/*

 * puff.c

 * Copyright (C) 2002-2004 Mark Adler

 * For conditions of distribution and use, see copyright notice in puff.h

 * version 1.8, 9 Jan 2004

 *

 * puff.c is a simple inflate written to be an unambiguous way to specify the

 * deflate format.  It is not written for speed but rather simplicity.  As a

 * side benefit, this code might actually be useful when small code is more

 * important than speed, such as bootstrap applications.  For typical deflate

 * data, zlib's inflate() is about four times as fast as puff().  zlib's

 * inflate compiles to around 20K on my machine, whereas puff.c compiles to

 * around 4K on my machine (a PowerPC using GNU cc).  If the faster decode()

 * function here is used, then puff() is only twice as slow as zlib's

 * inflate().

 *

 * All dynamically allocated memory comes from the stack.  The stack required

 * is less than 2K bytes.  This code is compatible with 16-bit int's and

 * assumes that long's are at least 32 bits.  puff.c uses the short data type,

 * assumed to be 16 bits, for arrays in order to to conserve memory.  The code

 * works whether integers are stored big endian or little endian.

 *

 * In the comments below are "Format notes" that describe the inflate process

 * and document some of the less obvious aspects of the format.  This source

 * code is meant to supplement RFC 1951, which formally describes the deflate

 * format:

 *

 *    http://www.zlib.org/rfc-deflate.html

 */

/*

 * Change history:

 *

 * 1.0  10 Feb 2002     - First version

 * 1.1  17 Feb 2002     - Clarifications of some comments and notes

 *                      - Update puff() dest and source pointers on negative

 *                        errors to facilitate debugging deflators

 *                      - Remove longest from struct huffman -- not needed

 *                      - Simplify offs[] index in construct()

 *                      - Add input size and checking, using longjmp() to

 *                        maintain easy readability

 *                      - Use short data type for large arrays

 *                      - Use pointers instead of long to specify source and

 *                        destination sizes to avoid arbitrary 4 GB limits

 * 1.2  17 Mar 2002     - Add faster version of decode(), doubles speed (!),

 *                        but leave simple version for readabilty

 *                      - Make sure invalid distances detected if pointers

 *                        are 16 bits

 *                      - Fix fixed codes table error

 *                      - Provide a scanning mode for determining size of

 *                        uncompressed data

 * 1.3  20 Mar 2002     - Go back to lengths for puff() parameters [Jean-loup]

 *                      - Add a puff.h file for the interface

 *                      - Add braces in puff() for else do [Jean-loup]

 *                      - Use indexes instead of pointers for readability

 * 1.4  31 Mar 2002     - Simplify construct() code set check

 *                      - Fix some comments

 *                      - Add FIXLCODES #define

 * 1.5   6 Apr 2002     - Minor comment fixes

 * 1.6   7 Aug 2002     - Minor format changes

 * 1.7   3 Mar 2003     - Added test code for distribution

 *                      - Added zlib-like license

 * 1.8   9 Jan 2004     - Added some comments on no distance codes case

 */

#include <setjmp.h>             /* for setjmp(), longjmp(), and jmp_buf */

#include "puff.h"               /* prototype for puff() */

#define local static            /* for local function definitions */

#define NIL ((unsigned char *)0)        /* for no output option */

/*

 * Maximums for allocations and loops.  It is not useful to change these --

 * they are fixed by the deflate format.

 */

#define MAXBITS 15              /* maximum bits in a code */

#define MAXLCODES 286           /* maximum number of literal/length codes */

#define MAXDCODES 30            /* maximum number of distance codes */

#define MAXCODES (MAXLCODES+MAXDCODES)  /* maximum codes lengths to read */

#define FIXLCODES 288           /* number of fixed literal/length codes */

/* input and output state */

struct state {

    /* output state */

    unsigned char *out;         /* output buffer */

    unsigned long outlen;       /* available space at out */

    unsigned long outcnt;       /* bytes written to out so far */

    /* input state */

    unsigned char *in;          /* input buffer */

    unsigned long inlen;        /* available input at in */

    unsigned long incnt;        /* bytes read so far */

    int bitbuf;                 /* bit buffer */

    int bitcnt;                 /* number of bits in bit buffer */

    /* input limit error return state for bits() and decode() */

    jmp_buf env;

};

/*

 * Return need bits from the input stream.  This always leaves less than

 * eight bits in the buffer.  bits() works properly for need == 0.

 *

 * Format notes:

 *

 * - Bits are stored in bytes from the least significant bit to the most

 *   significant bit.  Therefore bits are dropped from the bottom of the bit

 *   buffer, using shift right, and new bytes are appended to the top of the

 *   bit buffer, using shift left.

 */

local int bits(struct state *s, int need)

{

    long val;           /* bit accumulator (can use up to 20 bits) */

    /* load at least need bits into val */

    val = s->bitbuf;

    while (s->bitcnt < need) {

        if (s->incnt == s->inlen) longjmp(s->env, 1);   /* out of input */

        val |= (long)(s->in[s->incnt++]) << s->bitcnt;  /* load eight bits */

        s->bitcnt += 8;

    }

    /* drop need bits and update buffer, always zero to seven bits left */

    s->bitbuf = (int)(val >> need);

    s->bitcnt -= need;

    /* return need bits, zeroing the bits above that */

    return (int)(val & ((1L << need) - 1));

}

/*

 * Process a stored block.

 *

 * Format notes:

 *

 * - After the two-bit stored block type (00), the stored block length and

 *   stored bytes are byte-aligned for fast copying.  Therefore any leftover

 *   bits in the byte that has the last bit of the type, as many as seven, are

 *   discarded.  The value of the discarded bits are not defined and should not

 *   be checked against any expectation.

 *

 * - The second inverted copy of the stored block length does not have to be

 *   checked, but it's probably a good idea to do so anyway.

 *

 * - A stored block can have zero length.  This is sometimes used to byte-align

 *   subsets of the compressed data for random access or partial recovery.

 */

local int stored(struct state *s)

{

    unsigned len;       /* length of stored block */

    /* discard leftover bits from current byte (assumes s->bitcnt < 8) */

    s->bitbuf = 0;

    s->bitcnt = 0;

    /* get length and check against its one's complement */

    if (s->incnt + 4 > s->inlen) return 2;      /* not enough input */

    len = s->in[s->incnt++];

    len |= s->in[s->incnt++] << 8;

    if (s->in[s->incnt++] != (~len & 0xff) ||

        s->in[s->incnt++] != ((~len >> 8) & 0xff))

        return -2;                              /* didn't match complement! */

    /* copy len bytes from in to out */

    if (s->incnt + len > s->inlen) return 2;    /* not enough input */

    if (s->out != NIL) {

        if (s->outcnt + len > s->outlen)

            return 1;                           /* not enough output space */

        while (len--)

            s->out[s->outcnt++] = s->in[s->incnt++];

    }

    else {                                      /* just scanning */

        s->outcnt += len;

        s->incnt += len;

    }

    /* done with a valid stored block */

    return 0;

}

/*

 * Huffman code decoding tables.  count[1..MAXBITS] is the number of symbols of

 * each length, which for a canonical code are stepped through in order.

 * symbol[] are the symbol values in canonical order, where the number of

 * entries is the sum of the counts in count[].  The decoding process can be

 * seen in the function decode() below.

 */

struct huffman {

    short *count;       /* number of symbols of each length */

    short *symbol;      /* canonically ordered symbols */

};

/*

 * Decode a code from the stream s using huffman table h.  Return the symbol or

 * a negative value if there is an error.  If all of the lengths are zero, i.e.

 * an empty code, or if the code is incomplete and an invalid code is received,

 * then -9 is returned after reading MAXBITS bits.

 *

 * Format notes:

 *

 * - The codes as stored in the compressed data are bit-reversed relative to

 *   a simple integer ordering of codes of the same lengths.  Hence below the

 *   bits are pulled from the compressed data one at a time and used to

 *   build the code value reversed from what is in the stream in order to

 *   permit simple integer comparisons for decoding.  A table-based decoding

 *   scheme (as used in zlib) does not need to do this reversal.

 *

 * - The first code for the shortest length is all zeros.  Subsequent codes of

 *   the same length are simply integer increments of the previous code.  When

 *   moving up a length, a zero bit is appended to the code.  For a complete

 *   code, the last code of the longest length will be all ones.

 *

 * - Incomplete codes are handled by this decoder, since they are permitted

 *   in the deflate format.  See the format notes for fixed() and dynamic().

 */

#ifdef SLOW

local int decode(struct state *s, struct huffman *h)

{

    int len;            /* current number of bits in code */

    int code;           /* len bits being decoded */

    int first;          /* first code of length len */

    int count;          /* number of codes of length len */

    int index;          /* index of first code of length len in symbol table */

    code = first = index = 0;

    for (len = 1; len <= MAXBITS; len++) {

        code |= bits(s, 1);             /* get next bit */

        count = h->count[len];

        if (code < first + count)       /* if length len, return symbol */

            return h->symbol[index + (code - first)];

        index += count;                 /* else update for next length */

        first += count;

        first <<= 1;

        code <<= 1;

    }

    return -9;                          /* ran out of codes */

}

/*

 * A faster version of decode() for real applications of this code.   It's not

 * as readable, but it makes puff() twice as fast.  And it only makes the code

 * a few percent larger.

 */

#else /* !SLOW */

local int decode(struct state *s, struct huffman *h)

{

    int len;            /* current number of bits in code */

    int code;           /* len bits being decoded */

    int first;          /* first code of length len */

    int count;          /* number of codes of length len */

    int index;          /* index of first code of length len in symbol table */

    int bitbuf;         /* bits from stream */

    int left;           /* bits left in next or left to process */

    short *next;        /* next number of codes */

    bitbuf = s->bitbuf;

    left = s->bitcnt;

    code = first = index = 0;

    len = 1;

    next = h->count + 1;

    while (1) {

        while (left--) {

            code |= bitbuf & 1;

            bitbuf >>= 1;

            count = *next++;

            if (code < first + count) { /* if length len, return symbol */

                s->bitbuf = bitbuf;

                s->bitcnt = (s->bitcnt - len) & 7;

                return h->symbol[index + (code - first)];

            }

            index += count;             /* else update for next length */

            first += count;

            first <<= 1;

            code <<= 1;

            len++;

        }

        left = (MAXBITS+1) - len;

        if (left == 0) break;

        if (s->incnt == s->inlen) longjmp(s->env, 1);   /* out of input */

        bitbuf = s->in[s->incnt++];

        if (left > 8) left = 8;

    }

    return -9;                          /* ran out of codes */

}

#endif /* SLOW */

/*

 * Given the list of code lengths length[0..n-1] representing a canonical

 * Huffman code for n symbols, construct the tables required to decode those

 * codes.  Those tables are the number of codes of each length, and the symbols

 * sorted by length, retaining their original order within each length.  The

 * return value is zero for a complete code set, negative for an over-

 * subscribed code set, and positive for an incomplete code set.  The tables

 * can be used if the return value is zero or positive, but they cannot be used

 * if the return value is negative.  If the return value is zero, it is not

 * possible for decode() using that table to return an error--any stream of

 * enough bits will resolve to a symbol.  If the return value is positive, then

 * it is possible for decode() using that table to return an error for received

 * codes past the end of the incomplete lengths.

 *

 * Not used by decode(), but used for error checking, h->count[0] is the number

 * of the n symbols not in the code.  So n - h->count[0] is the number of

 * codes.  This is useful for checking for incomplete codes that have more than

 * one symbol, which is an error in a dynamic block.

 *

 * Assumption: for all i in 0..n-1, 0 <= length[i] <= MAXBITS

 * This is assured by the construction of the length arrays in dynamic() and

 * fixed() and is not verified by construct().

 *

 * Format notes:

 *

 * - Permitted and expected examples of incomplete codes are one of the fixed

 *   codes and any code with a single symbol which in deflate is coded as one

 *   bit instead of zero bits.  See the format notes for fixed() and dynamic().

 *

 * - Within a given code length, the symbols are kept in ascending order for

 *   the code bits definition.

 */

local int construct(struct huffman *h, short *length, int n)

{

    int symbol;         /* current symbol when stepping through length[] */

    int len;            /* current length when stepping through h->count[] */

    int left;           /* number of possible codes left of current length */

    short offs[MAXBITS+1];      /* offsets in symbol table for each length */

    /* count number of codes of each length */

    for (len = 0; len <= MAXBITS; len++)

        h->count[len] = 0;

    for (symbol = 0; symbol < n; symbol++)

        (h->count[length[symbol]])++;   /* assumes lengths are within bounds */

    if (h->count[0] == n)               /* no codes! */

        return 0;                       /* complete, but decode() will fail */

    /* check for an over-subscribed or incomplete set of lengths */

    left = 1;                           /* one possible code of zero length */

    for (len = 1; len <= MAXBITS; len++) {

        left <<= 1;                     /* one more bit, double codes left */

        left -= h->count[len];          /* deduct count from possible codes */

        if (left < 0) return left;      /* over-subscribed--return negative */

    }                                   /* left > 0 means incomplete */

    /* generate offsets into symbol table for each length for sorting */

    offs[1] = 0;

    for (len = 1; len < MAXBITS; len++)

        offs[len + 1] = offs[len] + h->count[len];

    /*

     * put symbols in table sorted by length, by symbol order within each

     * length

     */

    for (symbol = 0; symbol < n; symbol++)

        if (length[symbol] != 0)

            h->symbol[offs[length[symbol]]++] = symbol;

    /* return zero for complete set, positive for incomplete set */

    return left;

}

/*

 * Decode literal/length and distance codes until an end-of-block code.

 *

 * Format notes:

 *

 * - Compressed data that is after the block type if fixed or after the code

 *   description if dynamic is a combination of literals and length/distance

 *   pairs terminated by and end-of-block code.  Literals are simply Huffman

 *   coded bytes.  A length/distance pair is a coded length followed by a

 *   coded distance to represent a string that occurs earlier in the

 *   uncompressed data that occurs again at the current location.

 *

 * - Literals, lengths, and the end-of-block code are combined into a single

 *   code of up to 286 symbols.  They are 256 literals (0..255), 29 length

 *   symbols (257..285), and the end-of-block symbol (256).

 *

 * - There are 256 possible lengths (3..258), and so 29 symbols are not enough

 *   to represent all of those.  Lengths 3..10 and 258 are in fact represented

 *   by just a length symbol.  Lengths 11..257 are represented as a symbol and

 *   some number of extra bits that are added as an integer to the base length

 *   of the length symbol.  The number of extra bits is determined by the base

 *   length symbol.  These are in the static arrays below, lens[] for the base

 *   lengths and lext[] for the corresponding number of extra bits.

 *

 * - The reason that 258 gets its own symbol is that the longest length is used

 *   often in highly redundant files.  Note that 258 can also be coded as the

 *   base value 227 plus the maximum extra value of 31.  While a good deflate

 *   should never do this, it is not an error, and should be decoded properly.

 *

 * - If a length is decoded, including its extra bits if any, then it is

 *   followed a distance code.  There are up to 30 distance symbols.  Again

 *   there are many more possible distances (1..32768), so extra bits are added

 *   to a base value represented by the symbol.  The distances 1..4 get their

 *   own symbol, but the rest require extra bits.  The base distances and

 *   corresponding number of extra bits are below in the static arrays dist[]

 *   and dext[].

 *

 * - Literal bytes are simply written to the output.  A length/distance pair is

 *   an instruction to copy previously uncompressed bytes to the output.  The

 *   copy is from distance bytes back in the output stream, copying for length

 *   bytes.

 *

 * - Distances pointing before the beginning of the output data are not

 *   permitted.

 *

 * - Overlapped copies, where the length is greater than the distance, are

 *   allowed and common.  For example, a distance of one and a length of 258

 *   simply copies the last byte 258 times.  A distance of four and a length of

 *   twelve copies the last four bytes three times.  A simple forward copy

 *   ignoring whether the length is greater than the distance or not implements

 *   this correctly.  You should not use memcpy() since its behavior is not

 *   defined for overlapped arrays.  You should not use memmove() or bcopy()

 *   since though their behavior -is- defined for overlapping arrays, it is

 *   defined to do the wrong thing in this case.

 */

local int codes(struct state *s,

                struct huffman *lencode,

                struct huffman *distcode)

{

    int symbol;         /* decoded symbol */

    int len;            /* length for copy */

    unsigned dist;      /* distance for copy */

    static const short lens[29] = { /* Size base for length codes 257..285 */

        3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,

        35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258};

    static const short lext[29] = { /* Extra bits for length codes 257..285 */

        0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2,

        3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0};

    static const short dists[30] = { /* Offset base for distance codes 0..29 */

        1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,

        257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,

        8193, 12289, 16385, 24577};

    static const short dext[30] = { /* Extra bits for distance codes 0..29 */

        0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,

        7, 7, 8, 8, 9, 9, 10, 10, 11, 11,

        12, 12, 13, 13};

    /* decode literals and length/distance pairs */

    do {

        symbol = decode(s, lencode);

        if (symbol < 0) return symbol;  /* invalid symbol */

        if (symbol < 256) {             /* literal: symbol is the byte */

            /* write out the literal */

            if (s->out != NIL) {

                if (s->outcnt == s->outlen) return 1;

                s->out[s->outcnt] = symbol;

            }

            s->outcnt++;

        }

        else if (symbol > 256) {        /* length */

            /* get and compute length */

            symbol -= 257;

            if (symbol >= 29) return -9;        /* invalid fixed code */

            len = lens[symbol] + bits(s, lext[symbol]);

            /* get and check distance */

            symbol = decode(s, distcode);

            if (symbol < 0) return symbol;      /* invalid symbol */

            dist = dists[symbol] + bits(s, dext[symbol]);

            if (dist > s->outcnt)

                return -10;     /* distance too far back */

            /* copy length bytes from distance bytes back */

            if (s->out != NIL) {

                if (s->outcnt + len > s->outlen) return 1;

                while (len--) {

                    s->out[s->outcnt] = s->out[s->outcnt - dist];

                    s->outcnt++;

                }

            }

            else

                s->outcnt += len;

        }

    } while (symbol != 256);            /* end of block symbol */

    /* done with a valid fixed or dynamic block */

    return 0;

}

/*

 * Process a fixed codes block.

 *

 * Format notes:

 *

 * - This block type can be useful for compressing small amounts of data for

 *   which the size of the code descriptions in a dynamic block exceeds the

 *   benefit of custom codes for that block.  For fixed codes, no bits are

 *   spent on code descriptions.  Instead the code lengths for literal/length

 *   codes and distance codes are fixed.  The specific lengths for each symbol

 *   can be seen in the "for" loops below.

 *

 * - The literal/length code is complete, but has two symbols that are invalid

 *   and should result in an error if received.  This cannot be implemented

 *   simply as an incomplete code since those two symbols are in the "middle"

 *   of the code.  They are eight bits long and the longest literal/length\

 *   code is nine bits.  Therefore the code must be constructed with those

 *   symbols, and the invalid symbols must be detected after decoding.

 *

 * - The fixed distance codes also have two invalid symbols that should result

 *   in an error if received.  Since all of the distance codes are the same

 *   length, this can be implemented as an incomplete code.  Then the invalid

 *   codes are detected while decoding.

 */

local int fixed(struct state *s)

{

    static int virgin = 1;

    static short lencnt[MAXBITS+1], lensym[FIXLCODES];

    static short distcnt[MAXBITS+1], distsym[MAXDCODES];

    static struct huffman lencode = {lencnt, lensym};

    static struct huffman distcode = {distcnt, distsym};

    /* build fixed huffman tables if first call (may not be thread safe) */

    if (virgin) {

        int symbol;

        short lengths[FIXLCODES];

        /* literal/length table */

        for (symbol = 0; symbol < 144; symbol++)

            lengths[symbol] = 8;

        for (; symbol < 256; symbol++)

            lengths[symbol] = 9;

        for (; symbol < 280; symbol++)

            lengths[symbol] = 7;

        for (; symbol < FIXLCODES; symbol++)

            lengths[symbol] = 8;

        construct(&lencode, lengths, FIXLCODES);

        /* distance table */

        for (symbol = 0; symbol < MAXDCODES; symbol++)

            lengths[symbol] = 5;