--- JSON-XS/XS.xs 2007/03/25 02:37:00 1.16 +++ JSON-XS/XS.xs 2007/06/25 22:11:39 1.46 @@ -5,33 +5,61 @@ #include "assert.h" #include "string.h" #include "stdlib.h" +#include "stdio.h" -#define F_ASCII 0x00000001 -#define F_UTF8 0x00000002 -#define F_INDENT 0x00000004 -#define F_CANONICAL 0x00000008 -#define F_SPACE_BEFORE 0x00000010 -#define F_SPACE_AFTER 0x00000020 -#define F_ALLOW_NONREF 0x00000080 -#define F_SHRINK 0x00000100 - -// F_SKIPINVALID? -// F_EXECCODEREF? -// F_SELFCONVERT? +#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 // NYI +#define F_MAXDEPTH 0xf8000000UL +#define S_MAXDEPTH 27 +#define F_MAXSIZE 0x01f00000UL +#define S_MAXSIZE 20 + +#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 0 +#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 UTF8_MAX_LEN 11 // for perls UTF-X: max. number of octets per character -#define SHORT_STRING_LEN 512 // special-case strings of up to this size +#define SHORT_STRING_LEN 16384 // special-case strings of up to this size #define SB do { #define SE } while (0) -static HV *json_stash; // JSON::XS:: +#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) + +static HV *json_stash, *json_boolean_stash; // JSON::XS:: +static SV *json_true, *json_false; ///////////////////////////////////////////////////////////////////////////// // utility functions @@ -64,10 +92,10 @@ // 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 -static UV +inline UV decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen) { - if (s[0] > 0xdf || s[0] < 0xc2) + 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) { @@ -75,7 +103,10 @@ return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f); } else - return (UV)-1; + { + *clen = (STRLEN)-1; + return (UV)-1; + } } ///////////////////////////////////////////////////////////////////////////// @@ -87,24 +118,24 @@ char *cur; // SvPVX (sv) + current output position char *end; // SvEND (sv) SV *sv; // result scalar - UV flags; // F_* - int indent; // indentation level - int max_depth; // max. recursion level + U32 flags; // F_* + U32 indent; // indentation level + U32 maxdepth; // max. indentation/recursion level } enc_t; -static void +inline void need (enc_t *enc, STRLEN len) { - if (enc->cur + len >= enc->end) + 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); + enc->end = SvPVX (enc->sv) + SvLEN (enc->sv) - 1; } } -static void +inline void encode_ch (enc_t *enc, char ch) { need (enc, 1); @@ -122,15 +153,15 @@ { unsigned char ch = *(unsigned char *)str; - if (ch >= 0x20 && ch < 0x80) // most common case + if (expect_true (ch >= 0x20 && ch < 0x80)) // most common case { - if (ch == '"') // but with slow exceptions + if (expect_false (ch == '"')) // but with slow exceptions { need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = '"'; } - else if (ch == '\\') + else if (expect_false (ch == '\\')) { need (enc, len += 1); *enc->cur++ = '\\'; @@ -158,7 +189,6 @@ if (is_utf8) { - //uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY); 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); @@ -172,7 +202,7 @@ if (uch > 0x10FFFFUL) croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch); - if (uch < 0x80 || enc->flags & F_ASCII) + if (uch < 0x80 || enc->flags & F_ASCII || (enc->flags & F_LATIN1 && uch > 0xFF)) { if (uch > 0xFFFFUL) { @@ -196,6 +226,11 @@ str += clen; } + else if (enc->flags & F_LATIN1) + { + *enc->cur++ = uch; + str += clen; + } else if (is_utf8) { need (enc, len += clen); @@ -207,7 +242,7 @@ } else { - need (enc, len += UTF8_MAX_LEN - 1); // never more than 11 bytes needed + need (enc, len += UTF8_MAXBYTES - 1); // never more than 11 bytes needed enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0); ++str; } @@ -219,7 +254,7 @@ } } -static void +inline void encode_indent (enc_t *enc) { if (enc->flags & F_INDENT) @@ -232,14 +267,14 @@ } } -static void +inline void encode_space (enc_t *enc) { need (enc, 1); encode_ch (enc, ' '); } -static void +inline void encode_nl (enc_t *enc) { if (enc->flags & F_INDENT) @@ -249,7 +284,7 @@ } } -static void +inline void encode_comma (enc_t *enc) { encode_ch (enc, ','); @@ -267,6 +302,9 @@ { 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; @@ -342,6 +380,9 @@ { int count, i; + if (enc->indent >= enc->maxdepth) + croak ("data structure too deep (hit recursion limit)"); + encode_ch (enc, '{'); encode_nl (enc); ++enc->indent; if ((count = hv_iterinit (hv))) @@ -352,8 +393,13 @@ // that randomises hash orderings if (enc->flags & F_CANONICAL) { - HE *he, *hes [count]; // if your compiler dies here, you need to enable C99 mode int fast = 1; + HE *he; +#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))) @@ -398,7 +444,6 @@ } else { - SV *sv; HE *he = hv_iternext (hv); for (;;) @@ -419,6 +464,88 @@ --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))) + { + if (SvSTASH (sv) == json_boolean_stash) + { + if (SvIV (sv) == 0) + encode_str (enc, "false", 5, 0); + else + encode_str (enc, "true", 4, 0); + } + else + { +#if 0 + if (0 && sv_derived_from (rv, "JSON::Literal")) + { + // not yet + } +#endif + if (enc->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", 1); + + 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 reutrn value + // check anyways. + PUTBACK; + assert (1 == call_sv ((SV *)GvCV (to_json), G_SCALAR)); + SPAGAIN; + + encode_sv (enc, POPs); + + FREETMPS; + LEAVE; + } + else if (enc->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->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) + { + if (SvNIOK (sv) && SvIV (sv) == 0) + encode_str (enc, "false", 5, 0); + else if (SvNIOK (sv) && SvIV (sv) == 1) + encode_str (enc, "true", 4, 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) { @@ -434,35 +561,53 @@ } 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)) { - need (enc, 64); - enc->cur += - SvIsUV(sv) - ? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv)) - : snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv)); - } - else if (SvROK (sv)) - { - SV *rv = SvRV (sv); - - if (enc->indent >= enc->max_depth) - croak ("data structure too deep (hit recursion limit)"); - - switch (SvTYPE (rv)) + // we assume we can always read an IV as a UV + if (SvUV (sv) & ~(UV)0x7fff) { - case SVt_PVAV: encode_av (enc, (AV *)rv); break; - case SVt_PVHV: encode_hv (enc, (HV *)rv); break; - - default: - croak ("encountered %s, but JSON can only represent references to arrays or hashes", - SvPV_nolen (sv)); + // 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 @@ -471,26 +616,28 @@ } static SV * -encode_json (SV *scalar, UV flags) +encode_json (SV *scalar, U32 flags) { + enc_t enc; + if (!(flags & F_ALLOW_NONREF) && !SvROK (scalar)) croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)"); - enc_t enc; enc.flags = flags; enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE)); enc.cur = SvPVX (enc.sv); enc.end = SvEND (enc.sv); enc.indent = 0; - enc.max_depth = 0x7fffffffUL; + enc.maxdepth = DEC_DEPTH (flags); 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)); + *SvEND (enc.sv) = 0; // many xs functions expect a trailing 0 for text strings + + if (!(flags & (F_ASCII | F_LATIN1 | F_UTF8))) + SvUTF8_on (enc.sv); if (enc.flags & F_SHRINK) shrink (enc.sv); @@ -507,10 +654,12 @@ char *cur; // current parser pointer char *end; // end of input string const char *err; // parse error, if != 0 - UV flags; // F_* + U32 flags; // F_* + U32 depth; // recursion depth + U32 maxdepth; // recursion depth limit } dec_t; -static void +inline void decode_ws (dec_t *dec) { for (;;) @@ -526,12 +675,16 @@ } #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]; @@ -542,10 +695,10 @@ signed char d1, d2, d3, d4; unsigned char *cur = (unsigned char *)dec->cur; - d1 = decode_hexdigit [cur [0]]; if (d1 < 0) ERR ("four hexadecimal digits expected"); - d2 = decode_hexdigit [cur [1]]; if (d2 < 0) ERR ("four hexadecimal digits expected"); - d3 = decode_hexdigit [cur [2]]; if (d3 < 0) ERR ("four hexadecimal digits expected"); - d4 = decode_hexdigit [cur [3]]; if (d4 < 0) ERR ("four hexadecimal digits expected"); + 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; @@ -563,41 +716,44 @@ { SV *sv = 0; int utf8 = 0; + char *dec_cur = dec->cur; do { - char buf [SHORT_STRING_LEN + UTF8_MAX_LEN]; + char buf [SHORT_STRING_LEN + UTF8_MAXBYTES]; char *cur = buf; do { - unsigned char ch = *(unsigned char *)dec->cur++; + unsigned char ch = *(unsigned char *)dec_cur++; - if (ch == '"') + if (expect_false (ch == '"')) { - --dec->cur; + --dec_cur; break; } - else if (ch == '\\') + else if (expect_false (ch == '\\')) { - switch (*dec->cur) + switch (*dec_cur) { case '\\': case '/': - case '"': *cur++ = *dec->cur++; break; + 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 '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 = dec_cur; hi = decode_4hex (dec); + dec_cur = dec->cur; if (hi == (UV)-1) goto fail; @@ -605,12 +761,14 @@ if (hi >= 0xd800) if (hi < 0xdc00) { - if (dec->cur [0] != '\\' || dec->cur [1] != 'u') + if (dec_cur [0] != '\\' || dec_cur [1] != 'u') ERR ("missing low surrogate character in surrogate pair"); - dec->cur += 2; + dec_cur += 2; + dec->cur = dec_cur; lo = decode_4hex (dec); + dec_cur = dec->cur; if (lo == (UV)-1) goto fail; @@ -634,51 +792,57 @@ break; default: - --dec->cur; + --dec_cur; ERR ("illegal backslash escape sequence in string"); } } - else if (ch >= 0x20 && ch <= 0x7f) + else if (expect_true (ch >= 0x20 && ch <= 0x7f)) *cur++ = ch; else if (ch >= 0x80) { - --dec->cur; - STRLEN clen; - UV uch = decode_utf8 (dec->cur, dec->end - dec->cur, &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++; - } + *cur++ = *dec_cur++; while (--clen); utf8 = 1; } - else if (!ch) - ERR ("unexpected end of string while parsing json string"); else - ERR ("invalid character encountered"); + { + --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; + { + 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); + 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 != '"'); + while (*dec_cur != '"'); - ++dec->cur; + ++dec_cur; if (sv) { @@ -691,9 +855,11 @@ else sv = newSVpvn ("", 0); + dec->cur = dec_cur; return sv; fail: + dec->cur = dec_cur; return 0; } @@ -761,18 +927,41 @@ 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) + // special case the rather common 1..4-digit-int case, assumes 32 bit ints or so + if (*start == '-') + switch (dec->cur - start) { - if (uv < (UV)IV_MIN) - return newSViv (-(IV)uv); + 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 - return newSVuv (uv); + else + switch (dec->cur - start) + { + 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, 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); + + // here would likely be the place for bigint support + } } + // if we ever support bigint or bigfloat, this is the place for bigfloat return newSVnv (Atof (start)); fail: @@ -784,7 +973,9 @@ { AV *av = newAV (); + DEC_INC_DEPTH; decode_ws (dec); + if (*dec->cur == ']') ++dec->cur; else @@ -812,10 +1003,12 @@ ++dec->cur; } + DEC_DEC_DEPTH; return newRV_noinc ((SV *)av); fail: SvREFCNT_dec (av); + DEC_DEC_DEPTH; return 0; } @@ -824,31 +1017,70 @@ { HV *hv = newHV (); + DEC_INC_DEPTH; decode_ws (dec); + if (*dec->cur == '}') ++dec->cur; else for (;;) { - SV *key, *value; - decode_ws (dec); EXPECT_CH ('"'); - key = decode_str (dec); - if (!key) - goto fail; + // heuristic: assume that + // a) decode_str + hv_store_ent are abysmally slow + // b) most hash keys are short, simple ascii text + // so try to "fast-match" such strings to avoid + // the overhead of hv_store_ent. + { + SV *value; + char *p = dec->cur; + char *e = p + 24; // only try up to 24 bytes + + for (;;) + { + 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); EXPECT_CH (':'); - value = decode_sv (dec); - if (!value) - { - SvREFCNT_dec (key); - goto fail; - } + 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; - //TODO: optimise - hv_store_ent (hv, key, value, 0); + decode_ws (dec); EXPECT_CH (':'); + + value = decode_sv (dec); + if (!value) + goto fail; + + hv_store (hv, key, len, value, 0); + + break; + } + + ++p; + } + } decode_ws (dec); @@ -864,10 +1096,12 @@ ++dec->cur; } + DEC_DEC_DEPTH; return newRV_noinc ((SV *)hv); fail: SvREFCNT_dec (hv); + DEC_DEC_DEPTH; return 0; } @@ -875,6 +1109,9 @@ decode_sv (dec_t *dec) { decode_ws (dec); + + // the beauty of JSON: you need exactly one character lookahead + // to parse anything. switch (*dec->cur) { case '"': ++dec->cur; return decode_str (dec); @@ -890,7 +1127,7 @@ if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4)) { dec->cur += 4; - return newSViv (1); + return SvREFCNT_inc (json_true); } else ERR ("'true' expected"); @@ -901,7 +1138,7 @@ if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5)) { dec->cur += 5; - return newSViv (0); + return SvREFCNT_inc (json_false); } else ERR ("'false' expected"); @@ -920,7 +1157,7 @@ break; default: - ERR ("malformed json string, neither array, object, number, string or atom"); + ERR ("malformed JSON string, neither array, object, number, string or atom"); break; } @@ -929,10 +1166,19 @@ } static SV * -decode_json (SV *string, UV flags) +decode_json (SV *string, U32 flags, UV *offset_return) { + dec_t dec; + UV offset; SV *sv; + SvGETMAGIC (string); + SvUPGRADE (string, SVt_PV); + + if (flags & F_MAXSIZE && SvCUR (string) > DEC_SIZE (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 (flags)); + if (flags & F_UTF8) sv_utf8_downgrade (string, 0); else @@ -940,19 +1186,41 @@ 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.flags = flags; + dec.cur = SvPVX (string); + dec.end = SvEND (string); + dec.err = 0; + dec.depth = 0; + dec.maxdepth = DEC_DEPTH (dec.flags); + *dec.end = 0; // this should basically be a nop, too, but make sure it's there 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.flags & F_UTF8 + ? dec.cur - SvPVX (string) + : utf8_distance (dec.cur, SvPVX (string)); + + if (offset_return) + *offset_return = offset; + } + if (!sv) { - IV offset = dec.flags & F_UTF8 - ? dec.cur - SvPVX (string) - : utf8_distance (dec.cur, SvPVX (string)); SV *uni = sv_newmortal (); // horrible hack to silence warning inside pv_uni_display @@ -964,7 +1232,7 @@ pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ); LEAVE; - croak ("%s, at character offset %d (%s)", + croak ("%s, at character offset %d [\"%s\"]", dec.err, (int)offset, dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)"); @@ -987,17 +1255,18 @@ { int i; - memset (decode_hexdigit, 0xff, 256); - for (i = 10; i--; ) - decode_hexdigit ['0' + i] = 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; - for (i = 7; i--; ) - { - decode_hexdigit ['a' + i] = 10 + i; - decode_hexdigit ['A' + i] = 10 + i; - } + json_stash = gv_stashpv ("JSON::XS" , 1); + json_boolean_stash = gv_stashpv ("JSON::XS::Boolean", 1); - json_stash = gv_stashpv ("JSON::XS", 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 @@ -1010,15 +1279,18 @@ SV *ascii (SV *self, int enable = 1) ALIAS: - ascii = F_ASCII - utf8 = F_UTF8 - indent = F_INDENT - canonical = F_CANONICAL - space_before = F_SPACE_BEFORE - space_after = F_SPACE_AFTER - pretty = F_PRETTY - allow_nonref = F_ALLOW_NONREF - shrink = F_SHRINK + 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 CODE: { UV *uv = SvJSON (self); @@ -1032,13 +1304,59 @@ OUTPUT: RETVAL +SV *max_depth (SV *self, UV max_depth = 0x80000000UL) + CODE: +{ + UV *uv = SvJSON (self); + UV log2 = 0; + + if (max_depth > 0x80000000UL) max_depth = 0x80000000UL; + + while ((1UL << log2) < max_depth) + ++log2; + + *uv = *uv & ~F_MAXDEPTH | (log2 << S_MAXDEPTH); + + RETVAL = newSVsv (self); +} + OUTPUT: + RETVAL + +SV *max_size (SV *self, UV max_size = 0) + CODE: +{ + UV *uv = SvJSON (self); + UV log2 = 0; + + if (max_size > 0x80000000UL) max_size = 0x80000000UL; + if (max_size == 1) max_size = 2; + + while ((1UL << log2) < max_size) + ++log2; + + *uv = *uv & ~F_MAXSIZE | (log2 << S_MAXSIZE); + + 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))); + XPUSHs (decode_json (jsonstr, *SvJSON (self), 0)); + +void decode_prefix (SV *self, SV *jsonstr) + PPCODE: +{ + UV offset; + EXTEND (SP, 2); + PUSHs (decode_json (jsonstr, *SvJSON (self), &offset)); + PUSHs (sv_2mortal (newSVuv (offset))); +} PROTOTYPES: ENABLE @@ -1046,11 +1364,11 @@ ALIAS: objToJson = 0 PPCODE: - XPUSHs (encode_json (scalar, F_UTF8)); + XPUSHs (encode_json (scalar, F_DEFAULT | F_UTF8)); void from_json (SV *jsonstr) ALIAS: jsonToObj = 0 PPCODE: - XPUSHs (decode_json (jsonstr, F_UTF8)); + XPUSHs (decode_json (jsonstr, F_DEFAULT | F_UTF8, 0));