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 1.34 by root, Wed May 23 22:07:13 2007 UTC vs.
Revision 1.41 by root, Sat Jun 23 22:53:16 2007 UTC

… …
7#include "stdlib.h" 7#include "stdlib.h"
8#include "stdio.h" 8#include "stdio.h"
9 9
10#if defined(__BORLANDC__) || defined(_MSC_VER) 10#if defined(__BORLANDC__) || defined(_MSC_VER)
11# define snprintf _snprintf // C compilers have this in stdio.h 11# define snprintf _snprintf // C compilers have this in stdio.h
12#endif
13
14// some old perls do not have this, try to make it work, no
15// guarentees, though.
16#ifndef UTF8_MAXBYTES
17# define UTF8_MAXBYTES 13
12#endif 18#endif
13 19
14#define F_ASCII 0x00000001UL 20#define F_ASCII 0x00000001UL
15#define F_LATIN1 0x00000002UL 21#define F_LATIN1 0x00000002UL
16#define F_UTF8 0x00000004UL 22#define F_UTF8 0x00000004UL
… …
32#define F_DEFAULT (9UL << S_MAXDEPTH) 38#define F_DEFAULT (9UL << S_MAXDEPTH)
33 39
34#define INIT_SIZE 32 // initial scalar size to be allocated 40#define INIT_SIZE 32 // initial scalar size to be allocated
35#define INDENT_STEP 3 // spaces per indentation level 41#define INDENT_STEP 3 // spaces per indentation level
36 42
37#define SHORT_STRING_LEN 512 // special-case strings of up to this size 43#define SHORT_STRING_LEN 16384 // special-case strings of up to this size
38 44
39#define SB do { 45#define SB do {
40#define SE } while (0) 46#define SE } while (0)
47
48#if __GNUC__ >= 3
49# define expect(expr,value) __builtin_expect ((expr),(value))
50# define inline inline
51#else
52# define expect(expr,value) (expr)
53# define inline static
54#endif
55
56#define expect_false(expr) expect ((expr) != 0, 0)
57#define expect_true(expr) expect ((expr) != 0, 1)
41 58
42static HV *json_stash; // JSON::XS:: 59static HV *json_stash; // JSON::XS::
43 60
44///////////////////////////////////////////////////////////////////////////// 61/////////////////////////////////////////////////////////////////////////////
45// utility functions 62// utility functions
… …
70// decode an utf-8 character and return it, or (UV)-1 in 87// decode an utf-8 character and return it, or (UV)-1 in
71// case of an error. 88// case of an error.
72// we special-case "safe" characters from U+80 .. U+7FF, 89// we special-case "safe" characters from U+80 .. U+7FF,
73// but use the very good perl function to parse anything else. 90// but use the very good perl function to parse anything else.
74// note that we never call this function for a ascii codepoints 91// note that we never call this function for a ascii codepoints
75static UV 92inline UV
76decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen) 93decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen)
77{ 94{
78 if (s[0] > 0xdf || s[0] < 0xc2) 95 if (expect_false (s[0] > 0xdf || s[0] < 0xc2))
79 return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY); 96 return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY);
80 else if (len > 1 && s[1] >= 0x80 && s[1] <= 0xbf) 97 else if (len > 1 && s[1] >= 0x80 && s[1] <= 0xbf)
81 { 98 {
82 *clen = 2; 99 *clen = 2;
83 return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f); 100 return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f);
… …
101 U32 flags; // F_* 118 U32 flags; // F_*
102 U32 indent; // indentation level 119 U32 indent; // indentation level
103 U32 maxdepth; // max. indentation/recursion level 120 U32 maxdepth; // max. indentation/recursion level
104} enc_t; 121} enc_t;
105 122
106static void 123inline void
107need (enc_t *enc, STRLEN len) 124need (enc_t *enc, STRLEN len)
108{ 125{
109 if (enc->cur + len >= enc->end) 126 if (expect_false (enc->cur + len >= enc->end))
110 { 127 {
111 STRLEN cur = enc->cur - SvPVX (enc->sv); 128 STRLEN cur = enc->cur - SvPVX (enc->sv);
112 SvGROW (enc->sv, cur + len + 1); 129 SvGROW (enc->sv, cur + len + 1);
113 enc->cur = SvPVX (enc->sv) + cur; 130 enc->cur = SvPVX (enc->sv) + cur;
114 enc->end = SvPVX (enc->sv) + SvLEN (enc->sv) - 1; 131 enc->end = SvPVX (enc->sv) + SvLEN (enc->sv) - 1;
115 } 132 }
116} 133}
117 134
118static void 135inline void
119encode_ch (enc_t *enc, char ch) 136encode_ch (enc_t *enc, char ch)
120{ 137{
121 need (enc, 1); 138 need (enc, 1);
122 *enc->cur++ = ch; 139 *enc->cur++ = ch;
123} 140}
… …
131 148
132 while (str < end) 149 while (str < end)
133 { 150 {
134 unsigned char ch = *(unsigned char *)str; 151 unsigned char ch = *(unsigned char *)str;
135 152
136 if (ch >= 0x20 && ch < 0x80) // most common case 153 if (expect_true (ch >= 0x20 && ch < 0x80)) // most common case
137 { 154 {
138 if (ch == '"') // but with slow exceptions 155 if (expect_false (ch == '"')) // but with slow exceptions
139 { 156 {
140 need (enc, len += 1); 157 need (enc, len += 1);
141 *enc->cur++ = '\\'; 158 *enc->cur++ = '\\';
142 *enc->cur++ = '"'; 159 *enc->cur++ = '"';
143 } 160 }
144 else if (ch == '\\') 161 else if (expect_false (ch == '\\'))
145 { 162 {
146 need (enc, len += 1); 163 need (enc, len += 1);
147 *enc->cur++ = '\\'; 164 *enc->cur++ = '\\';
148 *enc->cur++ = '\\'; 165 *enc->cur++ = '\\';
149 } 166 }
… …
167 STRLEN clen; 184 STRLEN clen;
168 UV uch; 185 UV uch;
169 186
170 if (is_utf8) 187 if (is_utf8)
171 { 188 {
172 //uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
173 uch = decode_utf8 (str, end - str, &clen); 189 uch = decode_utf8 (str, end - str, &clen);
174 if (clen == (STRLEN)-1) 190 if (clen == (STRLEN)-1)
175 croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str); 191 croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str);
176 } 192 }
177 else 193 else
… …
233 249
234 --len; 250 --len;
235 } 251 }
236} 252}
237 253
238static void 254inline void
239encode_indent (enc_t *enc) 255encode_indent (enc_t *enc)
240{ 256{
241 if (enc->flags & F_INDENT) 257 if (enc->flags & F_INDENT)
242 { 258 {
243 int spaces = enc->indent * INDENT_STEP; 259 int spaces = enc->indent * INDENT_STEP;
… …
246 memset (enc->cur, ' ', spaces); 262 memset (enc->cur, ' ', spaces);
247 enc->cur += spaces; 263 enc->cur += spaces;
248 } 264 }
249} 265}
250 266
251static void 267inline void
252encode_space (enc_t *enc) 268encode_space (enc_t *enc)
253{ 269{
254 need (enc, 1); 270 need (enc, 1);
255 encode_ch (enc, ' '); 271 encode_ch (enc, ' ');
256} 272}
257 273
258static void 274inline void
259encode_nl (enc_t *enc) 275encode_nl (enc_t *enc)
260{ 276{
261 if (enc->flags & F_INDENT) 277 if (enc->flags & F_INDENT)
262 { 278 {
263 need (enc, 1); 279 need (enc, 1);
264 encode_ch (enc, '\n'); 280 encode_ch (enc, '\n');
265 } 281 }
266} 282}
267 283
268static void 284inline void
269encode_comma (enc_t *enc) 285encode_comma (enc_t *enc)
270{ 286{
271 encode_ch (enc, ','); 287 encode_ch (enc, ',');
272 288
273 if (enc->flags & F_INDENT) 289 if (enc->flags & F_INDENT)
… …
486 encode_str (enc, str, len, SvUTF8 (sv)); 502 encode_str (enc, str, len, SvUTF8 (sv));
487 encode_ch (enc, '"'); 503 encode_ch (enc, '"');
488 } 504 }
489 else if (SvNOKp (sv)) 505 else if (SvNOKp (sv))
490 { 506 {
507 // trust that perl will do the right thing w.r.t. JSON syntax.
491 need (enc, NV_DIG + 32); 508 need (enc, NV_DIG + 32);
492 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur); 509 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
493 enc->cur += strlen (enc->cur); 510 enc->cur += strlen (enc->cur);
494 } 511 }
495 else if (SvIOKp (sv)) 512 else if (SvIOKp (sv))
496 { 513 {
497 need (enc, 64); 514 // we assume we can always read an IV as a UV
515 if (SvUV (sv) & ~(UV)0x7fff)
516 {
517 // large integer, use the (rather slow) snprintf way.
518 need (enc, sizeof (UV) * 3);
498 enc->cur += 519 enc->cur +=
499 SvIsUV(sv) 520 SvIsUV(sv)
500 ? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv)) 521 ? snprintf (enc->cur, sizeof (UV) * 3, "%"UVuf, (UV)SvUVX (sv))
501 : snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv)); 522 : snprintf (enc->cur, sizeof (UV) * 3, "%"IVdf, (IV)SvIVX (sv));
523 }
524 else
525 {
526 // optimise the "small number case"
527 // code will likely be branchless and use only a single multiplication
528 I32 i = SvIV (sv);
529 U32 u;
530 char digit, nz = 0;
531
532 need (enc, 6);
533
534 *enc->cur = '-'; enc->cur += i < 0 ? 1 : 0;
535 u = i < 0 ? -i : i;
536
537 // convert to 4.28 fixed-point representation
538 u = u * ((0xfffffff + 10000) / 10000); // 10**5, 5 fractional digits
539
540 // now output digit by digit, each time masking out the integer part
541 // and multiplying by 5 while moving the decimal point one to the right,
542 // resulting in a net multiplication by 10.
543 // we always write the digit to memory but conditionally increment
544 // the pointer, to ease the usage of conditional move instructions.
545 digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffff) * 5;
546 digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffff) * 5;
547 digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffff) * 5;
548 digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffff) * 5;
549 digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0'
550 }
502 } 551 }
503 else if (SvROK (sv)) 552 else if (SvROK (sv))
504 encode_rv (enc, SvRV (sv)); 553 encode_rv (enc, SvRV (sv));
505 else if (!SvOK (sv)) 554 else if (!SvOK (sv))
506 encode_str (enc, "null", 4, 0); 555 encode_str (enc, "null", 4, 0);
… …
551 U32 flags; // F_* 600 U32 flags; // F_*
552 U32 depth; // recursion depth 601 U32 depth; // recursion depth
553 U32 maxdepth; // recursion depth limit 602 U32 maxdepth; // recursion depth limit
554} dec_t; 603} dec_t;
555 604
556static void 605inline void
557decode_ws (dec_t *dec) 606decode_ws (dec_t *dec)
558{ 607{
559 for (;;) 608 for (;;)
560 { 609 {
561 char ch = *dec->cur; 610 char ch = *dec->cur;
… …
587decode_4hex (dec_t *dec) 636decode_4hex (dec_t *dec)
588{ 637{
589 signed char d1, d2, d3, d4; 638 signed char d1, d2, d3, d4;
590 unsigned char *cur = (unsigned char *)dec->cur; 639 unsigned char *cur = (unsigned char *)dec->cur;
591 640
592 d1 = decode_hexdigit [cur [0]]; if (d1 < 0) ERR ("four hexadecimal digits expected"); 641 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"); 642 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"); 643 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"); 644 d4 = decode_hexdigit [cur [3]]; if (expect_false (d4 < 0)) ERR ("exactly four hexadecimal digits expected");
596 645
597 dec->cur += 4; 646 dec->cur += 4;
598 647
599 return ((UV)d1) << 12 648 return ((UV)d1) << 12
600 | ((UV)d2) << 8 649 | ((UV)d2) << 8
… …
608static SV * 657static SV *
609decode_str (dec_t *dec) 658decode_str (dec_t *dec)
610{ 659{
611 SV *sv = 0; 660 SV *sv = 0;
612 int utf8 = 0; 661 int utf8 = 0;
662 char *dec_cur = dec->cur;
613 663
614 do 664 do
615 { 665 {
616 char buf [SHORT_STRING_LEN + UTF8_MAXBYTES]; 666 char buf [SHORT_STRING_LEN + UTF8_MAXBYTES];
617 char *cur = buf; 667 char *cur = buf;
618 668
619 do 669 do
620 { 670 {
621 unsigned char ch = *(unsigned char *)dec->cur++; 671 unsigned char ch = *(unsigned char *)dec_cur++;
622 672
623 if (ch == '"') 673 if (expect_false (ch == '"'))
624 { 674 {
625 --dec->cur; 675 --dec_cur;
626 break; 676 break;
627 } 677 }
628 else if (ch == '\\') 678 else if (expect_false (ch == '\\'))
629 { 679 {
630 switch (*dec->cur) 680 switch (*dec_cur)
631 { 681 {
632 case '\\': 682 case '\\':
633 case '/': 683 case '/':
634 case '"': *cur++ = *dec->cur++; break; 684 case '"': *cur++ = *dec_cur++; break;
635 685
636 case 'b': ++dec->cur; *cur++ = '\010'; break; 686 case 'b': ++dec_cur; *cur++ = '\010'; break;
637 case 't': ++dec->cur; *cur++ = '\011'; break; 687 case 't': ++dec_cur; *cur++ = '\011'; break;
638 case 'n': ++dec->cur; *cur++ = '\012'; break; 688 case 'n': ++dec_cur; *cur++ = '\012'; break;
639 case 'f': ++dec->cur; *cur++ = '\014'; break; 689 case 'f': ++dec_cur; *cur++ = '\014'; break;
640 case 'r': ++dec->cur; *cur++ = '\015'; break; 690 case 'r': ++dec_cur; *cur++ = '\015'; break;
641 691
642 case 'u': 692 case 'u':
643 { 693 {
644 UV lo, hi; 694 UV lo, hi;
645 ++dec->cur; 695 ++dec_cur;
646 696
697 dec->cur = dec_cur;
647 hi = decode_4hex (dec); 698 hi = decode_4hex (dec);
699 dec_cur = dec->cur;
648 if (hi == (UV)-1) 700 if (hi == (UV)-1)
649 goto fail; 701 goto fail;
650 702
651 // possibly a surrogate pair 703 // possibly a surrogate pair
652 if (hi >= 0xd800) 704 if (hi >= 0xd800)
653 if (hi < 0xdc00) 705 if (hi < 0xdc00)
654 { 706 {
655 if (dec->cur [0] != '\\' || dec->cur [1] != 'u') 707 if (dec_cur [0] != '\\' || dec_cur [1] != 'u')
656 ERR ("missing low surrogate character in surrogate pair"); 708 ERR ("missing low surrogate character in surrogate pair");
657 709
658 dec->cur += 2; 710 dec_cur += 2;
659 711
712 dec->cur = dec_cur;
660 lo = decode_4hex (dec); 713 lo = decode_4hex (dec);
714 dec_cur = dec->cur;
661 if (lo == (UV)-1) 715 if (lo == (UV)-1)
662 goto fail; 716 goto fail;
663 717
664 if (lo < 0xdc00 || lo >= 0xe000) 718 if (lo < 0xdc00 || lo >= 0xe000)
665 ERR ("surrogate pair expected"); 719 ERR ("surrogate pair expected");
… …
679 *cur++ = hi; 733 *cur++ = hi;
680 } 734 }
681 break; 735 break;
682 736
683 default: 737 default:
684 --dec->cur; 738 --dec_cur;
685 ERR ("illegal backslash escape sequence in string"); 739 ERR ("illegal backslash escape sequence in string");
686 } 740 }
687 } 741 }
688 else if (ch >= 0x20 && ch <= 0x7f) 742 else if (expect_true (ch >= 0x20 && ch <= 0x7f))
689 *cur++ = ch; 743 *cur++ = ch;
690 else if (ch >= 0x80) 744 else if (ch >= 0x80)
691 { 745 {
692 STRLEN clen; 746 STRLEN clen;
693 UV uch; 747 UV uch;
694 748
695 --dec->cur; 749 --dec_cur;
696 750
697 uch = decode_utf8 (dec->cur, dec->end - dec->cur, &clen); 751 uch = decode_utf8 (dec_cur, dec->end - dec_cur, &clen);
698 if (clen == (STRLEN)-1) 752 if (clen == (STRLEN)-1)
699 ERR ("malformed UTF-8 character in JSON string"); 753 ERR ("malformed UTF-8 character in JSON string");
700 754
701 do 755 do
702 *cur++ = *dec->cur++; 756 *cur++ = *dec_cur++;
703 while (--clen); 757 while (--clen);
704 758
705 utf8 = 1; 759 utf8 = 1;
706 } 760 }
707 else 761 else
708 { 762 {
709 --dec->cur; 763 --dec_cur;
710 764
711 if (!ch) 765 if (!ch)
712 ERR ("unexpected end of string while parsing JSON string"); 766 ERR ("unexpected end of string while parsing JSON string");
713 else 767 else
714 ERR ("invalid character encountered while parsing JSON string"); 768 ERR ("invalid character encountered while parsing JSON string");
… …
727 } 781 }
728 else 782 else
729 sv = newSVpvn (buf, len); 783 sv = newSVpvn (buf, len);
730 } 784 }
731 } 785 }
732 while (*dec->cur != '"'); 786 while (*dec_cur != '"');
733 787
734 ++dec->cur; 788 ++dec_cur;
735 789
736 if (sv) 790 if (sv)
737 { 791 {
738 SvPOK_only (sv); 792 SvPOK_only (sv);
739 *SvEND (sv) = 0; 793 *SvEND (sv) = 0;
… …
742 SvUTF8_on (sv); 796 SvUTF8_on (sv);
743 } 797 }
744 else 798 else
745 sv = newSVpvn ("", 0); 799 sv = newSVpvn ("", 0);
746 800
801 dec->cur = dec_cur;
747 return sv; 802 return sv;
748 803
749fail: 804fail:
805 dec->cur = dec_cur;
750 return 0; 806 return 0;
751} 807}
752 808
753static SV * 809static SV *
754decode_num (dec_t *dec) 810decode_num (dec_t *dec)
… …
812 is_nv = 1; 868 is_nv = 1;
813 } 869 }
814 870
815 if (!is_nv) 871 if (!is_nv)
816 { 872 {
817 UV uv; 873 // 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); 874 if (*start == '-')
819 if (numtype & IS_NUMBER_IN_UV) 875 switch (dec->cur - start)
820 if (numtype & IS_NUMBER_NEG)
821 { 876 {
822 if (uv < (UV)IV_MIN) 877 case 2: return newSViv (-( start [1] - '0' * 1));
823 return newSViv (-(IV)uv); 878 case 3: return newSViv (-( start [1] * 10 + start [2] - '0' * 11));
879 case 4: return newSViv (-( start [1] * 100 + start [2] * 10 + start [3] - '0' * 111));
880 case 5: return newSViv (-(start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 1111));
824 } 881 }
882 else
883 switch (dec->cur - start)
884 {
885 case 1: return newSViv ( start [0] - '0' * 1);
886 case 2: return newSViv ( start [0] * 10 + start [1] - '0' * 11);
887 case 3: return newSViv ( start [0] * 100 + start [1] * 10 + start [2] - '0' * 111);
888 case 4: return newSViv ( start [0] * 1000 + start [1] * 100 + start [2] * 10 + start [3] - '0' * 1111);
889 }
890
891 {
892 UV uv;
893 int numtype = grok_number (start, dec->cur - start, &uv);
894 if (numtype & IS_NUMBER_IN_UV)
895 if (numtype & IS_NUMBER_NEG)
896 {
897 if (uv < (UV)IV_MIN)
898 return newSViv (-(IV)uv);
899 }
825 else 900 else
826 return newSVuv (uv); 901 return newSVuv (uv);
902
903 // here would likely be the place for bigint support
827 } 904 }
905 }
828 906
907 // if we ever support bigint or bigfloat, this is the place for bigfloat
829 return newSVnv (Atof (start)); 908 return newSVnv (Atof (start));
830 909
831fail: 910fail:
832 return 0; 911 return 0;
833} 912}
… …
934 1013
935static SV * 1014static SV *
936decode_sv (dec_t *dec) 1015decode_sv (dec_t *dec)
937{ 1016{
938 decode_ws (dec); 1017 decode_ws (dec);
1018
1019 // the beauty of JSON: you need exactly one character lookahead
1020 // to parse anything.
939 switch (*dec->cur) 1021 switch (*dec->cur)
940 { 1022 {
941 case '"': ++dec->cur; return decode_str (dec); 1023 case '"': ++dec->cur; return decode_str (dec);
942 case '[': ++dec->cur; return decode_av (dec); 1024 case '[': ++dec->cur; return decode_av (dec);
943 case '{': ++dec->cur; return decode_hv (dec); 1025 case '{': ++dec->cur; return decode_hv (dec);
… …
1072MODULE = JSON::XS PACKAGE = JSON::XS 1154MODULE = JSON::XS PACKAGE = JSON::XS
1073 1155
1074BOOT: 1156BOOT:
1075{ 1157{
1076 int i; 1158 int i;
1077
1078 memset (decode_hexdigit, 0xff, 256);
1079 1159
1080 for (i = 0; i < 256; ++i) 1160 for (i = 0; i < 256; ++i)
1081 decode_hexdigit [i] = 1161 decode_hexdigit [i] =
1082 i >= '0' && i <= '9' ? i - '0' 1162 i >= '0' && i <= '9' ? i - '0'
1083 : i >= 'a' && i <= 'f' ? i - 'a' + 10 1163 : i >= 'a' && i <= 'f' ? i - 'a' + 10

Diff Legend

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