xref: /netbsd/common/dist/zlib/inftrees.c (revision d9a827c3)
1 /*	$NetBSD: inftrees.c,v 1.4 2022/10/15 19:49:32 christos Exp $	*/
2 
3 /* inftrees.c -- generate Huffman trees for efficient decoding
4  * Copyright (C) 1995-2022 Mark Adler
5  * For conditions of distribution and use, see copyright notice in zlib.h
6  */
7 
8 #include "zutil.h"
9 #include "inftrees.h"
10 
11 #define MAXBITS 15
12 
13 const char inflate_copyright[] =
14    " inflate 1.2.13 Copyright 1995-2022 Mark Adler ";
15 /*
16   If you use the zlib library in a product, an acknowledgment is welcome
17   in the documentation of your product. If for some reason you cannot
18   include such an acknowledgment, I would appreciate that you keep this
19   copyright string in the executable of your product.
20  */
21 
22 /*
23    Build a set of tables to decode the provided canonical Huffman code.
24    The code lengths are lens[0..codes-1].  The result starts at *table,
25    whose indices are 0..2^bits-1.  work is a writable array of at least
26    lens shorts, which is used as a work area.  type is the type of code
27    to be generated, CODES, LENS, or DISTS.  On return, zero is success,
28    -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
29    on return points to the next available entry's address.  bits is the
30    requested root table index bits, and on return it is the actual root
31    table index bits.  It will differ if the request is greater than the
32    longest code or if it is less than the shortest code.
33  */
inflate_table(type,lens,codes,table,bits,work)34 int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work)
35 codetype type;
36 unsigned short FAR *lens;
37 unsigned codes;
38 code FAR * FAR *table;
39 unsigned FAR *bits;
40 unsigned short FAR *work;
41 {
42     unsigned len;               /* a code's length in bits */
43     unsigned sym;               /* index of code symbols */
44     unsigned mmin, mmax;        /* minimum and maximum code lengths */
45     unsigned root;              /* number of index bits for root table */
46     unsigned curr;              /* number of index bits for current table */
47     unsigned drop;              /* code bits to drop for sub-table */
48     int left;                   /* number of prefix codes available */
49     unsigned used;              /* code entries in table used */
50     unsigned huff;              /* Huffman code */
51     unsigned incr;              /* for incrementing code, index */
52     unsigned fill;              /* index for replicating entries */
53     unsigned low;               /* low bits for current root entry */
54     unsigned mask;              /* mask for low root bits */
55     code here;                  /* table entry for duplication */
56     code FAR *next;             /* next available space in table */
57     const unsigned short FAR *base;     /* base value table to use */
58     const unsigned short FAR *extra;    /* extra bits table to use */
59     unsigned match;             /* use base and extra for symbol >= match */
60     unsigned short count[MAXBITS+1];    /* number of codes of each length */
61     unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
62     static const unsigned short lbase[31] = { /* Length codes 257..285 base */
63         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
64         35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
65     static const unsigned short lext[31] = { /* Length codes 257..285 extra */
66         16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
67         19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 194, 65};
68     static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
69         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
70         257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
71         8193, 12289, 16385, 24577, 0, 0};
72     static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
73         16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
74         23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
75         28, 28, 29, 29, 64, 64};
76 
77     /*
78        Process a set of code lengths to create a canonical Huffman code.  The
79        code lengths are lens[0..codes-1].  Each length corresponds to the
80        symbols 0..codes-1.  The Huffman code is generated by first sorting the
81        symbols by length from short to long, and retaining the symbol order
82        for codes with equal lengths.  Then the code starts with all zero bits
83        for the first code of the shortest length, and the codes are integer
84        increments for the same length, and zeros are appended as the length
85        increases.  For the deflate format, these bits are stored backwards
86        from their more natural integer increment ordering, and so when the
87        decoding tables are built in the large loop below, the integer codes
88        are incremented backwards.
89 
90        This routine assumes, but does not check, that all of the entries in
91        lens[] are in the range 0..MAXBITS.  The caller must assure this.
92        1..MAXBITS is interpreted as that code length.  zero means that that
93        symbol does not occur in this code.
94 
95        The codes are sorted by computing a count of codes for each length,
96        creating from that a table of starting indices for each length in the
97        sorted table, and then entering the symbols in order in the sorted
98        table.  The sorted table is work[], with that space being provided by
99        the caller.
100 
101        The length counts are used for other purposes as well, i.e. finding
102        the minimum and maximum length codes, determining if there are any
103        codes at all, checking for a valid set of lengths, and looking ahead
104        at length counts to determine sub-table sizes when building the
105        decoding tables.
106      */
107 
108     /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
109     for (len = 0; len <= MAXBITS; len++)
110         count[len] = 0;
111     for (sym = 0; sym < codes; sym++)
112         count[lens[sym]]++;
113 
114     /* bound code lengths, force root to be within code lengths */
115     root = *bits;
116     for (mmax = MAXBITS; mmax >= 1; mmax--)
117         if (count[mmax] != 0) break;
118     if (root > mmax) root = mmax;
119     if (mmax == 0) {                     /* no symbols to code at all */
120         here.op = (unsigned char)64;    /* invalid code marker */
121         here.bits = (unsigned char)1;
122         here.val = (unsigned short)0;
123         *(*table)++ = here;             /* make a table to force an error */
124         *(*table)++ = here;
125         *bits = 1;
126         return 0;     /* no symbols, but wait for decoding to report error */
127     }
128     for (mmin = 1; mmin <= MAXBITS; mmin++)
129         if (count[mmin] != 0) break;
130     if (root < mmin) root = mmin;
131 
132     /* check for an over-subscribed or incomplete set of lengths */
133     left = 1;
134     for (len = 1; len <= MAXBITS; len++) {
135         left <<= 1;
136         left -= count[len];
137         if (left < 0) return -1;        /* over-subscribed */
138     }
139     if (left > 0 && (type == CODES || mmax != 1))
140         return -1;                      /* incomplete set */
141 
142     /* generate offsets into symbol table for each length for sorting */
143     offs[1] = 0;
144     for (len = 1; len < MAXBITS; len++)
145         offs[len + 1] = offs[len] + count[len];
146 
147     /* sort symbols by length, by symbol order within each length */
148     for (sym = 0; sym < codes; sym++)
149         if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
150 
151     /*
152        Create and fill in decoding tables.  In this loop, the table being
153        filled is at next and has curr index bits.  The code being used is huff
154        with length len.  That code is converted to an index by dropping drop
155        bits off of the bottom.  For codes where len is less than drop + curr,
156        those top drop + curr - len bits are incremented through all values to
157        fill the table with replicated entries.
158 
159        root is the number of index bits for the root table.  When len exceeds
160        root, sub-tables are created pointed to by the root entry with an index
161        of the low root bits of huff.  This is saved in low to check for when a
162        new sub-table should be started.  drop is zero when the root table is
163        being filled, and drop is root when sub-tables are being filled.
164 
165        When a new sub-table is needed, it is necessary to look ahead in the
166        code lengths to determine what size sub-table is needed.  The length
167        counts are used for this, and so count[] is decremented as codes are
168        entered in the tables.
169 
170        used keeps track of how many table entries have been allocated from the
171        provided *table space.  It is checked for LENS and DIST tables against
172        the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
173        the initial root table size constants.  See the comments in inftrees.h
174        for more information.
175 
176        sym increments through all symbols, and the loop terminates when
177        all codes of length mmax, i.e. all codes, have been processed.  This
178        routine permits incomplete codes, so another loop after this one fills
179        in the rest of the decoding tables with invalid code markers.
180      */
181 
182     /* set up for code type */
183     switch (type) {
184     case CODES:
185         base = extra = work;    /* dummy value--not used */
186         match = 20;
187         break;
188     case LENS:
189         base = lbase;
190         extra = lext;
191         match = 257;
192         break;
193     default:    /* DISTS */
194         base = dbase;
195         extra = dext;
196         match = 0;
197     }
198 
199     /* initialize state for loop */
200     huff = 0;                   /* starting code */
201     sym = 0;                    /* starting code symbol */
202     len = mmin;                  /* starting code length */
203     next = *table;              /* current table to fill in */
204     curr = root;                /* current table index bits */
205     drop = 0;                   /* current bits to drop from code for index */
206     low = (unsigned)(-1);       /* trigger new sub-table when len > root */
207     used = 1U << root;          /* use root table entries */
208     mask = used - 1;            /* mask for comparing low */
209 
210     /* check available table space */
211     if ((type == LENS && used > ENOUGH_LENS) ||
212         (type == DISTS && used > ENOUGH_DISTS))
213         return 1;
214 
215     /* process all codes and make table entries */
216     for (;;) {
217         /* create table entry */
218         here.bits = (unsigned char)(len - drop);
219         if (work[sym] + 1U < match) {
220             here.op = (unsigned char)0;
221             here.val = work[sym];
222         }
223         else if (work[sym] >= match) {
224             here.op = (unsigned char)(extra[work[sym] - match]);
225             here.val = base[work[sym] - match];
226         }
227         else {
228             here.op = (unsigned char)(32 + 64);         /* end of block */
229             here.val = 0;
230         }
231 
232         /* replicate for those indices with low len bits equal to huff */
233         incr = 1U << (len - drop);
234         fill = 1U << curr;
235         mmin = fill;                 /* save offset to next table */
236         do {
237             fill -= incr;
238             next[(huff >> drop) + fill] = here;
239         } while (fill != 0);
240 
241         /* backwards increment the len-bit code huff */
242         incr = 1U << (len - 1);
243         while (huff & incr)
244             incr >>= 1;
245         if (incr != 0) {
246             huff &= incr - 1;
247             huff += incr;
248         }
249         else
250             huff = 0;
251 
252         /* go to next symbol, update count, len */
253         sym++;
254         if (--(count[len]) == 0) {
255             if (len == mmax) break;
256             len = lens[work[sym]];
257         }
258 
259         /* create new sub-table if needed */
260         if (len > root && (huff & mask) != low) {
261             /* if first time, transition to sub-tables */
262             if (drop == 0)
263                 drop = root;
264 
265             /* increment past last table */
266             next += mmin;            /* here mmin is 1 << curr */
267 
268             /* determine length of next table */
269             curr = len - drop;
270             left = (int)(1 << curr);
271             while (curr + drop < mmax) {
272                 left -= count[curr + drop];
273                 if (left <= 0) break;
274                 curr++;
275                 left <<= 1;
276             }
277 
278             /* check for enough space */
279             used += 1U << curr;
280             if ((type == LENS && used > ENOUGH_LENS) ||
281                 (type == DISTS && used > ENOUGH_DISTS))
282                 return 1;
283 
284             /* point entry in root table to sub-table */
285             low = huff & mask;
286             (*table)[low].op = (unsigned char)curr;
287             (*table)[low].bits = (unsigned char)root;
288             (*table)[low].val = (unsigned short)(next - *table);
289         }
290     }
291 
292     /* fill in remaining table entry if code is incomplete (guaranteed to have
293        at most one remaining entry, since if the code is incomplete, the
294        maximum code length that was allowed to get this far is one bit) */
295     if (huff != 0) {
296         here.op = (unsigned char)64;            /* invalid code marker */
297         here.bits = (unsigned char)(len - drop);
298         here.val = (unsigned short)0;
299         next[huff] = here;
300     }
301 
302     /* set return parameters */
303     *table += used;
304     *bits = root;
305     return 0;
306 }
307