--- JSON-XS/XS.xs 2007/06/23 22:56:08 1.42 +++ JSON-XS/XS.xs 2008/03/26 01:40:43 1.81 @@ -2,10 +2,12 @@ #include "perl.h" #include "XSUB.h" -#include "assert.h" -#include "string.h" -#include "stdlib.h" -#include "stdio.h" +#include +#include +#include +#include +#include +#include #if defined(__BORLANDC__) || defined(_MSC_VER) # define snprintf _snprintf // C compilers have this in stdio.h @@ -17,22 +19,29 @@ # 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_MAXDEPTH 0xf8000000UL -#define S_MAXDEPTH 27 +#define IVUV_MAXCHARS (sizeof (UV) * CHAR_BIT * 28 / 93 + 2) -#define DEC_DEPTH(flags) (1UL << ((flags & F_MAXDEPTH) >> S_MAXDEPTH)) +#define F_ASCII 0x00000001UL +#define F_LATIN1 0x00000002UL +#define F_UTF8 0x00000004UL +#define F_INDENT 0x00000008UL +#define F_CANONICAL 0x00000010UL +#define F_SPACE_BEFORE 0x00000020UL +#define F_SPACE_AFTER 0x00000040UL +#define F_ALLOW_NONREF 0x00000100UL +#define F_SHRINK 0x00000200UL +#define F_ALLOW_BLESSED 0x00000400UL +#define F_CONV_BLESSED 0x00000800UL +#define F_RELAXED 0x00001000UL + +#define F_MAXDEPTH 0xf8000000UL +#define S_MAXDEPTH 27 +#define F_MAXSIZE 0x01f00000UL +#define S_MAXSIZE 20 +#define F_HOOK 0x00080000UL // some hooks exist, so slow-path processing -// F_SELFCONVERT? <=> to_json/toJson -// F_BLESSED? <=> { $__class__$ => } +#define DEC_DEPTH(flags) (1UL << ((flags & F_MAXDEPTH) >> S_MAXDEPTH)) +#define DEC_SIZE(flags) (1UL << ((flags & F_MAXSIZE ) >> S_MAXSIZE )) #define F_PRETTY F_INDENT | F_SPACE_BEFORE | F_SPACE_AFTER #define F_DEFAULT (9UL << S_MAXDEPTH) @@ -46,31 +55,56 @@ #define SE } while (0) #if __GNUC__ >= 3 -# define expect(expr,value) __builtin_expect ((expr),(value)) -# define inline inline +# define expect(expr,value) __builtin_expect ((expr), (value)) +# define INLINE static inline #else # define expect(expr,value) (expr) -# define inline static +# 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:: +#define IN_RANGE_INC(type,val,beg,end) \ + ((unsigned type)((unsigned type)(val) - (unsigned type)(beg)) \ + <= (unsigned type)((unsigned type)(end) - (unsigned type)(beg))) + +#ifdef USE_ITHREADS +# define JSON_SLOW 1 +# define JSON_STASH (json_stash ? json_stash : gv_stashpv ("JSON::XS", 1)) +#else +# define JSON_SLOW 0 +# define JSON_STASH json_stash +#endif -///////////////////////////////////////////////////////////////////////////// -// utility functions +static HV *json_stash, *json_boolean_stash; // JSON::XS:: +static SV *json_true, *json_false; -static UV * -SvJSON (SV *sv) -{ - if (!(SvROK (sv) && SvOBJECT (SvRV (sv)) && SvSTASH (SvRV (sv)) == json_stash)) - croak ("object is not of type JSON::XS"); +enum { + INCR_M_WS = 0, // initial whitespace skipping, must be 0 + INCR_M_STR, // inside string + INCR_M_BS, // inside backslash + INCR_M_JSON // outside anything, count nesting +}; + +#define INCR_DONE(json) (!(json)->incr_nest && (json)->incr_mode == INCR_M_JSON) + +typedef struct { + U32 flags; + SV *cb_object; + HV *cb_sk_object; + + // for the incremental parser + SV *incr_text; // the source text so far + STRLEN incr_pos; // the current offset into the text + int incr_nest; // {[]}-nesting level + int incr_mode; +} JSON; - return &SvUVX (SvRV (sv)); -} +///////////////////////////////////////////////////////////////////////////// +// utility functions -static void +INLINE void shrink (SV *sv) { sv_utf8_downgrade (sv, 1); @@ -89,21 +123,42 @@ // we special-case "safe" characters from U+80 .. U+7FF, // but use the very good perl function to parse anything else. // note that we never call this function for a ascii codepoints -inline UV +INLINE UV decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen) { - if (expect_false (s[0] > 0xdf || s[0] < 0xc2)) - return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY); - else if (len > 1 && s[1] >= 0x80 && s[1] <= 0xbf) + if (expect_true (len >= 2 + && IN_RANGE_INC (char, s[0], 0xc2, 0xdf) + && IN_RANGE_INC (char, s[1], 0x80, 0xbf))) { *clen = 2; return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f); } else - { - *clen = (STRLEN)-1; - return (UV)-1; - } + return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY); +} + +// likewise for encoding, also never called for ascii codepoints +// this function takes advantage of this fact, although current gccs +// seem to optimise the check for >= 0x80 away anyways +INLINE unsigned char * +encode_utf8 (unsigned char *s, UV ch) +{ + if (expect_false (ch < 0x000080)) + *s++ = ch; + else if (expect_true (ch < 0x000800)) + *s++ = 0xc0 | ( ch >> 6), + *s++ = 0x80 | ( ch & 0x3f); + else if ( ch < 0x010000) + *s++ = 0xe0 | ( ch >> 12), + *s++ = 0x80 | ((ch >> 6) & 0x3f), + *s++ = 0x80 | ( ch & 0x3f); + else if ( ch < 0x110000) + *s++ = 0xf0 | ( ch >> 18), + *s++ = 0x80 | ((ch >> 12) & 0x3f), + *s++ = 0x80 | ((ch >> 6) & 0x3f), + *s++ = 0x80 | ( ch & 0x3f); + + return s; } ///////////////////////////////////////////////////////////////////////////// @@ -115,12 +170,13 @@ char *cur; // SvPVX (sv) + current output position char *end; // SvEND (sv) SV *sv; // result scalar - U32 flags; // F_* + JSON json; U32 indent; // indentation level U32 maxdepth; // max. indentation/recursion level + UV limit; // escape character values >= this value when encoding } enc_t; -inline void +INLINE void need (enc_t *enc, STRLEN len) { if (expect_false (enc->cur + len >= enc->end)) @@ -132,7 +188,7 @@ } } -inline void +INLINE void encode_ch (enc_t *enc, char ch) { need (enc, 1); @@ -196,13 +252,13 @@ clen = 1; } - if (uch > 0x10FFFFUL) - croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch); - - if (uch < 0x80 || enc->flags & F_ASCII || (enc->flags & F_LATIN1 && uch > 0xFF)) + if (uch < 0x80/*0x20*/ || uch >= enc->limit) { - if (uch > 0xFFFFUL) + if (uch >= 0x10000UL) { + if (uch >= 0x110000UL) + croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch); + need (enc, len += 11); sprintf (enc->cur, "\\u%04x\\u%04x", (int)((uch - 0x10000) / 0x400 + 0xD800), @@ -223,7 +279,7 @@ str += clen; } - else if (enc->flags & F_LATIN1) + else if (enc->json.flags & F_LATIN1) { *enc->cur++ = uch; str += clen; @@ -240,7 +296,7 @@ else { need (enc, len += UTF8_MAXBYTES - 1); // never more than 11 bytes needed - enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0); + enc->cur = encode_utf8 (enc->cur, uch); ++str; } } @@ -251,10 +307,10 @@ } } -inline void +INLINE void encode_indent (enc_t *enc) { - if (enc->flags & F_INDENT) + if (enc->json.flags & F_INDENT) { int spaces = enc->indent * INDENT_STEP; @@ -264,31 +320,31 @@ } } -inline void +INLINE void encode_space (enc_t *enc) { need (enc, 1); encode_ch (enc, ' '); } -inline void +INLINE void encode_nl (enc_t *enc) { - if (enc->flags & F_INDENT) + if (enc->json.flags & F_INDENT) { need (enc, 1); encode_ch (enc, '\n'); } } -inline void +INLINE void encode_comma (enc_t *enc) { encode_ch (enc, ','); - if (enc->flags & F_INDENT) + if (enc->json.flags & F_INDENT) encode_nl (enc); - else if (enc->flags & F_SPACE_AFTER) + else if (enc->json.flags & F_SPACE_AFTER) encode_space (enc); } @@ -302,26 +358,35 @@ if (enc->indent >= enc->maxdepth) croak ("data structure too deep (hit recursion limit)"); - encode_ch (enc, '['); encode_nl (enc); - ++enc->indent; - - for (i = 0; i <= len; ++i) + encode_ch (enc, '['); + + if (len >= 0) { - encode_indent (enc); - encode_sv (enc, *av_fetch (av, i, 0)); + encode_nl (enc); ++enc->indent; - if (i < len) - encode_comma (enc); - } + for (i = 0; i <= len; ++i) + { + SV **svp = av_fetch (av, i, 0); - encode_nl (enc); + encode_indent (enc); + + if (svp) + encode_sv (enc, *svp); + else + encode_str (enc, "null", 4, 0); + + if (i < len) + encode_comma (enc); + } - --enc->indent; - encode_indent (enc); encode_ch (enc, ']'); + encode_nl (enc); --enc->indent; encode_indent (enc); + } + + encode_ch (enc, ']'); } static void -encode_he (enc_t *enc, HE *he) +encode_hk (enc_t *enc, HE *he) { encode_ch (enc, '"'); @@ -341,10 +406,9 @@ encode_ch (enc, '"'); - if (enc->flags & F_SPACE_BEFORE) encode_space (enc); + if (enc->json.flags & F_SPACE_BEFORE) encode_space (enc); encode_ch (enc, ':'); - if (enc->flags & F_SPACE_AFTER ) encode_space (enc); - encode_sv (enc, HeVAL (he)); + if (enc->json.flags & F_SPACE_AFTER ) encode_space (enc); } // compare hash entries, used when all keys are bytestrings @@ -359,8 +423,8 @@ STRLEN la = HeKLEN (a); STRLEN lb = HeKLEN (b); - if (!(cmp = memcmp (HeKEY (a), HeKEY (b), la < lb ? la : lb))) - cmp = la - lb; + if (!(cmp = memcmp (HeKEY (b), HeKEY (a), lb < la ? lb : la))) + cmp = lb - la; return cmp; } @@ -369,29 +433,44 @@ static int he_cmp_slow (const void *a, const void *b) { - return sv_cmp (HeSVKEY_force (*(HE **)a), HeSVKEY_force (*(HE **)b)); + return sv_cmp (HeSVKEY_force (*(HE **)b), HeSVKEY_force (*(HE **)a)); } static void encode_hv (enc_t *enc, HV *hv) { - int count, i; + HE *he; if (enc->indent >= enc->maxdepth) croak ("data structure too deep (hit recursion limit)"); - encode_ch (enc, '{'); encode_nl (enc); ++enc->indent; + encode_ch (enc, '{'); - 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->json.flags & F_CANONICAL) { - // 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) + int count = hv_iterinit (hv); + + if (SvMAGICAL (hv)) + { + // need to count by iterating. could improve by dynamically building the vector below + // but I don't care for the speed of this special case. + // note also that we will run into undefined behaviour when the two iterations + // do not result in the same count, something I might care for in some later release. + + count = 0; + while (hv_iternext (hv)) + ++count; + + hv_iterinit (hv); + } + + if (count) { - int fast = 1; - HE *he; + int i, fast = 1; #if defined(__BORLANDC__) || defined(_MSC_VER) HE **hes = _alloca (count * sizeof (HE)); #else @@ -428,37 +507,46 @@ LEAVE; } - for (i = 0; i < count; ++i) + encode_nl (enc); ++enc->indent; + + while (count--) { encode_indent (enc); - encode_he (enc, hes [i]); + he = hes [count]; + encode_hk (enc, he); + encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he)); - if (i < count - 1) + if (count) encode_comma (enc); } - encode_nl (enc); + encode_nl (enc); --enc->indent; encode_indent (enc); } - else - { - HE *he = hv_iternext (hv); + } + else + { + if (hv_iterinit (hv) || SvMAGICAL (hv)) + if ((he = hv_iternext (hv))) + { + encode_nl (enc); ++enc->indent; - for (;;) - { - encode_indent (enc); - encode_he (enc, he); + for (;;) + { + encode_indent (enc); + encode_hk (enc, he); + encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he)); - if (!(he = hv_iternext (hv))) - break; + if (!(he = hv_iternext (hv))) + break; - encode_comma (enc); - } + encode_comma (enc); + } - encode_nl (enc); - } + encode_nl (enc); --enc->indent; encode_indent (enc); + } } - --enc->indent; encode_indent (enc); encode_ch (enc, '}'); + encode_ch (enc, '}'); } // encode objects, arrays and special \0=false and \1=true values. @@ -470,16 +558,81 @@ SvGETMAGIC (sv); svt = SvTYPE (sv); - if (svt == SVt_PVHV) + if (expect_false (SvOBJECT (sv))) + { + HV *stash = !JSON_SLOW || json_boolean_stash + ? json_boolean_stash + : gv_stashpv ("JSON::XS::Boolean", 1); + + if (SvSTASH (sv) == stash) + { + if (SvIV (sv)) + encode_str (enc, "true", 4, 0); + else + encode_str (enc, "false", 5, 0); + } + else + { +#if 0 + if (0 && sv_derived_from (rv, "JSON::Literal")) + { + // not yet + } +#endif + if (enc->json.flags & F_CONV_BLESSED) + { + // we re-bless the reference to get overload and other niceties right + GV *to_json = gv_fetchmethod_autoload (SvSTASH (sv), "TO_JSON", 0); + + if (to_json) + { + dSP; + + ENTER; SAVETMPS; PUSHMARK (SP); + XPUSHs (sv_bless (sv_2mortal (newRV_inc (sv)), SvSTASH (sv))); + + // calling with G_SCALAR ensures that we always get a 1 return value + PUTBACK; + call_sv ((SV *)GvCV (to_json), G_SCALAR); + SPAGAIN; + + // catch this surprisingly common error + if (SvROK (TOPs) && SvRV (TOPs) == sv) + croak ("%s::TO_JSON method returned same object as was passed instead of a new one", HvNAME (SvSTASH (sv))); + + sv = POPs; + PUTBACK; + + encode_sv (enc, sv); + + FREETMPS; LEAVE; + } + else if (enc->json.flags & F_ALLOW_BLESSED) + encode_str (enc, "null", 4, 0); + else + croak ("encountered object '%s', but neither allow_blessed enabled nor TO_JSON method available on it", + SvPV_nolen (sv_2mortal (newRV_inc (sv)))); + } + else if (enc->json.flags & F_ALLOW_BLESSED) + encode_str (enc, "null", 4, 0); + else + croak ("encountered object '%s', but neither allow_blessed nor convert_blessed settings are enabled", + SvPV_nolen (sv_2mortal (newRV_inc (sv)))); + } + } + else if (svt == SVt_PVHV) encode_hv (enc, (HV *)sv); else if (svt == SVt_PVAV) encode_av (enc, (AV *)sv); else if (svt < SVt_PVAV) { - if (SvNIOK (sv) && SvIV (sv) == 0) - encode_str (enc, "false", 5, 0); - else if (SvNIOK (sv) && SvIV (sv) == 1) + STRLEN len = 0; + char *pv = svt ? SvPV (sv, len) : 0; + + if (len == 1 && *pv == '1') encode_str (enc, "true", 4, 0); + else if (len == 1 && *pv == '0') + encode_str (enc, "false", 5, 0); else croak ("cannot encode reference to scalar '%s' unless the scalar is 0 or 1", SvPV_nolen (sv_2mortal (newRV_inc (sv)))); @@ -511,21 +664,16 @@ } else if (SvIOKp (sv)) { - // we assume we can always read an IV as a UV - if (SvUV (sv) & ~(UV)0x7fff) - { - // large integer, use the (rather slow) snprintf way. - need (enc, sizeof (UV) * 3); - enc->cur += - SvIsUV(sv) - ? snprintf (enc->cur, sizeof (UV) * 3, "%"UVuf, (UV)SvUVX (sv)) - : snprintf (enc->cur, sizeof (UV) * 3, "%"IVdf, (IV)SvIVX (sv)); - } - else + // we assume we can always read an IV as a UV and vice versa + // we assume two's complement + // we assume no aliasing issues in the union + if (SvIsUV (sv) ? SvUVX (sv) <= 59000 + : SvIVX (sv) <= 59000 && SvIVX (sv) >= -59000) { // optimise the "small number case" // code will likely be branchless and use only a single multiplication - I32 i = SvIV (sv); + // works for numbers up to 59074 + I32 i = SvIVX (sv); U32 u; char digit, nz = 0; @@ -541,13 +689,22 @@ // 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; + // the pointer, to enable the use of conditional move instructions. + digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffffUL) * 5; + digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffffUL) * 5; + digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffffUL) * 5; + digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffffUL) * 5; digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0' } + else + { + // large integer, use the (rather slow) snprintf way. + need (enc, IVUV_MAXCHARS); + enc->cur += + SvIsUV(sv) + ? snprintf (enc->cur, IVUV_MAXCHARS, "%"UVuf, (UV)SvUVX (sv)) + : snprintf (enc->cur, IVUV_MAXCHARS, "%"IVdf, (IV)SvIVX (sv)); + } } else if (SvROK (sv)) encode_rv (enc, SvRV (sv)); @@ -559,19 +716,22 @@ } static SV * -encode_json (SV *scalar, U32 flags) +encode_json (SV *scalar, JSON *json) { enc_t enc; - if (!(flags & F_ALLOW_NONREF) && !SvROK (scalar)) + if (!(json->flags & F_ALLOW_NONREF) && !SvROK (scalar)) croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)"); - enc.flags = flags; + enc.json = *json; enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE)); enc.cur = SvPVX (enc.sv); enc.end = SvEND (enc.sv); enc.indent = 0; - enc.maxdepth = DEC_DEPTH (flags); + enc.maxdepth = DEC_DEPTH (enc.json.flags); + enc.limit = enc.json.flags & F_ASCII ? 0x000080UL + : enc.json.flags & F_LATIN1 ? 0x000100UL + : 0x110000UL; SvPOK_only (enc.sv); encode_sv (&enc, scalar); @@ -579,10 +739,10 @@ 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))) + if (!(enc.json.flags & (F_ASCII | F_LATIN1 | F_UTF8))) SvUTF8_on (enc.sv); - if (enc.flags & F_SHRINK) + if (enc.json.flags & F_SHRINK) shrink (enc.sv); return enc.sv; @@ -597,21 +757,41 @@ char *cur; // current parser pointer char *end; // end of input string const char *err; // parse error, if != 0 - U32 flags; // F_* + JSON json; U32 depth; // recursion depth U32 maxdepth; // recursion depth limit } dec_t; -inline void +INLINE void +decode_comment (dec_t *dec) +{ + // only '#'-style comments allowed a.t.m. + + while (*dec->cur && *dec->cur != 0x0a && *dec->cur != 0x0d) + ++dec->cur; +} + +INLINE void decode_ws (dec_t *dec) { for (;;) { char ch = *dec->cur; - if (ch > 0x20 - || (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)) - break; + if (ch > 0x20) + { + if (expect_false (ch == '#')) + { + if (dec->json.flags & F_RELAXED) + decode_comment (dec); + else + break; + } + else + break; + } + else if (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09) + break; // parse error, but let higher level handle it, gives better error messages ++dec->cur; } @@ -727,7 +907,7 @@ { utf8 = 1; - cur = (char *)uvuni_to_utf8_flags (cur, hi, 0); + cur = encode_utf8 (cur, hi); } else *cur++ = hi; @@ -739,7 +919,7 @@ ERR ("illegal backslash escape sequence in string"); } } - else if (expect_true (ch >= 0x20 && ch <= 0x7f)) + else if (expect_true (ch >= 0x20 && ch < 0x80)) *cur++ = ch; else if (ch >= 0x80) { @@ -870,27 +1050,31 @@ if (!is_nv) { - // special case the rather common 1..4-digit-int case, assumes 32 bit ints or so + int len = dec->cur - start; + + // special case the rather common 1..5-digit-int case if (*start == '-') - switch (dec->cur - start) + switch (len) { - case 2: return newSViv (-( start [1] - '0' * 1)); - case 3: return newSViv (-( start [1] * 10 + start [2] - '0' * 11)); - case 4: return newSViv (-( start [1] * 100 + start [2] * 10 + start [3] - '0' * 111)); - case 5: return newSViv (-(start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 1111)); + 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)); + case 6: return newSViv (-(start [1] * 10000 + start [2] * 1000 + start [3] * 100 + start [4] * 10 + start [5] - '0' * 11111)); } else - switch (dec->cur - start) + switch (len) { - case 1: return newSViv ( start [0] - '0' * 1); - case 2: return newSViv ( start [0] * 10 + start [1] - '0' * 11); - case 3: return newSViv ( start [0] * 100 + start [1] * 10 + start [2] - '0' * 111); - case 4: return newSViv ( start [0] * 1000 + start [1] * 100 + start [2] * 10 + start [3] - '0' * 1111); + 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); + case 5: return newSViv ( start [0] * 10000 + start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 11111); } { UV uv; - int numtype = grok_number (start, dec->cur - start, &uv); + int numtype = grok_number (start, len, &uv); if (numtype & IS_NUMBER_IN_UV) if (numtype & IS_NUMBER_NEG) { @@ -899,12 +1083,24 @@ } else return newSVuv (uv); - - // here would likely be the place for bigint support } + + len -= *start == '-' ? 1 : 0; + + // does not fit into IV or UV, try NV + if ((sizeof (NV) == sizeof (double) && DBL_DIG >= len) + #if defined (LDBL_DIG) + || (sizeof (NV) == sizeof (long double) && LDBL_DIG >= len) + #endif + ) + // fits into NV without loss of precision + return newSVnv (Atof (start)); + + // everything else fails, convert it to a string + return newSVpvn (start, dec->cur - start); } - // if we ever support bigint or bigfloat, this is the place for bigfloat + // loss of precision here return newSVnv (Atof (start)); fail: @@ -944,6 +1140,14 @@ ERR (", or ] expected while parsing array"); ++dec->cur; + + decode_ws (dec); + + if (*dec->cur == ']' && dec->json.flags & F_RELAXED) + { + ++dec->cur; + break; + } } DEC_DEC_DEPTH; @@ -958,6 +1162,7 @@ static SV * decode_hv (dec_t *dec) { + SV *sv; HV *hv = newHV (); DEC_INC_DEPTH; @@ -968,25 +1173,65 @@ else for (;;) { - SV *key, *value; + EXPECT_CH ('"'); - decode_ws (dec); EXPECT_CH ('"'); + // heuristic: assume that + // a) decode_str + hv_store_ent are abysmally slow. + // b) most hash keys are short, simple ascii text. + // => try to "fast-match" such strings to avoid + // the overhead of decode_str + hv_store_ent. + { + SV *value; + char *p = dec->cur; + char *e = p + 24; // only try up to 24 bytes - key = decode_str (dec); - if (!key) - goto fail; + for (;;) + { + // the >= 0x80 is false on most architectures + if (p == e || *p < 0x20 || *p >= 0x80 || *p == '\\') + { + // slow path, back up and use decode_str + SV *key = decode_str (dec); + if (!key) + goto fail; + + decode_ws (dec); EXPECT_CH (':'); + + decode_ws (dec); + value = decode_sv (dec); + if (!value) + { + SvREFCNT_dec (key); + goto fail; + } - decode_ws (dec); EXPECT_CH (':'); + hv_store_ent (hv, key, value, 0); + SvREFCNT_dec (key); - value = decode_sv (dec); - if (!value) - { - SvREFCNT_dec (key); - goto fail; - } + break; + } + else if (*p == '"') + { + // fast path, got a simple key + char *key = dec->cur; + int len = p - key; + dec->cur = p + 1; + + decode_ws (dec); EXPECT_CH (':'); - hv_store_ent (hv, key, value, 0); - SvREFCNT_dec (key); + decode_ws (dec); + value = decode_sv (dec); + if (!value) + goto fail; + + hv_store (hv, key, len, value, 0); + + break; + } + + ++p; + } + } decode_ws (dec); @@ -1000,10 +1245,78 @@ ERR (", or } expected while parsing object/hash"); ++dec->cur; + + decode_ws (dec); + + if (*dec->cur == '}' && dec->json.flags & F_RELAXED) + { + ++dec->cur; + break; + } } DEC_DEC_DEPTH; - return newRV_noinc ((SV *)hv); + sv = newRV_noinc ((SV *)hv); + + // check filter callbacks + if (dec->json.flags & F_HOOK) + { + if (dec->json.cb_sk_object && HvKEYS (hv) == 1) + { + HE *cb, *he; + + hv_iterinit (hv); + he = hv_iternext (hv); + hv_iterinit (hv); + + // the next line creates a mortal sv each time its called. + // might want to optimise this for common cases. + cb = hv_fetch_ent (dec->json.cb_sk_object, hv_iterkeysv (he), 0, 0); + + if (cb) + { + dSP; + int count; + + ENTER; SAVETMPS; PUSHMARK (SP); + XPUSHs (HeVAL (he)); + + PUTBACK; count = call_sv (HeVAL (cb), G_ARRAY); SPAGAIN; + + if (count == 1) + { + sv = newSVsv (POPs); + FREETMPS; LEAVE; + return sv; + } + + FREETMPS; LEAVE; + } + } + + if (dec->json.cb_object) + { + dSP; + int count; + + ENTER; SAVETMPS; PUSHMARK (SP); + XPUSHs (sv_2mortal (sv)); + + PUTBACK; count = call_sv (dec->json.cb_object, G_ARRAY); SPAGAIN; + + if (count == 1) + { + sv = newSVsv (POPs); + FREETMPS; LEAVE; + return sv; + } + + SvREFCNT_inc (sv); + FREETMPS; LEAVE; + } + } + + return sv; fail: SvREFCNT_dec (hv); @@ -1014,15 +1327,13 @@ static SV * decode_sv (dec_t *dec) { - decode_ws (dec); - // the beauty of JSON: you need exactly one character lookahead - // to parse anything. + // to parse everything. 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 '[': ++dec->cur; return decode_av (dec); + case '{': ++dec->cur; return decode_hv (dec); case '-': case '0': case '1': case '2': case '3': case '4': @@ -1033,7 +1344,10 @@ if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4)) { dec->cur += 4; - return newSViv (1); +#if JSON_SLOW + json_true = get_sv ("JSON::XS::true", 1); SvREADONLY_on (json_true); +#endif + return SvREFCNT_inc (json_true); } else ERR ("'true' expected"); @@ -1044,7 +1358,10 @@ if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5)) { dec->cur += 5; - return newSViv (0); +#if JSON_SLOW + json_false = get_sv ("JSON::XS::false", 1); SvREADONLY_on (json_false); +#endif + return SvREFCNT_inc (json_false); } else ERR ("'false' expected"); @@ -1072,30 +1389,39 @@ } static SV * -decode_json (SV *string, U32 flags, UV *offset_return) +decode_json (SV *string, JSON *json, STRLEN *offset_return) { dec_t dec; - UV offset; + STRLEN offset; SV *sv; SvGETMAGIC (string); SvUPGRADE (string, SVt_PV); - if (flags & F_UTF8) + if (json->flags & F_MAXSIZE && SvCUR (string) > DEC_SIZE (json->flags)) + croak ("attempted decode of JSON text of %lu bytes size, but max_size is set to %lu", + (unsigned long)SvCUR (string), (unsigned long)DEC_SIZE (json->flags)); + + if (json->flags & F_UTF8) sv_utf8_downgrade (string, 0); else sv_utf8_upgrade (string); SvGROW (string, SvCUR (string) + 1); // should basically be a NOP - dec.flags = flags; + dec.json = *json; dec.cur = SvPVX (string); dec.end = SvEND (string); dec.err = 0; dec.depth = 0; - dec.maxdepth = DEC_DEPTH (dec.flags); + dec.maxdepth = DEC_DEPTH (dec.json.flags); + + if (dec.json.cb_object || dec.json.cb_sk_object) + dec.json.flags |= F_HOOK; *dec.end = 0; // this should basically be a nop, too, but make sure it's there + + decode_ws (&dec); sv = decode_sv (&dec); if (!(offset_return || !sv)) @@ -1113,7 +1439,7 @@ if (offset_return || !sv) { - offset = dec.flags & F_UTF8 + offset = dec.json.flags & F_UTF8 ? dec.cur - SvPVX (string) : utf8_distance (dec.cur, SvPVX (string)); @@ -1142,13 +1468,129 @@ sv = sv_2mortal (sv); - if (!(dec.flags & F_ALLOW_NONREF) && !SvROK (sv)) + if (!(dec.json.flags & F_ALLOW_NONREF) && !SvROK (sv)) croak ("JSON text must be an object or array (but found number, string, true, false or null, use allow_nonref to allow this)"); return sv; } ///////////////////////////////////////////////////////////////////////////// +// incremental parser + +static void +incr_parse (JSON *self) +{ + const char *p = SvPVX (self->incr_text) + self->incr_pos; + + for (;;) + { + //printf ("loop pod %d *p<%c><%s>, mode %d nest %d\n", p - SvPVX (self->incr_text), *p, p, self->incr_mode, self->incr_nest);//D + switch (self->incr_mode) + { + // only used for intiial whitespace skipping + case INCR_M_WS: + for (;;) + { + if (*p > 0x20) + { + self->incr_mode = INCR_M_JSON; + goto incr_m_json; + } + else if (!*p) + goto interrupt; + + ++p; + } + + // skip a single char inside a string (for \\-processing) + case INCR_M_BS: + if (!*p) + goto interrupt; + + ++p; + self->incr_mode = INCR_M_STR; + goto incr_m_str; + + // inside a string + case INCR_M_STR: + incr_m_str: + for (;;) + { + if (*p == '"') + { + ++p; + self->incr_mode = INCR_M_JSON; + + if (!self->incr_nest) + goto interrupt; + + goto incr_m_json; + } + else if (*p == '\\') + { + ++p; // "virtually" consumes character after \ + + if (!*p) // if at end of string we have to switch modes + { + self->incr_mode = INCR_M_BS; + goto interrupt; + } + } + else if (!*p) + goto interrupt; + + ++p; + } + + // after initial ws, outside string + case INCR_M_JSON: + incr_m_json: + for (;;) + { + switch (*p++) + { + case 0: + --p; + goto interrupt; + + case 0x09: + case 0x0a: + case 0x0d: + case 0x20: + if (!self->incr_nest) + { + --p; // do not eat the whitespace, let the next round do it + goto interrupt; + } + break; + + case '"': + self->incr_mode = INCR_M_STR; + goto incr_m_str; + + case '[': + case '{': + ++self->incr_nest; + break; + + case ']': + case '}': + if (!--self->incr_nest) + goto interrupt; + } + } + } + + modechange: + ; + } + +interrupt: + self->incr_pos = p - SvPVX (self->incr_text); + //printf ("return pos %d mode %d nest %d\n", self->incr_pos, self->incr_mode, self->incr_nest);//D +} + +///////////////////////////////////////////////////////////////////////////// // XS interface functions MODULE = JSON::XS PACKAGE = JSON::XS @@ -1164,46 +1606,78 @@ : i >= 'A' && i <= 'F' ? i - 'A' + 10 : -1; - json_stash = gv_stashpv ("JSON::XS", 1); + json_stash = gv_stashpv ("JSON::XS" , 1); + json_boolean_stash = gv_stashpv ("JSON::XS::Boolean", 1); + + json_true = get_sv ("JSON::XS::true" , 1); SvREADONLY_on (json_true ); + json_false = get_sv ("JSON::XS::false", 1); SvREADONLY_on (json_false); } PROTOTYPES: DISABLE -SV *new (char *dummy) +void CLONE (...) CODE: - RETVAL = sv_bless (newRV_noinc (newSVuv (F_DEFAULT)), json_stash); - OUTPUT: - RETVAL + json_stash = 0; + json_boolean_stash = 0; -SV *ascii (SV *self, int enable = 1) +void new (char *klass) + PPCODE: +{ + SV *pv = NEWSV (0, sizeof (JSON)); + SvPOK_only (pv); + Zero (SvPVX (pv), 1, JSON); + ((JSON *)SvPVX (pv))->flags = F_DEFAULT; + XPUSHs (sv_2mortal (sv_bless ( + newRV_noinc (pv), + strEQ (klass, "JSON::XS") ? JSON_STASH : gv_stashpv (klass, 1) + ))); +} + +void ascii (JSON *self, int enable = 1) ALIAS: - ascii = F_ASCII - latin1 = F_LATIN1 - utf8 = F_UTF8 - indent = F_INDENT - canonical = F_CANONICAL - space_before = F_SPACE_BEFORE - space_after = F_SPACE_AFTER - pretty = F_PRETTY - allow_nonref = F_ALLOW_NONREF - shrink = F_SHRINK - CODE: + ascii = F_ASCII + latin1 = F_LATIN1 + utf8 = F_UTF8 + indent = F_INDENT + canonical = F_CANONICAL + space_before = F_SPACE_BEFORE + space_after = F_SPACE_AFTER + pretty = F_PRETTY + allow_nonref = F_ALLOW_NONREF + shrink = F_SHRINK + allow_blessed = F_ALLOW_BLESSED + convert_blessed = F_CONV_BLESSED + relaxed = F_RELAXED + PPCODE: { - UV *uv = SvJSON (self); if (enable) - *uv |= ix; + self->flags |= ix; else - *uv &= ~ix; + self->flags &= ~ix; - RETVAL = newSVsv (self); + XPUSHs (ST (0)); } - OUTPUT: - RETVAL -SV *max_depth (SV *self, UV max_depth = 0x80000000UL) - CODE: +void get_ascii (JSON *self) + ALIAS: + get_ascii = F_ASCII + get_latin1 = F_LATIN1 + get_utf8 = F_UTF8 + get_indent = F_INDENT + get_canonical = F_CANONICAL + get_space_before = F_SPACE_BEFORE + get_space_after = F_SPACE_AFTER + get_allow_nonref = F_ALLOW_NONREF + get_shrink = F_SHRINK + get_allow_blessed = F_ALLOW_BLESSED + get_convert_blessed = F_CONV_BLESSED + get_relaxed = F_RELAXED + PPCODE: + XPUSHs (boolSV (self->flags & ix)); + +void max_depth (JSON *self, UV max_depth = 0x80000000UL) + PPCODE: { - UV *uv = SvJSON (self); UV log2 = 0; if (max_depth > 0x80000000UL) max_depth = 0x80000000UL; @@ -1211,41 +1685,186 @@ while ((1UL << log2) < max_depth) ++log2; - *uv = *uv & ~F_MAXDEPTH | (log2 << S_MAXDEPTH); + self->flags = self->flags & ~F_MAXDEPTH | (log2 << S_MAXDEPTH); - RETVAL = newSVsv (self); + XPUSHs (ST (0)); } + +U32 get_max_depth (JSON *self) + CODE: + RETVAL = DEC_DEPTH (self->flags); OUTPUT: RETVAL -void encode (SV *self, SV *scalar) +void max_size (JSON *self, UV max_size = 0) PPCODE: - XPUSHs (encode_json (scalar, *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; + + self->flags = self->flags & ~F_MAXSIZE | (log2 << S_MAXSIZE); + + XPUSHs (ST (0)); +} + +int get_max_size (JSON *self) + CODE: + RETVAL = DEC_SIZE (self->flags); + OUTPUT: + RETVAL -void decode (SV *self, SV *jsonstr) +void filter_json_object (JSON *self, SV *cb = &PL_sv_undef) PPCODE: - XPUSHs (decode_json (jsonstr, *SvJSON (self), 0)); +{ + SvREFCNT_dec (self->cb_object); + self->cb_object = SvOK (cb) ? newSVsv (cb) : 0; + + XPUSHs (ST (0)); +} -void decode_prefix (SV *self, SV *jsonstr) +void filter_json_single_key_object (JSON *self, SV *key, SV *cb = &PL_sv_undef) PPCODE: { - UV offset; + if (!self->cb_sk_object) + self->cb_sk_object = newHV (); + + if (SvOK (cb)) + hv_store_ent (self->cb_sk_object, key, newSVsv (cb), 0); + else + { + hv_delete_ent (self->cb_sk_object, key, G_DISCARD, 0); + + if (!HvKEYS (self->cb_sk_object)) + { + SvREFCNT_dec (self->cb_sk_object); + self->cb_sk_object = 0; + } + } + + XPUSHs (ST (0)); +} + +void encode (JSON *self, SV *scalar) + PPCODE: + XPUSHs (encode_json (scalar, self)); + +void decode (JSON *self, SV *jsonstr) + PPCODE: + XPUSHs (decode_json (jsonstr, self, 0)); + +void decode_prefix (JSON *self, SV *jsonstr) + PPCODE: +{ + STRLEN offset; EXTEND (SP, 2); - PUSHs (decode_json (jsonstr, *SvJSON (self), &offset)); + PUSHs (decode_json (jsonstr, self, &offset)); PUSHs (sv_2mortal (newSVuv (offset))); } +void incr_parse (JSON *self, SV *jsonstr = 0) + PPCODE: +{ + if (!self->incr_text) + self->incr_text = newSVpvn ("", 0); + + // append data, if any + if (jsonstr) + { + if (SvUTF8 (jsonstr) && !SvUTF8 (self->incr_text)) + { + /* utf-8-ness differs, need to upgrade */ + sv_utf8_upgrade (self->incr_text); + + if (self->incr_pos) + self->incr_pos = utf8_hop ((U8 *)SvPVX (self->incr_text), self->incr_pos) + - (U8 *)SvPVX (self->incr_text); + } + + { + STRLEN len; + const char *str = SvPV (jsonstr, len); + SvGROW (self->incr_text, SvCUR (self->incr_text) + len + 1); + Move (str, SvEND (self->incr_text), len, char); + SvCUR_set (self->incr_text, SvCUR (self->incr_text) + len); + *SvEND (self->incr_text) = 0; // this should basically be a nop, too, but make sure it's there + } + } + + if (GIMME_V != G_VOID) + do + { + STRLEN offset; + + incr_parse (self); + + if (!INCR_DONE (self)) + break; + + XPUSHs (decode_json (self->incr_text, self, &offset)); + + sv_chop (self->incr_text, SvPV_nolen (self->incr_text) + offset); + self->incr_pos -= offset; + self->incr_nest = 0; + self->incr_mode = 0; + } + while (GIMME_V == G_ARRAY); +} + +SV *incr_text (JSON *self) + ATTRS: lvalue + CODE: +{ + if (self->incr_pos) + croak ("incr_text can only be called after a successful incr_parse call in scalar context"); + + RETVAL = self->incr_text ? SvREFCNT_inc (self->incr_text) : &PL_sv_undef; +} + OUTPUT: + RETVAL + +void incr_skip (JSON *self) + CODE: +{ + if (self->incr_pos) + { + sv_chop (self->incr_text, SvPV_nolen (self->incr_text) + self->incr_pos); + self->incr_pos = 0; + self->incr_nest = 0; + self->incr_mode = 0; + } +} + +void DESTROY (JSON *self) + CODE: + SvREFCNT_dec (self->cb_sk_object); + SvREFCNT_dec (self->cb_object); + SvREFCNT_dec (self->incr_text); + PROTOTYPES: ENABLE -void to_json (SV *scalar) +void encode_json (SV *scalar) ALIAS: - objToJson = 0 + to_json_ = 0 + encode_json = F_UTF8 PPCODE: - XPUSHs (encode_json (scalar, F_DEFAULT | F_UTF8)); +{ + JSON json = { F_DEFAULT | ix }; + XPUSHs (encode_json (scalar, &json)); +} -void from_json (SV *jsonstr) +void decode_json (SV *jsonstr) ALIAS: - jsonToObj = 0 + from_json_ = 0 + decode_json = F_UTF8 PPCODE: - XPUSHs (decode_json (jsonstr, F_DEFAULT | F_UTF8, 0)); +{ + JSON json = { F_DEFAULT | ix }; + XPUSHs (decode_json (jsonstr, &json, 0)); +} +