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
11const char inflate_copyright[] =
12   " inflate 1.3.1 Copyright 1995-2024 Mark Adler ";
13/*
14  If you use the zlib library in a product, an acknowledgment is welcome
15  in the documentation of your product. If for some reason you cannot
16  include such an acknowledgment, I would appreciate that you keep this
17  copyright string in the executable of your product.
18 */
19
20/*
21   Build a set of tables to decode the provided canonical Huffman code.
22   The code lengths are lens[0..codes-1].  The result starts at *table,
23   whose indices are 0..2^bits-1.  work is a writable array of at least
24   lens shorts, which is used as a work area.  type is the type of code
25   to be generated, CODES, LENS, or DISTS.  On return, zero is success,
26   -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
27   on return points to the next available entry's address.  bits is the
28   requested root table index bits, and on return it is the actual root
29   table index bits.  It will differ if the request is greater than the
30   longest code or if it is less than the shortest code.
31 */
32int ZLIB_INTERNAL inflate_table(codetype type, unsigned short FAR *lens,
33                                unsigned codes, code FAR * FAR *table,
34                                unsigned FAR *bits, unsigned short FAR *work) {
35    unsigned len;               /* a code's length in bits */
36    unsigned sym;               /* index of code symbols */
37    unsigned min, max;          /* minimum and maximum code lengths */
38    unsigned root;              /* number of index bits for root table */
39    unsigned curr;              /* number of index bits for current table */
40    unsigned drop;              /* code bits to drop for sub-table */
41    int left;                   /* number of prefix codes available */
42    unsigned used;              /* code entries in table used */
43    unsigned huff;              /* Huffman code */
44    unsigned incr;              /* for incrementing code, index */
45    unsigned fill;              /* index for replicating entries */
46    unsigned low;               /* low bits for current root entry */
47    unsigned mask;              /* mask for low root bits */
48    code here;                  /* table entry for duplication */
49    code FAR *next;             /* next available space in table */
50    const unsigned short FAR *base;     /* base value table to use */
51    const unsigned short FAR *extra;    /* extra bits table to use */
52    unsigned match;             /* use base and extra for symbol >= match */
53    unsigned short count[MAXBITS+1];    /* number of codes of each length */
54    unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
55    static const unsigned short lbase[31] = { /* Length codes 257..285 base */
56        3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
57        35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
58    static const unsigned short lext[31] = { /* Length codes 257..285 extra */
59        16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
60        19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 203, 77};
61    static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
62        1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
63        257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
64        8193, 12289, 16385, 24577, 0, 0};
65    static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
66        16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
67        23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
68        28, 28, 29, 29, 64, 64};
69
70    /*
71       Process a set of code lengths to create a canonical Huffman code.  The
72       code lengths are lens[0..codes-1].  Each length corresponds to the
73       symbols 0..codes-1.  The Huffman code is generated by first sorting the
74       symbols by length from short to long, and retaining the symbol order
75       for codes with equal lengths.  Then the code starts with all zero bits
76       for the first code of the shortest length, and the codes are integer
77       increments for the same length, and zeros are appended as the length
78       increases.  For the deflate format, these bits are stored backwards
79       from their more natural integer increment ordering, and so when the
80       decoding tables are built in the large loop below, the integer codes
81       are incremented backwards.
82
83       This routine assumes, but does not check, that all of the entries in
84       lens[] are in the range 0..MAXBITS.  The caller must assure this.
85       1..MAXBITS is interpreted as that code length.  zero means that that
86       symbol does not occur in this code.
87
88       The codes are sorted by computing a count of codes for each length,
89       creating from that a table of starting indices for each length in the
90       sorted table, and then entering the symbols in order in the sorted
91       table.  The sorted table is work[], with that space being provided by
92       the caller.
93
94       The length counts are used for other purposes as well, i.e. finding
95       the minimum and maximum length codes, determining if there are any
96       codes at all, checking for a valid set of lengths, and looking ahead
97       at length counts to determine sub-table sizes when building the
98       decoding tables.
99     */
100
101    /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
102    for (len = 0; len <= MAXBITS; len++)
103        count[len] = 0;
104    for (sym = 0; sym < codes; sym++)
105        count[lens[sym]]++;
106
107    /* bound code lengths, force root to be within code lengths */
108    root = *bits;
109    for (max = MAXBITS; max >= 1; max--)
110        if (count[max] != 0) break;
111    if (root > max) root = max;
112    if (max == 0) {                     /* no symbols to code at all */
113        here.op = (unsigned char)64;    /* invalid code marker */
114        here.bits = (unsigned char)1;
115        here.val = (unsigned short)0;
116        *(*table)++ = here;             /* make a table to force an error */
117        *(*table)++ = here;
118        *bits = 1;
119        return 0;     /* no symbols, but wait for decoding to report error */
120    }
121    for (min = 1; min < max; min++)
122        if (count[min] != 0) break;
123    if (root < min) root = min;
124
125    /* check for an over-subscribed or incomplete set of lengths */
126    left = 1;
127    for (len = 1; len <= MAXBITS; len++) {
128        left <<= 1;
129        left -= count[len];
130        if (left < 0) return -1;        /* over-subscribed */
131    }
132    if (left > 0 && (type == CODES || max != 1))
133        return -1;                      /* incomplete set */
134
135    /* generate offsets into symbol table for each length for sorting */
136    offs[1] = 0;
137    for (len = 1; len < MAXBITS; len++)
138        offs[len + 1] = offs[len] + count[len];
139
140    /* sort symbols by length, by symbol order within each length */
141    for (sym = 0; sym < codes; sym++)
142        if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
143
144    /*
145       Create and fill in decoding tables.  In this loop, the table being
146       filled is at next and has curr index bits.  The code being used is huff
147       with length len.  That code is converted to an index by dropping drop
148       bits off of the bottom.  For codes where len is less than drop + curr,
149       those top drop + curr - len bits are incremented through all values to
150       fill the table with replicated entries.
151
152       root is the number of index bits for the root table.  When len exceeds
153       root, sub-tables are created pointed to by the root entry with an index
154       of the low root bits of huff.  This is saved in low to check for when a
155       new sub-table should be started.  drop is zero when the root table is
156       being filled, and drop is root when sub-tables are being filled.
157
158       When a new sub-table is needed, it is necessary to look ahead in the
159       code lengths to determine what size sub-table is needed.  The length
160       counts are used for this, and so count[] is decremented as codes are
161       entered in the tables.
162
163       used keeps track of how many table entries have been allocated from the
164       provided *table space.  It is checked for LENS and DIST tables against
165       the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
166       the initial root table size constants.  See the comments in inftrees.h
167       for more information.
168
169       sym increments through all symbols, and the loop terminates when
170       all codes of length max, i.e. all codes, have been processed.  This
171       routine permits incomplete codes, so another loop after this one fills
172       in the rest of the decoding tables with invalid code markers.
173     */
174
175    /* set up for code type */
176    switch (type) {
177    case CODES:
178        base = extra = work;    /* dummy value--not used */
179        match = 20;
180        break;
181    case LENS:
182        base = lbase;
183        extra = lext;
184        match = 257;
185        break;
186    default:    /* DISTS */
187        base = dbase;
188        extra = dext;
189        match = 0;
190    }
191
192    /* initialize state for loop */
193    huff = 0;                   /* starting code */
194    sym = 0;                    /* starting code symbol */
195    len = min;                  /* starting code length */
196    next = *table;              /* current table to fill in */
197    curr = root;                /* current table index bits */
198    drop = 0;                   /* current bits to drop from code for index */
199    low = (unsigned)(-1);       /* trigger new sub-table when len > root */
200    used = 1U << root;          /* use root table entries */
201    mask = used - 1;            /* mask for comparing low */
202
203    /* check available table space */
204    if ((type == LENS && used > ENOUGH_LENS) ||
205        (type == DISTS && used > ENOUGH_DISTS))
206        return 1;
207
208    /* process all codes and make table entries */
209    for (;;) {
210        /* create table entry */
211        here.bits = (unsigned char)(len - drop);
212        if (work[sym] + 1U < match) {
213            here.op = (unsigned char)0;
214            here.val = work[sym];
215        }
216        else if (work[sym] >= match) {
217            here.op = (unsigned char)(extra[work[sym] - match]);
218            here.val = base[work[sym] - match];
219        }
220        else {
221            here.op = (unsigned char)(32 + 64);         /* end of block */
222            here.val = 0;
223        }
224
225        /* replicate for those indices with low len bits equal to huff */
226        incr = 1U << (len - drop);
227        fill = 1U << curr;
228        min = fill;                 /* save offset to next table */
229        do {
230            fill -= incr;
231            next[(huff >> drop) + fill] = here;
232        } while (fill != 0);
233
234        /* backwards increment the len-bit code huff */
235        incr = 1U << (len - 1);
236        while (huff & incr)
237            incr >>= 1;
238        if (incr != 0) {
239            huff &= incr - 1;
240            huff += incr;
241        }
242        else
243            huff = 0;
244
245        /* go to next symbol, update count, len */
246        sym++;
247        if (--(count[len]) == 0) {
248            if (len == max) break;
249            len = lens[work[sym]];
250        }
251
252        /* create new sub-table if needed */
253        if (len > root && (huff & mask) != low) {
254            /* if first time, transition to sub-tables */
255            if (drop == 0)
256                drop = root;
257
258            /* increment past last table */
259            next += min;            /* here min is 1 << curr */
260
261            /* determine length of next table */
262            curr = len - drop;
263            left = (int)(1 << curr);
264            while (curr + drop < max) {
265                left -= count[curr + drop];
266                if (left <= 0) break;
267                curr++;
268                left <<= 1;
269            }
270
271            /* check for enough space */
272            used += 1U << curr;
273            if ((type == LENS && used > ENOUGH_LENS) ||
274                (type == DISTS && used > ENOUGH_DISTS))
275                return 1;
276
277            /* point entry in root table to sub-table */
278            low = huff & mask;
279            (*table)[low].op = (unsigned char)curr;
280            (*table)[low].bits = (unsigned char)root;
281            (*table)[low].val = (unsigned short)(next - *table);
282        }
283    }
284
285    /* fill in remaining table entry if code is incomplete (guaranteed to have
286       at most one remaining entry, since if the code is incomplete, the
287       maximum code length that was allowed to get this far is one bit) */
288    if (huff != 0) {
289        here.op = (unsigned char)64;            /* invalid code marker */
290        here.bits = (unsigned char)(len - drop);
291        here.val = (unsigned short)0;
292        next[huff] = here;
293    }
294
295    /* set return parameters */
296    *table += used;
297    *bits = root;
298    return 0;
299}
300