ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/JSON-XS/XS.xs
(Generate patch)

Comparing JSON-XS/XS.xs (file contents):
Revision 1.33 by root, Wed May 23 20:26:40 2007 UTC vs.
Revision 1.40 by root, Tue Jun 12 01:27:02 2007 UTC

… …
32#define F_DEFAULT (9UL << S_MAXDEPTH) 32#define F_DEFAULT (9UL << S_MAXDEPTH)
33 33
34#define INIT_SIZE 32 // initial scalar size to be allocated 34#define INIT_SIZE 32 // initial scalar size to be allocated
35#define INDENT_STEP 3 // spaces per indentation level 35#define INDENT_STEP 3 // spaces per indentation level
36 36
37#define SHORT_STRING_LEN 512 // special-case strings of up to this size 37#define SHORT_STRING_LEN 16384 // special-case strings of up to this size
38 38
39#define SB do { 39#define SB do {
40#define SE } while (0) 40#define SE } while (0)
41
42#if __GNUC__ >= 3
43# define expect(expr,value) __builtin_expect ((expr),(value))
44# define inline inline
45#else
46# define expect(expr,value) (expr)
47# define inline static
48#endif
49
50#define expect_false(expr) expect ((expr) != 0, 0)
51#define expect_true(expr) expect ((expr) != 0, 1)
41 52
42static HV *json_stash; // JSON::XS:: 53static HV *json_stash; // JSON::XS::
43 54
44///////////////////////////////////////////////////////////////////////////// 55/////////////////////////////////////////////////////////////////////////////
45// utility functions 56// utility functions
… …
70// decode an utf-8 character and return it, or (UV)-1 in 81// decode an utf-8 character and return it, or (UV)-1 in
71// case of an error. 82// case of an error.
72// we special-case "safe" characters from U+80 .. U+7FF, 83// we special-case "safe" characters from U+80 .. U+7FF,
73// but use the very good perl function to parse anything else. 84// but use the very good perl function to parse anything else.
74// note that we never call this function for a ascii codepoints 85// note that we never call this function for a ascii codepoints
75static UV 86inline UV
76decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen) 87decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen)
77{ 88{
78 if (s[0] > 0xdf || s[0] < 0xc2) 89 if (expect_false (s[0] > 0xdf || s[0] < 0xc2))
79 return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY); 90 return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY);
80 else if (len > 1 && s[1] >= 0x80 && s[1] <= 0xbf) 91 else if (len > 1 && s[1] >= 0x80 && s[1] <= 0xbf)
81 { 92 {
82 *clen = 2; 93 *clen = 2;
83 return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f); 94 return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f);
… …
101 U32 flags; // F_* 112 U32 flags; // F_*
102 U32 indent; // indentation level 113 U32 indent; // indentation level
103 U32 maxdepth; // max. indentation/recursion level 114 U32 maxdepth; // max. indentation/recursion level
104} enc_t; 115} enc_t;
105 116
106static void 117inline void
107need (enc_t *enc, STRLEN len) 118need (enc_t *enc, STRLEN len)
108{ 119{
109 if (enc->cur + len >= enc->end) 120 if (expect_false (enc->cur + len >= enc->end))
110 { 121 {
111 STRLEN cur = enc->cur - SvPVX (enc->sv); 122 STRLEN cur = enc->cur - SvPVX (enc->sv);
112 SvGROW (enc->sv, cur + len + 1); 123 SvGROW (enc->sv, cur + len + 1);
113 enc->cur = SvPVX (enc->sv) + cur; 124 enc->cur = SvPVX (enc->sv) + cur;
114 enc->end = SvPVX (enc->sv) + SvLEN (enc->sv) - 1; 125 enc->end = SvPVX (enc->sv) + SvLEN (enc->sv) - 1;
115 } 126 }
116} 127}
117 128
118static void 129inline void
119encode_ch (enc_t *enc, char ch) 130encode_ch (enc_t *enc, char ch)
120{ 131{
121 need (enc, 1); 132 need (enc, 1);
122 *enc->cur++ = ch; 133 *enc->cur++ = ch;
123} 134}
… …
131 142
132 while (str < end) 143 while (str < end)
133 { 144 {
134 unsigned char ch = *(unsigned char *)str; 145 unsigned char ch = *(unsigned char *)str;
135 146
136 if (ch >= 0x20 && ch < 0x80) // most common case 147 if (expect_true (ch >= 0x20 && ch < 0x80)) // most common case
137 { 148 {
138 if (ch == '"') // but with slow exceptions 149 if (expect_false (ch == '"')) // but with slow exceptions
139 { 150 {
140 need (enc, len += 1); 151 need (enc, len += 1);
141 *enc->cur++ = '\\'; 152 *enc->cur++ = '\\';
142 *enc->cur++ = '"'; 153 *enc->cur++ = '"';
143 } 154 }
144 else if (ch == '\\') 155 else if (expect_false (ch == '\\'))
145 { 156 {
146 need (enc, len += 1); 157 need (enc, len += 1);
147 *enc->cur++ = '\\'; 158 *enc->cur++ = '\\';
148 *enc->cur++ = '\\'; 159 *enc->cur++ = '\\';
149 } 160 }
… …
167 STRLEN clen; 178 STRLEN clen;
168 UV uch; 179 UV uch;
169 180
170 if (is_utf8) 181 if (is_utf8)
171 { 182 {
172 //uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
173 uch = decode_utf8 (str, end - str, &clen); 183 uch = decode_utf8 (str, end - str, &clen);
174 if (clen == (STRLEN)-1) 184 if (clen == (STRLEN)-1)
175 croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str); 185 croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str);
176 } 186 }
177 else 187 else
… …
233 243
234 --len; 244 --len;
235 } 245 }
236} 246}
237 247
238static void 248inline void
239encode_indent (enc_t *enc) 249encode_indent (enc_t *enc)
240{ 250{
241 if (enc->flags & F_INDENT) 251 if (enc->flags & F_INDENT)
242 { 252 {
243 int spaces = enc->indent * INDENT_STEP; 253 int spaces = enc->indent * INDENT_STEP;
… …
246 memset (enc->cur, ' ', spaces); 256 memset (enc->cur, ' ', spaces);
247 enc->cur += spaces; 257 enc->cur += spaces;
248 } 258 }
249} 259}
250 260
251static void 261inline void
252encode_space (enc_t *enc) 262encode_space (enc_t *enc)
253{ 263{
254 need (enc, 1); 264 need (enc, 1);
255 encode_ch (enc, ' '); 265 encode_ch (enc, ' ');
256} 266}
257 267
258static void 268inline void
259encode_nl (enc_t *enc) 269encode_nl (enc_t *enc)
260{ 270{
261 if (enc->flags & F_INDENT) 271 if (enc->flags & F_INDENT)
262 { 272 {
263 need (enc, 1); 273 need (enc, 1);
264 encode_ch (enc, '\n'); 274 encode_ch (enc, '\n');
265 } 275 }
266} 276}
267 277
268static void 278inline void
269encode_comma (enc_t *enc) 279encode_comma (enc_t *enc)
270{ 280{
271 encode_ch (enc, ','); 281 encode_ch (enc, ',');
272 282
273 if (enc->flags & F_INDENT) 283 if (enc->flags & F_INDENT)
… …
374 // that randomises hash orderings 384 // that randomises hash orderings
375 if (enc->flags & F_CANONICAL) 385 if (enc->flags & F_CANONICAL)
376 { 386 {
377 int fast = 1; 387 int fast = 1;
378 HE *he; 388 HE *he;
379#if WIN32 389#if defined(__BORLANDC__) || defined(_MSC_VER)
380 HE **hes = _alloca (count * sizeof (HE)); 390 HE **hes = _alloca (count * sizeof (HE));
381#else 391#else
382 HE *hes [count]; // if your compiler dies here, you need to enable C99 mode 392 HE *hes [count]; // if your compiler dies here, you need to enable C99 mode
383#endif 393#endif
384 394
… …
486 encode_str (enc, str, len, SvUTF8 (sv)); 496 encode_str (enc, str, len, SvUTF8 (sv));
487 encode_ch (enc, '"'); 497 encode_ch (enc, '"');
488 } 498 }
489 else if (SvNOKp (sv)) 499 else if (SvNOKp (sv))
490 { 500 {
501 // trust that perl will do the right thing w.r.t. JSON syntax.
491 need (enc, NV_DIG + 32); 502 need (enc, NV_DIG + 32);
492 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur); 503 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
493 enc->cur += strlen (enc->cur); 504 enc->cur += strlen (enc->cur);
494 } 505 }
495 else if (SvIOKp (sv)) 506 else if (SvIOKp (sv))
496 { 507 {
497 need (enc, 64); 508 // we assume we can always read an IV as a UV
509 if (SvUV (sv) & ~(UV)0x7fff)
510 {
511 // large integer, use the (rather slow) snprintf way.
512 need (enc, sizeof (UV) * 3);
498 enc->cur += 513 enc->cur +=
499 SvIsUV(sv) 514 SvIsUV(sv)
500 ? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv)) 515 ? snprintf (enc->cur, sizeof (UV) * 3, "%"UVuf, (UV)SvUVX (sv))
501 : snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv)); 516 : snprintf (enc->cur, sizeof (UV) * 3, "%"IVdf, (IV)SvIVX (sv));
517 }
518 else
519 {
520 // optimise the "small number case"
521 // code will likely be branchless and use only a single multiplication
522 I32 i = SvIV (sv);
523 U32 u;
524 char digit, nz = 0;
525
526 need (enc, 6);
527
528 *enc->cur = '-'; enc->cur += i < 0 ? 1 : 0;
529 u = i < 0 ? -i : i;
530
531 // convert to 4.28 fixed-point representation
532 u = u * ((0xfffffff + 10000) / 10000); // 10**5, 5 fractional digits
533
534 // now output digit by digit, each time masking out the integer part
535 // and multiplying by 5 while moving the decimal point one to the right,
536 // resulting in a net multiplication by 10.
537 // we always write the digit to memory but conditionally increment
538 // the pointer, to ease the usage of conditional move instructions.
539 digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffff) * 5;
540 digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffff) * 5;
541 digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffff) * 5;
542 digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffff) * 5;
543 digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0'
544 }
502 } 545 }
503 else if (SvROK (sv)) 546 else if (SvROK (sv))
504 encode_rv (enc, SvRV (sv)); 547 encode_rv (enc, SvRV (sv));
505 else if (!SvOK (sv)) 548 else if (!SvOK (sv))
506 encode_str (enc, "null", 4, 0); 549 encode_str (enc, "null", 4, 0);
… …
551 U32 flags; // F_* 594 U32 flags; // F_*
552 U32 depth; // recursion depth 595 U32 depth; // recursion depth
553 U32 maxdepth; // recursion depth limit 596 U32 maxdepth; // recursion depth limit
554} dec_t; 597} dec_t;
555 598
556static void 599inline void
557decode_ws (dec_t *dec) 600decode_ws (dec_t *dec)
558{ 601{
559 for (;;) 602 for (;;)
560 { 603 {
561 char ch = *dec->cur; 604 char ch = *dec->cur;
… …
587decode_4hex (dec_t *dec) 630decode_4hex (dec_t *dec)
588{ 631{
589 signed char d1, d2, d3, d4; 632 signed char d1, d2, d3, d4;
590 unsigned char *cur = (unsigned char *)dec->cur; 633 unsigned char *cur = (unsigned char *)dec->cur;
591 634
592 d1 = decode_hexdigit [cur [0]]; if (d1 < 0) ERR ("four hexadecimal digits expected"); 635 d1 = decode_hexdigit [cur [0]]; if (expect_false (d1 < 0)) ERR ("exactly four hexadecimal digits expected");
593 d2 = decode_hexdigit [cur [1]]; if (d2 < 0) ERR ("four hexadecimal digits expected"); 636 d2 = decode_hexdigit [cur [1]]; if (expect_false (d2 < 0)) ERR ("exactly four hexadecimal digits expected");
594 d3 = decode_hexdigit [cur [2]]; if (d3 < 0) ERR ("four hexadecimal digits expected"); 637 d3 = decode_hexdigit [cur [2]]; if (expect_false (d3 < 0)) ERR ("exactly four hexadecimal digits expected");
595 d4 = decode_hexdigit [cur [3]]; if (d4 < 0) ERR ("four hexadecimal digits expected"); 638 d4 = decode_hexdigit [cur [3]]; if (expect_false (d4 < 0)) ERR ("exactly four hexadecimal digits expected");
596 639
597 dec->cur += 4; 640 dec->cur += 4;
598 641
599 return ((UV)d1) << 12 642 return ((UV)d1) << 12
600 | ((UV)d2) << 8 643 | ((UV)d2) << 8
… …
608static SV * 651static SV *
609decode_str (dec_t *dec) 652decode_str (dec_t *dec)
610{ 653{
611 SV *sv = 0; 654 SV *sv = 0;
612 int utf8 = 0; 655 int utf8 = 0;
656 char *dec_cur = dec->cur;
613 657
614 do 658 do
615 { 659 {
616 char buf [SHORT_STRING_LEN + UTF8_MAXBYTES]; 660 char buf [SHORT_STRING_LEN + UTF8_MAXBYTES];
617 char *cur = buf; 661 char *cur = buf;
618 662
619 do 663 do
620 { 664 {
621 unsigned char ch = *(unsigned char *)dec->cur++; 665 unsigned char ch = *(unsigned char *)dec_cur++;
622 666
623 if (ch == '"') 667 if (expect_false (ch == '"'))
624 { 668 {
625 --dec->cur; 669 --dec_cur;
626 break; 670 break;
627 } 671 }
628 else if (ch == '\\') 672 else if (expect_false (ch == '\\'))
629 { 673 {
630 switch (*dec->cur) 674 switch (*dec_cur)
631 { 675 {
632 case '\\': 676 case '\\':
633 case '/': 677 case '/':
634 case '"': *cur++ = *dec->cur++; break; 678 case '"': *cur++ = *dec_cur++; break;
635 679
636 case 'b': ++dec->cur; *cur++ = '\010'; break; 680 case 'b': ++dec_cur; *cur++ = '\010'; break;
637 case 't': ++dec->cur; *cur++ = '\011'; break; 681 case 't': ++dec_cur; *cur++ = '\011'; break;
638 case 'n': ++dec->cur; *cur++ = '\012'; break; 682 case 'n': ++dec_cur; *cur++ = '\012'; break;
639 case 'f': ++dec->cur; *cur++ = '\014'; break; 683 case 'f': ++dec_cur; *cur++ = '\014'; break;
640 case 'r': ++dec->cur; *cur++ = '\015'; break; 684 case 'r': ++dec_cur; *cur++ = '\015'; break;
641 685
642 case 'u': 686 case 'u':
643 { 687 {
644 UV lo, hi; 688 UV lo, hi;
645 ++dec->cur; 689 ++dec_cur;
646 690
691 dec->cur = dec_cur;
647 hi = decode_4hex (dec); 692 hi = decode_4hex (dec);
693 dec_cur = dec->cur;
648 if (hi == (UV)-1) 694 if (hi == (UV)-1)
649 goto fail; 695 goto fail;
650 696
651 // possibly a surrogate pair 697 // possibly a surrogate pair
652 if (hi >= 0xd800) 698 if (hi >= 0xd800)
653 if (hi < 0xdc00) 699 if (hi < 0xdc00)
654 { 700 {
655 if (dec->cur [0] != '\\' || dec->cur [1] != 'u') 701 if (dec_cur [0] != '\\' || dec_cur [1] != 'u')
656 ERR ("missing low surrogate character in surrogate pair"); 702 ERR ("missing low surrogate character in surrogate pair");
657 703
658 dec->cur += 2; 704 dec_cur += 2;
659 705
706 dec->cur = dec_cur;
660 lo = decode_4hex (dec); 707 lo = decode_4hex (dec);
708 dec_cur = dec->cur;
661 if (lo == (UV)-1) 709 if (lo == (UV)-1)
662 goto fail; 710 goto fail;
663 711
664 if (lo < 0xdc00 || lo >= 0xe000) 712 if (lo < 0xdc00 || lo >= 0xe000)
665 ERR ("surrogate pair expected"); 713 ERR ("surrogate pair expected");
… …
679 *cur++ = hi; 727 *cur++ = hi;
680 } 728 }
681 break; 729 break;
682 730
683 default: 731 default:
684 --dec->cur; 732 --dec_cur;
685 ERR ("illegal backslash escape sequence in string"); 733 ERR ("illegal backslash escape sequence in string");
686 } 734 }
687 } 735 }
688 else if (ch >= 0x20 && ch <= 0x7f) 736 else if (expect_true (ch >= 0x20 && ch <= 0x7f))
689 *cur++ = ch; 737 *cur++ = ch;
690 else if (ch >= 0x80) 738 else if (ch >= 0x80)
691 { 739 {
692 STRLEN clen; 740 STRLEN clen;
693 UV uch; 741 UV uch;
694 742
695 --dec->cur; 743 --dec_cur;
696 744
697 uch = decode_utf8 (dec->cur, dec->end - dec->cur, &clen); 745 uch = decode_utf8 (dec_cur, dec->end - dec_cur, &clen);
698 if (clen == (STRLEN)-1) 746 if (clen == (STRLEN)-1)
699 ERR ("malformed UTF-8 character in JSON string"); 747 ERR ("malformed UTF-8 character in JSON string");
700 748
701 do 749 do
702 *cur++ = *dec->cur++; 750 *cur++ = *dec_cur++;
703 while (--clen); 751 while (--clen);
704 752
705 utf8 = 1; 753 utf8 = 1;
706 } 754 }
707 else 755 else
708 { 756 {
709 --dec->cur; 757 --dec_cur;
710 758
711 if (!ch) 759 if (!ch)
712 ERR ("unexpected end of string while parsing JSON string"); 760 ERR ("unexpected end of string while parsing JSON string");
713 else 761 else
714 ERR ("invalid character encountered while parsing JSON string"); 762 ERR ("invalid character encountered while parsing JSON string");
… …
727 } 775 }
728 else 776 else
729 sv = newSVpvn (buf, len); 777 sv = newSVpvn (buf, len);
730 } 778 }
731 } 779 }
732 while (*dec->cur != '"'); 780 while (*dec_cur != '"');
733 781
734 ++dec->cur; 782 ++dec_cur;
735 783
736 if (sv) 784 if (sv)
737 { 785 {
738 SvPOK_only (sv); 786 SvPOK_only (sv);
739 *SvEND (sv) = 0; 787 *SvEND (sv) = 0;
… …
742 SvUTF8_on (sv); 790 SvUTF8_on (sv);
743 } 791 }
744 else 792 else
745 sv = newSVpvn ("", 0); 793 sv = newSVpvn ("", 0);
746 794
795 dec->cur = dec_cur;
747 return sv; 796 return sv;
748 797
749fail: 798fail:
799 dec->cur = dec_cur;
750 return 0; 800 return 0;
751} 801}
752 802
753static SV * 803static SV *
754decode_num (dec_t *dec) 804decode_num (dec_t *dec)
… …
812 is_nv = 1; 862 is_nv = 1;
813 } 863 }
814 864
815 if (!is_nv) 865 if (!is_nv)
816 { 866 {
817 UV uv; 867 // special case the rather common 1..4-digit-int case, assumes 32 bit ints or so
818 int numtype = grok_number (start, dec->cur - start, &uv); 868 if (*start == '-')
819 if (numtype & IS_NUMBER_IN_UV) 869 switch (dec->cur - start)
820 if (numtype & IS_NUMBER_NEG)
821 { 870 {
822 if (uv < (UV)IV_MIN) 871 case 2: return newSViv (-( start [1] - '0' * 1));
823 return newSViv (-(IV)uv); 872 case 3: return newSViv (-( start [1] * 10 + start [2] - '0' * 11));
873 case 4: return newSViv (-( start [1] * 100 + start [2] * 10 + start [3] - '0' * 111));
874 case 5: return newSViv (-(start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 1111));
824 } 875 }
876 else
877 switch (dec->cur - start)
878 {
879 case 1: return newSViv ( start [0] - '0' * 1);
880 case 2: return newSViv ( start [0] * 10 + start [1] - '0' * 11);
881 case 3: return newSViv ( start [0] * 100 + start [1] * 10 + start [2] - '0' * 111);
882 case 4: return newSViv ( start [0] * 1000 + start [1] * 100 + start [2] * 10 + start [3] - '0' * 1111);
883 }
884
885 {
886 UV uv;
887 int numtype = grok_number (start, dec->cur - start, &uv);
888 if (numtype & IS_NUMBER_IN_UV)
889 if (numtype & IS_NUMBER_NEG)
890 {
891 if (uv < (UV)IV_MIN)
892 return newSViv (-(IV)uv);
893 }
825 else 894 else
826 return newSVuv (uv); 895 return newSVuv (uv);
896
897 // here would likely be the place for bigint support
827 } 898 }
899 }
828 900
901 // if we ever support bigint or bigfloat, this is the place for bigfloat
829 return newSVnv (Atof (start)); 902 return newSVnv (Atof (start));
830 903
831fail: 904fail:
832 return 0; 905 return 0;
833} 906}
… …
934 1007
935static SV * 1008static SV *
936decode_sv (dec_t *dec) 1009decode_sv (dec_t *dec)
937{ 1010{
938 decode_ws (dec); 1011 decode_ws (dec);
1012
1013 // the beauty of JSON: you need exactly one character lookahead
1014 // to parse anything.
939 switch (*dec->cur) 1015 switch (*dec->cur)
940 { 1016 {
941 case '"': ++dec->cur; return decode_str (dec); 1017 case '"': ++dec->cur; return decode_str (dec);
942 case '[': ++dec->cur; return decode_av (dec); 1018 case '[': ++dec->cur; return decode_av (dec);
943 case '{': ++dec->cur; return decode_hv (dec); 1019 case '{': ++dec->cur; return decode_hv (dec);
… …
1072MODULE = JSON::XS PACKAGE = JSON::XS 1148MODULE = JSON::XS PACKAGE = JSON::XS
1073 1149
1074BOOT: 1150BOOT:
1075{ 1151{
1076 int i; 1152 int i;
1077
1078 memset (decode_hexdigit, 0xff, 256);
1079 1153
1080 for (i = 0; i < 256; ++i) 1154 for (i = 0; i < 256; ++i)
1081 decode_hexdigit [i] = 1155 decode_hexdigit [i] =
1082 i >= '0' && i <= '9' ? i - '0' 1156 i >= '0' && i <= '9' ? i - '0'
1083 : i >= 'a' && i <= 'f' ? i - 'a' + 10 1157 : i >= 'a' && i <= 'f' ? i - 'a' + 10

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines