JSON-XS/XS.xs

#include "EXTERN.h"
#include "perl.h"
#include "XSUB.h"

#include "assert.h"
#include "string.h"
#include "stdlib.h"

#define F_ASCII        0x00000001
#define F_UTF8         0x00000002
#define F_INDENT       0x00000004
#define F_CANONICAL    0x00000008
#define F_SPACE_BEFORE 0x00000010
#define F_SPACE_AFTER  0x00000020
#define F_ALLOW_NONREF 0x00000080
#define F_SHRINK       0x00000100

#define F_PRETTY    F_INDENT | F_SPACE_BEFORE | F_SPACE_AFTER
#define F_DEFAULT   0

#define INIT_SIZE   32 // initial scalar size to be allocated

#define SB do {
#define SE } while (0)

static HV *json_stash;

// structure used for encoding JSON
typedef struct
{
  char *cur;
  STRLEN len; // SvLEN (sv)
  char *end;  // SvEND (sv)
  SV *sv;
  UV flags;
  int max_recurse;
  int indent;
} enc_t;

// structure used for decoding JSON
typedef struct
{
  char *cur;
  char *end;
  const char *err;
  UV flags;
} dec_t;

static UV *
SvJSON (SV *sv)
{
  if (!(SvROK (sv) && SvOBJECT (SvRV (sv)) && SvSTASH (SvRV (sv)) == json_stash))
    croak ("object is not of type JSON::XS");

  return &SvUVX (SvRV (sv));
}

static void
shrink (SV *sv)
{
  sv_utf8_downgrade (sv, 1);
#ifdef SvPV_shrink_to_cur
  SvPV_shrink_to_cur (sv);
#endif
}

/////////////////////////////////////////////////////////////////////////////

static void
need (enc_t *enc, STRLEN len)
{
  if (enc->cur + len >= enc->end)
    {
      STRLEN cur = enc->cur - SvPVX (enc->sv);
      SvGROW (enc->sv, cur + len + 1);
      enc->cur = SvPVX (enc->sv) + cur;
      enc->end = SvPVX (enc->sv) + SvLEN (enc->sv);
    }
}

static void
encode_ch (enc_t *enc, char ch)
{
  need (enc, 1);
  *enc->cur++ = ch;
}

static void
encode_str (enc_t *enc, char *str, STRLEN len, int is_utf8)
{
  char *end = str + len;

  need (enc, len);

  while (str < end)
    {
      unsigned char ch = *(unsigned char *)str;

      if (ch >= 0x20 && ch < 0x80) // most common case
        {
          if (ch == '"') // but with slow exceptions
            {
              need (enc, len += 1);
              *enc->cur++ = '\\';
              *enc->cur++ = '"';
            }
          else if (ch == '\\')
            {
              need (enc, len += 1);
              *enc->cur++ = '\\';
              *enc->cur++ = '\\';
            }
          else
            *enc->cur++ = ch;

          ++str;
        }
      else
        {
          switch (ch)
            {
              case '\010': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'b'; ++str; break;
              case '\011': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 't'; ++str; break;
              case '\012': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'n'; ++str; break;
              case '\014': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'f'; ++str; break;
              case '\015': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'r'; ++str; break;

              default:
                {
                  STRLEN clen;
                  UV uch;

                  if (is_utf8)
                    {
                      uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
                      if (clen == (STRLEN)-1)
                        croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str);
                    }
                  else
                    {
                      uch = ch;
                      clen = 1;
                    }

                  if (uch > 0x10FFFFUL)
                    croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch);

                  if (uch < 0x80 || enc->flags & F_ASCII)
                    {
                      if (uch > 0xFFFFUL)
                        {
                          need (enc, len += 11);
                          sprintf (enc->cur, "\\u%04x\\u%04x",
                                   (int)((uch - 0x10000) / 0x400 + 0xD800),
                                   (int)((uch - 0x10000) % 0x400 + 0xDC00));
                          enc->cur += 12;
                        }
                      else
                        {
                          static char hexdigit [16] = "0123456789abcdef";
                          need (enc, len += 5);
                          *enc->cur++ = '\\';
                          *enc->cur++ = 'u';
                          *enc->cur++ = hexdigit [ uch >> 12      ];
                          *enc->cur++ = hexdigit [(uch >>  8) & 15];
                          *enc->cur++ = hexdigit [(uch >>  4) & 15];
                          *enc->cur++ = hexdigit [(uch >>  0) & 15];
                        }

                      str += clen;
                    }
                  else if (is_utf8)
                    {
                      need (enc, len += clen);
                      do
                        {
                          *enc->cur++ = *str++;
                        }
                      while (--clen);
                    }
                  else
                    {
                      need (enc, len += 10); // never more than 11 bytes needed
                      enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);
                      ++str;
                    }
                }
            }
        }

      --len;
    }
}

#define INDENT SB \
  if (enc->flags & F_INDENT)            \
    {                                   \
      int i_;                           \
      need (enc, enc->indent);          \
      for (i_ = enc->indent * 3; i_--; )\
        encode_ch (enc, ' ');           \
    }                                   \
  SE

#define SPACE SB need (enc, 1); encode_ch (enc, ' '); SE
#define NL    SB if (enc->flags & F_INDENT) { need (enc, 1); encode_ch (enc, '\n'); } SE
#define COMMA SB \
  encode_ch (enc, ',');                 \
  if (enc->flags & F_INDENT)            \
    NL;                                 \
  else if (enc->flags & F_SPACE_AFTER)  \
    SPACE;                              \
  SE

static void encode_sv (enc_t *enc, SV *sv);

static void
encode_av (enc_t *enc, AV *av)
{
  int i, len = av_len (av);

  encode_ch (enc, '['); NL;
  ++enc->indent;

  for (i = 0; i <= len; ++i)
    {
      INDENT;
      encode_sv (enc, *av_fetch (av, i, 0));

      if (i < len)
        COMMA;
    }

  NL;

  --enc->indent;
  INDENT; encode_ch (enc, ']');
}

static void
encode_he (enc_t *enc, HE *he)
{
  encode_ch (enc, '"');

  if (HeKLEN (he) == HEf_SVKEY)
    {
      SV *sv = HeSVKEY (he);
      STRLEN len;
      char *str;
      
      SvGETMAGIC (sv);
      str = SvPV (sv, len);

      encode_str (enc, str, len, SvUTF8 (sv));
    }
  else
    encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he));

  encode_ch (enc, '"');

  if (enc->flags & F_SPACE_BEFORE) SPACE;
  encode_ch (enc, ':');
  if (enc->flags & F_SPACE_AFTER ) SPACE;
  encode_sv (enc, HeVAL (he));
}

// compare hash entries, used when all keys are bytestrings
static int
he_cmp_fast (const void *a_, const void *b_)
{
  int cmp;

  HE *a = *(HE **)a_;
  HE *b = *(HE **)b_;

  STRLEN la = HeKLEN (a);
  STRLEN lb = HeKLEN (b);

  if (!(cmp == memcmp (HeKEY (a), HeKEY (b), la < lb ? la : lb)))
    cmp = la < lb ? -1 : la == lb ? 0 : 1;

  return cmp;
}

// compare hash entries, used when some keys are sv's or utf-x
static int
he_cmp_slow (const void *a, const void *b)
{
  return sv_cmp (HeSVKEY_force (*(HE **)a), HeSVKEY_force (*(HE **)b));
}

static void
encode_hv (enc_t *enc, HV *hv)
{
  int count, i;

  encode_ch (enc, '{'); NL; ++enc->indent;

  if ((count = hv_iterinit (hv)))
    {
      // for canonical output we have to sort by keys first
      // actually, this is mostly due to the stupid so-called
      // security workaround added somewhere in 5.8.x.
      // that randomises hash orderings
      if (enc->flags & F_CANONICAL)
        {
          HE *he, *hes [count];
          int fast = 1;

          i = 0;
          while ((he = hv_iternext (hv)))
            {
              hes [i++] = he;
              if (HeKLEN (he) < 0 || HeKUTF8 (he))
                fast = 0;
            }

          assert (i == count);

          if (fast)
            qsort (hes, count, sizeof (HE *), he_cmp_fast);
          else
            {
              // hack to forcefully disable "use bytes"
              COP cop = *PL_curcop;
              cop.op_private = 0;

              ENTER;
              SAVETMPS;

              SAVEVPTR (PL_curcop);
              PL_curcop = &cop;

              qsort (hes, count, sizeof (HE *), he_cmp_slow);

              FREETMPS;
              LEAVE;
            }

          for (i = 0; i < count; ++i)
            {
              INDENT;
              encode_he (enc, hes [i]);

              if (i < count - 1)
                COMMA;
            }

          NL;
        }
      else
        {
          SV *sv;
          HE *he = hv_iternext (hv);

          for (;;)
            {
              INDENT;
              encode_he (enc, he);

              if (!(he = hv_iternext (hv)))
                break;

              COMMA;
            }

          NL;
        }
    }

  --enc->indent; INDENT; encode_ch (enc, '}');
}

static void
encode_sv (enc_t *enc, SV *sv)
{
  SvGETMAGIC (sv);

  if (SvPOKp (sv))
    {
      STRLEN len;
      char *str = SvPV (sv, len);
      encode_ch (enc, '"');
      encode_str (enc, str, len, SvUTF8 (sv));
      encode_ch (enc, '"');
    }
  else if (SvNOKp (sv))
    {
      need (enc, NV_DIG + 32);
      Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
      enc->cur += strlen (enc->cur);
    }
  else if (SvIOKp (sv))
    {
      need (enc, 64);
      enc->cur += 
         SvIsUV(sv)
            ? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv))
            : snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv));
    }
  else if (SvROK (sv))
    {
      SV *rv = SvRV (sv);

      if (!--enc->max_recurse)
        croak ("data structure too deep (hit recursion limit)");

      switch (SvTYPE (rv))
        {
          case SVt_PVAV: encode_av (enc, (AV *)rv); break;
          case SVt_PVHV: encode_hv (enc, (HV *)rv); break;

          default:
            croak ("encountered %s, but JSON can only represent references to arrays or hashes",
                   SvPV_nolen (sv));
        }
    }
  else if (!SvOK (sv))
    encode_str (enc, "null", 4, 0);
  else
    croak ("encountered perl type (%s,0x%x) that JSON cannot handle, you might want to report this",
           SvPV_nolen (sv), SvFLAGS (sv));
}

static SV *
encode_json (SV *scalar, UV flags)
{
  if (!(flags & F_ALLOW_NONREF) && !SvROK (scalar))
    croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)");

  enc_t enc;
  enc.flags       = flags;
  enc.sv          = sv_2mortal (NEWSV (0, INIT_SIZE));
  enc.cur         = SvPVX (enc.sv);
  enc.end         = SvEND (enc.sv);
  enc.max_recurse = 0;
  enc.indent      = 0;

  SvPOK_only (enc.sv);
  encode_sv (&enc, scalar);

  if (!(flags & (F_ASCII | F_UTF8)))
    SvUTF8_on (enc.sv);

  SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv));

  if (enc.flags & F_SHRINK)
    shrink (enc.sv);

  return enc.sv;
}

/////////////////////////////////////////////////////////////////////////////

#define WS \
  for (;;)                              \
    {                                   \
      char ch = *dec->cur;              \
      if (ch > 0x20                     \
          || (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)) \
        break;                          \
      ++dec->cur;                       \
    }

#define ERR(reason) SB dec->err = reason; goto fail; SE
#define EXPECT_CH(ch) SB \
  if (*dec->cur != ch)          \
    ERR (# ch " expected");     \
  ++dec->cur;                   \
  SE

static SV *decode_sv (dec_t *dec);

static signed char decode_hexdigit[256];

static UV
decode_4hex (dec_t *dec)
{
  signed char d1, d2, d3, d4;

  d1 = decode_hexdigit [((unsigned char *)dec->cur) [0]];
  if (d1 < 0) ERR ("four hexadecimal digits expected");
  d2 = decode_hexdigit [((unsigned char *)dec->cur) [1]];
  if (d2 < 0) ERR ("four hexadecimal digits expected");
  d3 = decode_hexdigit [((unsigned char *)dec->cur) [2]];
  if (d3 < 0) ERR ("four hexadecimal digits expected");
  d4 = decode_hexdigit [((unsigned char *)dec->cur) [3]];
  if (d4 < 0) ERR ("four hexadecimal digits expected");

  dec->cur += 4;

  return ((UV)d1) << 12
       | ((UV)d2) <<  8
       | ((UV)d3) <<  4
       | ((UV)d4);

fail:
  return (UV)-1;
}

#define APPEND_GROW(n) SB \
  if (cur + (n) >= end)                         \
    {                                           \
      STRLEN ofs = cur - SvPVX (sv);            \
      SvGROW (sv, ofs + (n) + 1);               \
      cur = SvPVX (sv) + ofs;                   \
      end = SvEND (sv);                         \
    }                                           \
  SE

#define APPEND_CH(ch) SB \
  APPEND_GROW (1);      \
  *cur++ = (ch);        \
  SE

static SV *
decode_str (dec_t *dec)
{
  SV *sv = NEWSV (0,2);
  int utf8 = 0;
  char *cur = SvPVX (sv);
  char *end = SvEND (sv);

  for (;;)
    {
      unsigned char ch = *(unsigned char *)dec->cur;

      if (ch == '"')
        break;
      else if (ch == '\\')
        {
          switch (*++dec->cur)
            {
              case '\\':
              case '/':
              case '"': APPEND_CH (*dec->cur++); break;

              case 'b': APPEND_CH ('\010'); ++dec->cur; break;
              case 't': APPEND_CH ('\011'); ++dec->cur; break;
              case 'n': APPEND_CH ('\012'); ++dec->cur; break;
              case 'f': APPEND_CH ('\014'); ++dec->cur; break;
              case 'r': APPEND_CH ('\015'); ++dec->cur; break;

              case 'u':
                {
                  UV lo, hi;
                  ++dec->cur;

                  hi = decode_4hex (dec);
                  if (hi == (UV)-1)
                    goto fail;

                  // possibly a surrogate pair
                  if (hi >= 0xd800 && hi < 0xdc00)
                    {
                      if (dec->cur [0] != '\\' || dec->cur [1] != 'u')
                        ERR ("missing low surrogate character in surrogate pair");

                      dec->cur += 2;

                      lo = decode_4hex (dec);
                      if (lo == (UV)-1)
                        goto fail;

                      if (lo < 0xdc00 || lo >= 0xe000)
                        ERR ("surrogate pair expected");

                      hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000;
                    }
                  else if (hi >= 0xdc00 && hi < 0xe000)
                    ERR ("missing high surrogate character in surrogate pair");

                  if (hi >= 0x80)
                    {
                      utf8 = 1;

                      APPEND_GROW (4); // at most 4 bytes for 21 bits
                      cur = (char *)uvuni_to_utf8_flags (cur, hi, 0);
                    }
                  else
                    APPEND_CH (hi);
                }
                break;

              default:
                --dec->cur;
                ERR ("illegal backslash escape sequence in string");
            }
        }
      else if (ch >= 0x20 && ch <= 0x7f)
        APPEND_CH (*dec->cur++);
      else if (ch >= 0x80)
        {
          STRLEN clen;
          UV uch = utf8n_to_uvuni (dec->cur, dec->end - dec->cur, &clen, UTF8_CHECK_ONLY);
          if (clen == (STRLEN)-1)
            ERR ("malformed UTF-8 character in JSON string");

          APPEND_GROW (clen);
          do
            {
              *cur++ = *dec->cur++;
            }
          while (--clen);

          utf8 = 1;
        }
      else if (dec->cur == dec->end)
        ERR ("unexpected end of string while parsing json string");
      else
        ERR ("invalid character encountered");
    }

  ++dec->cur;

  SvCUR_set (sv, cur - SvPVX (sv));

  SvPOK_only (sv);
  *SvEND (sv) = 0;

  if (utf8)
    SvUTF8_on (sv);

  if (dec->flags & F_SHRINK)
    shrink (sv);

  return sv;

fail:
  SvREFCNT_dec (sv);
  return 0;
}

static SV *
decode_num (dec_t *dec)
{
  int is_nv = 0;
  char *start = dec->cur;

  // [minus]
  if (*dec->cur == '-')
    ++dec->cur;

  if (*dec->cur == '0')
    {
      ++dec->cur;
      if (*dec->cur >= '0' && *dec->cur <= '9')
         ERR ("malformed number (leading zero must not be followed by another digit)");
    }
  else if (*dec->cur < '0' || *dec->cur > '9')
    ERR ("malformed number (no digits after initial minus)");
  else
    do
      {
        ++dec->cur;
      }
    while (*dec->cur >= '0' && *dec->cur <= '9');

  // [frac]
  if (*dec->cur == '.')
    {
      ++dec->cur;

      if (*dec->cur < '0' || *dec->cur > '9')
        ERR ("malformed number (no digits after decimal point)");

      do
        {
          ++dec->cur;
        }
      while (*dec->cur >= '0' && *dec->cur <= '9');

      is_nv = 1;
    }

  // [exp]
  if (*dec->cur == 'e' || *dec->cur == 'E')
    {
      ++dec->cur;

      if (*dec->cur == '-' || *dec->cur == '+')
        ++dec->cur;

      if (*dec->cur < '0' || *dec->cur > '9')
        ERR ("malformed number (no digits after exp sign)");

      do
        {
          ++dec->cur;
        }
      while (*dec->cur >= '0' && *dec->cur <= '9');

      is_nv = 1;
    }

  if (!is_nv)
    {
      UV uv;
      int numtype = grok_number (start, dec->cur - start, &uv);
      if (numtype & IS_NUMBER_IN_UV)
        if (numtype & IS_NUMBER_NEG)
          {
            if (uv < (UV)IV_MIN)
              return newSViv (-(IV)uv);
          }
        else
          return newSVuv (uv);
    }

  return newSVnv (Atof (start));

fail:
  return 0;
}

static SV *
decode_av (dec_t *dec)
{
  AV *av = newAV ();

  WS;
  if (*dec->cur == ']')
    ++dec->cur;
  else
    for (;;)
      {
        SV *value;

        value = decode_sv (dec);
        if (!value)
          goto fail;

        av_push (av, value);

        WS;

        if (*dec->cur == ']')
          {
            ++dec->cur;
            break;
          }
        
        if (*dec->cur != ',')
          ERR (", or ] expected while parsing array");

        ++dec->cur;
      }

  return newRV_noinc ((SV *)av);

fail:
  SvREFCNT_dec (av);
  return 0;
}

static SV *
decode_hv (dec_t *dec)
{
  HV *hv = newHV ();

  WS;
  if (*dec->cur == '}')
    ++dec->cur;
  else
    for (;;)
      {
        SV *key, *value;

        WS; EXPECT_CH ('"');

        key = decode_str (dec);
        if (!key)
          goto fail;

        WS; EXPECT_CH (':');

        value = decode_sv (dec);
        if (!value)
          {
            SvREFCNT_dec (key);
            goto fail;
          }

        //TODO: optimise
        hv_store_ent (hv, key, value, 0);

        WS;

        if (*dec->cur == '}')
          {
            ++dec->cur;
            break;
          }

        if (*dec->cur != ',')
          ERR (", or } expected while parsing object/hash");

        ++dec->cur;
      }

  return newRV_noinc ((SV *)hv);

fail:
  SvREFCNT_dec (hv);
  return 0;
}

static SV *
decode_sv (dec_t *dec)
{
  WS;
  switch (*dec->cur)
    {
      case '"': ++dec->cur; return decode_str (dec); 
      case '[': ++dec->cur; return decode_av (dec); 
      case '{': ++dec->cur; return decode_hv (dec);

      case '-':
      case '0': case '1': case '2': case '3': case '4':
      case '5': case '6': case '7': case '8': case '9':
        return decode_num (dec);

      case 't':
        if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4))
          {
            dec->cur += 4;
            return newSViv (1);
          }
        else
          ERR ("'true' expected");

        break;

      case 'f':
        if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5))
          {
            dec->cur += 5;
            return newSViv (0);
          }
        else
          ERR ("'false' expected");

        break;

      case 'n':
        if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4))
          {
            dec->cur += 4;
            return newSVsv (&PL_sv_undef);
          }
        else
          ERR ("'null' expected");

        break;

      default:
        ERR ("malformed json string, neither array, object, number, string or atom");
        break;
    }

fail:
  return 0;
}

static SV *
decode_json (SV *string, UV flags)
{
  SV *sv;

  if (flags & F_UTF8)
    sv_utf8_downgrade (string, 0);
  else
    sv_utf8_upgrade (string);

  SvGROW (string, SvCUR (string) + 1); // should basically be a NOP

  dec_t dec;
  dec.flags = flags;
  dec.cur   = SvPVX (string);
  dec.end   = SvEND (string);
  dec.err   = 0;

  sv = decode_sv (&dec);

  if (!sv)
    {
      IV offset = dec.flags & F_UTF8
                  ? dec.cur - SvPVX (string)
                  : utf8_distance (dec.cur, SvPVX (string));
      SV *uni = sv_newmortal ();

      // horrible hack to silence warning inside pv_uni_display
      COP cop = *PL_curcop;
      cop.cop_warnings = pWARN_NONE;
      ENTER;
      SAVEVPTR (PL_curcop);
      PL_curcop = &cop;
      pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ);
      LEAVE;

      croak ("%s, at character offset %d (%s)",
             dec.err,
             (int)offset,
             dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)");
    }

  sv = sv_2mortal (sv);

  if (!(dec.flags & F_ALLOW_NONREF) && !SvROK (sv))
    croak ("JSON text must be an object or array (but found number, string, true, false or null, use allow_nonref to allow this)");

  return sv;
}

MODULE = JSON::XS               PACKAGE = JSON::XS

BOOT:
{
        int i;

        memset (decode_hexdigit, 0xff, 256);
        for (i = 10; i--; )
          decode_hexdigit ['0' + i] = i;

        for (i = 7; i--; )
          {
            decode_hexdigit ['a' + i] = 10 + i;
            decode_hexdigit ['A' + i] = 10 + i;
          }

        json_stash = gv_stashpv ("JSON::XS", 1);
}

PROTOTYPES: DISABLE

SV *new (char *dummy)
        CODE:
        RETVAL = sv_bless (newRV_noinc (newSVuv (F_DEFAULT)), json_stash);
        OUTPUT:
        RETVAL

SV *ascii (SV *self, int enable = 1)
        ALIAS:
        ascii        = F_ASCII
        utf8         = F_UTF8
        indent       = F_INDENT
        canonical    = F_CANONICAL
        space_before = F_SPACE_BEFORE
        space_after  = F_SPACE_AFTER
        pretty       = F_PRETTY
        allow_nonref = F_ALLOW_NONREF
        shrink       = F_SHRINK
        CODE:
{
        UV *uv = SvJSON (self);
        if (enable)
          *uv |=  ix;
        else
          *uv &= ~ix;

        RETVAL = newSVsv (self);
}
        OUTPUT:
        RETVAL

void encode (SV *self, SV *scalar)
        PPCODE:
        XPUSHs (encode_json (scalar, *SvJSON (self)));

void decode (SV *self, SV *jsonstr)
        PPCODE:
        XPUSHs (decode_json (jsonstr, *SvJSON (self)));

PROTOTYPES: ENABLE

void to_json (SV *scalar)
        PPCODE:
        XPUSHs (encode_json (scalar, F_UTF8));

void from_json (SV *jsonstr)
        PPCODE:
        XPUSHs (decode_json (jsonstr, F_UTF8));

Revision:	1.10
Committed:	Sat Mar 24 01:15:22 2007 UTC (17 years, 1 month ago) by root
Branch:	MAIN
CVS Tags:	rel-0_31
Changes since 1.9:	+2 -4 lines
Log Message:	* empty log message *
#	User	Rev	Content
1	root	1.1	#include "EXTERN.h"
2			#include "perl.h"
3			#include "XSUB.h"
4
5			#include "assert.h"
6			#include "string.h"
7			#include "stdlib.h"
8
9			#define F_ASCII 0x00000001
10			#define F_UTF8 0x00000002
11			#define F_INDENT 0x00000004
12			#define F_CANONICAL 0x00000008
13			#define F_SPACE_BEFORE 0x00000010
14			#define F_SPACE_AFTER 0x00000020
15	root	1.3	#define F_ALLOW_NONREF 0x00000080
16	root	1.6	#define F_SHRINK 0x00000100
17	root	1.1
18	root	1.2	#define F_PRETTY F_INDENT \| F_SPACE_BEFORE \| F_SPACE_AFTER
19	root	1.1	#define F_DEFAULT 0
20
21			#define INIT_SIZE 32 // initial scalar size to be allocated
22
23			#define SB do {
24			#define SE } while (0)
25
26			static HV *json_stash;
27
28			// structure used for encoding JSON
29			typedef struct
30			{
31			char *cur;
32			STRLEN len; // SvLEN (sv)
33			char *end; // SvEND (sv)
34			SV *sv;
35			UV flags;
36			int max_recurse;
37			int indent;
38			} enc_t;
39
40			// structure used for decoding JSON
41			typedef struct
42			{
43			char *cur;
44			char *end;
45	root	1.4	const char *err;
46	root	1.1	UV flags;
47			} dec_t;
48
49			static UV *
50			SvJSON (SV *sv)
51			{
52			if (!(SvROK (sv) && SvOBJECT (SvRV (sv)) && SvSTASH (SvRV (sv)) == json_stash))
53			croak ("object is not of type JSON::XS");
54
55			return &SvUVX (SvRV (sv));
56			}
57
58	root	1.7	static void
59			shrink (SV *sv)
60			{
61			sv_utf8_downgrade (sv, 1);
62			#ifdef SvPV_shrink_to_cur
63			SvPV_shrink_to_cur (sv);
64			#endif
65			}
66
67	root	1.1	/////////////////////////////////////////////////////////////////////////////
68
69			static void
70			need (enc_t *enc, STRLEN len)
71			{
72			if (enc->cur + len >= enc->end)
73			{
74			STRLEN cur = enc->cur - SvPVX (enc->sv);
75			SvGROW (enc->sv, cur + len + 1);
76			enc->cur = SvPVX (enc->sv) + cur;
77	root	1.4	enc->end = SvPVX (enc->sv) + SvLEN (enc->sv);
78	root	1.1	}
79			}
80
81			static void
82			encode_ch (enc_t *enc, char ch)
83			{
84			need (enc, 1);
85			*enc->cur++ = ch;
86			}
87
88			static void
89			encode_str (enc_t enc, char str, STRLEN len, int is_utf8)
90			{
91			char *end = str + len;
92
93	root	1.4	need (enc, len);
94
95	root	1.1	while (str < end)
96			{
97			unsigned char ch = (unsigned char )str;
98	root	1.4
99	root	1.6	if (ch >= 0x20 && ch < 0x80) // most common case
100	root	1.4	{
101	root	1.6	if (ch == '"') // but with slow exceptions
102			{
103			need (enc, len += 1);
104			*enc->cur++ = '\\';
105			*enc->cur++ = '"';
106			}
107			else if (ch == '\\')
108			{
109			need (enc, len += 1);
110			*enc->cur++ = '\\';
111			*enc->cur++ = '\\';
112			}
113			else
114			*enc->cur++ = ch;
115
116	root	1.4	++str;
117	root	1.1	}
118			else
119			{
120	root	1.6	switch (ch)
121	root	1.1	{
122	root	1.6	case '\010': need (enc, len += 1); enc->cur++ = '\\'; enc->cur++ = 'b'; ++str; break;
123			case '\011': need (enc, len += 1); enc->cur++ = '\\'; enc->cur++ = 't'; ++str; break;
124			case '\012': need (enc, len += 1); enc->cur++ = '\\'; enc->cur++ = 'n'; ++str; break;
125			case '\014': need (enc, len += 1); enc->cur++ = '\\'; enc->cur++ = 'f'; ++str; break;
126			case '\015': need (enc, len += 1); enc->cur++ = '\\'; enc->cur++ = 'r'; ++str; break;
127	root	1.1
128	root	1.6	default:
129	root	1.1	{
130	root	1.6	STRLEN clen;
131			UV uch;
132
133			if (is_utf8)
134			{
135			uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
136			if (clen == (STRLEN)-1)
137	root	1.9	croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str);
138	root	1.6	}
139			else
140			{
141			uch = ch;
142			clen = 1;
143			}
144
145	root	1.9	if (uch > 0x10FFFFUL)
146			croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch);
147
148	root	1.6	if (uch < 0x80 \|\| enc->flags & F_ASCII)
149			{
150			if (uch > 0xFFFFUL)
151			{
152			need (enc, len += 11);
153			sprintf (enc->cur, "\\u%04x\\u%04x",
154	root	1.10	(int)((uch - 0x10000) / 0x400 + 0xD800),
155			(int)((uch - 0x10000) % 0x400 + 0xDC00));
156	root	1.6	enc->cur += 12;
157			}
158			else
159			{
160			static char hexdigit [16] = "0123456789abcdef";
161			need (enc, len += 5);
162			*enc->cur++ = '\\';
163			*enc->cur++ = 'u';
164			*enc->cur++ = hexdigit [ uch >> 12 ];
165			*enc->cur++ = hexdigit [(uch >> 8) & 15];
166			*enc->cur++ = hexdigit [(uch >> 4) & 15];
167			*enc->cur++ = hexdigit [(uch >> 0) & 15];
168			}
169	root	1.4
170	root	1.6	str += clen;
171			}
172			else if (is_utf8)
173			{
174			need (enc, len += clen);
175			do
176			{
177			enc->cur++ = str++;
178			}
179			while (--clen);
180			}
181			else
182			{
183	root	1.7	need (enc, len += 10); // never more than 11 bytes needed
184	root	1.6	enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);
185			++str;
186			}
187	root	1.5	}
188	root	1.4	}
189	root	1.1	}
190
191			--len;
192			}
193			}
194
195			#define INDENT SB \
196			if (enc->flags & F_INDENT) \
197			{ \
198			int i_; \
199			need (enc, enc->indent); \
200			for (i_ = enc->indent * 3; i_--; )\
201			encode_ch (enc, ' '); \
202			} \
203			SE
204
205	root	1.2	#define SPACE SB need (enc, 1); encode_ch (enc, ' '); SE
206	root	1.1	#define NL SB if (enc->flags & F_INDENT) { need (enc, 1); encode_ch (enc, '\n'); } SE
207			#define COMMA SB \
208			encode_ch (enc, ','); \
209			if (enc->flags & F_INDENT) \
210			NL; \
211			else if (enc->flags & F_SPACE_AFTER) \
212			SPACE; \
213			SE
214
215			static void encode_sv (enc_t enc, SV sv);
216
217			static void
218			encode_av (enc_t enc, AV av)
219			{
220			int i, len = av_len (av);
221
222			encode_ch (enc, '['); NL;
223			++enc->indent;
224
225			for (i = 0; i <= len; ++i)
226			{
227			INDENT;
228			encode_sv (enc, *av_fetch (av, i, 0));
229
230			if (i < len)
231			COMMA;
232			}
233
234			NL;
235
236			--enc->indent;
237			INDENT; encode_ch (enc, ']');
238			}
239
240			static void
241			encode_he (enc_t enc, HE he)
242			{
243			encode_ch (enc, '"');
244
245			if (HeKLEN (he) == HEf_SVKEY)
246			{
247			SV *sv = HeSVKEY (he);
248			STRLEN len;
249	root	1.4	char *str;
250
251			SvGETMAGIC (sv);
252			str = SvPV (sv, len);
253	root	1.1
254			encode_str (enc, str, len, SvUTF8 (sv));
255			}
256			else
257			encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he));
258
259			encode_ch (enc, '"');
260
261			if (enc->flags & F_SPACE_BEFORE) SPACE;
262			encode_ch (enc, ':');
263			if (enc->flags & F_SPACE_AFTER ) SPACE;
264			encode_sv (enc, HeVAL (he));
265			}
266
267			// compare hash entries, used when all keys are bytestrings
268			static int
269			he_cmp_fast (const void a_, const void b_)
270			{
271			int cmp;
272
273			HE a = (HE **)a_;
274			HE b = (HE **)b_;
275
276			STRLEN la = HeKLEN (a);
277			STRLEN lb = HeKLEN (b);
278
279			if (!(cmp == memcmp (HeKEY (a), HeKEY (b), la < lb ? la : lb)))
280			cmp = la < lb ? -1 : la == lb ? 0 : 1;
281
282			return cmp;
283			}
284
285			// compare hash entries, used when some keys are sv's or utf-x
286			static int
287			he_cmp_slow (const void a, const void b)
288			{
289			return sv_cmp (HeSVKEY_force ((HE )a), HeSVKEY_force ((HE **)b));
290			}
291
292			static void
293			encode_hv (enc_t enc, HV hv)
294			{
295			int count, i;
296
297			encode_ch (enc, '{'); NL; ++enc->indent;
298
299			if ((count = hv_iterinit (hv)))
300			{
301			// for canonical output we have to sort by keys first
302			// actually, this is mostly due to the stupid so-called
303			// security workaround added somewhere in 5.8.x.
304			// that randomises hash orderings
305			if (enc->flags & F_CANONICAL)
306			{
307			HE he, hes [count];
308			int fast = 1;
309
310			i = 0;
311			while ((he = hv_iternext (hv)))
312			{
313			hes [i++] = he;
314			if (HeKLEN (he) < 0 \|\| HeKUTF8 (he))
315			fast = 0;
316			}
317
318			assert (i == count);
319
320			if (fast)
321			qsort (hes, count, sizeof (HE *), he_cmp_fast);
322			else
323			{
324	root	1.8	// hack to forcefully disable "use bytes"
325			COP cop = *PL_curcop;
326	root	1.1	cop.op_private = 0;
327	root	1.8
328			ENTER;
329			SAVETMPS;
330
331			SAVEVPTR (PL_curcop);
332	root	1.1	PL_curcop = &cop;
333
334			qsort (hes, count, sizeof (HE *), he_cmp_slow);
335	root	1.8
336	root	1.1	FREETMPS;
337	root	1.8	LEAVE;
338	root	1.1	}
339
340			for (i = 0; i < count; ++i)
341			{
342			INDENT;
343			encode_he (enc, hes [i]);
344
345			if (i < count - 1)
346			COMMA;
347			}
348
349			NL;
350			}
351			else
352			{
353			SV *sv;
354			HE *he = hv_iternext (hv);
355
356			for (;;)
357			{
358			INDENT;
359			encode_he (enc, he);
360
361			if (!(he = hv_iternext (hv)))
362			break;
363
364			COMMA;
365			}
366
367			NL;
368			}
369			}
370
371			--enc->indent; INDENT; encode_ch (enc, '}');
372			}
373
374			static void
375			encode_sv (enc_t enc, SV sv)
376			{
377	root	1.4	SvGETMAGIC (sv);
378
379	root	1.1	if (SvPOKp (sv))
380			{
381			STRLEN len;
382			char *str = SvPV (sv, len);
383			encode_ch (enc, '"');
384			encode_str (enc, str, len, SvUTF8 (sv));
385			encode_ch (enc, '"');
386			}
387			else if (SvNOKp (sv))
388			{
389			need (enc, NV_DIG + 32);
390			Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
391			enc->cur += strlen (enc->cur);
392			}
393			else if (SvIOKp (sv))
394			{
395			need (enc, 64);
396			enc->cur +=
397			SvIsUV(sv)
398			? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv))
399			: snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv));
400			}
401			else if (SvROK (sv))
402			{
403	root	1.9	SV *rv = SvRV (sv);
404
405	root	1.1	if (!--enc->max_recurse)
406			croak ("data structure too deep (hit recursion limit)");
407
408	root	1.9	switch (SvTYPE (rv))
409	root	1.1	{
410	root	1.9	case SVt_PVAV: encode_av (enc, (AV *)rv); break;
411			case SVt_PVHV: encode_hv (enc, (HV *)rv); break;
412	root	1.1
413			default:
414	root	1.9	croak ("encountered %s, but JSON can only represent references to arrays or hashes",
415			SvPV_nolen (sv));
416	root	1.1	}
417			}
418			else if (!SvOK (sv))
419			encode_str (enc, "null", 4, 0);
420			else
421	root	1.9	croak ("encountered perl type (%s,0x%x) that JSON cannot handle, you might want to report this",
422			SvPV_nolen (sv), SvFLAGS (sv));
423	root	1.1	}
424
425			static SV *
426			encode_json (SV *scalar, UV flags)
427			{
428	root	1.3	if (!(flags & F_ALLOW_NONREF) && !SvROK (scalar))
429	root	1.9	croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)");
430	root	1.3
431	root	1.1	enc_t enc;
432			enc.flags = flags;
433			enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE));
434			enc.cur = SvPVX (enc.sv);
435			enc.end = SvEND (enc.sv);
436			enc.max_recurse = 0;
437			enc.indent = 0;
438
439			SvPOK_only (enc.sv);
440			encode_sv (&enc, scalar);
441
442			if (!(flags & (F_ASCII \| F_UTF8)))
443			SvUTF8_on (enc.sv);
444
445			SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv));
446	root	1.6
447			if (enc.flags & F_SHRINK)
448	root	1.7	shrink (enc.sv);
449
450	root	1.1	return enc.sv;
451			}
452
453			/////////////////////////////////////////////////////////////////////////////
454
455			#define WS \
456			for (;;) \
457			{ \
458			char ch = *dec->cur; \
459			if (ch > 0x20 \
460			\|\| (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)) \
461			break; \
462			++dec->cur; \
463			}
464
465			#define ERR(reason) SB dec->err = reason; goto fail; SE
466			#define EXPECT_CH(ch) SB \
467			if (*dec->cur != ch) \
468			ERR (# ch " expected"); \
469			++dec->cur; \
470			SE
471
472			static SV decode_sv (dec_t dec);
473
474			static signed char decode_hexdigit[256];
475
476			static UV
477			decode_4hex (dec_t *dec)
478			{
479			signed char d1, d2, d3, d4;
480
481			d1 = decode_hexdigit [((unsigned char *)dec->cur) [0]];
482			if (d1 < 0) ERR ("four hexadecimal digits expected");
483			d2 = decode_hexdigit [((unsigned char *)dec->cur) [1]];
484			if (d2 < 0) ERR ("four hexadecimal digits expected");
485			d3 = decode_hexdigit [((unsigned char *)dec->cur) [2]];
486			if (d3 < 0) ERR ("four hexadecimal digits expected");
487			d4 = decode_hexdigit [((unsigned char *)dec->cur) [3]];
488			if (d4 < 0) ERR ("four hexadecimal digits expected");
489
490			dec->cur += 4;
491
492			return ((UV)d1) << 12
493			\| ((UV)d2) << 8
494			\| ((UV)d3) << 4
495			\| ((UV)d4);
496
497			fail:
498			return (UV)-1;
499			}
500
501	root	1.4	#define APPEND_GROW(n) SB \
502			if (cur + (n) >= end) \
503			{ \
504			STRLEN ofs = cur - SvPVX (sv); \
505			SvGROW (sv, ofs + (n) + 1); \
506			cur = SvPVX (sv) + ofs; \
507			end = SvEND (sv); \
508			} \
509			SE
510
511			#define APPEND_CH(ch) SB \
512			APPEND_GROW (1); \
513			*cur++ = (ch); \
514			SE
515
516	root	1.1	static SV *
517			decode_str (dec_t *dec)
518			{
519			SV *sv = NEWSV (0,2);
520			int utf8 = 0;
521	root	1.4	char *cur = SvPVX (sv);
522			char *end = SvEND (sv);
523	root	1.1
524			for (;;)
525			{
526			unsigned char ch = (unsigned char )dec->cur;
527
528			if (ch == '"')
529			break;
530			else if (ch == '\\')
531			{
532			switch (*++dec->cur)
533			{
534			case '\\':
535			case '/':
536			case '"': APPEND_CH (*dec->cur++); break;
537
538			case 'b': APPEND_CH ('\010'); ++dec->cur; break;
539			case 't': APPEND_CH ('\011'); ++dec->cur; break;
540			case 'n': APPEND_CH ('\012'); ++dec->cur; break;
541			case 'f': APPEND_CH ('\014'); ++dec->cur; break;
542			case 'r': APPEND_CH ('\015'); ++dec->cur; break;
543
544			case 'u':
545			{
546			UV lo, hi;
547			++dec->cur;
548
549			hi = decode_4hex (dec);
550			if (hi == (UV)-1)
551			goto fail;
552
553			// possibly a surrogate pair
554			if (hi >= 0xd800 && hi < 0xdc00)
555			{
556			if (dec->cur [0] != '\\' \|\| dec->cur [1] != 'u')
557	root	1.5	ERR ("missing low surrogate character in surrogate pair");
558	root	1.1
559			dec->cur += 2;
560
561			lo = decode_4hex (dec);
562			if (lo == (UV)-1)
563			goto fail;
564
565			if (lo < 0xdc00 \|\| lo >= 0xe000)
566			ERR ("surrogate pair expected");
567
568			hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000;
569			}
570	root	1.5	else if (hi >= 0xdc00 && hi < 0xe000)
571			ERR ("missing high surrogate character in surrogate pair");
572	root	1.1
573			if (hi >= 0x80)
574			{
575			utf8 = 1;
576
577	root	1.4	APPEND_GROW (4); // at most 4 bytes for 21 bits
578			cur = (char *)uvuni_to_utf8_flags (cur, hi, 0);
579	root	1.1	}
580			else
581			APPEND_CH (hi);
582			}
583			break;
584	root	1.5
585			default:
586			--dec->cur;
587			ERR ("illegal backslash escape sequence in string");
588	root	1.1	}
589			}
590			else if (ch >= 0x20 && ch <= 0x7f)
591			APPEND_CH (*dec->cur++);
592			else if (ch >= 0x80)
593			{
594			STRLEN clen;
595			UV uch = utf8n_to_uvuni (dec->cur, dec->end - dec->cur, &clen, UTF8_CHECK_ONLY);
596	root	1.5	if (clen == (STRLEN)-1)
597	root	1.7	ERR ("malformed UTF-8 character in JSON string");
598	root	1.1
599	root	1.4	APPEND_GROW (clen);
600	root	1.5	do
601			{
602			cur++ = dec->cur++;
603			}
604			while (--clen);
605
606			utf8 = 1;
607	root	1.1	}
608	root	1.5	else if (dec->cur == dec->end)
609			ERR ("unexpected end of string while parsing json string");
610	root	1.1	else
611			ERR ("invalid character encountered");
612			}
613
614			++dec->cur;
615
616	root	1.4	SvCUR_set (sv, cur - SvPVX (sv));
617
618	root	1.1	SvPOK_only (sv);
619			*SvEND (sv) = 0;
620
621			if (utf8)
622			SvUTF8_on (sv);
623
624	root	1.6	if (dec->flags & F_SHRINK)
625	root	1.7	shrink (sv);
626	root	1.6
627	root	1.1	return sv;
628
629			fail:
630			SvREFCNT_dec (sv);
631			return 0;
632			}
633
634			static SV *
635			decode_num (dec_t *dec)
636			{
637			int is_nv = 0;
638			char *start = dec->cur;
639
640			// [minus]
641			if (*dec->cur == '-')
642			++dec->cur;
643
644			if (*dec->cur == '0')
645			{
646			++dec->cur;
647			if (dec->cur >= '0' && dec->cur <= '9')
648			ERR ("malformed number (leading zero must not be followed by another digit)");
649			}
650	root	1.5	else if (dec->cur < '0' \|\| dec->cur > '9')
651			ERR ("malformed number (no digits after initial minus)");
652			else
653			do
654			{
655			++dec->cur;
656			}
657			while (dec->cur >= '0' && dec->cur <= '9');
658	root	1.1
659			// [frac]
660			if (*dec->cur == '.')
661			{
662	root	1.5	++dec->cur;
663
664			if (dec->cur < '0' \|\| dec->cur > '9')
665			ERR ("malformed number (no digits after decimal point)");
666	root	1.1
667			do
668			{
669			++dec->cur;
670			}
671			while (dec->cur >= '0' && dec->cur <= '9');
672	root	1.5
673			is_nv = 1;
674	root	1.1	}
675
676			// [exp]
677			if (dec->cur == 'e' \|\| dec->cur == 'E')
678			{
679	root	1.5	++dec->cur;
680	root	1.1
681			if (dec->cur == '-' \|\| dec->cur == '+')
682			++dec->cur;
683
684	root	1.5	if (dec->cur < '0' \|\| dec->cur > '9')
685			ERR ("malformed number (no digits after exp sign)");
686
687			do
688			{
689			++dec->cur;
690			}
691			while (dec->cur >= '0' && dec->cur <= '9');
692
693			is_nv = 1;
694	root	1.1	}
695
696			if (!is_nv)
697			{
698			UV uv;
699			int numtype = grok_number (start, dec->cur - start, &uv);
700			if (numtype & IS_NUMBER_IN_UV)
701			if (numtype & IS_NUMBER_NEG)
702			{
703			if (uv < (UV)IV_MIN)
704			return newSViv (-(IV)uv);
705			}
706			else
707			return newSVuv (uv);
708			}
709
710			return newSVnv (Atof (start));
711
712			fail:
713			return 0;
714			}
715
716			static SV *
717			decode_av (dec_t *dec)
718			{
719			AV *av = newAV ();
720
721	root	1.5	WS;
722			if (*dec->cur == ']')
723			++dec->cur;
724			else
725			for (;;)
726			{
727			SV *value;
728	root	1.1
729	root	1.5	value = decode_sv (dec);
730			if (!value)
731			goto fail;
732	root	1.1
733	root	1.5	av_push (av, value);
734	root	1.1
735	root	1.5	WS;
736	root	1.1
737	root	1.5	if (*dec->cur == ']')
738			{
739			++dec->cur;
740			break;
741			}
742
743			if (*dec->cur != ',')
744			ERR (", or ] expected while parsing array");
745	root	1.1
746	root	1.5	++dec->cur;
747			}
748	root	1.1
749			return newRV_noinc ((SV *)av);
750
751			fail:
752			SvREFCNT_dec (av);
753			return 0;
754			}
755
756			static SV *
757			decode_hv (dec_t *dec)
758			{
759			HV *hv = newHV ();
760
761	root	1.5	WS;
762			if (*dec->cur == '}')
763			++dec->cur;
764			else
765			for (;;)
766			{
767			SV key, value;
768	root	1.1
769	root	1.5	WS; EXPECT_CH ('"');
770	root	1.1
771	root	1.5	key = decode_str (dec);
772			if (!key)
773			goto fail;
774	root	1.1
775	root	1.5	WS; EXPECT_CH (':');
776	root	1.1
777	root	1.5	value = decode_sv (dec);
778			if (!value)
779			{
780			SvREFCNT_dec (key);
781			goto fail;
782			}
783	root	1.1
784	root	1.5	//TODO: optimise
785			hv_store_ent (hv, key, value, 0);
786	root	1.1
787	root	1.5	WS;
788	root	1.1
789	root	1.5	if (*dec->cur == '}')
790			{
791			++dec->cur;
792			break;
793			}
794	root	1.1
795	root	1.5	if (*dec->cur != ',')
796			ERR (", or } expected while parsing object/hash");
797	root	1.1
798	root	1.5	++dec->cur;
799			}
800	root	1.1
801			return newRV_noinc ((SV *)hv);
802
803			fail:
804			SvREFCNT_dec (hv);
805			return 0;
806			}
807
808			static SV *
809			decode_sv (dec_t *dec)
810			{
811			WS;
812			switch (*dec->cur)
813			{
814			case '"': ++dec->cur; return decode_str (dec);
815			case '[': ++dec->cur; return decode_av (dec);
816			case '{': ++dec->cur; return decode_hv (dec);
817
818			case '-':
819			case '0': case '1': case '2': case '3': case '4':
820			case '5': case '6': case '7': case '8': case '9':
821			return decode_num (dec);
822
823			case 't':
824			if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4))
825			{
826			dec->cur += 4;
827			return newSViv (1);
828			}
829			else
830			ERR ("'true' expected");
831
832			break;
833
834			case 'f':
835			if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5))
836			{
837			dec->cur += 5;
838			return newSViv (0);
839			}
840			else
841			ERR ("'false' expected");
842
843			break;
844
845			case 'n':
846			if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4))
847			{
848			dec->cur += 4;
849	root	1.5	return newSVsv (&PL_sv_undef);
850	root	1.1	}
851			else
852			ERR ("'null' expected");
853
854			break;
855
856			default:
857	root	1.7	ERR ("malformed json string, neither array, object, number, string or atom");
858	root	1.1	break;
859			}
860
861			fail:
862			return 0;
863			}
864
865			static SV *
866			decode_json (SV *string, UV flags)
867			{
868			SV *sv;
869
870	root	1.5	if (flags & F_UTF8)
871			sv_utf8_downgrade (string, 0);
872			else
873	root	1.1	sv_utf8_upgrade (string);
874
875			SvGROW (string, SvCUR (string) + 1); // should basically be a NOP
876
877			dec_t dec;
878			dec.flags = flags;
879			dec.cur = SvPVX (string);
880			dec.end = SvEND (string);
881			dec.err = 0;
882
883			sv = decode_sv (&dec);
884
885			if (!sv)
886			{
887	root	1.7	IV offset = dec.flags & F_UTF8
888			? dec.cur - SvPVX (string)
889			: utf8_distance (dec.cur, SvPVX (string));
890	root	1.1	SV *uni = sv_newmortal ();
891	root	1.8
892	root	1.5	// horrible hack to silence warning inside pv_uni_display
893	root	1.8	COP cop = *PL_curcop;
894	root	1.5	cop.cop_warnings = pWARN_NONE;
895	root	1.8	ENTER;
896	root	1.5	SAVEVPTR (PL_curcop);
897			PL_curcop = &cop;
898	root	1.8	pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ);
899			LEAVE;
900	root	1.1
901	root	1.5	croak ("%s, at character offset %d (%s)",
902	root	1.1	dec.err,
903			(int)offset,
904			dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)");
905			}
906
907	root	1.3	sv = sv_2mortal (sv);
908
909			if (!(dec.flags & F_ALLOW_NONREF) && !SvROK (sv))
910	root	1.9	croak ("JSON text must be an object or array (but found number, string, true, false or null, use allow_nonref to allow this)");
911	root	1.3
912			return sv;
913	root	1.1	}
914
915			MODULE = JSON::XS PACKAGE = JSON::XS
916
917			BOOT:
918			{
919			int i;
920
921			memset (decode_hexdigit, 0xff, 256);
922			for (i = 10; i--; )
923			decode_hexdigit ['0' + i] = i;
924
925	root	1.4	for (i = 7; i--; )
926	root	1.1	{
927			decode_hexdigit ['a' + i] = 10 + i;
928			decode_hexdigit ['A' + i] = 10 + i;
929			}
930
931			json_stash = gv_stashpv ("JSON::XS", 1);
932			}
933
934	root	1.4	PROTOTYPES: DISABLE
935
936	root	1.1	SV new (char dummy)
937			CODE:
938			RETVAL = sv_bless (newRV_noinc (newSVuv (F_DEFAULT)), json_stash);
939			OUTPUT:
940			RETVAL
941
942	root	1.6	SV ascii (SV self, int enable = 1)
943	root	1.1	ALIAS:
944			ascii = F_ASCII
945			utf8 = F_UTF8
946			indent = F_INDENT
947			canonical = F_CANONICAL
948			space_before = F_SPACE_BEFORE
949			space_after = F_SPACE_AFTER
950	root	1.2	pretty = F_PRETTY
951	root	1.3	allow_nonref = F_ALLOW_NONREF
952	root	1.6	shrink = F_SHRINK
953	root	1.1	CODE:
954			{
955			UV *uv = SvJSON (self);
956			if (enable)
957			*uv \|= ix;
958			else
959			*uv &= ~ix;
960
961			RETVAL = newSVsv (self);
962			}
963			OUTPUT:
964			RETVAL
965
966			void encode (SV self, SV scalar)
967			PPCODE:
968			XPUSHs (encode_json (scalar, *SvJSON (self)));
969
970	root	1.2	void decode (SV self, SV jsonstr)
971	root	1.1	PPCODE:
972	root	1.2	XPUSHs (decode_json (jsonstr, *SvJSON (self)));
973
974	root	1.4	PROTOTYPES: ENABLE
975
976	root	1.2	void to_json (SV *scalar)
977			PPCODE:
978			XPUSHs (encode_json (scalar, F_UTF8));
979
980			void from_json (SV *jsonstr)
981			PPCODE:
982			XPUSHs (decode_json (jsonstr, F_UTF8));
983	root	1.1