--- JSON-XS/XS.xs 2007/04/06 20:37:22 1.25 +++ JSON-XS/XS.xs 2007/06/23 23:49:29 1.43 @@ -5,15 +5,27 @@ #include "assert.h" #include "string.h" #include "stdlib.h" +#include "stdio.h" + +#if defined(__BORLANDC__) || defined(_MSC_VER) +# define snprintf _snprintf // C compilers have this in stdio.h +#endif + +// some old perls do not have this, try to make it work, no +// guarentees, though. if it breaks, you get to keep the pieces. +#ifndef UTF8_MAXBYTES +# define UTF8_MAXBYTES 13 +#endif #define F_ASCII 0x00000001UL -#define F_UTF8 0x00000002UL -#define F_INDENT 0x00000004UL -#define F_CANONICAL 0x00000008UL -#define F_SPACE_BEFORE 0x00000010UL -#define F_SPACE_AFTER 0x00000020UL -#define F_ALLOW_NONREF 0x00000080UL -#define F_SHRINK 0x00000100UL +#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_MAXDEPTH 0xf8000000UL #define S_MAXDEPTH 27 @@ -28,13 +40,24 @@ #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) +#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::XS:: +static SV *json_true, *json_false; ///////////////////////////////////////////////////////////////////////////// // utility functions @@ -67,10 +90,10 @@ // 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) { @@ -98,19 +121,19 @@ 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); @@ -128,15 +151,15 @@ { 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++ = '\\'; @@ -164,7 +187,6 @@ 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); @@ -178,7 +200,7 @@ 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) { @@ -202,6 +224,11 @@ str += clen; } + else if (enc->flags & F_LATIN1) + { + *enc->cur++ = uch; + str += clen; + } else if (is_utf8) { need (enc, len += clen); @@ -213,7 +240,7 @@ } 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; } @@ -225,7 +252,7 @@ } } -static void +inline void encode_indent (enc_t *enc) { if (enc->flags & F_INDENT) @@ -238,14 +265,14 @@ } } -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) @@ -255,7 +282,7 @@ } } -static void +inline void encode_comma (enc_t *enc) { encode_ch (enc, ','); @@ -364,8 +391,13 @@ // 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))) @@ -473,17 +505,50 @@ } 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)); + // we assume we can always read an IV as a UV + if (SvUV (sv) & ~(UV)0x7fff) + { + // large integer, use the (rather slow) snprintf way. + need (enc, sizeof (UV) * 3); + enc->cur += + SvIsUV(sv) + ? snprintf (enc->cur, sizeof (UV) * 3, "%"UVuf, (UV)SvUVX (sv)) + : snprintf (enc->cur, sizeof (UV) * 3, "%"IVdf, (IV)SvIVX (sv)); + } + else + { + // optimise the "small number case" + // code will likely be branchless and use only a single multiplication + I32 i = SvIV (sv); + U32 u; + char digit, nz = 0; + + need (enc, 6); + + *enc->cur = '-'; enc->cur += i < 0 ? 1 : 0; + u = i < 0 ? -i : i; + + // convert to 4.28 fixed-point representation + u = u * ((0xfffffff + 10000) / 10000); // 10**5, 5 fractional digits + + // now output digit by digit, each time masking out the integer part + // and multiplying by 5 while moving the decimal point one to the right, + // resulting in a net multiplication by 10. + // we always write the digit to memory but conditionally increment + // the pointer, to ease the usage of conditional move instructions. + digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffff) * 5; + digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffff) * 5; + digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffff) * 5; + digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffff) * 5; + digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0' + } } else if (SvROK (sv)) encode_rv (enc, SvRV (sv)); @@ -512,10 +577,11 @@ 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); @@ -537,7 +603,7 @@ U32 maxdepth; // recursion depth limit } dec_t; -static void +inline void decode_ws (dec_t *dec) { for (;;) @@ -573,10 +639,10 @@ 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; @@ -594,41 +660,44 @@ { 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; @@ -636,12 +705,14 @@ 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; @@ -665,32 +736,32 @@ 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) { STRLEN clen; UV uch; - --dec->cur; + --dec_cur; - uch = decode_utf8 (dec->cur, dec->end - dec->cur, &clen); + 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 { - --dec->cur; + --dec_cur; if (!ch) ERR ("unexpected end of string while parsing JSON string"); @@ -713,9 +784,9 @@ sv = newSVpvn (buf, len); } } - while (*dec->cur != '"'); + while (*dec_cur != '"'); - ++dec->cur; + ++dec_cur; if (sv) { @@ -728,9 +799,11 @@ else sv = newSVpvn ("", 0); + dec->cur = dec_cur; return sv; fail: + dec->cur = dec_cur; return 0; } @@ -798,18 +871,41 @@ 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: @@ -920,6 +1016,9 @@ 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); @@ -935,7 +1034,7 @@ 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"); @@ -946,7 +1045,7 @@ 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"); @@ -974,11 +1073,13 @@ } static SV * -decode_json (SV *string, U32 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_UTF8) @@ -995,14 +1096,34 @@ dec.depth = 0; dec.maxdepth = DEC_DEPTH (dec.flags); - *dec.end = 0; // this should basically be a nop, too, but make sure its there + *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 @@ -1037,8 +1158,6 @@ { int i; - memset (decode_hexdigit, 0xff, 256); - for (i = 0; i < 256; ++i) decode_hexdigit [i] = i >= '0' && i <= '9' ? i - '0' @@ -1047,6 +1166,9 @@ : -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 @@ -1060,6 +1182,7 @@ SV *ascii (SV *self, int enable = 1) ALIAS: ascii = F_ASCII + latin1 = F_LATIN1 utf8 = F_UTF8 indent = F_INDENT canonical = F_CANONICAL @@ -1105,7 +1228,16 @@ 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 @@ -1119,5 +1251,5 @@ ALIAS: jsonToObj = 0 PPCODE: - XPUSHs (decode_json (jsonstr, F_DEFAULT | F_UTF8)); + XPUSHs (decode_json (jsonstr, F_DEFAULT | F_UTF8, 0));