |
michael@0 | 307 | next = *table; |
michael@0 | 308 | here.bits = (unsigned char)len; |
michael@0 | 309 | } |
michael@0 | 310 | |
michael@0 | 311 | /* put invalid code marker in table */ |
michael@0 | 312 | next[huff >> drop] = here; |
michael@0 | 313 | |
michael@0 | 314 | /* backwards increment the len-bit code huff */ |
michael@0 | 315 | incr = 1U << (len - 1); |
michael@0 | 316 | while (huff & incr) |
michael@0 | 317 | incr >>= 1; |
michael@0 | 318 | if (incr != 0) { |
michael@0 | 319 | huff &= incr - 1; |
michael@0 | 320 | huff += incr; |
michael@0 | 321 | } |
michael@0 | 322 | else |
michael@0 | 323 | huff = 0; |
michael@0 | 324 | } |
michael@0 | 325 | |
michael@0 | 326 | /* set return parameters */ |
michael@0 | 327 | *table += used; |
michael@0 | 328 | *bits = root; |
michael@0 | 329 | return 0; |
michael@0 | 330 | } |