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_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)
{
  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_dump (sv);//D
  return sv_2mortal (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
        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:	1.2
Committed:	Thu Mar 22 17:28:50 2007 UTC (17 years, 2 months ago) by root
Branch:	MAIN
Changes since 1.1:	+13 -3 lines
Log Message:	added preliminary documentation
#	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
17	#define F_PRETTY F_INDENT \| F_SPACE_BEFORE \| F_SPACE_AFTER
18	#define F_DEFAULT 0
19
20	#define INIT_SIZE 32 // initial scalar size to be allocated
21
22	#define SB do {
23	#define SE } while (0)
24
25	static HV *json_stash;
26
27	// structure used for encoding JSON
28	typedef struct
29	{
30	char *cur;
31	STRLEN len; // SvLEN (sv)
32	char *end; // SvEND (sv)
33	SV *sv;
34	UV flags;
35	int max_recurse;
36	int indent;
37	} enc_t;
38
39	// structure used for decoding JSON
40	typedef struct
41	{
42	char *cur;
43	char *end;
44	char *err;
45	UV flags;
46	} dec_t;
47
48	static UV *
49	SvJSON (SV *sv)
50	{
51	if (!(SvROK (sv) && SvOBJECT (SvRV (sv)) && SvSTASH (SvRV (sv)) == json_stash))
52	croak ("object is not of type JSON::XS");
53
54	return &SvUVX (SvRV (sv));
55	}
56
57	/////////////////////////////////////////////////////////////////////////////
58
59	static void
60	need (enc_t *enc, STRLEN len)
61	{
62	if (enc->cur + len >= enc->end)
63	{
64	STRLEN cur = enc->cur - SvPVX (enc->sv);
65	SvGROW (enc->sv, cur + len + 1);
66	enc->cur = SvPVX (enc->sv) + cur;
67	enc->end = SvEND (enc->sv);
68	}
69	}
70
71	static void
72	encode_ch (enc_t *enc, char ch)
73	{
74	need (enc, 1);
75	*enc->cur++ = ch;
76	}
77
78	static void
79	encode_str (enc_t enc, char str, STRLEN len, int is_utf8)
80	{
81	char *end = str + len;
82
83	while (str < end)
84	{
85	unsigned char ch = (unsigned char )str;
86	if (ch >= 0x20 && ch < 0x80) // most common case
87	{
88	*enc->cur++ = ch;
89	str++;
90	}
91	else
92	{
93	STRLEN clen;
94	UV uch;
95
96	if (is_utf8)
97	{
98	uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
99	if (clen < 0)
100	croak ("malformed UTF-8 character in string, cannot convert to JSON");
101	}
102	else
103	{
104	uch = ch;
105	clen = 1;
106	}
107
108	need (enc, len += 6);
109
110	if (uch < 0xa0 \|\| enc->flags & F_ASCII)
111	{
112	if (uch > 0xFFFFUL)
113	{
114	len += 6;
115	need (enc, len += 6);
116	sprintf (enc->cur, "\\u%04x\\u%04x",
117	(uch - 0x10000) / 0x400 + 0xD800,
118	(uch - 0x10000) % 0x400 + 0xDC00);
119	enc->cur += 12;
120	}
121	else
122	{
123	sprintf (enc->cur, "\\u%04x", uch);
124	enc->cur += 6;
125	}
126	}
127	else if (is_utf8)
128	{
129	memcpy (enc->cur, str, clen);
130	enc->cur += clen;
131	}
132	else
133	enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);
134
135	str += clen;
136	}
137
138	--len;
139	}
140	}
141
142	#define INDENT SB \
143	if (enc->flags & F_INDENT) \
144	{ \
145	int i_; \
146	need (enc, enc->indent); \
147	for (i_ = enc->indent * 3; i_--; )\
148	encode_ch (enc, ' '); \
149	} \
150	SE
151
152	#define SPACE SB need (enc, 1); encode_ch (enc, ' '); SE
153	#define NL SB if (enc->flags & F_INDENT) { need (enc, 1); encode_ch (enc, '\n'); } SE
154	#define COMMA SB \
155	encode_ch (enc, ','); \
156	if (enc->flags & F_INDENT) \
157	NL; \
158	else if (enc->flags & F_SPACE_AFTER) \
159	SPACE; \
160	SE
161
162	static void encode_sv (enc_t enc, SV sv);
163
164	static void
165	encode_av (enc_t enc, AV av)
166	{
167	int i, len = av_len (av);
168
169	encode_ch (enc, '['); NL;
170	++enc->indent;
171
172	for (i = 0; i <= len; ++i)
173	{
174	INDENT;
175	encode_sv (enc, *av_fetch (av, i, 0));
176
177	if (i < len)
178	COMMA;
179	}
180
181	NL;
182
183	--enc->indent;
184	INDENT; encode_ch (enc, ']');
185	}
186
187	static void
188	encode_he (enc_t enc, HE he)
189	{
190	encode_ch (enc, '"');
191
192	if (HeKLEN (he) == HEf_SVKEY)
193	{
194	SV *sv = HeSVKEY (he);
195	STRLEN len;
196	char *str = SvPV (sv, len);
197
198	encode_str (enc, str, len, SvUTF8 (sv));
199	}
200	else
201	encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he));
202
203	encode_ch (enc, '"');
204
205	if (enc->flags & F_SPACE_BEFORE) SPACE;
206	encode_ch (enc, ':');
207	if (enc->flags & F_SPACE_AFTER ) SPACE;
208	encode_sv (enc, HeVAL (he));
209	}
210
211	// compare hash entries, used when all keys are bytestrings
212	static int
213	he_cmp_fast (const void a_, const void b_)
214	{
215	int cmp;
216
217	HE a = (HE **)a_;
218	HE b = (HE **)b_;
219
220	STRLEN la = HeKLEN (a);
221	STRLEN lb = HeKLEN (b);
222
223	if (!(cmp == memcmp (HeKEY (a), HeKEY (b), la < lb ? la : lb)))
224	cmp = la < lb ? -1 : la == lb ? 0 : 1;
225
226	return cmp;
227	}
228
229	// compare hash entries, used when some keys are sv's or utf-x
230	static int
231	he_cmp_slow (const void a, const void b)
232	{
233	return sv_cmp (HeSVKEY_force ((HE )a), HeSVKEY_force ((HE **)b));
234	}
235
236	static void
237	encode_hv (enc_t enc, HV hv)
238	{
239	int count, i;
240
241	encode_ch (enc, '{'); NL; ++enc->indent;
242
243	if ((count = hv_iterinit (hv)))
244	{
245	// for canonical output we have to sort by keys first
246	// actually, this is mostly due to the stupid so-called
247	// security workaround added somewhere in 5.8.x.
248	// that randomises hash orderings
249	if (enc->flags & F_CANONICAL)
250	{
251	HE he, hes [count];
252	int fast = 1;
253
254	i = 0;
255	while ((he = hv_iternext (hv)))
256	{
257	hes [i++] = he;
258	if (HeKLEN (he) < 0 \|\| HeKUTF8 (he))
259	fast = 0;
260	}
261
262	assert (i == count);
263
264	if (fast)
265	qsort (hes, count, sizeof (HE *), he_cmp_fast);
266	else
267	{
268	// hack to disable "use bytes"
269	COP *oldcop = PL_curcop, cop;
270	cop.op_private = 0;
271	PL_curcop = &cop;
272
273	SAVETMPS;
274	qsort (hes, count, sizeof (HE *), he_cmp_slow);
275	FREETMPS;
276
277	PL_curcop = oldcop;
278	}
279
280	for (i = 0; i < count; ++i)
281	{
282	INDENT;
283	encode_he (enc, hes [i]);
284
285	if (i < count - 1)
286	COMMA;
287	}
288
289	NL;
290	}
291	else
292	{
293	SV *sv;
294	HE *he = hv_iternext (hv);
295
296	for (;;)
297	{
298	INDENT;
299	encode_he (enc, he);
300
301	if (!(he = hv_iternext (hv)))
302	break;
303
304	COMMA;
305	}
306
307	NL;
308	}
309	}
310
311	--enc->indent; INDENT; encode_ch (enc, '}');
312	}
313
314	static void
315	encode_sv (enc_t enc, SV sv)
316	{
317	if (SvPOKp (sv))
318	{
319	STRLEN len;
320	char *str = SvPV (sv, len);
321	encode_ch (enc, '"');
322	encode_str (enc, str, len, SvUTF8 (sv));
323	encode_ch (enc, '"');
324	}
325	else if (SvNOKp (sv))
326	{
327	need (enc, NV_DIG + 32);
328	Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
329	enc->cur += strlen (enc->cur);
330	}
331	else if (SvIOKp (sv))
332	{
333	need (enc, 64);
334	enc->cur +=
335	SvIsUV(sv)
336	? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv))
337	: snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv));
338	}
339	else if (SvROK (sv))
340	{
341	if (!--enc->max_recurse)
342	croak ("data structure too deep (hit recursion limit)");
343
344	sv = SvRV (sv);
345
346	switch (SvTYPE (sv))
347	{
348	case SVt_PVAV: encode_av (enc, (AV *)sv); break;
349	case SVt_PVHV: encode_hv (enc, (HV *)sv); break;
350
351	default:
352	croak ("JSON can only represent references to arrays or hashes");
353	}
354	}
355	else if (!SvOK (sv))
356	encode_str (enc, "null", 4, 0);
357	else
358	croak ("encountered perl type that JSON cannot handle");
359	}
360
361	static SV *
362	encode_json (SV *scalar, UV flags)
363	{
364	enc_t enc;
365	enc.flags = flags;
366	enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE));
367	enc.cur = SvPVX (enc.sv);
368	enc.end = SvEND (enc.sv);
369	enc.max_recurse = 0;
370	enc.indent = 0;
371
372	SvPOK_only (enc.sv);
373	encode_sv (&enc, scalar);
374
375	if (!(flags & (F_ASCII \| F_UTF8)))
376	SvUTF8_on (enc.sv);
377
378	SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv));
379	return enc.sv;
380	}
381
382	/////////////////////////////////////////////////////////////////////////////
383
384	#define WS \
385	for (;;) \
386	{ \
387	char ch = *dec->cur; \
388	if (ch > 0x20 \
389	\|\| (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)) \
390	break; \
391	++dec->cur; \
392	}
393
394	#define ERR(reason) SB dec->err = reason; goto fail; SE
395	#define EXPECT_CH(ch) SB \
396	if (*dec->cur != ch) \
397	ERR (# ch " expected"); \
398	++dec->cur; \
399	SE
400
401	static SV decode_sv (dec_t dec);
402
403	#define APPEND_CH(ch) SB \
404	SvGROW (sv, cur + 1 + 1); \
405	SvPVX (sv)[cur++] = (ch); \
406	SE
407
408	static signed char decode_hexdigit[256];
409
410	static UV
411	decode_4hex (dec_t *dec)
412	{
413	signed char d1, d2, d3, d4;
414
415	d1 = decode_hexdigit [((unsigned char *)dec->cur) [0]];
416	if (d1 < 0) ERR ("four hexadecimal digits expected");
417	d2 = decode_hexdigit [((unsigned char *)dec->cur) [1]];
418	if (d2 < 0) ERR ("four hexadecimal digits expected");
419	d3 = decode_hexdigit [((unsigned char *)dec->cur) [2]];
420	if (d3 < 0) ERR ("four hexadecimal digits expected");
421	d4 = decode_hexdigit [((unsigned char *)dec->cur) [3]];
422	if (d4 < 0) ERR ("four hexadecimal digits expected");
423
424	dec->cur += 4;
425
426	return ((UV)d1) << 12
427	\| ((UV)d2) << 8
428	\| ((UV)d3) << 4
429	\| ((UV)d4);
430
431	fail:
432	return (UV)-1;
433	}
434
435	static SV *
436	decode_str (dec_t *dec)
437	{
438	SV *sv = NEWSV (0,2);
439	STRLEN cur = 0;
440	int utf8 = 0;
441
442	for (;;)
443	{
444	unsigned char ch = (unsigned char )dec->cur;
445
446	if (ch == '"')
447	break;
448	else if (ch == '\\')
449	{
450	switch (*++dec->cur)
451	{
452	case '\\':
453	case '/':
454	case '"': APPEND_CH (*dec->cur++); break;
455
456	case 'b': APPEND_CH ('\010'); ++dec->cur; break;
457	case 't': APPEND_CH ('\011'); ++dec->cur; break;
458	case 'n': APPEND_CH ('\012'); ++dec->cur; break;
459	case 'f': APPEND_CH ('\014'); ++dec->cur; break;
460	case 'r': APPEND_CH ('\015'); ++dec->cur; break;
461
462	case 'u':
463	{
464	UV lo, hi;
465	++dec->cur;
466
467	hi = decode_4hex (dec);
468	if (hi == (UV)-1)
469	goto fail;
470
471	// possibly a surrogate pair
472	if (hi >= 0xd800 && hi < 0xdc00)
473	{
474	if (dec->cur [0] != '\\' \|\| dec->cur [1] != 'u')
475	ERR ("illegal surrogate character");
476
477	dec->cur += 2;
478
479	lo = decode_4hex (dec);
480	if (lo == (UV)-1)
481	goto fail;
482
483	if (lo < 0xdc00 \|\| lo >= 0xe000)
484	ERR ("surrogate pair expected");
485
486	hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000;
487	}
488	else if (lo >= 0xdc00 && lo < 0xe000)
489	ERR ("illegal surrogate character");
490
491	if (hi >= 0x80)
492	{
493	utf8 = 1;
494
495	SvGROW (sv, cur + 4 + 1); // at most 4 bytes for 21 bits
496	cur = (char *)uvuni_to_utf8_flags (SvPVX (sv) + cur, hi, 0) - SvPVX (sv);
497	}
498	else
499	APPEND_CH (hi);
500	}
501	break;
502	}
503	}
504	else if (ch >= 0x20 && ch <= 0x7f)
505	APPEND_CH (*dec->cur++);
506	else if (ch >= 0x80)
507	{
508	STRLEN clen;
509	UV uch = utf8n_to_uvuni (dec->cur, dec->end - dec->cur, &clen, UTF8_CHECK_ONLY);
510	if (clen < 0)
511	ERR ("malformed UTF-8 character in string, cannot convert to JSON");
512
513	SvGROW (sv, cur + clen + 1); // at most 4 bytes for 21 bits
514	memcpy (SvPVX (sv) + cur, dec->cur, clen);
515	dec->cur += clen;
516	}
517	else
518	ERR ("invalid character encountered");
519	}
520
521	++dec->cur;
522
523	SvPOK_only (sv);
524
525	SvCUR_set (sv, cur);
526	*SvEND (sv) = 0;
527
528	if (utf8)
529	SvUTF8_on (sv);
530
531	return sv;
532
533	fail:
534	SvREFCNT_dec (sv);
535	return 0;
536	}
537
538	static SV *
539	decode_num (dec_t *dec)
540	{
541	int is_nv = 0;
542	char *start = dec->cur;
543
544	// [minus]
545	if (*dec->cur == '-')
546	++dec->cur;
547
548	if (*dec->cur == '0')
549	{
550	++dec->cur;
551	if (dec->cur >= '0' && dec->cur <= '9')
552	ERR ("malformed number (leading zero must not be followed by another digit)");
553	}
554
555	// int
556	while (dec->cur >= '0' && dec->cur <= '9')
557	++dec->cur;
558
559	// [frac]
560	if (*dec->cur == '.')
561	{
562	is_nv = 1;
563
564	do
565	{
566	++dec->cur;
567	}
568	while (dec->cur >= '0' && dec->cur <= '9');
569	}
570
571	// [exp]
572	if (dec->cur == 'e' \|\| dec->cur == 'E')
573	{
574	is_nv = 1;
575
576	++dec->cur;
577	if (dec->cur == '-' \|\| dec->cur == '+')
578	++dec->cur;
579
580	while (dec->cur >= '0' && dec->cur <= '9')
581	++dec->cur;
582	}
583
584	if (!is_nv)
585	{
586	UV uv;
587	int numtype = grok_number (start, dec->cur - start, &uv);
588	if (numtype & IS_NUMBER_IN_UV)
589	if (numtype & IS_NUMBER_NEG)
590	{
591	if (uv < (UV)IV_MIN)
592	return newSViv (-(IV)uv);
593	}
594	else
595	return newSVuv (uv);
596	}
597
598	return newSVnv (Atof (start));
599
600	fail:
601	return 0;
602	}
603
604	static SV *
605	decode_av (dec_t *dec)
606	{
607	AV *av = newAV ();
608
609	for (;;)
610	{
611	SV *value;
612
613	value = decode_sv (dec);
614	if (!value)
615	goto fail;
616
617	av_push (av, value);
618
619	WS;
620
621	if (*dec->cur == ']')
622	{
623	++dec->cur;
624	break;
625	}
626
627	if (*dec->cur != ',')
628	ERR (", or ] expected while parsing array");
629
630	++dec->cur;
631	}
632
633	return newRV_noinc ((SV *)av);
634
635	fail:
636	SvREFCNT_dec (av);
637	return 0;
638	}
639
640	static SV *
641	decode_hv (dec_t *dec)
642	{
643	HV *hv = newHV ();
644
645	for (;;)
646	{
647	SV key, value;
648
649	WS; EXPECT_CH ('"');
650
651	key = decode_str (dec);
652	if (!key)
653	goto fail;
654
655	WS; EXPECT_CH (':');
656
657	value = decode_sv (dec);
658	if (!value)
659	{
660	SvREFCNT_dec (key);
661	goto fail;
662	}
663
664	//TODO: optimise
665	hv_store_ent (hv, key, value, 0);
666
667	WS;
668
669	if (*dec->cur == '}')
670	{
671	++dec->cur;
672	break;
673	}
674
675	if (*dec->cur != ',')
676	ERR (", or } expected while parsing object/hash");
677
678	++dec->cur;
679	}
680
681	return newRV_noinc ((SV *)hv);
682
683	fail:
684	SvREFCNT_dec (hv);
685	return 0;
686	}
687
688	static SV *
689	decode_sv (dec_t *dec)
690	{
691	WS;
692	switch (*dec->cur)
693	{
694	case '"': ++dec->cur; return decode_str (dec);
695	case '[': ++dec->cur; return decode_av (dec);
696	case '{': ++dec->cur; return decode_hv (dec);
697
698	case '-':
699	case '0': case '1': case '2': case '3': case '4':
700	case '5': case '6': case '7': case '8': case '9':
701	return decode_num (dec);
702
703	case 't':
704	if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4))
705	{
706	dec->cur += 4;
707	return newSViv (1);
708	}
709	else
710	ERR ("'true' expected");
711
712	break;
713
714	case 'f':
715	if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5))
716	{
717	dec->cur += 5;
718	return newSViv (0);
719	}
720	else
721	ERR ("'false' expected");
722
723	break;
724
725	case 'n':
726	if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4))
727	{
728	dec->cur += 4;
729	return newSViv (1);
730	}
731	else
732	ERR ("'null' expected");
733
734	break;
735
736	default:
737	ERR ("malformed json string");
738	break;
739	}
740
741	fail:
742	return 0;
743	}
744
745	static SV *
746	decode_json (SV *string, UV flags)
747	{
748	SV *sv;
749
750	if (!(flags & F_UTF8))
751	sv_utf8_upgrade (string);
752
753	SvGROW (string, SvCUR (string) + 1); // should basically be a NOP
754
755	dec_t dec;
756	dec.flags = flags;
757	dec.cur = SvPVX (string);
758	dec.end = SvEND (string);
759	dec.err = 0;
760
761	*dec.end = 1; // invalid anywhere
762	sv = decode_sv (&dec);
763	*dec.end = 0;
764
765	if (!sv)
766	{
767	IV offset = utf8_distance (dec.cur, SvPVX (string));
768	SV *uni = sv_newmortal ();
769
770	pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ);
771	croak ("%s, at character %d (%s)",
772	dec.err,
773	(int)offset,
774	dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)");
775	}
776
777	sv_dump (sv);//D
778	return sv_2mortal (sv);
779	}
780
781	MODULE = JSON::XS PACKAGE = JSON::XS
782
783	BOOT:
784	{
785	int i;
786
787	memset (decode_hexdigit, 0xff, 256);
788	for (i = 10; i--; )
789	decode_hexdigit ['0' + i] = i;
790
791	for (i = 6; --i; )
792	{
793	decode_hexdigit ['a' + i] = 10 + i;
794	decode_hexdigit ['A' + i] = 10 + i;
795	}
796
797	json_stash = gv_stashpv ("JSON::XS", 1);
798	}
799
800	SV new (char dummy)
801	CODE:
802	RETVAL = sv_bless (newRV_noinc (newSVuv (F_DEFAULT)), json_stash);
803	OUTPUT:
804	RETVAL
805
806	SV ascii (SV self, int enable)
807	ALIAS:
808	ascii = F_ASCII
809	utf8 = F_UTF8
810	indent = F_INDENT
811	canonical = F_CANONICAL
812	space_before = F_SPACE_BEFORE
813	space_after = F_SPACE_AFTER
814	json_rpc = F_JSON_RPC
815	pretty = F_PRETTY
816	CODE:
817	{
818	UV *uv = SvJSON (self);
819	if (enable)
820	*uv \|= ix;
821	else
822	*uv &= ~ix;
823
824	RETVAL = newSVsv (self);
825	}
826	OUTPUT:
827	RETVAL
828
829	void encode (SV self, SV scalar)
830	PPCODE:
831	XPUSHs (encode_json (scalar, *SvJSON (self)));
832
833	void decode (SV self, SV jsonstr)
834	PPCODE:
835	XPUSHs (decode_json (jsonstr, *SvJSON (self)));
836
837	void to_json (SV *scalar)
838	PPCODE:
839	XPUSHs (encode_json (scalar, F_UTF8));
840
841	void from_json (SV *jsonstr)
842	PPCODE:
843	XPUSHs (decode_json (jsonstr, F_UTF8));
844