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_JSON_RPC     0x00000040
#define F_ALLOW_NONREF 0x00000080

#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;
  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
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 = SvEND (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;

  while (str < end)
    {
      unsigned char ch = *(unsigned char *)str;
      if (ch >= 0x20 && ch < 0x80) // most common case
        {
          *enc->cur++ = ch;
          str++;
        }
      else
        {
          STRLEN clen;
          UV uch;

          if (is_utf8)
            {
              uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
              if (clen < 0)
                croak ("malformed UTF-8 character in string, cannot convert to JSON");
            }
          else
            {
              uch = ch;
              clen = 1;
            }

          need (enc, len += 6);

          if (uch < 0xa0 || enc->flags & F_ASCII)
            {
              if (uch > 0xFFFFUL)
                {
                  len += 6;
                  need (enc, len += 6);
                  sprintf (enc->cur, "\\u%04x\\u%04x",
                           (uch - 0x10000) / 0x400 + 0xD800,
                           (uch - 0x10000) % 0x400 + 0xDC00);
                  enc->cur += 12;
                }
              else
                {
                  sprintf (enc->cur, "\\u%04x", uch);
                  enc->cur += 6;
                }
            }
          else if (is_utf8)
            {
              memcpy (enc->cur, str, clen);
              enc->cur += clen;
            }
          else
            enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);

          str += clen;
        }

      --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 = 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 disable "use bytes"
              COP *oldcop = PL_curcop, cop;
              cop.op_private = 0;
              PL_curcop = &cop;

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

              PL_curcop = oldcop;
            }

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

      sv = SvRV (sv);

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

          default:
            croak ("JSON can only represent references to arrays or hashes");
        }
    }
  else if (!SvOK (sv))
    encode_str (enc, "null", 4, 0);
  else
    croak ("encountered perl type that JSON cannot handle");
}

static SV *
encode_json (SV *scalar, UV flags)
{
  if (!(flags & F_ALLOW_NONREF) && !SvROK (scalar))
    croak ("hash- or arraref required (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));
  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);

#define APPEND_CH(ch) SB \
  SvGROW (sv, cur + 1 + 1);     \
  SvPVX (sv)[cur++] = (ch);     \
  SE

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;
}

static SV *
decode_str (dec_t *dec)
{
  SV *sv = NEWSV (0,2);
  STRLEN cur = 0;
  int utf8 = 0;

  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 ("illegal surrogate character");

                      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 (lo >= 0xdc00 && lo < 0xe000)
                    ERR ("illegal surrogate character");

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

                      SvGROW (sv, cur + 4 + 1); // at most 4 bytes for 21 bits
                      cur = (char *)uvuni_to_utf8_flags (SvPVX (sv) + cur, hi, 0) - SvPVX (sv);
                    }
                  else
                    APPEND_CH (hi);
                }
                break;
            }
        }
      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 < 0)
            ERR ("malformed UTF-8 character in string, cannot convert to JSON");

          SvGROW (sv, cur + clen + 1); // at most 4 bytes for 21 bits
          memcpy (SvPVX (sv) + cur, dec->cur, clen);
          dec->cur += clen;
        }
      else
        ERR ("invalid character encountered");
    }

  ++dec->cur;

  SvPOK_only (sv);

  SvCUR_set (sv, cur);
  *SvEND (sv) = 0;

  if (utf8)
    SvUTF8_on (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)");
    }

  // int
  while (*dec->cur >= '0' && *dec->cur <= '9')
    ++dec->cur;

  // [frac]
  if (*dec->cur == '.')
    {
      is_nv = 1;

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

  // [exp]
  if (*dec->cur == 'e' || *dec->cur == 'E')
    {
      is_nv = 1;

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

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

  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 ();

  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 ();

  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 newSViv (1);
          }
        else
          ERR ("'null' expected");

        break;

      default:
        ERR ("malformed json string");
        break;
    }

fail:
  return 0;
}

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

  if (!(flags & F_UTF8))
    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;

  *dec.end = 1; // invalid anywhere
  sv = decode_sv (&dec);
  *dec.end = 0;

  if (!sv)
    {
      IV offset = utf8_distance (dec.cur, SvPVX (string));
      SV *uni = sv_newmortal ();

      pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ);
      croak ("%s, at character %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 object or array expected (but number, string, true, false or null found, 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 = 6; --i; )
          {
            decode_hexdigit ['a' + i] = 10 + i;
            decode_hexdigit ['A' + i] = 10 + i;
          }

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

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

SV *ascii (SV *self, int enable)
        ALIAS:
        ascii        = F_ASCII
        utf8         = F_UTF8
        indent       = F_INDENT
        canonical    = F_CANONICAL
        space_before = F_SPACE_BEFORE
        space_after  = F_SPACE_AFTER
        json_rpc     = F_JSON_RPC
        pretty       = F_PRETTY
        allow_nonref = F_ALLOW_NONREF
        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)));

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:	1.3
Committed:	Thu Mar 22 18:10:29 2007 UTC (17 years, 2 months ago) by root
Branch:	MAIN
Changes since 1.2:	+11 -2 lines
Log Message:	rant
#	Content
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	#define F_JSON_RPC 0x00000040
16	#define F_ALLOW_NONREF 0x00000080
17
18	#define F_PRETTY F_INDENT \| F_SPACE_BEFORE \| F_SPACE_AFTER
19	#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	char *err;
46	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	/////////////////////////////////////////////////////////////////////////////
59
60	static void
61	need (enc_t *enc, STRLEN len)
62	{
63	if (enc->cur + len >= enc->end)
64	{
65	STRLEN cur = enc->cur - SvPVX (enc->sv);
66	SvGROW (enc->sv, cur + len + 1);
67	enc->cur = SvPVX (enc->sv) + cur;
68	enc->end = SvEND (enc->sv);
69	}
70	}
71
72	static void
73	encode_ch (enc_t *enc, char ch)
74	{
75	need (enc, 1);
76	*enc->cur++ = ch;
77	}
78
79	static void
80	encode_str (enc_t enc, char str, STRLEN len, int is_utf8)
81	{
82	char *end = str + len;
83
84	while (str < end)
85	{
86	unsigned char ch = (unsigned char )str;
87	if (ch >= 0x20 && ch < 0x80) // most common case
88	{
89	*enc->cur++ = ch;
90	str++;
91	}
92	else
93	{
94	STRLEN clen;
95	UV uch;
96
97	if (is_utf8)
98	{
99	uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
100	if (clen < 0)
101	croak ("malformed UTF-8 character in string, cannot convert to JSON");
102	}
103	else
104	{
105	uch = ch;
106	clen = 1;
107	}
108
109	need (enc, len += 6);
110
111	if (uch < 0xa0 \|\| enc->flags & F_ASCII)
112	{
113	if (uch > 0xFFFFUL)
114	{
115	len += 6;
116	need (enc, len += 6);
117	sprintf (enc->cur, "\\u%04x\\u%04x",
118	(uch - 0x10000) / 0x400 + 0xD800,
119	(uch - 0x10000) % 0x400 + 0xDC00);
120	enc->cur += 12;
121	}
122	else
123	{
124	sprintf (enc->cur, "\\u%04x", uch);
125	enc->cur += 6;
126	}
127	}
128	else if (is_utf8)
129	{
130	memcpy (enc->cur, str, clen);
131	enc->cur += clen;
132	}
133	else
134	enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);
135
136	str += clen;
137	}
138
139	--len;
140	}
141	}
142
143	#define INDENT SB \
144	if (enc->flags & F_INDENT) \
145	{ \
146	int i_; \
147	need (enc, enc->indent); \
148	for (i_ = enc->indent * 3; i_--; )\
149	encode_ch (enc, ' '); \
150	} \
151	SE
152
153	#define SPACE SB need (enc, 1); encode_ch (enc, ' '); SE
154	#define NL SB if (enc->flags & F_INDENT) { need (enc, 1); encode_ch (enc, '\n'); } SE
155	#define COMMA SB \
156	encode_ch (enc, ','); \
157	if (enc->flags & F_INDENT) \
158	NL; \
159	else if (enc->flags & F_SPACE_AFTER) \
160	SPACE; \
161	SE
162
163	static void encode_sv (enc_t enc, SV sv);
164
165	static void
166	encode_av (enc_t enc, AV av)
167	{
168	int i, len = av_len (av);
169
170	encode_ch (enc, '['); NL;
171	++enc->indent;
172
173	for (i = 0; i <= len; ++i)
174	{
175	INDENT;
176	encode_sv (enc, *av_fetch (av, i, 0));
177
178	if (i < len)
179	COMMA;
180	}
181
182	NL;
183
184	--enc->indent;
185	INDENT; encode_ch (enc, ']');
186	}
187
188	static void
189	encode_he (enc_t enc, HE he)
190	{
191	encode_ch (enc, '"');
192
193	if (HeKLEN (he) == HEf_SVKEY)
194	{
195	SV *sv = HeSVKEY (he);
196	STRLEN len;
197	char *str = SvPV (sv, len);
198
199	encode_str (enc, str, len, SvUTF8 (sv));
200	}
201	else
202	encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he));
203
204	encode_ch (enc, '"');
205
206	if (enc->flags & F_SPACE_BEFORE) SPACE;
207	encode_ch (enc, ':');
208	if (enc->flags & F_SPACE_AFTER ) SPACE;
209	encode_sv (enc, HeVAL (he));
210	}
211
212	// compare hash entries, used when all keys are bytestrings
213	static int
214	he_cmp_fast (const void a_, const void b_)
215	{
216	int cmp;
217
218	HE a = (HE **)a_;
219	HE b = (HE **)b_;
220
221	STRLEN la = HeKLEN (a);
222	STRLEN lb = HeKLEN (b);
223
224	if (!(cmp == memcmp (HeKEY (a), HeKEY (b), la < lb ? la : lb)))
225	cmp = la < lb ? -1 : la == lb ? 0 : 1;
226
227	return cmp;
228	}
229
230	// compare hash entries, used when some keys are sv's or utf-x
231	static int
232	he_cmp_slow (const void a, const void b)
233	{
234	return sv_cmp (HeSVKEY_force ((HE )a), HeSVKEY_force ((HE **)b));
235	}
236
237	static void
238	encode_hv (enc_t enc, HV hv)
239	{
240	int count, i;
241
242	encode_ch (enc, '{'); NL; ++enc->indent;
243
244	if ((count = hv_iterinit (hv)))
245	{
246	// for canonical output we have to sort by keys first
247	// actually, this is mostly due to the stupid so-called
248	// security workaround added somewhere in 5.8.x.
249	// that randomises hash orderings
250	if (enc->flags & F_CANONICAL)
251	{
252	HE he, hes [count];
253	int fast = 1;
254
255	i = 0;
256	while ((he = hv_iternext (hv)))
257	{
258	hes [i++] = he;
259	if (HeKLEN (he) < 0 \|\| HeKUTF8 (he))
260	fast = 0;
261	}
262
263	assert (i == count);
264
265	if (fast)
266	qsort (hes, count, sizeof (HE *), he_cmp_fast);
267	else
268	{
269	// hack to disable "use bytes"
270	COP *oldcop = PL_curcop, cop;
271	cop.op_private = 0;
272	PL_curcop = &cop;
273
274	SAVETMPS;
275	qsort (hes, count, sizeof (HE *), he_cmp_slow);
276	FREETMPS;
277
278	PL_curcop = oldcop;
279	}
280
281	for (i = 0; i < count; ++i)
282	{
283	INDENT;
284	encode_he (enc, hes [i]);
285
286	if (i < count - 1)
287	COMMA;
288	}
289
290	NL;
291	}
292	else
293	{
294	SV *sv;
295	HE *he = hv_iternext (hv);
296
297	for (;;)
298	{
299	INDENT;
300	encode_he (enc, he);
301
302	if (!(he = hv_iternext (hv)))
303	break;
304
305	COMMA;
306	}
307
308	NL;
309	}
310	}
311
312	--enc->indent; INDENT; encode_ch (enc, '}');
313	}
314
315	static void
316	encode_sv (enc_t enc, SV sv)
317	{
318	if (SvPOKp (sv))
319	{
320	STRLEN len;
321	char *str = SvPV (sv, len);
322	encode_ch (enc, '"');
323	encode_str (enc, str, len, SvUTF8 (sv));
324	encode_ch (enc, '"');
325	}
326	else if (SvNOKp (sv))
327	{
328	need (enc, NV_DIG + 32);
329	Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
330	enc->cur += strlen (enc->cur);
331	}
332	else if (SvIOKp (sv))
333	{
334	need (enc, 64);
335	enc->cur +=
336	SvIsUV(sv)
337	? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv))
338	: snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv));
339	}
340	else if (SvROK (sv))
341	{
342	if (!--enc->max_recurse)
343	croak ("data structure too deep (hit recursion limit)");
344
345	sv = SvRV (sv);
346
347	switch (SvTYPE (sv))
348	{
349	case SVt_PVAV: encode_av (enc, (AV *)sv); break;
350	case SVt_PVHV: encode_hv (enc, (HV *)sv); break;
351
352	default:
353	croak ("JSON can only represent references to arrays or hashes");
354	}
355	}
356	else if (!SvOK (sv))
357	encode_str (enc, "null", 4, 0);
358	else
359	croak ("encountered perl type that JSON cannot handle");
360	}
361
362	static SV *
363	encode_json (SV *scalar, UV flags)
364	{
365	if (!(flags & F_ALLOW_NONREF) && !SvROK (scalar))
366	croak ("hash- or arraref required (not a simple scalar, use allow_nonref to allow this)");
367
368	enc_t enc;
369	enc.flags = flags;
370	enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE));
371	enc.cur = SvPVX (enc.sv);
372	enc.end = SvEND (enc.sv);
373	enc.max_recurse = 0;
374	enc.indent = 0;
375
376	SvPOK_only (enc.sv);
377	encode_sv (&enc, scalar);
378
379	if (!(flags & (F_ASCII \| F_UTF8)))
380	SvUTF8_on (enc.sv);
381
382	SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv));
383	return enc.sv;
384	}
385
386	/////////////////////////////////////////////////////////////////////////////
387
388	#define WS \
389	for (;;) \
390	{ \
391	char ch = *dec->cur; \
392	if (ch > 0x20 \
393	\|\| (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)) \
394	break; \
395	++dec->cur; \
396	}
397
398	#define ERR(reason) SB dec->err = reason; goto fail; SE
399	#define EXPECT_CH(ch) SB \
400	if (*dec->cur != ch) \
401	ERR (# ch " expected"); \
402	++dec->cur; \
403	SE
404
405	static SV decode_sv (dec_t dec);
406
407	#define APPEND_CH(ch) SB \
408	SvGROW (sv, cur + 1 + 1); \
409	SvPVX (sv)[cur++] = (ch); \
410	SE
411
412	static signed char decode_hexdigit[256];
413
414	static UV
415	decode_4hex (dec_t *dec)
416	{
417	signed char d1, d2, d3, d4;
418
419	d1 = decode_hexdigit [((unsigned char *)dec->cur) [0]];
420	if (d1 < 0) ERR ("four hexadecimal digits expected");
421	d2 = decode_hexdigit [((unsigned char *)dec->cur) [1]];
422	if (d2 < 0) ERR ("four hexadecimal digits expected");
423	d3 = decode_hexdigit [((unsigned char *)dec->cur) [2]];
424	if (d3 < 0) ERR ("four hexadecimal digits expected");
425	d4 = decode_hexdigit [((unsigned char *)dec->cur) [3]];
426	if (d4 < 0) ERR ("four hexadecimal digits expected");
427
428	dec->cur += 4;
429
430	return ((UV)d1) << 12
431	\| ((UV)d2) << 8
432	\| ((UV)d3) << 4
433	\| ((UV)d4);
434
435	fail:
436	return (UV)-1;
437	}
438
439	static SV *
440	decode_str (dec_t *dec)
441	{
442	SV *sv = NEWSV (0,2);
443	STRLEN cur = 0;
444	int utf8 = 0;
445
446	for (;;)
447	{
448	unsigned char ch = (unsigned char )dec->cur;
449
450	if (ch == '"')
451	break;
452	else if (ch == '\\')
453	{
454	switch (*++dec->cur)
455	{
456	case '\\':
457	case '/':
458	case '"': APPEND_CH (*dec->cur++); break;
459
460	case 'b': APPEND_CH ('\010'); ++dec->cur; break;
461	case 't': APPEND_CH ('\011'); ++dec->cur; break;
462	case 'n': APPEND_CH ('\012'); ++dec->cur; break;
463	case 'f': APPEND_CH ('\014'); ++dec->cur; break;
464	case 'r': APPEND_CH ('\015'); ++dec->cur; break;
465
466	case 'u':
467	{
468	UV lo, hi;
469	++dec->cur;
470
471	hi = decode_4hex (dec);
472	if (hi == (UV)-1)
473	goto fail;
474
475	// possibly a surrogate pair
476	if (hi >= 0xd800 && hi < 0xdc00)
477	{
478	if (dec->cur [0] != '\\' \|\| dec->cur [1] != 'u')
479	ERR ("illegal surrogate character");
480
481	dec->cur += 2;
482
483	lo = decode_4hex (dec);
484	if (lo == (UV)-1)
485	goto fail;
486
487	if (lo < 0xdc00 \|\| lo >= 0xe000)
488	ERR ("surrogate pair expected");
489
490	hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000;
491	}
492	else if (lo >= 0xdc00 && lo < 0xe000)
493	ERR ("illegal surrogate character");
494
495	if (hi >= 0x80)
496	{
497	utf8 = 1;
498
499	SvGROW (sv, cur + 4 + 1); // at most 4 bytes for 21 bits
500	cur = (char *)uvuni_to_utf8_flags (SvPVX (sv) + cur, hi, 0) - SvPVX (sv);
501	}
502	else
503	APPEND_CH (hi);
504	}
505	break;
506	}
507	}
508	else if (ch >= 0x20 && ch <= 0x7f)
509	APPEND_CH (*dec->cur++);
510	else if (ch >= 0x80)
511	{
512	STRLEN clen;
513	UV uch = utf8n_to_uvuni (dec->cur, dec->end - dec->cur, &clen, UTF8_CHECK_ONLY);
514	if (clen < 0)
515	ERR ("malformed UTF-8 character in string, cannot convert to JSON");
516
517	SvGROW (sv, cur + clen + 1); // at most 4 bytes for 21 bits
518	memcpy (SvPVX (sv) + cur, dec->cur, clen);
519	dec->cur += clen;
520	}
521	else
522	ERR ("invalid character encountered");
523	}
524
525	++dec->cur;
526
527	SvPOK_only (sv);
528
529	SvCUR_set (sv, cur);
530	*SvEND (sv) = 0;
531
532	if (utf8)
533	SvUTF8_on (sv);
534
535	return sv;
536
537	fail:
538	SvREFCNT_dec (sv);
539	return 0;
540	}
541
542	static SV *
543	decode_num (dec_t *dec)
544	{
545	int is_nv = 0;
546	char *start = dec->cur;
547
548	// [minus]
549	if (*dec->cur == '-')
550	++dec->cur;
551
552	if (*dec->cur == '0')
553	{
554	++dec->cur;
555	if (dec->cur >= '0' && dec->cur <= '9')
556	ERR ("malformed number (leading zero must not be followed by another digit)");
557	}
558
559	// int
560	while (dec->cur >= '0' && dec->cur <= '9')
561	++dec->cur;
562
563	// [frac]
564	if (*dec->cur == '.')
565	{
566	is_nv = 1;
567
568	do
569	{
570	++dec->cur;
571	}
572	while (dec->cur >= '0' && dec->cur <= '9');
573	}
574
575	// [exp]
576	if (dec->cur == 'e' \|\| dec->cur == 'E')
577	{
578	is_nv = 1;
579
580	++dec->cur;
581	if (dec->cur == '-' \|\| dec->cur == '+')
582	++dec->cur;
583
584	while (dec->cur >= '0' && dec->cur <= '9')
585	++dec->cur;
586	}
587
588	if (!is_nv)
589	{
590	UV uv;
591	int numtype = grok_number (start, dec->cur - start, &uv);
592	if (numtype & IS_NUMBER_IN_UV)
593	if (numtype & IS_NUMBER_NEG)
594	{
595	if (uv < (UV)IV_MIN)
596	return newSViv (-(IV)uv);
597	}
598	else
599	return newSVuv (uv);
600	}
601
602	return newSVnv (Atof (start));
603
604	fail:
605	return 0;
606	}
607
608	static SV *
609	decode_av (dec_t *dec)
610	{
611	AV *av = newAV ();
612
613	for (;;)
614	{
615	SV *value;
616
617	value = decode_sv (dec);
618	if (!value)
619	goto fail;
620
621	av_push (av, value);
622
623	WS;
624
625	if (*dec->cur == ']')
626	{
627	++dec->cur;
628	break;
629	}
630
631	if (*dec->cur != ',')
632	ERR (", or ] expected while parsing array");
633
634	++dec->cur;
635	}
636
637	return newRV_noinc ((SV *)av);
638
639	fail:
640	SvREFCNT_dec (av);
641	return 0;
642	}
643
644	static SV *
645	decode_hv (dec_t *dec)
646	{
647	HV *hv = newHV ();
648
649	for (;;)
650	{
651	SV key, value;
652
653	WS; EXPECT_CH ('"');
654
655	key = decode_str (dec);
656	if (!key)
657	goto fail;
658
659	WS; EXPECT_CH (':');
660
661	value = decode_sv (dec);
662	if (!value)
663	{
664	SvREFCNT_dec (key);
665	goto fail;
666	}
667
668	//TODO: optimise
669	hv_store_ent (hv, key, value, 0);
670
671	WS;
672
673	if (*dec->cur == '}')
674	{
675	++dec->cur;
676	break;
677	}
678
679	if (*dec->cur != ',')
680	ERR (", or } expected while parsing object/hash");
681
682	++dec->cur;
683	}
684
685	return newRV_noinc ((SV *)hv);
686
687	fail:
688	SvREFCNT_dec (hv);
689	return 0;
690	}
691
692	static SV *
693	decode_sv (dec_t *dec)
694	{
695	WS;
696	switch (*dec->cur)
697	{
698	case '"': ++dec->cur; return decode_str (dec);
699	case '[': ++dec->cur; return decode_av (dec);
700	case '{': ++dec->cur; return decode_hv (dec);
701
702	case '-':
703	case '0': case '1': case '2': case '3': case '4':
704	case '5': case '6': case '7': case '8': case '9':
705	return decode_num (dec);
706
707	case 't':
708	if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4))
709	{
710	dec->cur += 4;
711	return newSViv (1);
712	}
713	else
714	ERR ("'true' expected");
715
716	break;
717
718	case 'f':
719	if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5))
720	{
721	dec->cur += 5;
722	return newSViv (0);
723	}
724	else
725	ERR ("'false' expected");
726
727	break;
728
729	case 'n':
730	if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4))
731	{
732	dec->cur += 4;
733	return newSViv (1);
734	}
735	else
736	ERR ("'null' expected");
737
738	break;
739
740	default:
741	ERR ("malformed json string");
742	break;
743	}
744
745	fail:
746	return 0;
747	}
748
749	static SV *
750	decode_json (SV *string, UV flags)
751	{
752	SV *sv;
753
754	if (!(flags & F_UTF8))
755	sv_utf8_upgrade (string);
756
757	SvGROW (string, SvCUR (string) + 1); // should basically be a NOP
758
759	dec_t dec;
760	dec.flags = flags;
761	dec.cur = SvPVX (string);
762	dec.end = SvEND (string);
763	dec.err = 0;
764
765	*dec.end = 1; // invalid anywhere
766	sv = decode_sv (&dec);
767	*dec.end = 0;
768
769	if (!sv)
770	{
771	IV offset = utf8_distance (dec.cur, SvPVX (string));
772	SV *uni = sv_newmortal ();
773
774	pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ);
775	croak ("%s, at character %d (%s)",
776	dec.err,
777	(int)offset,
778	dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)");
779	}
780
781	sv = sv_2mortal (sv);
782
783	if (!(dec.flags & F_ALLOW_NONREF) && !SvROK (sv))
784	croak ("JSON object or array expected (but number, string, true, false or null found, use allow_nonref to allow this)");
785
786	return sv;
787	}
788
789	MODULE = JSON::XS PACKAGE = JSON::XS
790
791	BOOT:
792	{
793	int i;
794
795	memset (decode_hexdigit, 0xff, 256);
796	for (i = 10; i--; )
797	decode_hexdigit ['0' + i] = i;
798
799	for (i = 6; --i; )
800	{
801	decode_hexdigit ['a' + i] = 10 + i;
802	decode_hexdigit ['A' + i] = 10 + i;
803	}
804
805	json_stash = gv_stashpv ("JSON::XS", 1);
806	}
807
808	SV new (char dummy)
809	CODE:
810	RETVAL = sv_bless (newRV_noinc (newSVuv (F_DEFAULT)), json_stash);
811	OUTPUT:
812	RETVAL
813
814	SV ascii (SV self, int enable)
815	ALIAS:
816	ascii = F_ASCII
817	utf8 = F_UTF8
818	indent = F_INDENT
819	canonical = F_CANONICAL
820	space_before = F_SPACE_BEFORE
821	space_after = F_SPACE_AFTER
822	json_rpc = F_JSON_RPC
823	pretty = F_PRETTY
824	allow_nonref = F_ALLOW_NONREF
825	CODE:
826	{
827	UV *uv = SvJSON (self);
828	if (enable)
829	*uv \|= ix;
830	else
831	*uv &= ~ix;
832
833	RETVAL = newSVsv (self);
834	}
835	OUTPUT:
836	RETVAL
837
838	void encode (SV self, SV scalar)
839	PPCODE:
840	XPUSHs (encode_json (scalar, *SvJSON (self)));
841
842	void decode (SV self, SV jsonstr)
843	PPCODE:
844	XPUSHs (decode_json (jsonstr, *SvJSON (self)));
845
846	void to_json (SV *scalar)
847	PPCODE:
848	XPUSHs (encode_json (scalar, F_UTF8));
849
850	void from_json (SV *jsonstr)
851	PPCODE:
852	XPUSHs (decode_json (jsonstr, F_UTF8));
853