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 1.2 by root, Thu Mar 22 17:28:50 2007 UTC vs.
Revision 1.59 by root, Mon Aug 13 16:14:20 2007 UTC

1#include "EXTERN.h" 1#include "EXTERN.h"
2#include "perl.h" 2#include "perl.h"
3#include "XSUB.h" 3#include "XSUB.h"
4 4
5#include "assert.h" 5#include <assert.h>
6#include "string.h" 6#include <string.h>
7#include "stdlib.h" 7#include <stdlib.h>
8#include <stdio.h>
9#include <float.h>
8 10
11#if defined(__BORLANDC__) || defined(_MSC_VER)
12# define snprintf _snprintf // C compilers have this in stdio.h
13#endif
14
15// some old perls do not have this, try to make it work, no
16// guarentees, though. if it breaks, you get to keep the pieces.
17#ifndef UTF8_MAXBYTES
18# define UTF8_MAXBYTES 13
19#endif
20
9#define F_ASCII 0x00000001 21#define F_ASCII 0x00000001UL
22#define F_LATIN1 0x00000002UL
10#define F_UTF8 0x00000002 23#define F_UTF8 0x00000004UL
11#define F_INDENT 0x00000004 24#define F_INDENT 0x00000008UL
12#define F_CANONICAL 0x00000008 25#define F_CANONICAL 0x00000010UL
13#define F_SPACE_BEFORE 0x00000010 26#define F_SPACE_BEFORE 0x00000020UL
14#define F_SPACE_AFTER 0x00000020 27#define F_SPACE_AFTER 0x00000040UL
15#define F_JSON_RPC 0x00000040 28#define F_ALLOW_NONREF 0x00000100UL
29#define F_SHRINK 0x00000200UL
30#define F_ALLOW_BLESSED 0x00000400UL
31#define F_CONV_BLESSED 0x00000800UL
32#define F_MAXDEPTH 0xf8000000UL
33#define S_MAXDEPTH 27
34#define F_MAXSIZE 0x01f00000UL
35#define S_MAXSIZE 20
36#define F_HOOK 0x00080000UL // some hooks exist, so slow-path processing
37
38#define DEC_DEPTH(flags) (1UL << ((flags & F_MAXDEPTH) >> S_MAXDEPTH))
39#define DEC_SIZE(flags) (1UL << ((flags & F_MAXSIZE ) >> S_MAXSIZE ))
16 40
17#define F_PRETTY F_INDENT | F_SPACE_BEFORE | F_SPACE_AFTER 41#define F_PRETTY F_INDENT | F_SPACE_BEFORE | F_SPACE_AFTER
18#define F_DEFAULT 0 42#define F_DEFAULT (9UL << S_MAXDEPTH)
19 43
20#define INIT_SIZE 32 // initial scalar size to be allocated 44#define INIT_SIZE 32 // initial scalar size to be allocated
45#define INDENT_STEP 3 // spaces per indentation level
46
47#define SHORT_STRING_LEN 16384 // special-case strings of up to this size
21 48
22#define SB do { 49#define SB do {
23#define SE } while (0) 50#define SE } while (0)
24 51
25static HV *json_stash; 52#if __GNUC__ >= 3
53# define expect(expr,value) __builtin_expect ((expr),(value))
54# define inline inline
55#else
56# define expect(expr,value) (expr)
57# define inline static
58#endif
59
60#define expect_false(expr) expect ((expr) != 0, 0)
61#define expect_true(expr) expect ((expr) != 0, 1)
62
63#ifdef USE_ITHREADS
64# define JSON_SLOW 1
65# define JSON_STASH (json_stash ? json_stash : gv_stashpv ("JSON::XS", 1))
66#else
67# define JSON_SLOW 0
68# define JSON_STASH json_stash
69#endif
70
71static HV *json_stash, *json_boolean_stash; // JSON::XS::
72static SV *json_true, *json_false;
73
74typedef struct {
75 U32 flags;
76 SV *cb_object;
77 HV *cb_sk_object;
78} JSON;
79
80/////////////////////////////////////////////////////////////////////////////
81// utility functions
82
83inline void
84shrink (SV *sv)
85{
86 sv_utf8_downgrade (sv, 1);
87 if (SvLEN (sv) > SvCUR (sv) + 1)
88 {
89#ifdef SvPV_shrink_to_cur
90 SvPV_shrink_to_cur (sv);
91#elif defined (SvPV_renew)
92 SvPV_renew (sv, SvCUR (sv) + 1);
93#endif
94 }
95}
96
97// decode an utf-8 character and return it, or (UV)-1 in
98// case of an error.
99// we special-case "safe" characters from U+80 .. U+7FF,
100// but use the very good perl function to parse anything else.
101// note that we never call this function for a ascii codepoints
102inline UV
103decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen)
104{
105 if (expect_false (s[0] > 0xdf || s[0] < 0xc2))
106 return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY);
107 else if (len > 1 && s[1] >= 0x80 && s[1] <= 0xbf)
108 {
109 *clen = 2;
110 return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f);
111 }
112 else
113 {
114 *clen = (STRLEN)-1;
115 return (UV)-1;
116 }
117}
118
119/////////////////////////////////////////////////////////////////////////////
120// encoder
26 121
27// structure used for encoding JSON 122// structure used for encoding JSON
28typedef struct 123typedef struct
29{ 124{
30 char *cur; 125 char *cur; // SvPVX (sv) + current output position
31 STRLEN len; // SvLEN (sv)
32 char *end; // SvEND (sv) 126 char *end; // SvEND (sv)
33 SV *sv; 127 SV *sv; // result scalar
34 UV flags; 128 JSON json;
35 int max_recurse; 129 U32 indent; // indentation level
36 int indent; 130 U32 maxdepth; // max. indentation/recursion level
37} enc_t; 131} enc_t;
38 132
39// structure used for decoding JSON 133inline void
40typedef struct
41{
42 char *cur;
43 char *end;
44 char *err;
45 UV flags;
46} dec_t;
47
48static UV *
49SvJSON (SV *sv)
50{
51 if (!(SvROK (sv) && SvOBJECT (SvRV (sv)) && SvSTASH (SvRV (sv)) == json_stash))
52 croak ("object is not of type JSON::XS");
53
54 return &SvUVX (SvRV (sv));
55}
56
57/////////////////////////////////////////////////////////////////////////////
58
59static void
60need (enc_t *enc, STRLEN len) 134need (enc_t *enc, STRLEN len)
61{ 135{
62 if (enc->cur + len >= enc->end) 136 if (expect_false (enc->cur + len >= enc->end))
63 { 137 {
64 STRLEN cur = enc->cur - SvPVX (enc->sv); 138 STRLEN cur = enc->cur - SvPVX (enc->sv);
65 SvGROW (enc->sv, cur + len + 1); 139 SvGROW (enc->sv, cur + len + 1);
66 enc->cur = SvPVX (enc->sv) + cur; 140 enc->cur = SvPVX (enc->sv) + cur;
67 enc->end = SvEND (enc->sv); 141 enc->end = SvPVX (enc->sv) + SvLEN (enc->sv) - 1;
68 } 142 }
69} 143}
70 144
71static void 145inline void
72encode_ch (enc_t *enc, char ch) 146encode_ch (enc_t *enc, char ch)
73{ 147{
74 need (enc, 1); 148 need (enc, 1);
75 *enc->cur++ = ch; 149 *enc->cur++ = ch;
76} 150}
… …
78static void 152static void
79encode_str (enc_t *enc, char *str, STRLEN len, int is_utf8) 153encode_str (enc_t *enc, char *str, STRLEN len, int is_utf8)
80{ 154{
81 char *end = str + len; 155 char *end = str + len;
82 156
157 need (enc, len);
158
83 while (str < end) 159 while (str < end)
84 { 160 {
85 unsigned char ch = *(unsigned char *)str; 161 unsigned char ch = *(unsigned char *)str;
162
86 if (ch >= 0x20 && ch < 0x80) // most common case 163 if (expect_true (ch >= 0x20 && ch < 0x80)) // most common case
87 { 164 {
165 if (expect_false (ch == '"')) // but with slow exceptions
166 {
167 need (enc, len += 1);
168 *enc->cur++ = '\\';
169 *enc->cur++ = '"';
170 }
171 else if (expect_false (ch == '\\'))
172 {
173 need (enc, len += 1);
174 *enc->cur++ = '\\';
175 *enc->cur++ = '\\';
176 }
177 else
88 *enc->cur++ = ch; 178 *enc->cur++ = ch;
179
89 str++; 180 ++str;
90 } 181 }
91 else 182 else
92 { 183 {
93 STRLEN clen; 184 switch (ch)
94 UV uch;
95
96 if (is_utf8)
97 {
98 uch = utf8n_to_uvuni (str, end - str, &clen, UTF8_CHECK_ONLY);
99 if (clen < 0)
100 croak ("malformed UTF-8 character in string, cannot convert to JSON");
101 } 185 {
102 else 186 case '\010': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'b'; ++str; break;
103 { 187 case '\011': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 't'; ++str; break;
104 uch = ch; 188 case '\012': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'n'; ++str; break;
105 clen = 1; 189 case '\014': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'f'; ++str; break;
106 } 190 case '\015': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'r'; ++str; break;
107 191
108 need (enc, len += 6); 192 default:
109
110 if (uch < 0xa0 || enc->flags & F_ASCII)
111 {
112 if (uch > 0xFFFFUL)
113 { 193 {
194 STRLEN clen;
195 UV uch;
196
197 if (is_utf8)
198 {
199 uch = decode_utf8 (str, end - str, &clen);
200 if (clen == (STRLEN)-1)
201 croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str);
202 }
203 else
204 {
205 uch = ch;
114 len += 6; 206 clen = 1;
207 }
208
209 if (uch > 0x10FFFFUL)
210 croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch);
211
212 if (uch < 0x80 || enc->json.flags & F_ASCII || (enc->json.flags & F_LATIN1 && uch > 0xFF))
213 {
214 if (uch > 0xFFFFUL)
215 {
115 need (enc, len += 6); 216 need (enc, len += 11);
116 sprintf (enc->cur, "\\u%04x\\u%04x", 217 sprintf (enc->cur, "\\u%04x\\u%04x",
117 (uch - 0x10000) / 0x400 + 0xD800, 218 (int)((uch - 0x10000) / 0x400 + 0xD800),
118 (uch - 0x10000) % 0x400 + 0xDC00); 219 (int)((uch - 0x10000) % 0x400 + 0xDC00));
119 enc->cur += 12; 220 enc->cur += 12;
221 }
222 else
223 {
224 static char hexdigit [16] = "0123456789abcdef";
225 need (enc, len += 5);
226 *enc->cur++ = '\\';
227 *enc->cur++ = 'u';
228 *enc->cur++ = hexdigit [ uch >> 12 ];
229 *enc->cur++ = hexdigit [(uch >> 8) & 15];
230 *enc->cur++ = hexdigit [(uch >> 4) & 15];
231 *enc->cur++ = hexdigit [(uch >> 0) & 15];
232 }
233
234 str += clen;
235 }
236 else if (enc->json.flags & F_LATIN1)
237 {
238 *enc->cur++ = uch;
239 str += clen;
240 }
241 else if (is_utf8)
242 {
243 need (enc, len += clen);
244 do
245 {
246 *enc->cur++ = *str++;
247 }
248 while (--clen);
249 }
250 else
251 {
252 need (enc, len += UTF8_MAXBYTES - 1); // never more than 11 bytes needed
253 enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);
254 ++str;
255 }
120 } 256 }
121 else
122 {
123 sprintf (enc->cur, "\\u%04x", uch);
124 enc->cur += 6;
125 }
126 }
127 else if (is_utf8)
128 { 257 }
129 memcpy (enc->cur, str, clen);
130 enc->cur += clen;
131 }
132 else
133 enc->cur = uvuni_to_utf8_flags (enc->cur, uch, 0);
134
135 str += clen;
136 } 258 }
137 259
138 --len; 260 --len;
139 } 261 }
140} 262}
141 263
142#define INDENT SB \ 264inline void
265encode_indent (enc_t *enc)
266{
143 if (enc->flags & F_INDENT) \ 267 if (enc->json.flags & F_INDENT)
144 { \ 268 {
145 int i_; \ 269 int spaces = enc->indent * INDENT_STEP;
146 need (enc, enc->indent); \ 270
147 for (i_ = enc->indent * 3; i_--; )\ 271 need (enc, spaces);
272 memset (enc->cur, ' ', spaces);
273 enc->cur += spaces;
274 }
275}
276
277inline void
278encode_space (enc_t *enc)
279{
280 need (enc, 1);
281 encode_ch (enc, ' ');
282}
283
284inline void
285encode_nl (enc_t *enc)
286{
287 if (enc->json.flags & F_INDENT)
288 {
289 need (enc, 1);
148 encode_ch (enc, ' '); \ 290 encode_ch (enc, '\n');
149 } \ 291 }
150 SE 292}
151 293
152#define SPACE SB need (enc, 1); encode_ch (enc, ' '); SE 294inline void
153#define NL SB if (enc->flags & F_INDENT) { need (enc, 1); encode_ch (enc, '\n'); } SE 295encode_comma (enc_t *enc)
154#define COMMA SB \ 296{
155 encode_ch (enc, ','); \ 297 encode_ch (enc, ',');
298
156 if (enc->flags & F_INDENT) \ 299 if (enc->json.flags & F_INDENT)
157 NL; \ 300 encode_nl (enc);
158 else if (enc->flags & F_SPACE_AFTER) \ 301 else if (enc->json.flags & F_SPACE_AFTER)
159 SPACE; \ 302 encode_space (enc);
160 SE 303}
161 304
162static void encode_sv (enc_t *enc, SV *sv); 305static void encode_sv (enc_t *enc, SV *sv);
163 306
164static void 307static void
165encode_av (enc_t *enc, AV *av) 308encode_av (enc_t *enc, AV *av)
166{ 309{
167 int i, len = av_len (av); 310 int i, len = av_len (av);
168 311
169 encode_ch (enc, '['); NL; 312 if (enc->indent >= enc->maxdepth)
313 croak ("data structure too deep (hit recursion limit)");
314
315 encode_ch (enc, '['); encode_nl (enc);
170 ++enc->indent; 316 ++enc->indent;
171 317
172 for (i = 0; i <= len; ++i) 318 for (i = 0; i <= len; ++i)
173 { 319 {
174 INDENT; 320 SV **svp = av_fetch (av, i, 0);
175 encode_sv (enc, *av_fetch (av, i, 0)); 321
322 encode_indent (enc);
323
324 if (svp)
325 encode_sv (enc, *svp);
326 else
327 encode_str (enc, "null", 4, 0);
176 328
177 if (i < len) 329 if (i < len)
178 COMMA; 330 encode_comma (enc);
179 } 331 }
180 332
181 NL; 333 encode_nl (enc);
182 334
183 --enc->indent; 335 --enc->indent;
184 INDENT; encode_ch (enc, ']'); 336 encode_indent (enc); encode_ch (enc, ']');
185} 337}
186 338
187static void 339static void
188encode_he (enc_t *enc, HE *he) 340encode_he (enc_t *enc, HE *he)
189{ 341{
… …
191 343
192 if (HeKLEN (he) == HEf_SVKEY) 344 if (HeKLEN (he) == HEf_SVKEY)
193 { 345 {
194 SV *sv = HeSVKEY (he); 346 SV *sv = HeSVKEY (he);
195 STRLEN len; 347 STRLEN len;
348 char *str;
349
350 SvGETMAGIC (sv);
196 char *str = SvPV (sv, len); 351 str = SvPV (sv, len);
197 352
198 encode_str (enc, str, len, SvUTF8 (sv)); 353 encode_str (enc, str, len, SvUTF8 (sv));
199 } 354 }
200 else 355 else
201 encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he)); 356 encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he));
202 357
203 encode_ch (enc, '"'); 358 encode_ch (enc, '"');
204 359
205 if (enc->flags & F_SPACE_BEFORE) SPACE; 360 if (enc->json.flags & F_SPACE_BEFORE) encode_space (enc);
206 encode_ch (enc, ':'); 361 encode_ch (enc, ':');
207 if (enc->flags & F_SPACE_AFTER ) SPACE; 362 if (enc->json.flags & F_SPACE_AFTER ) encode_space (enc);
208 encode_sv (enc, HeVAL (he)); 363 encode_sv (enc, HeVAL (he));
209} 364}
210 365
211// compare hash entries, used when all keys are bytestrings 366// compare hash entries, used when all keys are bytestrings
212static int 367static int
… …
218 HE *b = *(HE **)b_; 373 HE *b = *(HE **)b_;
219 374
220 STRLEN la = HeKLEN (a); 375 STRLEN la = HeKLEN (a);
221 STRLEN lb = HeKLEN (b); 376 STRLEN lb = HeKLEN (b);
222 377
223 if (!(cmp == memcmp (HeKEY (a), HeKEY (b), la < lb ? la : lb))) 378 if (!(cmp = memcmp (HeKEY (a), HeKEY (b), la < lb ? la : lb)))
224 cmp = la < lb ? -1 : la == lb ? 0 : 1; 379 cmp = la - lb;
225 380
226 return cmp; 381 return cmp;
227} 382}
228 383
229// compare hash entries, used when some keys are sv's or utf-x 384// compare hash entries, used when some keys are sv's or utf-x
… …
236static void 391static void
237encode_hv (enc_t *enc, HV *hv) 392encode_hv (enc_t *enc, HV *hv)
238{ 393{
239 int count, i; 394 int count, i;
240 395
396 if (enc->indent >= enc->maxdepth)
397 croak ("data structure too deep (hit recursion limit)");
398
241 encode_ch (enc, '{'); NL; ++enc->indent; 399 encode_ch (enc, '{'); encode_nl (enc); ++enc->indent;
242 400
243 if ((count = hv_iterinit (hv))) 401 if ((count = hv_iterinit (hv)))
244 { 402 {
245 // for canonical output we have to sort by keys first 403 // for canonical output we have to sort by keys first
246 // actually, this is mostly due to the stupid so-called 404 // actually, this is mostly due to the stupid so-called
247 // security workaround added somewhere in 5.8.x. 405 // security workaround added somewhere in 5.8.x.
248 // that randomises hash orderings 406 // that randomises hash orderings
249 if (enc->flags & F_CANONICAL) 407 if (enc->json.flags & F_CANONICAL)
250 { 408 {
251 HE *he, *hes [count];
252 int fast = 1; 409 int fast = 1;
410 HE *he;
411#if defined(__BORLANDC__) || defined(_MSC_VER)
412 HE **hes = _alloca (count * sizeof (HE));
413#else
414 HE *hes [count]; // if your compiler dies here, you need to enable C99 mode
415#endif
253 416
254 i = 0; 417 i = 0;
255 while ((he = hv_iternext (hv))) 418 while ((he = hv_iternext (hv)))
256 { 419 {
257 hes [i++] = he; 420 hes [i++] = he;
… …
263 426
264 if (fast) 427 if (fast)
265 qsort (hes, count, sizeof (HE *), he_cmp_fast); 428 qsort (hes, count, sizeof (HE *), he_cmp_fast);
266 else 429 else
267 { 430 {
268 // hack to disable "use bytes" 431 // hack to forcefully disable "use bytes"
269 COP *oldcop = PL_curcop, cop; 432 COP cop = *PL_curcop;
270 cop.op_private = 0; 433 cop.op_private = 0;
434
435 ENTER;
436 SAVETMPS;
437
438 SAVEVPTR (PL_curcop);
271 PL_curcop = &cop; 439 PL_curcop = &cop;
272 440
273 SAVETMPS;
274 qsort (hes, count, sizeof (HE *), he_cmp_slow); 441 qsort (hes, count, sizeof (HE *), he_cmp_slow);
442
275 FREETMPS; 443 FREETMPS;
276 444 LEAVE;
277 PL_curcop = oldcop;
278 } 445 }
279 446
280 for (i = 0; i < count; ++i) 447 for (i = 0; i < count; ++i)
281 { 448 {
282 INDENT; 449 encode_indent (enc);
283 encode_he (enc, hes [i]); 450 encode_he (enc, hes [i]);
284 451
285 if (i < count - 1) 452 if (i < count - 1)
286 COMMA; 453 encode_comma (enc);
287 }
288
289 NL; 454 }
455
456 encode_nl (enc);
290 } 457 }
291 else 458 else
292 { 459 {
293 SV *sv;
294 HE *he = hv_iternext (hv); 460 HE *he = hv_iternext (hv);
295 461
296 for (;;) 462 for (;;)
297 { 463 {
298 INDENT; 464 encode_indent (enc);
299 encode_he (enc, he); 465 encode_he (enc, he);
300 466
301 if (!(he = hv_iternext (hv))) 467 if (!(he = hv_iternext (hv)))
302 break; 468 break;
303 469
304 COMMA; 470 encode_comma (enc);
305 }
306
307 NL; 471 }
472
473 encode_nl (enc);
308 } 474 }
309 } 475 }
310 476
311 --enc->indent; INDENT; encode_ch (enc, '}'); 477 --enc->indent; encode_indent (enc); encode_ch (enc, '}');
478}
479
480// encode objects, arrays and special \0=false and \1=true values.
481static void
482encode_rv (enc_t *enc, SV *sv)
483{
484 svtype svt;
485
486 SvGETMAGIC (sv);
487 svt = SvTYPE (sv);
488
489 if (expect_false (SvOBJECT (sv)))
490 {
491 HV *stash = !JSON_SLOW || json_boolean_stash
492 ? json_boolean_stash
493 : gv_stashpv ("JSON::XS::Boolean", 1);
494
495 if (SvSTASH (sv) == stash)
496 {
497 if (SvIV (sv))
498 encode_str (enc, "true", 4, 0);
499 else
500 encode_str (enc, "false", 5, 0);
501 }
502 else
503 {
504#if 0
505 if (0 && sv_derived_from (rv, "JSON::Literal"))
506 {
507 // not yet
508 }
509#endif
510 if (enc->json.flags & F_CONV_BLESSED)
511 {
512 // we re-bless the reference to get overload and other niceties right
513 GV *to_json = gv_fetchmethod_autoload (SvSTASH (sv), "TO_JSON", 0);
514
515 if (to_json)
516 {
517 dSP;
518
519 ENTER; SAVETMPS; PUSHMARK (SP);
520 XPUSHs (sv_bless (sv_2mortal (newRV_inc (sv)), SvSTASH (sv)));
521
522 // calling with G_SCALAR ensures that we always get a 1 return value
523 PUTBACK;
524 call_sv ((SV *)GvCV (to_json), G_SCALAR);
525 SPAGAIN;
526
527 // catch this surprisingly common error
528 if (SvROK (TOPs) && SvRV (TOPs) == sv)
529 croak ("%s::TO_JSON method returned same object as was passed instead of a new one", HvNAME (SvSTASH (sv)));
530
531 sv = POPs;
532 PUTBACK;
533
534 encode_sv (enc, sv);
535
536 FREETMPS; LEAVE;
537 }
538 else if (enc->json.flags & F_ALLOW_BLESSED)
539 encode_str (enc, "null", 4, 0);
540 else
541 croak ("encountered object '%s', but neither allow_blessed enabled nor TO_JSON method available on it",
542 SvPV_nolen (sv_2mortal (newRV_inc (sv))));
543 }
544 else if (enc->json.flags & F_ALLOW_BLESSED)
545 encode_str (enc, "null", 4, 0);
546 else
547 croak ("encountered object '%s', but neither allow_blessed nor convert_blessed settings are enabled",
548 SvPV_nolen (sv_2mortal (newRV_inc (sv))));
549 }
550 }
551 else if (svt == SVt_PVHV)
552 encode_hv (enc, (HV *)sv);
553 else if (svt == SVt_PVAV)
554 encode_av (enc, (AV *)sv);
555 else if (svt < SVt_PVAV)
556 {
557 STRLEN len = 0;
558 char *pv = svt ? SvPV (sv, len) : 0;
559
560 if (len == 1 && *pv == '1')
561 encode_str (enc, "true", 4, 0);
562 else if (len == 1 && *pv == '0')
563 encode_str (enc, "false", 5, 0);
564 else
565 croak ("cannot encode reference to scalar '%s' unless the scalar is 0 or 1",
566 SvPV_nolen (sv_2mortal (newRV_inc (sv))));
567 }
568 else
569 croak ("encountered %s, but JSON can only represent references to arrays or hashes",
570 SvPV_nolen (sv_2mortal (newRV_inc (sv))));
312} 571}
313 572
314static void 573static void
315encode_sv (enc_t *enc, SV *sv) 574encode_sv (enc_t *enc, SV *sv)
316{ 575{
576 SvGETMAGIC (sv);
577
317 if (SvPOKp (sv)) 578 if (SvPOKp (sv))
318 { 579 {
319 STRLEN len; 580 STRLEN len;
320 char *str = SvPV (sv, len); 581 char *str = SvPV (sv, len);
321 encode_ch (enc, '"'); 582 encode_ch (enc, '"');
322 encode_str (enc, str, len, SvUTF8 (sv)); 583 encode_str (enc, str, len, SvUTF8 (sv));
323 encode_ch (enc, '"'); 584 encode_ch (enc, '"');
324 } 585 }
325 else if (SvNOKp (sv)) 586 else if (SvNOKp (sv))
326 { 587 {
588 // trust that perl will do the right thing w.r.t. JSON syntax.
327 need (enc, NV_DIG + 32); 589 need (enc, NV_DIG + 32);
328 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur); 590 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
329 enc->cur += strlen (enc->cur); 591 enc->cur += strlen (enc->cur);
330 } 592 }
331 else if (SvIOKp (sv)) 593 else if (SvIOKp (sv))
332 { 594 {
333 need (enc, 64); 595 // we assume we can always read an IV as a UV
596 if (SvUV (sv) & ~(UV)0x7fff)
597 {
598 // large integer, use the (rather slow) snprintf way.
599 need (enc, sizeof (UV) * 3);
334 enc->cur += 600 enc->cur +=
335 SvIsUV(sv) 601 SvIsUV(sv)
336 ? snprintf (enc->cur, 64, "%"UVuf, (UV)SvUVX (sv)) 602 ? snprintf (enc->cur, sizeof (UV) * 3, "%"UVuf, (UV)SvUVX (sv))
337 : snprintf (enc->cur, 64, "%"IVdf, (IV)SvIVX (sv)); 603 : snprintf (enc->cur, sizeof (UV) * 3, "%"IVdf, (IV)SvIVX (sv));
604 }
605 else
606 {
607 // optimise the "small number case"
608 // code will likely be branchless and use only a single multiplication
609 I32 i = SvIV (sv);
610 U32 u;
611 char digit, nz = 0;
612
613 need (enc, 6);
614
615 *enc->cur = '-'; enc->cur += i < 0 ? 1 : 0;
616 u = i < 0 ? -i : i;
617
618 // convert to 4.28 fixed-point representation
619 u = u * ((0xfffffff + 10000) / 10000); // 10**5, 5 fractional digits
620
621 // now output digit by digit, each time masking out the integer part
622 // and multiplying by 5 while moving the decimal point one to the right,
623 // resulting in a net multiplication by 10.
624 // we always write the digit to memory but conditionally increment
625 // the pointer, to ease the usage of conditional move instructions.
626 digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffff) * 5;
627 digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffff) * 5;
628 digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffff) * 5;
629 digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffff) * 5;
630 digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0'
631 }
338 } 632 }
339 else if (SvROK (sv)) 633 else if (SvROK (sv))
340 { 634 encode_rv (enc, SvRV (sv));
341 if (!--enc->max_recurse)
342 croak ("data structure too deep (hit recursion limit)");
343
344 sv = SvRV (sv);
345
346 switch (SvTYPE (sv))
347 {
348 case SVt_PVAV: encode_av (enc, (AV *)sv); break;
349 case SVt_PVHV: encode_hv (enc, (HV *)sv); break;
350
351 default:
352 croak ("JSON can only represent references to arrays or hashes");
353 }
354 }
355 else if (!SvOK (sv)) 635 else if (!SvOK (sv))
356 encode_str (enc, "null", 4, 0); 636 encode_str (enc, "null", 4, 0);
357 else 637 else
358 croak ("encountered perl type that JSON cannot handle"); 638 croak ("encountered perl type (%s,0x%x) that JSON cannot handle, you might want to report this",
639 SvPV_nolen (sv), SvFLAGS (sv));
359} 640}
360 641
361static SV * 642static SV *
362encode_json (SV *scalar, UV flags) 643encode_json (SV *scalar, JSON *json)
363{ 644{
364 enc_t enc; 645 enc_t enc;
365 enc.flags = flags; 646
647 if (!(json->flags & F_ALLOW_NONREF) && !SvROK (scalar))
648 croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)");
649
650 enc.json = *json;
366 enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE)); 651 enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE));
367 enc.cur = SvPVX (enc.sv); 652 enc.cur = SvPVX (enc.sv);
368 enc.end = SvEND (enc.sv); 653 enc.end = SvEND (enc.sv);
369 enc.max_recurse = 0;
370 enc.indent = 0; 654 enc.indent = 0;
655 enc.maxdepth = DEC_DEPTH (enc.json.flags);
371 656
372 SvPOK_only (enc.sv); 657 SvPOK_only (enc.sv);
373 encode_sv (&enc, scalar); 658 encode_sv (&enc, scalar);
374 659
660 SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv));
661 *SvEND (enc.sv) = 0; // many xs functions expect a trailing 0 for text strings
662
375 if (!(flags & (F_ASCII | F_UTF8))) 663 if (!(enc.json.flags & (F_ASCII | F_LATIN1 | F_UTF8)))
376 SvUTF8_on (enc.sv); 664 SvUTF8_on (enc.sv);
377 665
378 SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv)); 666 if (enc.json.flags & F_SHRINK)
667 shrink (enc.sv);
668
379 return enc.sv; 669 return enc.sv;
380} 670}
381 671
382///////////////////////////////////////////////////////////////////////////// 672/////////////////////////////////////////////////////////////////////////////
673// decoder
383 674
384#define WS \ 675// structure used for decoding JSON
676typedef struct
677{
678 char *cur; // current parser pointer
679 char *end; // end of input string
680 const char *err; // parse error, if != 0
681 JSON json;
682 U32 depth; // recursion depth
683 U32 maxdepth; // recursion depth limit
684} dec_t;
685
686inline void
687decode_ws (dec_t *dec)
688{
385 for (;;) \ 689 for (;;)
386 { \ 690 {
387 char ch = *dec->cur; \ 691 char ch = *dec->cur;
692
388 if (ch > 0x20 \ 693 if (ch > 0x20
389 || (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)) \ 694 || (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09))
390 break; \ 695 break;
696
391 ++dec->cur; \ 697 ++dec->cur;
392 } 698 }
699}
393 700
394#define ERR(reason) SB dec->err = reason; goto fail; SE 701#define ERR(reason) SB dec->err = reason; goto fail; SE
702
395#define EXPECT_CH(ch) SB \ 703#define EXPECT_CH(ch) SB \
396 if (*dec->cur != ch) \ 704 if (*dec->cur != ch) \
397 ERR (# ch " expected"); \ 705 ERR (# ch " expected"); \
398 ++dec->cur; \ 706 ++dec->cur; \
399 SE 707 SE
400 708
709#define DEC_INC_DEPTH if (++dec->depth > dec->maxdepth) ERR ("json datastructure exceeds maximum nesting level (set a higher max_depth)")
710#define DEC_DEC_DEPTH --dec->depth
711
401static SV *decode_sv (dec_t *dec); 712static SV *decode_sv (dec_t *dec);
402
403#define APPEND_CH(ch) SB \
404 SvGROW (sv, cur + 1 + 1); \
405 SvPVX (sv)[cur++] = (ch); \
406 SE
407 713
408static signed char decode_hexdigit[256]; 714static signed char decode_hexdigit[256];
409 715
410static UV 716static UV
411decode_4hex (dec_t *dec) 717decode_4hex (dec_t *dec)
412{ 718{
413 signed char d1, d2, d3, d4; 719 signed char d1, d2, d3, d4;
720 unsigned char *cur = (unsigned char *)dec->cur;
414 721
415 d1 = decode_hexdigit [((unsigned char *)dec->cur) [0]]; 722 d1 = decode_hexdigit [cur [0]]; if (expect_false (d1 < 0)) ERR ("exactly four hexadecimal digits expected");
416 if (d1 < 0) ERR ("four hexadecimal digits expected"); 723 d2 = decode_hexdigit [cur [1]]; if (expect_false (d2 < 0)) ERR ("exactly four hexadecimal digits expected");
417 d2 = decode_hexdigit [((unsigned char *)dec->cur) [1]]; 724 d3 = decode_hexdigit [cur [2]]; if (expect_false (d3 < 0)) ERR ("exactly four hexadecimal digits expected");
418 if (d2 < 0) ERR ("four hexadecimal digits expected"); 725 d4 = decode_hexdigit [cur [3]]; if (expect_false (d4 < 0)) ERR ("exactly four hexadecimal digits expected");
419 d3 = decode_hexdigit [((unsigned char *)dec->cur) [2]];
420 if (d3 < 0) ERR ("four hexadecimal digits expected");
421 d4 = decode_hexdigit [((unsigned char *)dec->cur) [3]];
422 if (d4 < 0) ERR ("four hexadecimal digits expected");
423 726
424 dec->cur += 4; 727 dec->cur += 4;
425 728
426 return ((UV)d1) << 12 729 return ((UV)d1) << 12
427 | ((UV)d2) << 8 730 | ((UV)d2) << 8
… …
433} 736}
434 737
435static SV * 738static SV *
436decode_str (dec_t *dec) 739decode_str (dec_t *dec)
437{ 740{
438 SV *sv = NEWSV (0,2); 741 SV *sv = 0;
439 STRLEN cur = 0;
440 int utf8 = 0; 742 int utf8 = 0;
743 char *dec_cur = dec->cur;
441 744
442 for (;;) 745 do
443 { 746 {
444 unsigned char ch = *(unsigned char *)dec->cur; 747 char buf [SHORT_STRING_LEN + UTF8_MAXBYTES];
748 char *cur = buf;
445 749
446 if (ch == '"') 750 do
447 break;
448 else if (ch == '\\')
449 { 751 {
450 switch (*++dec->cur) 752 unsigned char ch = *(unsigned char *)dec_cur++;
753
754 if (expect_false (ch == '"'))
755 {
756 --dec_cur;
757 break;
451 { 758 }
452 case '\\': 759 else if (expect_false (ch == '\\'))
453 case '/': 760 {
454 case '"': APPEND_CH (*dec->cur++); break; 761 switch (*dec_cur)
455
456 case 'b': APPEND_CH ('\010'); ++dec->cur; break;
457 case 't': APPEND_CH ('\011'); ++dec->cur; break;
458 case 'n': APPEND_CH ('\012'); ++dec->cur; break;
459 case 'f': APPEND_CH ('\014'); ++dec->cur; break;
460 case 'r': APPEND_CH ('\015'); ++dec->cur; break;
461
462 case 'u':
463 { 762 {
464 UV lo, hi; 763 case '\\':
465 ++dec->cur; 764 case '/':
765 case '"': *cur++ = *dec_cur++; break;
466 766
467 hi = decode_4hex (dec); 767 case 'b': ++dec_cur; *cur++ = '\010'; break;
468 if (hi == (UV)-1) 768 case 't': ++dec_cur; *cur++ = '\011'; break;
469 goto fail; 769 case 'n': ++dec_cur; *cur++ = '\012'; break;
770 case 'f': ++dec_cur; *cur++ = '\014'; break;
771 case 'r': ++dec_cur; *cur++ = '\015'; break;
470 772
471 // possibly a surrogate pair 773 case 'u':
472 if (hi >= 0xd800 && hi < 0xdc00)
473 { 774 {
474 if (dec->cur [0] != '\\' || dec->cur [1] != 'u') 775 UV lo, hi;
475 ERR ("illegal surrogate character"); 776 ++dec_cur;
476 777
477 dec->cur += 2; 778 dec->cur = dec_cur;
478
479 lo = decode_4hex (dec); 779 hi = decode_4hex (dec);
780 dec_cur = dec->cur;
480 if (lo == (UV)-1) 781 if (hi == (UV)-1)
481 goto fail; 782 goto fail;
482 783
784 // possibly a surrogate pair
785 if (hi >= 0xd800)
786 if (hi < 0xdc00)
787 {
788 if (dec_cur [0] != '\\' || dec_cur [1] != 'u')
789 ERR ("missing low surrogate character in surrogate pair");
790
791 dec_cur += 2;
792
793 dec->cur = dec_cur;
794 lo = decode_4hex (dec);
795 dec_cur = dec->cur;
796 if (lo == (UV)-1)
797 goto fail;
798
483 if (lo < 0xdc00 || lo >= 0xe000) 799 if (lo < 0xdc00 || lo >= 0xe000)
484 ERR ("surrogate pair expected"); 800 ERR ("surrogate pair expected");
485 801
486 hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000; 802 hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000;
803 }
804 else if (hi < 0xe000)
805 ERR ("missing high surrogate character in surrogate pair");
806
807 if (hi >= 0x80)
808 {
809 utf8 = 1;
810
811 cur = (char *)uvuni_to_utf8_flags (cur, hi, 0);
812 }
813 else
814 *cur++ = hi;
487 } 815 }
488 else if (lo >= 0xdc00 && lo < 0xe000)
489 ERR ("illegal surrogate character");
490
491 if (hi >= 0x80)
492 { 816 break;
493 utf8 = 1;
494 817
495 SvGROW (sv, cur + 4 + 1); // at most 4 bytes for 21 bits
496 cur = (char *)uvuni_to_utf8_flags (SvPVX (sv) + cur, hi, 0) - SvPVX (sv);
497 }
498 else 818 default:
499 APPEND_CH (hi); 819 --dec_cur;
820 ERR ("illegal backslash escape sequence in string");
500 } 821 }
822 }
823 else if (expect_true (ch >= 0x20 && ch <= 0x7f))
824 *cur++ = ch;
825 else if (ch >= 0x80)
826 {
827 STRLEN clen;
828 UV uch;
829
830 --dec_cur;
831
832 uch = decode_utf8 (dec_cur, dec->end - dec_cur, &clen);
833 if (clen == (STRLEN)-1)
834 ERR ("malformed UTF-8 character in JSON string");
835
836 do
837 *cur++ = *dec_cur++;
838 while (--clen);
839
840 utf8 = 1;
841 }
842 else
843 {
844 --dec_cur;
845
846 if (!ch)
847 ERR ("unexpected end of string while parsing JSON string");
501 break; 848 else
849 ERR ("invalid character encountered while parsing JSON string");
502 } 850 }
503 } 851 }
504 else if (ch >= 0x20 && ch <= 0x7f) 852 while (cur < buf + SHORT_STRING_LEN);
505 APPEND_CH (*dec->cur++); 853
506 else if (ch >= 0x80) 854 {
855 STRLEN len = cur - buf;
856
857 if (sv)
507 { 858 {
508 STRLEN clen; 859 SvGROW (sv, SvCUR (sv) + len + 1);
509 UV uch = utf8n_to_uvuni (dec->cur, dec->end - dec->cur, &clen, UTF8_CHECK_ONLY); 860 memcpy (SvPVX (sv) + SvCUR (sv), buf, len);
510 if (clen < 0) 861 SvCUR_set (sv, SvCUR (sv) + len);
511 ERR ("malformed UTF-8 character in string, cannot convert to JSON");
512
513 SvGROW (sv, cur + clen + 1); // at most 4 bytes for 21 bits
514 memcpy (SvPVX (sv) + cur, dec->cur, clen);
515 dec->cur += clen;
516 } 862 }
517 else 863 else
518 ERR ("invalid character encountered"); 864 sv = newSVpvn (buf, len);
519 } 865 }
866 }
867 while (*dec_cur != '"');
520 868
521 ++dec->cur; 869 ++dec_cur;
522 870
871 if (sv)
872 {
523 SvPOK_only (sv); 873 SvPOK_only (sv);
524
525 SvCUR_set (sv, cur);
526 *SvEND (sv) = 0; 874 *SvEND (sv) = 0;
527 875
528 if (utf8) 876 if (utf8)
529 SvUTF8_on (sv); 877 SvUTF8_on (sv);
878 }
879 else
880 sv = newSVpvn ("", 0);
530 881
882 dec->cur = dec_cur;
531 return sv; 883 return sv;
532 884
533fail: 885fail:
534 SvREFCNT_dec (sv); 886 dec->cur = dec_cur;
535 return 0; 887 return 0;
536} 888}
537 889
538static SV * 890static SV *
539decode_num (dec_t *dec) 891decode_num (dec_t *dec)
… …
549 { 901 {
550 ++dec->cur; 902 ++dec->cur;
551 if (*dec->cur >= '0' && *dec->cur <= '9') 903 if (*dec->cur >= '0' && *dec->cur <= '9')
552 ERR ("malformed number (leading zero must not be followed by another digit)"); 904 ERR ("malformed number (leading zero must not be followed by another digit)");
553 } 905 }
554 906 else if (*dec->cur < '0' || *dec->cur > '9')
555 // int 907 ERR ("malformed number (no digits after initial minus)");
908 else
909 do
910 {
911 ++dec->cur;
912 }
556 while (*dec->cur >= '0' && *dec->cur <= '9') 913 while (*dec->cur >= '0' && *dec->cur <= '9');
557 ++dec->cur;
558 914
559 // [frac] 915 // [frac]
560 if (*dec->cur == '.') 916 if (*dec->cur == '.')
561 { 917 {
562 is_nv = 1; 918 ++dec->cur;
919
920 if (*dec->cur < '0' || *dec->cur > '9')
921 ERR ("malformed number (no digits after decimal point)");
563 922
564 do 923 do
565 { 924 {
566 ++dec->cur; 925 ++dec->cur;
567 } 926 }
568 while (*dec->cur >= '0' && *dec->cur <= '9'); 927 while (*dec->cur >= '0' && *dec->cur <= '9');
928
929 is_nv = 1;
569 } 930 }
570 931
571 // [exp] 932 // [exp]
572 if (*dec->cur == 'e' || *dec->cur == 'E') 933 if (*dec->cur == 'e' || *dec->cur == 'E')
573 { 934 {
574 is_nv = 1;
575
576 ++dec->cur; 935 ++dec->cur;
936
577 if (*dec->cur == '-' || *dec->cur == '+') 937 if (*dec->cur == '-' || *dec->cur == '+')
578 ++dec->cur; 938 ++dec->cur;
579 939
940 if (*dec->cur < '0' || *dec->cur > '9')
941 ERR ("malformed number (no digits after exp sign)");
942
943 do
944 {
945 ++dec->cur;
946 }
580 while (*dec->cur >= '0' && *dec->cur <= '9') 947 while (*dec->cur >= '0' && *dec->cur <= '9');
581 ++dec->cur; 948
949 is_nv = 1;
582 } 950 }
583 951
584 if (!is_nv) 952 if (!is_nv)
585 { 953 {
586 UV uv; 954 int len = dec->cur - start;
587 int numtype = grok_number (start, dec->cur - start, &uv); 955
588 if (numtype & IS_NUMBER_IN_UV) 956 // special case the rather common 1..4-digit-int case, assumes 32 bit ints or so
589 if (numtype & IS_NUMBER_NEG) 957 if (*start == '-')
958 switch (len)
590 { 959 {
591 if (uv < (UV)IV_MIN) 960 case 2: return newSViv (-( start [1] - '0' * 1));
592 return newSViv (-(IV)uv); 961 case 3: return newSViv (-( start [1] * 10 + start [2] - '0' * 11));
962 case 4: return newSViv (-( start [1] * 100 + start [2] * 10 + start [3] - '0' * 111));
963 case 5: return newSViv (-(start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 1111));
593 } 964 }
965 else
966 switch (len)
967 {
968 case 1: return newSViv ( start [0] - '0' * 1);
969 case 2: return newSViv ( start [0] * 10 + start [1] - '0' * 11);
970 case 3: return newSViv ( start [0] * 100 + start [1] * 10 + start [2] - '0' * 111);
971 case 4: return newSViv ( start [0] * 1000 + start [1] * 100 + start [2] * 10 + start [3] - '0' * 1111);
972 }
973
974 {
975 UV uv;
976 int numtype = grok_number (start, len, &uv);
977 if (numtype & IS_NUMBER_IN_UV)
978 if (numtype & IS_NUMBER_NEG)
979 {
980 if (uv < (UV)IV_MIN)
981 return newSViv (-(IV)uv);
982 }
594 else 983 else
595 return newSVuv (uv); 984 return newSVuv (uv);
596 } 985 }
597 986
987 len -= *start == '-' ? 1 : 0;
988
989 // does not fit into IV or UV, try NV
990 if ((sizeof (NV) == sizeof (double) && DBL_DIG >= len)
991 #if defined (LDBL_DIG)
992 || (sizeof (NV) == sizeof (long double) && LDBL_DIG >= len)
993 #endif
994 )
995 // fits into NV without loss of precision
996 return newSVnv (Atof (start));
997
998 // everything else fails, convert it to a string
999 return newSVpvn (start, dec->cur - start);
1000 }
1001
1002 // loss of precision here
598 return newSVnv (Atof (start)); 1003 return newSVnv (Atof (start));
599 1004
600fail: 1005fail:
601 return 0; 1006 return 0;
602} 1007}
… …
604static SV * 1009static SV *
605decode_av (dec_t *dec) 1010decode_av (dec_t *dec)
606{ 1011{
607 AV *av = newAV (); 1012 AV *av = newAV ();
608 1013
1014 DEC_INC_DEPTH;
1015 decode_ws (dec);
1016
1017 if (*dec->cur == ']')
1018 ++dec->cur;
1019 else
609 for (;;) 1020 for (;;)
610 { 1021 {
611 SV *value; 1022 SV *value;
612 1023
613 value = decode_sv (dec); 1024 value = decode_sv (dec);
614 if (!value) 1025 if (!value)
615 goto fail; 1026 goto fail;
616 1027
617 av_push (av, value); 1028 av_push (av, value);
618 1029
619 WS; 1030 decode_ws (dec);
620 1031
621 if (*dec->cur == ']') 1032 if (*dec->cur == ']')
622 { 1033 {
623 ++dec->cur; 1034 ++dec->cur;
624 break; 1035 break;
1036 }
625 } 1037
626
627 if (*dec->cur != ',') 1038 if (*dec->cur != ',')
628 ERR (", or ] expected while parsing array"); 1039 ERR (", or ] expected while parsing array");
629 1040
630 ++dec->cur; 1041 ++dec->cur;
631 } 1042 }
632 1043
1044 DEC_DEC_DEPTH;
633 return newRV_noinc ((SV *)av); 1045 return newRV_noinc ((SV *)av);
634 1046
635fail: 1047fail:
636 SvREFCNT_dec (av); 1048 SvREFCNT_dec (av);
1049 DEC_DEC_DEPTH;
637 return 0; 1050 return 0;
638} 1051}
639 1052
640static SV * 1053static SV *
641decode_hv (dec_t *dec) 1054decode_hv (dec_t *dec)
642{ 1055{
1056 SV *sv;
643 HV *hv = newHV (); 1057 HV *hv = newHV ();
644 1058
1059 DEC_INC_DEPTH;
1060 decode_ws (dec);
1061
1062 if (*dec->cur == '}')
1063 ++dec->cur;
1064 else
645 for (;;) 1065 for (;;)
646 { 1066 {
647 SV *key, *value; 1067 decode_ws (dec); EXPECT_CH ('"');
648 1068
649 WS; EXPECT_CH ('"'); 1069 // heuristic: assume that
650 1070 // a) decode_str + hv_store_ent are abysmally slow.
651 key = decode_str (dec); 1071 // b) most hash keys are short, simple ascii text.
652 if (!key) 1072 // => try to "fast-match" such strings to avoid
653 goto fail; 1073 // the overhead of decode_str + hv_store_ent.
654
655 WS; EXPECT_CH (':');
656
657 value = decode_sv (dec);
658 if (!value)
659 { 1074 {
1075 SV *value;
1076 char *p = dec->cur;
1077 char *e = p + 24; // only try up to 24 bytes
1078
1079 for (;;)
1080 {
1081 // the >= 0x80 is true on most architectures
1082 if (p == e || *p < 0x20 || *p >= 0x80 || *p == '\\')
1083 {
1084 // slow path, back up and use decode_str
1085 SV *key = decode_str (dec);
1086 if (!key)
1087 goto fail;
1088
1089 decode_ws (dec); EXPECT_CH (':');
1090
1091 value = decode_sv (dec);
1092 if (!value)
1093 {
1094 SvREFCNT_dec (key);
1095 goto fail;
1096 }
1097
1098 hv_store_ent (hv, key, value, 0);
660 SvREFCNT_dec (key); 1099 SvREFCNT_dec (key);
1100
1101 break;
1102 }
1103 else if (*p == '"')
1104 {
1105 // fast path, got a simple key
1106 char *key = dec->cur;
1107 int len = p - key;
1108 dec->cur = p + 1;
1109
1110 decode_ws (dec); EXPECT_CH (':');
1111
1112 value = decode_sv (dec);
1113 if (!value)
661 goto fail; 1114 goto fail;
1115
1116 hv_store (hv, key, len, value, 0);
1117
1118 break;
1119 }
1120
1121 ++p;
1122 }
662 } 1123 }
663 1124
664 //TODO: optimise 1125 decode_ws (dec);
665 hv_store_ent (hv, key, value, 0);
666 1126
667 WS;
668
669 if (*dec->cur == '}') 1127 if (*dec->cur == '}')
1128 {
1129 ++dec->cur;
1130 break;
1131 }
1132
1133 if (*dec->cur != ',')
1134 ERR (", or } expected while parsing object/hash");
1135
1136 ++dec->cur;
1137 }
1138
1139 DEC_DEC_DEPTH;
1140 sv = newRV_noinc ((SV *)hv);
1141
1142 // check filter callbacks
1143 if (dec->json.flags & F_HOOK)
1144 {
1145 if (dec->json.cb_sk_object && HvKEYS (hv) == 1)
670 { 1146 {
671 ++dec->cur; 1147 HE *cb, *he;
672 break; 1148
1149 hv_iterinit (hv);
1150 he = hv_iternext (hv);
1151 hv_iterinit (hv);
1152
1153 // the next line creates a mortal sv each time its called.
1154 // might want to optimise this for common cases.
1155 cb = hv_fetch_ent (dec->json.cb_sk_object, hv_iterkeysv (he), 0, 0);
1156
1157 if (cb)
1158 {
1159 dSP;
1160 int count;
1161
1162 ENTER; SAVETMPS; PUSHMARK (SP);
1163 XPUSHs (HeVAL (he));
1164
1165 PUTBACK; count = call_sv (HeVAL (cb), G_ARRAY); SPAGAIN;
1166
1167 if (count == 1)
1168 {
1169 sv = newSVsv (POPs);
1170 FREETMPS; LEAVE;
1171 return sv;
1172 }
1173
1174 FREETMPS; LEAVE;
1175 }
673 } 1176 }
674 1177
675 if (*dec->cur != ',') 1178 if (dec->json.cb_object)
676 ERR (", or } expected while parsing object/hash"); 1179 {
1180 dSP;
1181 int count;
677 1182
678 ++dec->cur; 1183 ENTER; SAVETMPS; PUSHMARK (SP);
679 } 1184 XPUSHs (sv_2mortal (sv));
680 1185
681 return newRV_noinc ((SV *)hv); 1186 PUTBACK; count = call_sv (dec->json.cb_object, G_ARRAY); SPAGAIN;
1187
1188 if (count == 1)
1189 {
1190 sv = newSVsv (POPs);
1191 FREETMPS; LEAVE;
1192 return sv;
1193 }
1194
1195 SvREFCNT_inc (sv);
1196 FREETMPS; LEAVE;
1197 }
1198 }
1199
1200 return sv;
682 1201
683fail: 1202fail:
684 SvREFCNT_dec (hv); 1203 SvREFCNT_dec (hv);
1204 DEC_DEC_DEPTH;
685 return 0; 1205 return 0;
686} 1206}
687 1207
688static SV * 1208static SV *
689decode_sv (dec_t *dec) 1209decode_sv (dec_t *dec)
690{ 1210{
691 WS; 1211 decode_ws (dec);
1212
1213 // the beauty of JSON: you need exactly one character lookahead
1214 // to parse anything.
692 switch (*dec->cur) 1215 switch (*dec->cur)
693 { 1216 {
694 case '"': ++dec->cur; return decode_str (dec); 1217 case '"': ++dec->cur; return decode_str (dec);
695 case '[': ++dec->cur; return decode_av (dec); 1218 case '[': ++dec->cur; return decode_av (dec);
696 case '{': ++dec->cur; return decode_hv (dec); 1219 case '{': ++dec->cur; return decode_hv (dec);
… …
702 1225
703 case 't': 1226 case 't':
704 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4)) 1227 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4))
705 { 1228 {
706 dec->cur += 4; 1229 dec->cur += 4;
707 return newSViv (1); 1230 return SvREFCNT_inc (json_true);
708 } 1231 }
709 else 1232 else
710 ERR ("'true' expected"); 1233 ERR ("'true' expected");
711 1234
712 break; 1235 break;
713 1236
714 case 'f': 1237 case 'f':
715 if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5)) 1238 if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5))
716 { 1239 {
717 dec->cur += 5; 1240 dec->cur += 5;
718 return newSViv (0); 1241 return SvREFCNT_inc (json_false);
719 } 1242 }
720 else 1243 else
721 ERR ("'false' expected"); 1244 ERR ("'false' expected");
722 1245
723 break; 1246 break;
724 1247
725 case 'n': 1248 case 'n':
726 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4)) 1249 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4))
727 { 1250 {
728 dec->cur += 4; 1251 dec->cur += 4;
729 return newSViv (1); 1252 return newSVsv (&PL_sv_undef);
730 } 1253 }
731 else 1254 else
732 ERR ("'null' expected"); 1255 ERR ("'null' expected");
733 1256
734 break; 1257 break;
735 1258
736 default: 1259 default:
737 ERR ("malformed json string"); 1260 ERR ("malformed JSON string, neither array, object, number, string or atom");
738 break; 1261 break;
739 } 1262 }
740 1263
741fail: 1264fail:
742 return 0; 1265 return 0;
743} 1266}
744 1267
745static SV * 1268static SV *
746decode_json (SV *string, UV flags) 1269decode_json (SV *string, JSON *json, UV *offset_return)
747{ 1270{
1271 dec_t dec;
1272 UV offset;
748 SV *sv; 1273 SV *sv;
749 1274
1275 SvGETMAGIC (string);
1276 SvUPGRADE (string, SVt_PV);
1277
1278 if (json->flags & F_MAXSIZE && SvCUR (string) > DEC_SIZE (json->flags))
1279 croak ("attempted decode of JSON text of %lu bytes size, but max_size is set to %lu",
1280 (unsigned long)SvCUR (string), (unsigned long)DEC_SIZE (json->flags));
1281
750 if (!(flags & F_UTF8)) 1282 if (json->flags & F_UTF8)
1283 sv_utf8_downgrade (string, 0);
1284 else
751 sv_utf8_upgrade (string); 1285 sv_utf8_upgrade (string);
752 1286
753 SvGROW (string, SvCUR (string) + 1); // should basically be a NOP 1287 SvGROW (string, SvCUR (string) + 1); // should basically be a NOP
754 1288
755 dec_t dec; 1289 dec.json = *json;
756 dec.flags = flags;
757 dec.cur = SvPVX (string); 1290 dec.cur = SvPVX (string);
758 dec.end = SvEND (string); 1291 dec.end = SvEND (string);
759 dec.err = 0; 1292 dec.err = 0;
1293 dec.depth = 0;
1294 dec.maxdepth = DEC_DEPTH (dec.json.flags);
760 1295
761 *dec.end = 1; // invalid anywhere 1296 if (dec.json.cb_object || dec.json.cb_sk_object)
1297 dec.json.flags |= F_HOOK;
1298
1299 *dec.end = 0; // this should basically be a nop, too, but make sure it's there
762 sv = decode_sv (&dec); 1300 sv = decode_sv (&dec);
763 *dec.end = 0; 1301
1302 if (!(offset_return || !sv))
1303 {
1304 // check for trailing garbage
1305 decode_ws (&dec);
1306
1307 if (*dec.cur)
1308 {
1309 dec.err = "garbage after JSON object";
1310 SvREFCNT_dec (sv);
1311 sv = 0;
1312 }
1313 }
1314
1315 if (offset_return || !sv)
1316 {
1317 offset = dec.json.flags & F_UTF8
1318 ? dec.cur - SvPVX (string)
1319 : utf8_distance (dec.cur, SvPVX (string));
1320
1321 if (offset_return)
1322 *offset_return = offset;
1323 }
764 1324
765 if (!sv) 1325 if (!sv)
766 { 1326 {
767 IV offset = utf8_distance (dec.cur, SvPVX (string));
768 SV *uni = sv_newmortal (); 1327 SV *uni = sv_newmortal ();
769 1328
1329 // horrible hack to silence warning inside pv_uni_display
1330 COP cop = *PL_curcop;
1331 cop.cop_warnings = pWARN_NONE;
1332 ENTER;
1333 SAVEVPTR (PL_curcop);
1334 PL_curcop = &cop;
770 pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ); 1335 pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ);
1336 LEAVE;
1337
771 croak ("%s, at character %d (%s)", 1338 croak ("%s, at character offset %d [\"%s\"]",
772 dec.err, 1339 dec.err,
773 (int)offset, 1340 (int)offset,
774 dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)"); 1341 dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)");
775 } 1342 }
776 1343
777 sv_dump (sv);//D
778 return sv_2mortal (sv); 1344 sv = sv_2mortal (sv);
1345
1346 if (!(dec.json.flags & F_ALLOW_NONREF) && !SvROK (sv))
1347 croak ("JSON text must be an object or array (but found number, string, true, false or null, use allow_nonref to allow this)");
1348
1349 return sv;
779} 1350}
1351
1352/////////////////////////////////////////////////////////////////////////////
1353// XS interface functions
780 1354
781MODULE = JSON::XS PACKAGE = JSON::XS 1355MODULE = JSON::XS PACKAGE = JSON::XS
782 1356
783BOOT: 1357BOOT:
784{ 1358{
785 int i; 1359 int i;
786 1360
787 memset (decode_hexdigit, 0xff, 256);
788 for (i = 10; i--; ) 1361 for (i = 0; i < 256; ++i)
789 decode_hexdigit ['0' + i] = i; 1362 decode_hexdigit [i] =
1363 i >= '0' && i <= '9' ? i - '0'
1364 : i >= 'a' && i <= 'f' ? i - 'a' + 10
1365 : i >= 'A' && i <= 'F' ? i - 'A' + 10
1366 : -1;
790 1367
791 for (i = 6; --i; ) 1368 json_stash = gv_stashpv ("JSON::XS" , 1);
1369 json_boolean_stash = gv_stashpv ("JSON::XS::Boolean", 1);
1370
1371 json_true = get_sv ("JSON::XS::true" , 1); SvREADONLY_on (json_true );
1372 json_false = get_sv ("JSON::XS::false", 1); SvREADONLY_on (json_false);
1373}
1374
1375PROTOTYPES: DISABLE
1376
1377void CLONE (...)
1378 CODE:
1379 json_stash = 0;
1380 json_boolean_stash = 0;
1381
1382void new (char *klass)
1383 PPCODE:
1384{
1385 SV *pv = NEWSV (0, sizeof (JSON));
1386 SvPOK_only (pv);
1387 Zero (SvPVX (pv), 1, JSON);
1388 ((JSON *)SvPVX (pv))->flags = F_DEFAULT;
1389 XPUSHs (sv_2mortal (sv_bless (newRV_noinc (pv), JSON_STASH)));
1390}
1391
1392void ascii (JSON *self, int enable = 1)
1393 ALIAS:
1394 ascii = F_ASCII
1395 latin1 = F_LATIN1
1396 utf8 = F_UTF8
1397 indent = F_INDENT
1398 canonical = F_CANONICAL
1399 space_before = F_SPACE_BEFORE
1400 space_after = F_SPACE_AFTER
1401 pretty = F_PRETTY
1402 allow_nonref = F_ALLOW_NONREF
1403 shrink = F_SHRINK
1404 allow_blessed = F_ALLOW_BLESSED
1405 convert_blessed = F_CONV_BLESSED
1406 PPCODE:
1407{
1408 if (enable)
1409 self->flags |= ix;
1410 else
1411 self->flags &= ~ix;
1412
1413 XPUSHs (ST (0));
1414}
1415
1416void max_depth (JSON *self, UV max_depth = 0x80000000UL)
1417 PPCODE:
1418{
1419 UV log2 = 0;
1420
1421 if (max_depth > 0x80000000UL) max_depth = 0x80000000UL;
1422
1423 while ((1UL << log2) < max_depth)
1424 ++log2;
1425
1426 self->flags = self->flags & ~F_MAXDEPTH | (log2 << S_MAXDEPTH);
1427
1428 XPUSHs (ST (0));
1429}
1430
1431void max_size (JSON *self, UV max_size = 0)
1432 PPCODE:
1433{
1434 UV log2 = 0;
1435
1436 if (max_size > 0x80000000UL) max_size = 0x80000000UL;
1437 if (max_size == 1) max_size = 2;
1438
1439 while ((1UL << log2) < max_size)
1440 ++log2;
1441
1442 self->flags = self->flags & ~F_MAXSIZE | (log2 << S_MAXSIZE);
1443
1444 XPUSHs (ST (0));
1445}
1446
1447void filter_json_object (JSON *self, SV *cb = &PL_sv_undef)
1448 PPCODE:
1449{
1450 SvREFCNT_dec (self->cb_object);
1451 self->cb_object = SvOK (cb) ? newSVsv (cb) : 0;
1452
1453 XPUSHs (ST (0));
1454}
1455
1456void filter_json_single_key_object (JSON *self, SV *key, SV *cb = &PL_sv_undef)
1457 PPCODE:
1458{
1459 if (!self->cb_sk_object)
1460 self->cb_sk_object = newHV ();
1461
1462 if (SvOK (cb))
1463 hv_store_ent (self->cb_sk_object, key, newSVsv (cb), 0);
1464 else
792 { 1465 {
793 decode_hexdigit ['a' + i] = 10 + i; 1466 hv_delete_ent (self->cb_sk_object, key, G_DISCARD, 0);
794 decode_hexdigit ['A' + i] = 10 + i; 1467
1468 if (!HvKEYS (self->cb_sk_object))
1469 {
1470 SvREFCNT_dec (self->cb_sk_object);
1471 self->cb_sk_object = 0;
1472 }
795 } 1473 }
796 1474
797 json_stash = gv_stashpv ("JSON::XS", 1); 1475 XPUSHs (ST (0));
798} 1476}
799 1477
800SV *new (char *dummy) 1478void encode (JSON *self, SV *scalar)
1479 PPCODE:
1480 XPUSHs (encode_json (scalar, self));
1481
1482void decode (JSON *self, SV *jsonstr)
1483 PPCODE:
1484 XPUSHs (decode_json (jsonstr, self, 0));
1485
1486void decode_prefix (JSON *self, SV *jsonstr)
1487 PPCODE:
1488{
1489 UV offset;
1490 EXTEND (SP, 2);
1491 PUSHs (decode_json (jsonstr, self, &offset));
1492 PUSHs (sv_2mortal (newSVuv (offset)));
1493}
1494
1495void DESTROY (JSON *self)
801 CODE: 1496 CODE:
802 RETVAL = sv_bless (newRV_noinc (newSVuv (F_DEFAULT)), json_stash); 1497 SvREFCNT_dec (self->cb_sk_object);
803 OUTPUT: 1498 SvREFCNT_dec (self->cb_object);
804 RETVAL
805 1499
806SV *ascii (SV *self, int enable) 1500PROTOTYPES: ENABLE
807 ALIAS:
808 ascii = F_ASCII
809 utf8 = F_UTF8
810 indent = F_INDENT
811 canonical = F_CANONICAL
812 space_before = F_SPACE_BEFORE
813 space_after = F_SPACE_AFTER
814 json_rpc = F_JSON_RPC
815 pretty = F_PRETTY
816 CODE:
817{
818 UV *uv = SvJSON (self);
819 if (enable)
820 *uv |= ix;
821 else
822 *uv &= ~ix;
823
824 RETVAL = newSVsv (self);
825}
826 OUTPUT:
827 RETVAL
828
829void encode (SV *self, SV *scalar)
830 PPCODE:
831 XPUSHs (encode_json (scalar, *SvJSON (self)));
832
833void decode (SV *self, SV *jsonstr)
834 PPCODE:
835 XPUSHs (decode_json (jsonstr, *SvJSON (self)));
836 1501
837void to_json (SV *scalar) 1502void to_json (SV *scalar)
838 PPCODE: 1503 PPCODE:
1504{
1505 JSON json = { F_DEFAULT | F_UTF8 };
839 XPUSHs (encode_json (scalar, F_UTF8)); 1506 XPUSHs (encode_json (scalar, &json));
1507}
840 1508
841void from_json (SV *jsonstr) 1509void from_json (SV *jsonstr)
842 PPCODE: 1510 PPCODE:
1511{
1512 JSON json = { F_DEFAULT | F_UTF8 };
843 XPUSHs (decode_json (jsonstr, F_UTF8)); 1513 XPUSHs (decode_json (jsonstr, &json, 0));
1514}
844 1515

Diff Legend

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