JSON-XS/XS.xs

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

#include <assert.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <float.h>

#if defined(__BORLANDC__) || defined(_MSC_VER)
# define snprintf _snprintf // C compilers have this in stdio.h
#endif

// some old perls do not have this, try to make it work, no
// guarentees, though. if it breaks, you get to keep the pieces.
#ifndef UTF8_MAXBYTES
# define UTF8_MAXBYTES 13
#endif

#define F_ASCII          0x00000001UL
#define F_LATIN1         0x00000002UL
#define F_UTF8           0x00000004UL
#define F_INDENT         0x00000008UL
#define F_CANONICAL      0x00000010UL
#define F_SPACE_BEFORE   0x00000020UL
#define F_SPACE_AFTER    0x00000040UL
#define F_ALLOW_NONREF   0x00000100UL
#define F_SHRINK         0x00000200UL
#define F_ALLOW_BLESSED  0x00000400UL
#define F_CONV_BLESSED   0x00000800UL
#define F_RELAXED        0x00001000UL

#define F_MAXDEPTH       0xf8000000UL
#define S_MAXDEPTH       27
#define F_MAXSIZE        0x01f00000UL
#define S_MAXSIZE        20
#define F_HOOK           0x00080000UL // some hooks exist, so slow-path processing

#define DEC_DEPTH(flags) (1UL << ((flags & F_MAXDEPTH) >> S_MAXDEPTH))
#define DEC_SIZE(flags)  (1UL << ((flags & F_MAXSIZE ) >> S_MAXSIZE ))

#define F_PRETTY    F_INDENT | F_SPACE_BEFORE | F_SPACE_AFTER
#define F_DEFAULT   (9UL << S_MAXDEPTH)

#define INIT_SIZE   32 // initial scalar size to be allocated
#define INDENT_STEP 3  // spaces per indentation level

#define SHORT_STRING_LEN 16384 // special-case strings of up to this size

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

#if __GNUC__ >= 3
# define expect(expr,value)         __builtin_expect ((expr),(value))
# define inline                     inline
#else
# define expect(expr,value)         (expr)
# define inline                     static
#endif

#define expect_false(expr) expect ((expr) != 0, 0)
#define expect_true(expr)  expect ((expr) != 0, 1)

#ifdef USE_ITHREADS
# define JSON_SLOW 1
# define JSON_STASH (json_stash ? json_stash : gv_stashpv ("JSON::XS", 1))
#else
# define JSON_SLOW 0
# define JSON_STASH json_stash
#endif

static HV *json_stash, *json_boolean_stash; // JSON::XS::
static SV *json_true, *json_false;

typedef struct {
  U32 flags;
  SV *cb_object;
  HV *cb_sk_object;
} JSON;

/////////////////////////////////////////////////////////////////////////////
// utility functions

inline void
shrink (SV *sv)
{
  sv_utf8_downgrade (sv, 1);
  if (SvLEN (sv) > SvCUR (sv) + 1)
    {
#ifdef SvPV_shrink_to_cur
      SvPV_shrink_to_cur (sv);
#elif defined (SvPV_renew)
      SvPV_renew (sv, SvCUR (sv) + 1);
#endif
    }
}

// decode an utf-8 character and return it, or (UV)-1 in
// case of an error.
// we special-case "safe" characters from U+80 .. U+7FF,
// but use the very good perl function to parse anything else.
// note that we never call this function for a ascii codepoints
inline UV
decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen)
{
  if (expect_false (s[0] > 0xdf || s[0] < 0xc2))
    return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY);
  else if (len > 1 && s[1] >= 0x80 && s[1] <= 0xbf)
    {
      *clen = 2;
      return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f);
    }
  else
    {
      *clen = (STRLEN)-1;
      return (UV)-1;
    }
}

/////////////////////////////////////////////////////////////////////////////
// encoder

// structure used for encoding JSON
typedef struct
{
  char *cur;  // SvPVX (sv) + current output position
  char *end;  // SvEND (sv)
  SV *sv;     // result scalar
  JSON json;
  U32 indent; // indentation level
  U32 maxdepth; // max. indentation/recursion level
} enc_t;

inline void
need (enc_t *enc, STRLEN len)
{
  if (expect_false (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) - 1;
    }
}

inline 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 (expect_true (ch >= 0x20 && ch < 0x80)) // most common case
        {
          if (expect_false (ch == '"')) // but with slow exceptions
            {
              need (enc, len += 1);
              *enc->cur++ = '\\';
              *enc->cur++ = '"';
            }
          else if (expect_false (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 = decode_utf8 (str, end - str, &clen);
                      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->json.flags & F_ASCII || (enc->json.flags & F_LATIN1 && uch > 0xFF))
                    {
                      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 (enc->json.flags & F_LATIN1)
                    {
                      *enc->cur++ = uch;
                      str += clen;
                    }
                  else if (is_utf8)
                    {
                      need (enc, len += clen);
                      do
                        {
                          *enc->cur++ = *str++;
                        }
                      while (--clen);
                    }
                  else
                    {
                      need (enc, len += UTF8_MAXBYTES - 1); // never more than 11 bytes needed
                      enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);
                      ++str;
                    }
                }
            }
        }

      --len;
    }
}

inline void
encode_indent (enc_t *enc)
{
  if (enc->json.flags & F_INDENT)
    {
      int spaces = enc->indent * INDENT_STEP;

      need (enc, spaces);
      memset (enc->cur, ' ', spaces);
      enc->cur += spaces;
    }
}

inline void
encode_space (enc_t *enc)
{
  need (enc, 1);
  encode_ch (enc, ' ');
}

inline void
encode_nl (enc_t *enc)
{
  if (enc->json.flags & F_INDENT)
    {
      need (enc, 1);
      encode_ch (enc, '\n');
    }
}

inline void
encode_comma (enc_t *enc)
{
  encode_ch (enc, ',');

  if (enc->json.flags & F_INDENT)
    encode_nl (enc);
  else if (enc->json.flags & F_SPACE_AFTER)
    encode_space (enc);
}

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

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

  if (enc->indent >= enc->maxdepth)
    croak ("data structure too deep (hit recursion limit)");

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

  for (i = 0; i <= len; ++i)
    {
      SV **svp = av_fetch (av, i, 0);

      encode_indent (enc);

      if (svp)
        encode_sv (enc, *svp);
      else
        encode_str (enc, "null", 4, 0);

      if (i < len)
        encode_comma (enc);
    }

  encode_nl (enc);

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

static void
encode_hk (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->json.flags & F_SPACE_BEFORE) encode_space (enc);
  encode_ch (enc, ':');
  if (enc->json.flags & F_SPACE_AFTER ) encode_space (enc);
}

// 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 (b), HeKEY (a), lb < la ? lb : la)))
    cmp = lb - la;

  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 **)b), HeSVKEY_force (*(HE **)a));
}

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

  if (enc->indent >= enc->maxdepth)
    croak ("data structure too deep (hit recursion limit)");

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

  // 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->json.flags & F_CANONICAL)
    {
      int count = hv_iterinit (hv);

      if (SvMAGICAL (hv))
        {
          // need to count by iterating. could improve by dynamically building the vector below
          // but I don't care for the speed of this special case.
          // note also that we will run into undefined behaviour when the two iterations
          // do not result in the same count, something I might care for in some later release.

          count = 0;
          while (hv_iternext (hv))
            ++count;

          hv_iterinit (hv);
        }

      if (count)
        {
          int i, fast = 1;
#if defined(__BORLANDC__) || defined(_MSC_VER)
          HE **hes = _alloca (count * sizeof (HE));
#else
          HE *hes [count]; // if your compiler dies here, you need to enable C99 mode
#endif

          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;
            }

          while (count--)
            {
              encode_indent (enc);
              he = hes [count];
              encode_hk (enc, he);
              encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he));

              if (count)
                encode_comma (enc);
            }
        }
    }
  else
    {
      if (hv_iterinit (hv) || SvMAGICAL (hv))
        if ((he = hv_iternext (hv)))
          for (;;)
            {
              encode_indent (enc);
              encode_hk (enc, he);
              encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he));

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

              encode_comma (enc);
            }
    }

  encode_nl (enc);

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

// encode objects, arrays and special \0=false and \1=true values.
static void
encode_rv (enc_t *enc, SV *sv)
{
  svtype svt;

  SvGETMAGIC (sv);
  svt = SvTYPE (sv);

  if (expect_false (SvOBJECT (sv)))
    {
      HV *stash = !JSON_SLOW || json_boolean_stash
                  ? json_boolean_stash
                  : gv_stashpv ("JSON::XS::Boolean", 1);

      if (SvSTASH (sv) == stash)
        {
          if (SvIV (sv))
            encode_str (enc, "true", 4, 0);
          else
            encode_str (enc, "false", 5, 0);
        }
      else
        {
#if 0
          if (0 && sv_derived_from (rv, "JSON::Literal"))
            {
              // not yet
            }
#endif
          if (enc->json.flags & F_CONV_BLESSED)
            {
              // we re-bless the reference to get overload and other niceties right
              GV *to_json = gv_fetchmethod_autoload (SvSTASH (sv), "TO_JSON", 0);

              if (to_json)
                {
                  dSP;

                  ENTER; SAVETMPS; PUSHMARK (SP);
                  XPUSHs (sv_bless (sv_2mortal (newRV_inc (sv)), SvSTASH (sv)));

                  // calling with G_SCALAR ensures that we always get a 1 return value
                  PUTBACK;
                  call_sv ((SV *)GvCV (to_json), G_SCALAR);
                  SPAGAIN;

                  // catch this surprisingly common error
                  if (SvROK (TOPs) && SvRV (TOPs) == sv)
                    croak ("%s::TO_JSON method returned same object as was passed instead of a new one", HvNAME (SvSTASH (sv)));

                  sv = POPs;
                  PUTBACK;

                  encode_sv (enc, sv);

                  FREETMPS; LEAVE;
                }
              else if (enc->json.flags & F_ALLOW_BLESSED)
                encode_str (enc, "null", 4, 0);
              else
                croak ("encountered object '%s', but neither allow_blessed enabled nor TO_JSON method available on it",
                       SvPV_nolen (sv_2mortal (newRV_inc (sv))));
            }
          else if (enc->json.flags & F_ALLOW_BLESSED)
            encode_str (enc, "null", 4, 0);
          else
            croak ("encountered object '%s', but neither allow_blessed nor convert_blessed settings are enabled",
                   SvPV_nolen (sv_2mortal (newRV_inc (sv))));
        }
    }
  else if (svt == SVt_PVHV)
    encode_hv (enc, (HV *)sv);
  else if (svt == SVt_PVAV)
    encode_av (enc, (AV *)sv);
  else if (svt < SVt_PVAV)
    {
      STRLEN len = 0;
      char *pv = svt ? SvPV (sv, len) : 0;

      if (len == 1 && *pv == '1')
        encode_str (enc, "true", 4, 0);
      else if (len == 1 && *pv == '0')
        encode_str (enc, "false", 5, 0);
      else
        croak ("cannot encode reference to scalar '%s' unless the scalar is 0 or 1",
               SvPV_nolen (sv_2mortal (newRV_inc (sv))));
    }
  else
    croak ("encountered %s, but JSON can only represent references to arrays or hashes",
           SvPV_nolen (sv_2mortal (newRV_inc (sv))));
}

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))
    {
      // trust that perl will do the right thing w.r.t. JSON syntax.
      need (enc, NV_DIG + 32);
      Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
      enc->cur += strlen (enc->cur);
    }
  else if (SvIOKp (sv))
    {
      // we assume we can always read an IV as a UV
      if (SvUV (sv) & ~(UV)0x7fff)
        {
          // large integer, use the (rather slow) snprintf way.
          need (enc, sizeof (UV) * 3);
          enc->cur += 
             SvIsUV(sv)
                ? snprintf (enc->cur, sizeof (UV) * 3, "%"UVuf, (UV)SvUVX (sv))
                : snprintf (enc->cur, sizeof (UV) * 3, "%"IVdf, (IV)SvIVX (sv));
        }
      else
        {
          // optimise the "small number case"
          // code will likely be branchless and use only a single multiplication
          I32 i = SvIV (sv);
          U32 u;
          char digit, nz = 0;

          need (enc, 6);

          *enc->cur = '-'; enc->cur += i < 0 ? 1 : 0;
          u = i < 0 ? -i : i;

          // convert to 4.28 fixed-point representation
          u = u * ((0xfffffff + 10000) / 10000); // 10**5, 5 fractional digits

          // now output digit by digit, each time masking out the integer part
          // and multiplying by 5 while moving the decimal point one to the right,
          // resulting in a net multiplication by 10.
          // we always write the digit to memory but conditionally increment
          // the pointer, to ease the usage of conditional move instructions.
          digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffff) * 5;
          digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffff) * 5;
          digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffff) * 5;
          digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffff) * 5;
          digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0'
        }
    }
  else if (SvROK (sv))
    encode_rv (enc, SvRV (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, JSON *json)
{
  enc_t enc;

  if (!(json->flags & F_ALLOW_NONREF) && !SvROK (scalar))
    croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)");

  enc.json      = *json;
  enc.sv        = sv_2mortal (NEWSV (0, INIT_SIZE));
  enc.cur       = SvPVX (enc.sv);
  enc.end       = SvEND (enc.sv);
  enc.indent    = 0;
  enc.maxdepth  = DEC_DEPTH (enc.json.flags);

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

  SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv));
  *SvEND (enc.sv) = 0; // many xs functions expect a trailing 0 for text strings

  if (!(enc.json.flags & (F_ASCII | F_LATIN1 | F_UTF8)))
    SvUTF8_on (enc.sv);

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

  return enc.sv;
}

/////////////////////////////////////////////////////////////////////////////
// decoder

// structure used for decoding JSON
typedef struct
{
  char *cur; // current parser pointer
  char *end; // end of input string
  const char *err; // parse error, if != 0
  JSON json;
  U32 depth; // recursion depth
  U32 maxdepth; // recursion depth limit
} dec_t;

inline void
decode_ws (dec_t *dec)
{
  for (;;)
    {
      char ch = *dec->cur;

      if (ch > 0x20
          || (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)) 
        break;

      if (ch == '#' && dec->json.flags & F_RELAXED)
        ++dec->cur;

      ++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

#define DEC_INC_DEPTH if (++dec->depth > dec->maxdepth) ERR ("json datastructure exceeds maximum nesting level (set a higher max_depth)")
#define DEC_DEC_DEPTH --dec->depth

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;
  unsigned char *cur = (unsigned char *)dec->cur;

  d1 = decode_hexdigit [cur [0]]; if (expect_false (d1 < 0)) ERR ("exactly four hexadecimal digits expected");
  d2 = decode_hexdigit [cur [1]]; if (expect_false (d2 < 0)) ERR ("exactly four hexadecimal digits expected");
  d3 = decode_hexdigit [cur [2]]; if (expect_false (d3 < 0)) ERR ("exactly four hexadecimal digits expected");
  d4 = decode_hexdigit [cur [3]]; if (expect_false (d4 < 0)) ERR ("exactly four hexadecimal digits expected");

  dec->cur += 4;

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

fail:
  return (UV)-1;
}

static SV *
decode_str (dec_t *dec)
{
  SV *sv = 0;
  int utf8 = 0;
  char *dec_cur = dec->cur;

  do
    {
      char buf [SHORT_STRING_LEN + UTF8_MAXBYTES];
      char *cur = buf;

      do
        {
          unsigned char ch = *(unsigned char *)dec_cur++;

          if (expect_false (ch == '"'))
            {
              --dec_cur;
              break;
            }
          else if (expect_false (ch == '\\'))
            {
              switch (*dec_cur)
                {
                  case '\\':
                  case '/':
                  case '"': *cur++ = *dec_cur++; break;

                  case 'b': ++dec_cur; *cur++ = '\010'; break;
                  case 't': ++dec_cur; *cur++ = '\011'; break;
                  case 'n': ++dec_cur; *cur++ = '\012'; break;
                  case 'f': ++dec_cur; *cur++ = '\014'; break;
                  case 'r': ++dec_cur; *cur++ = '\015'; break;

                  case 'u':
                    {
                      UV lo, hi;
                      ++dec_cur;

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

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

                            dec_cur += 2;

                            dec->cur = dec_cur;
                            lo = decode_4hex (dec);
                            dec_cur = dec->cur;
                            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 < 0xe000)
                          ERR ("missing high surrogate character in surrogate pair");

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

                          cur = (char *)uvuni_to_utf8_flags (cur, hi, 0);
                        }
                      else
                        *cur++ = hi;
                    }
                    break;

                  default:
                    --dec_cur;
                    ERR ("illegal backslash escape sequence in string");
                }
            }
          else if (expect_true (ch >= 0x20 && ch <= 0x7f))
            *cur++ = ch;
          else if (ch >= 0x80)
            {
              STRLEN clen;
              UV uch;

              --dec_cur;

              uch = decode_utf8 (dec_cur, dec->end - dec_cur, &clen);
              if (clen == (STRLEN)-1)
                ERR ("malformed UTF-8 character in JSON string");

              do
                *cur++ = *dec_cur++;
              while (--clen);

              utf8 = 1;
            }
          else
            {
              --dec_cur;

              if (!ch)
                ERR ("unexpected end of string while parsing JSON string");
              else
                ERR ("invalid character encountered while parsing JSON string");
            }
        }
      while (cur < buf + SHORT_STRING_LEN);

      {
        STRLEN len = cur - buf;

        if (sv)
          {
            SvGROW (sv, SvCUR (sv) + len + 1);
            memcpy (SvPVX (sv) + SvCUR (sv), buf, len);
            SvCUR_set (sv, SvCUR (sv) + len);
          }
        else
          sv = newSVpvn (buf, len);
      }
    }
  while (*dec_cur != '"');

  ++dec_cur;

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

      if (utf8)
        SvUTF8_on (sv);
    }
  else
    sv = newSVpvn ("", 0);

  dec->cur = dec_cur;
  return sv;

fail:
  dec->cur = dec_cur;
  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)
    {
      int len = dec->cur - start;

      // special case the rather common 1..4-digit-int case, assumes 32 bit ints or so
      if (*start == '-')
        switch (len)
          {
            case 2: return newSViv (-(                                                      start [1] - '0' *    1));
            case 3: return newSViv (-(                                     start [1] * 10 + start [2] - '0' *   11));
            case 4: return newSViv (-(                   start [1] * 100 + start [2] * 10 + start [3] - '0' *  111));
            case 5: return newSViv (-(start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 1111));
          }
      else
        switch (len)
          {
            case 1: return newSViv (                                                        start [0] - '0' *    1);
            case 2: return newSViv (                                       start [0] * 10 + start [1] - '0' *   11);
            case 3: return newSViv (                     start [0] * 100 + start [1] * 10 + start [2] - '0' *  111);
            case 4: return newSViv (  start [0] * 1000 + start [1] * 100 + start [2] * 10 + start [3] - '0' * 1111);
          }

      {
        UV uv;
        int numtype = grok_number (start, len, &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);
      }

      len -= *start == '-' ? 1 : 0;

      // does not fit into IV or UV, try NV
      if ((sizeof (NV) == sizeof (double) && DBL_DIG >= len)
          #if defined (LDBL_DIG)
          || (sizeof (NV) == sizeof (long double) && LDBL_DIG >= len)
          #endif
         )
        // fits into NV without loss of precision
        return newSVnv (Atof (start));

      // everything else fails, convert it to a string
      return newSVpvn (start, dec->cur - start);
    }

  // loss of precision here
  return newSVnv (Atof (start));

fail:
  return 0;
}

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

  DEC_INC_DEPTH;
  decode_ws (dec);

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

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

        av_push (av, value);

        decode_ws (dec);

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

        ++dec->cur;

        decode_ws (dec);

        if (*dec->cur == ']' && dec->json.flags & F_RELAXED)
          {
            ++dec->cur;
            break;
          }
      }

  DEC_DEC_DEPTH;
  return newRV_noinc ((SV *)av);

fail:
  SvREFCNT_dec (av);
  DEC_DEC_DEPTH;
  return 0;
}

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

  DEC_INC_DEPTH;
  decode_ws (dec);

  if (*dec->cur == '}')
    ++dec->cur;
  else
    for (;;)
      {
        EXPECT_CH ('"');

        // heuristic: assume that
        // a) decode_str + hv_store_ent are abysmally slow.
        // b) most hash keys are short, simple ascii text.
        // => try to "fast-match" such strings to avoid
        // the overhead of decode_str + hv_store_ent.
        {
          SV *value;
          char *p = dec->cur;
          char *e = p + 24; // only try up to 24 bytes

          for (;;)
            {
              // the >= 0x80 is true on most architectures
              if (p == e || *p < 0x20 || *p >= 0x80 || *p == '\\')
                {
                  // slow path, back up and use decode_str
                  SV *key = decode_str (dec);
                  if (!key)
                    goto fail;

                  decode_ws (dec); EXPECT_CH (':');

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

                  hv_store_ent (hv, key, value, 0);
                  SvREFCNT_dec (key);

                  break;
                }
              else if (*p == '"')
                {
                  // fast path, got a simple key
                  char *key = dec->cur;
                  int len = p - key;
                  dec->cur = p + 1;

                  decode_ws (dec); EXPECT_CH (':');

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

                  hv_store (hv, key, len, value, 0);

                  break;
                }

              ++p;
            }
        }

        decode_ws (dec);

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

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

        ++dec->cur;

        decode_ws (dec);

        if (*dec->cur == '}' && dec->json.flags & F_RELAXED)
          {
            ++dec->cur;
            break;
          }
      }

  DEC_DEC_DEPTH;
  sv = newRV_noinc ((SV *)hv);

  // check filter callbacks
  if (dec->json.flags & F_HOOK)
    {
      if (dec->json.cb_sk_object && HvKEYS (hv) == 1)
        {
          HE *cb, *he;

          hv_iterinit (hv);
          he = hv_iternext (hv);
          hv_iterinit (hv);

          // the next line creates a mortal sv each time its called.
          // might want to optimise this for common cases.
          cb = hv_fetch_ent (dec->json.cb_sk_object, hv_iterkeysv (he), 0, 0);

          if (cb)
            {
              dSP;
              int count;

              ENTER; SAVETMPS; PUSHMARK (SP);
              XPUSHs (HeVAL (he));

              PUTBACK; count = call_sv (HeVAL (cb), G_ARRAY); SPAGAIN;

              if (count == 1)
                {
                  sv = newSVsv (POPs);
                  FREETMPS; LEAVE;
                  return sv;
                }

              FREETMPS; LEAVE;
            }
        }

      if (dec->json.cb_object)
        {
          dSP;
          int count;

          ENTER; SAVETMPS; PUSHMARK (SP);
          XPUSHs (sv_2mortal (sv));

          PUTBACK; count = call_sv (dec->json.cb_object, G_ARRAY); SPAGAIN;

          if (count == 1)
            {
              sv = newSVsv (POPs);
              FREETMPS; LEAVE;
              return sv;
            }

          SvREFCNT_inc (sv);
          FREETMPS; LEAVE;
        }
    }

  return sv;

fail:
  SvREFCNT_dec (hv);
  DEC_DEC_DEPTH;
  return 0;
}

static SV *
decode_sv (dec_t *dec)
{
  // the beauty of JSON: you need exactly one character lookahead
  // to parse anything.
  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;
#if JSON_SLOW
            json_true = get_sv ("JSON::XS::true", 1); SvREADONLY_on (json_true);
#endif
            return SvREFCNT_inc (json_true);
          }
        else
          ERR ("'true' expected");

        break;

      case 'f':
        if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5))
          {
            dec->cur += 5;
#if JSON_SLOW
            json_false = get_sv ("JSON::XS::false", 1); SvREADONLY_on (json_false);
#endif
            return SvREFCNT_inc (json_false);
          }
        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, JSON *json, UV *offset_return)
{
  dec_t dec;
  UV offset;
  SV *sv;

  SvGETMAGIC (string);
  SvUPGRADE (string, SVt_PV);

  if (json->flags & F_MAXSIZE && SvCUR (string) > DEC_SIZE (json->flags))
    croak ("attempted decode of JSON text of %lu bytes size, but max_size is set to %lu",
           (unsigned long)SvCUR (string), (unsigned long)DEC_SIZE (json->flags));

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

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

  dec.json     = *json;
  dec.cur      = SvPVX (string);
  dec.end      = SvEND (string);
  dec.err      = 0;
  dec.depth    = 0;
  dec.maxdepth = DEC_DEPTH (dec.json.flags);

  if (dec.json.cb_object || dec.json.cb_sk_object)
    dec.json.flags |= F_HOOK;

  *dec.end = 0; // this should basically be a nop, too, but make sure it's there

  decode_ws (&dec);
  sv = decode_sv (&dec);

  if (!(offset_return || !sv))
    {
      // check for trailing garbage
      decode_ws (&dec);

      if (*dec.cur)
        {
          dec.err = "garbage after JSON object";
          SvREFCNT_dec (sv);
          sv = 0;
        }
    }

  if (offset_return || !sv)
    {
      offset = dec.json.flags & F_UTF8
               ? dec.cur - SvPVX (string)
               : utf8_distance (dec.cur, SvPVX (string));

      if (offset_return)
        *offset_return = offset;
    }

  if (!sv)
    {
      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.json.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;
}

/////////////////////////////////////////////////////////////////////////////
// XS interface functions

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

BOOT:
{
        int i;

        for (i = 0; i < 256; ++i)
          decode_hexdigit [i] =
            i >= '0' && i <= '9' ? i - '0'
            : i >= 'a' && i <= 'f' ? i - 'a' + 10
            : i >= 'A' && i <= 'F' ? i - 'A' + 10
            : -1;

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

        json_true  = get_sv ("JSON::XS::true" , 1); SvREADONLY_on (json_true );
        json_false = get_sv ("JSON::XS::false", 1); SvREADONLY_on (json_false);
}

PROTOTYPES: DISABLE

void CLONE (...)
        CODE:
        json_stash         = 0;
        json_boolean_stash = 0;

void new (char *klass)
        PPCODE:
{
        SV *pv = NEWSV (0, sizeof (JSON));
        SvPOK_only (pv);
        Zero (SvPVX (pv), 1, JSON);
        ((JSON *)SvPVX (pv))->flags = F_DEFAULT;
        XPUSHs (sv_2mortal (sv_bless (newRV_noinc (pv), JSON_STASH)));
}

void ascii (JSON *self, int enable = 1)
        ALIAS:
        ascii           = F_ASCII
        latin1          = F_LATIN1
        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
        allow_blessed   = F_ALLOW_BLESSED
        convert_blessed = F_CONV_BLESSED
        relaxed         = F_RELAXED
        PPCODE:
{
        if (enable)
          self->flags |=  ix;
        else
          self->flags &= ~ix;

        XPUSHs (ST (0));
}

void max_depth (JSON *self, UV max_depth = 0x80000000UL)
        PPCODE:
{
        UV log2 = 0;

        if (max_depth > 0x80000000UL) max_depth = 0x80000000UL;

        while ((1UL << log2) < max_depth)
          ++log2;

        self->flags = self->flags & ~F_MAXDEPTH | (log2 << S_MAXDEPTH);

        XPUSHs (ST (0));
}

void max_size (JSON *self, UV max_size = 0)
        PPCODE:
{
        UV log2 = 0;

        if (max_size > 0x80000000UL) max_size = 0x80000000UL;
        if (max_size == 1)           max_size = 2;

        while ((1UL << log2) < max_size)
          ++log2;

        self->flags = self->flags & ~F_MAXSIZE | (log2 << S_MAXSIZE);

        XPUSHs (ST (0));
}

void filter_json_object (JSON *self, SV *cb = &PL_sv_undef)
        PPCODE:
{
        SvREFCNT_dec (self->cb_object);
        self->cb_object = SvOK (cb) ? newSVsv (cb) : 0;

        XPUSHs (ST (0));
}

void filter_json_single_key_object (JSON *self, SV *key, SV *cb = &PL_sv_undef)
        PPCODE:
{
        if (!self->cb_sk_object)
          self->cb_sk_object = newHV ();

        if (SvOK (cb))
          hv_store_ent (self->cb_sk_object, key, newSVsv (cb), 0);
        else
          {
            hv_delete_ent (self->cb_sk_object, key, G_DISCARD, 0);

            if (!HvKEYS (self->cb_sk_object))
              {
                SvREFCNT_dec (self->cb_sk_object);
                self->cb_sk_object = 0;
              }
          }

        XPUSHs (ST (0));
}

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

void decode (JSON *self, SV *jsonstr)
        PPCODE:
        XPUSHs (decode_json (jsonstr, self, 0));

void decode_prefix (JSON *self, SV *jsonstr)
        PPCODE:
{
        UV offset;
        EXTEND (SP, 2);
        PUSHs (decode_json (jsonstr, self, &offset));
        PUSHs (sv_2mortal (newSVuv (offset)));
}

void DESTROY (JSON *self)
        CODE:
        SvREFCNT_dec (self->cb_sk_object);
        SvREFCNT_dec (self->cb_object);

PROTOTYPES: ENABLE

void to_json (SV *scalar)
        PPCODE:
{
        JSON json = { F_DEFAULT | F_UTF8 };
        XPUSHs (encode_json (scalar, &json));
}

void from_json (SV *jsonstr)
        PPCODE:
{
        JSON json = { F_DEFAULT | F_UTF8 };
        XPUSHs (decode_json (jsonstr, &json, 0));
}

Revision:	1.63
Committed:	Mon Aug 27 01:49:01 2007 UTC (16 years, 8 months ago) by root
Branch:	MAIN
Changes since 1.62:	+27 -3 lines
Log Message:	* empty log message *