1 #include "EXTERN.h"
2 #include "perl.h"
3 #include "XSUB.h"
4
5 #include <assert.h>
6 #include <string.h>
7 #include <stdlib.h>
8 #include <stdio.h>
9 #include <limits.h>
10 #include <float.h>
11 #include <inttypes.h>
12
13 #if defined(__BORLANDC__) || defined(_MSC_VER)
14 # define snprintf _snprintf // C compilers have this in stdio.h
15 #endif
16
17 // some old perls do not have this, try to make it work, no
18 // guarantees, though. if it breaks, you get to keep the pieces.
19 #ifndef UTF8_MAXBYTES
20 # define UTF8_MAXBYTES 13
21 #endif
22
23 // compatibility with perl <5.18
24 #ifndef HvNAMELEN_get
25 # define HvNAMELEN_get(hv) strlen (HvNAME (hv))
26 #endif
27 #ifndef HvNAMELEN
28 # define HvNAMELEN(hv) HvNAMELEN_get (hv)
29 #endif
30 #ifndef HvNAMEUTF8
31 # define HvNAMEUTF8(hv) 0
32 #endif
33
34 // three extra for rounding, sign, and end of string
35 #define IVUV_MAXCHARS (sizeof (UV) * CHAR_BIT * 28 / 93 + 3)
36
37 #define F_ASCII 0x00000001UL
38 #define F_LATIN1 0x00000002UL
39 #define F_UTF8 0x00000004UL
40 #define F_INDENT 0x00000008UL
41 #define F_CANONICAL 0x00000010UL
42 #define F_SPACE_BEFORE 0x00000020UL
43 #define F_SPACE_AFTER 0x00000040UL
44 #define F_ALLOW_NONREF 0x00000100UL
45 #define F_SHRINK 0x00000200UL
46 #define F_ALLOW_BLESSED 0x00000400UL
47 #define F_CONV_BLESSED 0x00000800UL
48 #define F_RELAXED 0x00001000UL
49 #define F_ALLOW_UNKNOWN 0x00002000UL
50 #define F_ALLOW_TAGS 0x00004000UL
51 #define F_HOOK 0x00080000UL // some hooks exist, so slow-path processing
52
53 #define F_PRETTY F_INDENT | F_SPACE_BEFORE | F_SPACE_AFTER
54
55 #define INIT_SIZE 64 // initial scalar size to be allocated
56 #define INDENT_STEP 3 // spaces per indentation level
57
58 #define SHORT_STRING_LEN 16384 // special-case strings of up to this size
59
60 #define DECODE_WANTS_OCTETS(json) ((json)->flags & F_UTF8)
61
62 #define SB do {
63 #define SE } while (0)
64
65 #if __GNUC__ >= 3
66 # define expect(expr,value) __builtin_expect ((expr), (value))
67 # define INLINE static inline
68 #else
69 # define expect(expr,value) (expr)
70 # define INLINE static
71 #endif
72
73 #define expect_false(expr) expect ((expr) != 0, 0)
74 #define expect_true(expr) expect ((expr) != 0, 1)
75
76 #define IN_RANGE_INC(type,val,beg,end) \
77 ((unsigned type)((unsigned type)(val) - (unsigned type)(beg)) \
78 <= (unsigned type)((unsigned type)(end) - (unsigned type)(beg)))
79
80 #define ERR_NESTING_EXCEEDED "json text or perl structure exceeds maximum nesting level (max_depth set too low?)"
81
82 #ifdef USE_ITHREADS
83 # define JSON_STASH (expect_true (json_stash) ? json_stash : gv_stashpv ("JSON::XS", 1))
84 # define BOOL_STASH (expect_true (bool_stash) ? bool_stash : gv_stashpv ("Types::Serialiser::Boolean", 1))
85 # define GET_BOOL(value) (expect_true (bool_ ## value) ? bool_ ## value : get_bool ("Types::Serialiser::" # value))
86 #else
87 # define JSON_STASH json_stash
88 # define BOOL_STASH bool_stash
89 # define GET_BOOL(value) bool_ ## value
90 #endif
91
92 // the amount of HEs to allocate on the stack, when sorting keys
93 #define STACK_HES 64
94
95 static HV *json_stash, *bool_stash; // JSON::XS::, Types::Serialiser::Boolean::
96 static SV *bool_false, *bool_true;
97 static SV *sv_json;
98
99 enum {
100 INCR_M_WS = 0, // initial whitespace skipping, must be 0
101 INCR_M_TFN, // inside true/false/null
102 INCR_M_NUM, // inside number
103 INCR_M_STR, // inside string
104 INCR_M_BS, // inside backslash
105 INCR_M_C0, // inside comment in initial whitespace sequence
106 INCR_M_C1, // inside comment in other places
107 INCR_M_JSON // outside anything, count nesting
108 };
109
110 #define INCR_DONE(json) ((json)->incr_nest <= 0 && (json)->incr_mode == INCR_M_JSON)
111
112 typedef struct {
113 U32 flags;
114 U32 max_depth;
115 STRLEN max_size;
116
117 SV *cb_object;
118 HV *cb_sk_object;
119
120 // for the incremental parser
121 SV *incr_text; // the source text so far
122 STRLEN incr_pos; // the current offset into the text
123 int incr_nest; // {[]}-nesting level
124 unsigned char incr_mode;
125
126 SV *v_false, *v_true;
127 } JSON;
128
129 INLINE void
json_init(JSON * json)130 json_init (JSON *json)
131 {
132 static const JSON init = { F_ALLOW_NONREF, 512 };
133
134 *json = init;
135 }
136
137 /////////////////////////////////////////////////////////////////////////////
138 // utility functions
139
140 INLINE SV *
get_bool(const char * name)141 get_bool (const char *name)
142 {
143 SV *sv = get_sv (name, 1);
144
145 SvREADONLY_on (sv);
146 SvREADONLY_on (SvRV (sv));
147
148 return sv;
149 }
150
151 INLINE void
shrink(SV * sv)152 shrink (SV *sv)
153 {
154 sv_utf8_downgrade (sv, 1);
155
156 if (SvLEN (sv) > SvCUR (sv) + 1)
157 {
158 #ifdef SvPV_shrink_to_cur
159 SvPV_shrink_to_cur (sv);
160 #elif defined (SvPV_renew)
161 SvPV_renew (sv, SvCUR (sv) + 1);
162 #endif
163 }
164 }
165
166 /* adds two STRLENs together, slow, and with paranoia */
167 static STRLEN
strlen_sum(STRLEN l1,STRLEN l2)168 strlen_sum (STRLEN l1, STRLEN l2)
169 {
170 size_t sum = l1 + l2;
171
172 if (sum < (size_t)l2 || sum != (size_t)(STRLEN)sum)
173 croak ("JSON::XS: string size overflow");
174
175 return sum;
176 }
177
178 /* similar to SvGROW, but somewhat safer and guarantees exponential realloc strategy */
179 static char *
json_sv_grow(SV * sv,size_t len1,size_t len2)180 json_sv_grow (SV *sv, size_t len1, size_t len2)
181 {
182 len1 = strlen_sum (len1, len2);
183 len1 = strlen_sum (len1, len1 >> 1);
184
185 if (len1 > 4096 - 24)
186 len1 = (len1 | 4095) - 24;
187
188 return SvGROW (sv, len1);
189 }
190
191 // decode a utf-8 character and return it, or (UV)-1 in
192 // case of an error.
193 // we special-case "safe" characters from U+80 .. U+7FF,
194 // but use the very good perl function to parse anything else.
195 // note that we never call this function for a ascii codepoints
196 INLINE UV
decode_utf8(unsigned char * s,STRLEN len,STRLEN * clen)197 decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen)
198 {
199 if (expect_true (len >= 2
200 && IN_RANGE_INC (char, s[0], 0xc2, 0xdf)
201 && IN_RANGE_INC (char, s[1], 0x80, 0xbf)))
202 {
203 *clen = 2;
204 return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f);
205 }
206 else
207 return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY);
208 }
209
210 // likewise for encoding, also never called for ascii codepoints
211 // this function takes advantage of this fact, although current gccs
212 // seem to optimise the check for >= 0x80 away anyways
213 INLINE unsigned char *
encode_utf8(unsigned char * s,UV ch)214 encode_utf8 (unsigned char *s, UV ch)
215 {
216 if (expect_false (ch < 0x000080))
217 *s++ = ch;
218 else if (expect_true (ch < 0x000800))
219 *s++ = 0xc0 | ( ch >> 6),
220 *s++ = 0x80 | ( ch & 0x3f);
221 else if ( ch < 0x010000)
222 *s++ = 0xe0 | ( ch >> 12),
223 *s++ = 0x80 | ((ch >> 6) & 0x3f),
224 *s++ = 0x80 | ( ch & 0x3f);
225 else if ( ch < 0x110000)
226 *s++ = 0xf0 | ( ch >> 18),
227 *s++ = 0x80 | ((ch >> 12) & 0x3f),
228 *s++ = 0x80 | ((ch >> 6) & 0x3f),
229 *s++ = 0x80 | ( ch & 0x3f);
230
231 return s;
232 }
233
234 // convert offset pointer to character index, sv must be string
235 static STRLEN
ptr_to_index(SV * sv,char * offset)236 ptr_to_index (SV *sv, char *offset)
237 {
238 return SvUTF8 (sv)
239 ? utf8_distance (offset, SvPVX (sv))
240 : offset - SvPVX (sv);
241 }
242
243 /////////////////////////////////////////////////////////////////////////////
244 // fp hell
245
246 // scan a group of digits, and a trailing exponent
247 static void
json_atof_scan1(const char * s,NV * accum,int * expo,int postdp,int maxdepth)248 json_atof_scan1 (const char *s, NV *accum, int *expo, int postdp, int maxdepth)
249 {
250 UV uaccum = 0;
251 int eaccum = 0;
252
253 // if we recurse too deep, skip all remaining digits
254 // to avoid a stack overflow attack
255 if (expect_false (--maxdepth <= 0))
256 while (((U8)*s - '0') < 10)
257 ++s;
258
259 for (;;)
260 {
261 U8 dig = (U8)*s - '0';
262
263 if (expect_false (dig >= 10))
264 {
265 if (dig == (U8)((U8)'.' - (U8)'0'))
266 {
267 ++s;
268 json_atof_scan1 (s, accum, expo, 1, maxdepth);
269 }
270 else if ((dig | ' ') == 'e' - '0')
271 {
272 int exp2 = 0;
273 int neg = 0;
274
275 ++s;
276
277 if (*s == '-')
278 {
279 ++s;
280 neg = 1;
281 }
282 else if (*s == '+')
283 ++s;
284
285 while ((dig = (U8)*s - '0') < 10)
286 exp2 = exp2 * 10 + *s++ - '0';
287
288 *expo += neg ? -exp2 : exp2;
289 }
290
291 break;
292 }
293
294 ++s;
295
296 uaccum = uaccum * 10 + dig;
297 ++eaccum;
298
299 // if we have too many digits, then recurse for more
300 // we actually do this for rather few digits
301 if (uaccum >= (UV_MAX - 9) / 10)
302 {
303 if (postdp) *expo -= eaccum;
304 json_atof_scan1 (s, accum, expo, postdp, maxdepth);
305 if (postdp) *expo += eaccum;
306
307 break;
308 }
309 }
310
311 // this relies greatly on the quality of the pow ()
312 // implementation of the platform, but a good
313 // implementation is hard to beat.
314 // (IEEE 754 conformant ones are required to be exact)
315 if (postdp) *expo -= eaccum;
316 *accum += uaccum * Perl_pow (10., *expo);
317 *expo += eaccum;
318 }
319
320 static NV
json_atof(const char * s)321 json_atof (const char *s)
322 {
323 NV accum = 0.;
324 int expo = 0;
325 int neg = 0;
326
327 if (*s == '-')
328 {
329 ++s;
330 neg = 1;
331 }
332
333 // a recursion depth of ten gives us >>500 bits
334 json_atof_scan1 (s, &accum, &expo, 0, 10);
335
336 return neg ? -accum : accum;
337 }
338
339 // target of scalar reference is bool? -1 == nope, 0 == false, 1 == true
340 static int
ref_bool_type(SV * sv)341 ref_bool_type (SV *sv)
342 {
343 svtype svt = SvTYPE (sv);
344
345 if (svt < SVt_PVAV)
346 {
347 STRLEN len = 0;
348 char *pv = svt ? SvPV (sv, len) : 0;
349
350 if (len == 1)
351 if (*pv == '1')
352 return 1;
353 else if (*pv == '0')
354 return 0;
355 }
356
357 return -1;
358 }
359
360 // returns whether scalar is not a reference in the sense of allow_nonref
361 static int
json_nonref(SV * scalar)362 json_nonref (SV *scalar)
363 {
364 if (!SvROK (scalar))
365 return 1;
366
367 scalar = SvRV (scalar);
368
369 if (SvTYPE (scalar) >= SVt_PVMG)
370 {
371 if (SvSTASH (scalar) == bool_stash)
372 return 1;
373
374 if (!SvOBJECT (scalar) && ref_bool_type (scalar) >= 0)
375 return 1;
376 }
377
378 return 0;
379 }
380
381 /////////////////////////////////////////////////////////////////////////////
382 // encoder
383
384 // structure used for encoding JSON
385 typedef struct
386 {
387 char *cur; // SvPVX (sv) + current output position
388 char *end; // SvEND (sv)
389 SV *sv; // result scalar
390 JSON json;
391 U32 indent; // indentation level
392 UV limit; // escape character values >= this value when encoding
393 } enc_t;
394
395 INLINE void
need(enc_t * enc,STRLEN len)396 need (enc_t *enc, STRLEN len)
397 {
398 if (expect_false ((uintptr_t)(enc->end - enc->cur) < len))
399 {
400 STRLEN cur = enc->cur - (char *)SvPVX (enc->sv);
401 char *buf = json_sv_grow (enc->sv, cur, len);
402 enc->cur = buf + cur;
403 enc->end = buf + SvLEN (enc->sv) - 1;
404 }
405 }
406
407 INLINE void
encode_ch(enc_t * enc,char ch)408 encode_ch (enc_t *enc, char ch)
409 {
410 need (enc, 1);
411 *enc->cur++ = ch;
412 }
413
414 static void
encode_str(enc_t * enc,char * str,STRLEN len,int is_utf8)415 encode_str (enc_t *enc, char *str, STRLEN len, int is_utf8)
416 {
417 char *end = str + len;
418
419 need (enc, len);
420
421 while (str < end)
422 {
423 unsigned char ch = *(unsigned char *)str;
424
425 if (expect_true (ch >= 0x20 && ch < 0x80)) // most common case
426 {
427 if (expect_false (ch == '"')) // but with slow exceptions
428 {
429 need (enc, len + 1);
430 *enc->cur++ = '\\';
431 *enc->cur++ = '"';
432 }
433 else if (expect_false (ch == '\\'))
434 {
435 need (enc, len + 1);
436 *enc->cur++ = '\\';
437 *enc->cur++ = '\\';
438 }
439 else
440 *enc->cur++ = ch;
441
442 ++str;
443 }
444 else
445 {
446 switch (ch)
447 {
448 case '\010': need (enc, len + 1); *enc->cur++ = '\\'; *enc->cur++ = 'b'; ++str; break;
449 case '\011': need (enc, len + 1); *enc->cur++ = '\\'; *enc->cur++ = 't'; ++str; break;
450 case '\012': need (enc, len + 1); *enc->cur++ = '\\'; *enc->cur++ = 'n'; ++str; break;
451 case '\014': need (enc, len + 1); *enc->cur++ = '\\'; *enc->cur++ = 'f'; ++str; break;
452 case '\015': need (enc, len + 1); *enc->cur++ = '\\'; *enc->cur++ = 'r'; ++str; break;
453
454 default:
455 {
456 STRLEN clen;
457 UV uch;
458
459 if (is_utf8)
460 {
461 uch = decode_utf8 (str, end - str, &clen);
462 if (clen == (STRLEN)-1)
463 croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str);
464 }
465 else
466 {
467 uch = ch;
468 clen = 1;
469 }
470
471 if (uch < 0x80/*0x20*/ || uch >= enc->limit)
472 {
473 if (uch >= 0x10000UL)
474 {
475 if (uch >= 0x110000UL)
476 croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch);
477
478 need (enc, len + 11);
479 sprintf (enc->cur, "\\u%04x\\u%04x",
480 (int)((uch - 0x10000) / 0x400 + 0xD800),
481 (int)((uch - 0x10000) % 0x400 + 0xDC00));
482 enc->cur += 12;
483 }
484 else
485 {
486 need (enc, len + 5);
487 *enc->cur++ = '\\';
488 *enc->cur++ = 'u';
489 *enc->cur++ = PL_hexdigit [ uch >> 12 ];
490 *enc->cur++ = PL_hexdigit [(uch >> 8) & 15];
491 *enc->cur++ = PL_hexdigit [(uch >> 4) & 15];
492 *enc->cur++ = PL_hexdigit [(uch >> 0) & 15];
493 }
494
495 str += clen;
496 }
497 else if (enc->json.flags & F_LATIN1)
498 {
499 *enc->cur++ = uch;
500 str += clen;
501 }
502 else if (is_utf8)
503 {
504 need (enc, len + clen);
505 do
506 {
507 *enc->cur++ = *str++;
508 }
509 while (--clen);
510 }
511 else
512 {
513 need (enc, len + UTF8_MAXBYTES - 1); // never more than 11 bytes needed
514 enc->cur = encode_utf8 (enc->cur, uch);
515 ++str;
516 }
517 }
518 }
519 }
520
521 --len;
522 }
523 }
524
525 INLINE void
encode_indent(enc_t * enc)526 encode_indent (enc_t *enc)
527 {
528 if (enc->json.flags & F_INDENT)
529 {
530 int spaces = enc->indent * INDENT_STEP;
531
532 need (enc, spaces);
533 memset (enc->cur, ' ', spaces);
534 enc->cur += spaces;
535 }
536 }
537
538 INLINE void
encode_space(enc_t * enc)539 encode_space (enc_t *enc)
540 {
541 need (enc, 1);
542 encode_ch (enc, ' ');
543 }
544
545 INLINE void
encode_nl(enc_t * enc)546 encode_nl (enc_t *enc)
547 {
548 if (enc->json.flags & F_INDENT)
549 {
550 need (enc, 1);
551 encode_ch (enc, '\n');
552 }
553 }
554
555 INLINE void
encode_comma(enc_t * enc)556 encode_comma (enc_t *enc)
557 {
558 encode_ch (enc, ',');
559
560 if (enc->json.flags & F_INDENT)
561 encode_nl (enc);
562 else if (enc->json.flags & F_SPACE_AFTER)
563 encode_space (enc);
564 }
565
566 static void encode_sv (enc_t *enc, SV *sv);
567
568 static void
encode_av(enc_t * enc,AV * av)569 encode_av (enc_t *enc, AV *av)
570 {
571 int i, len = av_len (av);
572
573 if (enc->indent >= enc->json.max_depth)
574 croak (ERR_NESTING_EXCEEDED);
575
576 encode_ch (enc, '[');
577
578 if (len >= 0)
579 {
580 encode_nl (enc); ++enc->indent;
581
582 for (i = 0; i <= len; ++i)
583 {
584 SV **svp = av_fetch (av, i, 0);
585
586 encode_indent (enc);
587
588 if (svp)
589 encode_sv (enc, *svp);
590 else
591 encode_str (enc, "null", 4, 0);
592
593 if (i < len)
594 encode_comma (enc);
595 }
596
597 encode_nl (enc); --enc->indent; encode_indent (enc);
598 }
599
600 encode_ch (enc, ']');
601 }
602
603 static void
encode_hk(enc_t * enc,HE * he)604 encode_hk (enc_t *enc, HE *he)
605 {
606 encode_ch (enc, '"');
607
608 if (HeKLEN (he) == HEf_SVKEY)
609 {
610 SV *sv = HeSVKEY (he);
611 STRLEN len;
612 char *str;
613
614 SvGETMAGIC (sv);
615 str = SvPV (sv, len);
616
617 encode_str (enc, str, len, SvUTF8 (sv));
618 }
619 else
620 encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he));
621
622 encode_ch (enc, '"');
623
624 if (enc->json.flags & F_SPACE_BEFORE) encode_space (enc);
625 encode_ch (enc, ':');
626 if (enc->json.flags & F_SPACE_AFTER ) encode_space (enc);
627 }
628
629 // compare hash entries, used when all keys are bytestrings
630 static int
he_cmp_fast(const void * a_,const void * b_)631 he_cmp_fast (const void *a_, const void *b_)
632 {
633 int cmp;
634
635 HE *a = *(HE **)a_;
636 HE *b = *(HE **)b_;
637
638 STRLEN la = HeKLEN (a);
639 STRLEN lb = HeKLEN (b);
640
641 if (!(cmp = memcmp (HeKEY (b), HeKEY (a), lb < la ? lb : la)))
642 cmp = lb - la;
643
644 return cmp;
645 }
646
647 // compare hash entries, used when some keys are sv's or utf-x
648 static int
he_cmp_slow(const void * a,const void * b)649 he_cmp_slow (const void *a, const void *b)
650 {
651 return sv_cmp (HeSVKEY_force (*(HE **)b), HeSVKEY_force (*(HE **)a));
652 }
653
654 static void
encode_hv(enc_t * enc,HV * hv)655 encode_hv (enc_t *enc, HV *hv)
656 {
657 HE *he;
658
659 if (enc->indent >= enc->json.max_depth)
660 croak (ERR_NESTING_EXCEEDED);
661
662 encode_ch (enc, '{');
663
664 // for canonical output we have to sort by keys first
665 // actually, this is mostly due to the stupid so-called
666 // security workaround added somewhere in 5.8.x
667 // that randomises hash orderings
668 if (enc->json.flags & F_CANONICAL && !SvRMAGICAL (hv))
669 {
670 int count = hv_iterinit (hv);
671
672 if (SvMAGICAL (hv))
673 {
674 // need to count by iterating. could improve by dynamically building the vector below
675 // but I don't care for the speed of this special case.
676 // note also that we will run into undefined behaviour when the two iterations
677 // do not result in the same count, something I might care for in some later release.
678
679 count = 0;
680 while (hv_iternext (hv))
681 ++count;
682
683 hv_iterinit (hv);
684 }
685
686 if (count)
687 {
688 int i, fast = 1;
689 HE *hes_stack [STACK_HES];
690 HE **hes = hes_stack;
691
692 // allocate larger arrays on the heap
693 if (count > STACK_HES)
694 {
695 SV *sv = sv_2mortal (NEWSV (0, count * sizeof (*hes)));
696 hes = (HE **)SvPVX (sv);
697 }
698
699 i = 0;
700 while ((he = hv_iternext (hv)))
701 {
702 hes [i++] = he;
703 if (HeKLEN (he) < 0 || HeKUTF8 (he))
704 fast = 0;
705 }
706
707 assert (i == count);
708
709 if (fast)
710 qsort (hes, count, sizeof (HE *), he_cmp_fast);
711 else
712 {
713 // hack to forcefully disable "use bytes"
714 COP cop = *PL_curcop;
715 cop.op_private = 0;
716
717 ENTER;
718 SAVETMPS;
719
720 SAVEVPTR (PL_curcop);
721 PL_curcop = &cop;
722
723 qsort (hes, count, sizeof (HE *), he_cmp_slow);
724
725 FREETMPS;
726 LEAVE;
727 }
728
729 encode_nl (enc); ++enc->indent;
730
731 while (count--)
732 {
733 encode_indent (enc);
734 he = hes [count];
735 encode_hk (enc, he);
736 encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he));
737
738 if (count)
739 encode_comma (enc);
740 }
741
742 encode_nl (enc); --enc->indent; encode_indent (enc);
743 }
744 }
745 else
746 {
747 if (hv_iterinit (hv) || SvMAGICAL (hv))
748 if ((he = hv_iternext (hv)))
749 {
750 encode_nl (enc); ++enc->indent;
751
752 for (;;)
753 {
754 encode_indent (enc);
755 encode_hk (enc, he);
756 encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he));
757
758 if (!(he = hv_iternext (hv)))
759 break;
760
761 encode_comma (enc);
762 }
763
764 encode_nl (enc); --enc->indent; encode_indent (enc);
765 }
766 }
767
768 encode_ch (enc, '}');
769 }
770
771 // encode objects, arrays and special \0=false and \1=true values.
772 static void
encode_rv(enc_t * enc,SV * sv)773 encode_rv (enc_t *enc, SV *sv)
774 {
775 svtype svt;
776 GV *method;
777
778 SvGETMAGIC (sv);
779 svt = SvTYPE (sv);
780
781 if (expect_false (SvOBJECT (sv)))
782 {
783 HV *stash = SvSTASH (sv);
784
785 if (stash == bool_stash)
786 {
787 if (SvIV (sv)) encode_str (enc, "true" , 4, 0);
788 else encode_str (enc, "false", 5, 0);
789 }
790 else if ((enc->json.flags & F_ALLOW_TAGS) && (method = gv_fetchmethod_autoload (stash, "FREEZE", 0)))
791 {
792 int count;
793 dSP;
794
795 ENTER; SAVETMPS;
796 PUSHMARK (SP);
797 EXTEND (SP, 2);
798 // we re-bless the reference to get overload and other niceties right
799 PUSHs (sv_bless (sv_2mortal (newRV_inc (sv)), stash));
800 PUSHs (sv_json);
801
802 PUTBACK;
803 count = call_sv ((SV *)GvCV (method), G_ARRAY);
804 SPAGAIN;
805
806 // catch this surprisingly common error
807 if (SvROK (TOPs) && SvRV (TOPs) == sv)
808 croak ("%s::FREEZE method returned same object as was passed instead of a new one", HvNAME (SvSTASH (sv)));
809
810 encode_ch (enc, '(');
811 encode_ch (enc, '"');
812 encode_str (enc, HvNAME (stash), HvNAMELEN (stash), HvNAMEUTF8 (stash));
813 encode_ch (enc, '"');
814 encode_ch (enc, ')');
815 encode_ch (enc, '[');
816
817 if (count)
818 {
819 int i;
820
821 for (i = 0; i < count - 1; ++i)
822 {
823 encode_sv (enc, SP[i + 1 - count]);
824 encode_ch (enc, ',');
825 }
826
827 encode_sv (enc, TOPs);
828 SP -= count;
829 }
830
831 PUTBACK;
832
833 encode_ch (enc, ']');
834
835 FREETMPS; LEAVE;
836 }
837 else if ((enc->json.flags & F_CONV_BLESSED) && (method = gv_fetchmethod_autoload (stash, "TO_JSON", 0)))
838 {
839 dSP;
840
841 ENTER; SAVETMPS;
842 PUSHMARK (SP);
843 // we re-bless the reference to get overload and other niceties right
844 XPUSHs (sv_bless (sv_2mortal (newRV_inc (sv)), stash));
845
846 // calling with G_SCALAR ensures that we always get a 1 return value
847 PUTBACK;
848 call_sv ((SV *)GvCV (method), G_SCALAR);
849 SPAGAIN;
850
851 // catch this surprisingly common error
852 if (SvROK (TOPs) && SvRV (TOPs) == sv)
853 croak ("%s::TO_JSON method returned same object as was passed instead of a new one", HvNAME (SvSTASH (sv)));
854
855 sv = POPs;
856 PUTBACK;
857
858 encode_sv (enc, sv);
859
860 FREETMPS; LEAVE;
861 }
862 else if (enc->json.flags & F_ALLOW_BLESSED)
863 encode_str (enc, "null", 4, 0);
864 else
865 croak ("encountered object '%s', but neither allow_blessed, convert_blessed nor allow_tags settings are enabled (or TO_JSON/FREEZE method missing)",
866 SvPV_nolen (sv_2mortal (newRV_inc (sv))));
867 }
868 else if (svt == SVt_PVHV)
869 encode_hv (enc, (HV *)sv);
870 else if (svt == SVt_PVAV)
871 encode_av (enc, (AV *)sv);
872 else if (svt < SVt_PVAV)
873 {
874 int bool_type = ref_bool_type (sv);
875
876 if (bool_type == 1)
877 encode_str (enc, "true", 4, 0);
878 else if (bool_type == 0)
879 encode_str (enc, "false", 5, 0);
880 else if (enc->json.flags & F_ALLOW_UNKNOWN)
881 encode_str (enc, "null", 4, 0);
882 else
883 croak ("cannot encode reference to scalar '%s' unless the scalar is 0 or 1",
884 SvPV_nolen (sv_2mortal (newRV_inc (sv))));
885 }
886 else if (enc->json.flags & F_ALLOW_UNKNOWN)
887 encode_str (enc, "null", 4, 0);
888 else
889 croak ("encountered %s, but JSON can only represent references to arrays or hashes",
890 SvPV_nolen (sv_2mortal (newRV_inc (sv))));
891 }
892
893 static void
encode_sv(enc_t * enc,SV * sv)894 encode_sv (enc_t *enc, SV *sv)
895 {
896 SvGETMAGIC (sv);
897
898 if (SvPOKp (sv))
899 {
900 STRLEN len;
901 char *str = SvPV (sv, len);
902 encode_ch (enc, '"');
903 encode_str (enc, str, len, SvUTF8 (sv));
904 encode_ch (enc, '"');
905 }
906 else if (SvNOKp (sv))
907 {
908 // trust that perl will do the right thing w.r.t. JSON syntax.
909 need (enc, NV_DIG + 32);
910 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur);
911 enc->cur += strlen (enc->cur);
912 }
913 else if (SvIOKp (sv))
914 {
915 // we assume we can always read an IV as a UV and vice versa
916 // we assume two's complement
917 // we assume no aliasing issues in the union
918 if (SvIsUV (sv) ? SvUVX (sv) <= 59000
919 : SvIVX (sv) <= 59000 && SvIVX (sv) >= -59000)
920 {
921 // optimise the "small number case"
922 // code will likely be branchless and use only a single multiplication
923 // works for numbers up to 59074
924 I32 i = SvIVX (sv);
925 U32 u;
926 char digit, nz = 0;
927
928 need (enc, 6);
929
930 *enc->cur = '-'; enc->cur += i < 0 ? 1 : 0;
931 u = i < 0 ? -i : i;
932
933 // convert to 4.28 fixed-point representation
934 u = u * ((0xfffffff + 10000) / 10000); // 10**5, 5 fractional digits
935
936 // now output digit by digit, each time masking out the integer part
937 // and multiplying by 5 while moving the decimal point one to the right,
938 // resulting in a net multiplication by 10.
939 // we always write the digit to memory but conditionally increment
940 // the pointer, to enable the use of conditional move instructions.
941 digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffffUL) * 5;
942 digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffffUL) * 5;
943 digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffffUL) * 5;
944 digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffffUL) * 5;
945 digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0'
946 }
947 else
948 {
949 // large integer, use the (rather slow) snprintf way.
950 need (enc, IVUV_MAXCHARS);
951 enc->cur +=
952 SvIsUV(sv)
953 ? snprintf (enc->cur, IVUV_MAXCHARS, "%"UVuf, (UV)SvUVX (sv))
954 : snprintf (enc->cur, IVUV_MAXCHARS, "%"IVdf, (IV)SvIVX (sv));
955 }
956 }
957 else if (SvROK (sv))
958 encode_rv (enc, SvRV (sv));
959 else if (!SvOK (sv) || enc->json.flags & F_ALLOW_UNKNOWN)
960 encode_str (enc, "null", 4, 0);
961 else
962 croak ("encountered perl type (%s,0x%x) that JSON cannot handle, check your input data",
963 SvPV_nolen (sv), (unsigned int)SvFLAGS (sv));
964 }
965
966 static SV *
encode_json(SV * scalar,JSON * json)967 encode_json (SV *scalar, JSON *json)
968 {
969 enc_t enc;
970
971 if (!(json->flags & F_ALLOW_NONREF) && json_nonref (scalar))
972 croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)");
973
974 enc.json = *json;
975 enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE));
976 enc.cur = SvPVX (enc.sv);
977 enc.end = SvEND (enc.sv);
978 enc.indent = 0;
979 enc.limit = enc.json.flags & F_ASCII ? 0x000080UL
980 : enc.json.flags & F_LATIN1 ? 0x000100UL
981 : 0x110000UL;
982
983 SvPOK_only (enc.sv);
984 encode_sv (&enc, scalar);
985 encode_nl (&enc);
986
987 SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv));
988 *SvEND (enc.sv) = 0; // many xs functions expect a trailing 0 for text strings
989
990 if (!(enc.json.flags & (F_ASCII | F_LATIN1 | F_UTF8)))
991 SvUTF8_on (enc.sv);
992
993 if (enc.json.flags & F_SHRINK)
994 shrink (enc.sv);
995
996 return enc.sv;
997 }
998
999 /////////////////////////////////////////////////////////////////////////////
1000 // decoder
1001
1002 // structure used for decoding JSON
1003 typedef struct
1004 {
1005 char *cur; // current parser pointer
1006 char *end; // end of input string
1007 const char *err; // parse error, if != 0
1008 JSON json;
1009 U32 depth; // recursion depth
1010 U32 maxdepth; // recursion depth limit
1011 } dec_t;
1012
1013 INLINE void
decode_comment(dec_t * dec)1014 decode_comment (dec_t *dec)
1015 {
1016 // only '#'-style comments allowed a.t.m.
1017
1018 while (*dec->cur && *dec->cur != 0x0a && *dec->cur != 0x0d)
1019 ++dec->cur;
1020 }
1021
1022 INLINE void
decode_ws(dec_t * dec)1023 decode_ws (dec_t *dec)
1024 {
1025 for (;;)
1026 {
1027 char ch = *dec->cur;
1028
1029 if (ch > 0x20)
1030 {
1031 if (expect_false (ch == '#'))
1032 {
1033 if (dec->json.flags & F_RELAXED)
1034 decode_comment (dec);
1035 else
1036 break;
1037 }
1038 else
1039 break;
1040 }
1041 else if (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09)
1042 break; // parse error, but let higher level handle it, gives better error messages
1043 else
1044 ++dec->cur;
1045 }
1046 }
1047
1048 #define ERR(reason) SB dec->err = reason; goto fail; SE
1049
1050 #define EXPECT_CH(ch) SB \
1051 if (*dec->cur != ch) \
1052 ERR (# ch " expected"); \
1053 ++dec->cur; \
1054 SE
1055
1056 #define DEC_INC_DEPTH if (++dec->depth > dec->json.max_depth) ERR (ERR_NESTING_EXCEEDED)
1057 #define DEC_DEC_DEPTH --dec->depth
1058
1059 static SV *decode_sv (dec_t *dec);
1060
1061 static signed char decode_hexdigit[256];
1062
1063 static UV
decode_4hex(dec_t * dec)1064 decode_4hex (dec_t *dec)
1065 {
1066 signed char d1, d2, d3, d4;
1067 unsigned char *cur = (unsigned char *)dec->cur;
1068
1069 d1 = decode_hexdigit [cur [0]]; if (expect_false (d1 < 0)) ERR ("exactly four hexadecimal digits expected");
1070 d2 = decode_hexdigit [cur [1]]; if (expect_false (d2 < 0)) ERR ("exactly four hexadecimal digits expected");
1071 d3 = decode_hexdigit [cur [2]]; if (expect_false (d3 < 0)) ERR ("exactly four hexadecimal digits expected");
1072 d4 = decode_hexdigit [cur [3]]; if (expect_false (d4 < 0)) ERR ("exactly four hexadecimal digits expected");
1073
1074 dec->cur += 4;
1075
1076 return ((UV)d1) << 12
1077 | ((UV)d2) << 8
1078 | ((UV)d3) << 4
1079 | ((UV)d4);
1080
1081 fail:
1082 return (UV)-1;
1083 }
1084
1085 static SV *
decode_str(dec_t * dec)1086 decode_str (dec_t *dec)
1087 {
1088 SV *sv = 0;
1089 int utf8 = 0;
1090 char *dec_cur = dec->cur;
1091
1092 do
1093 {
1094 char buf [SHORT_STRING_LEN + UTF8_MAXBYTES];
1095 char *cur = buf;
1096
1097 do
1098 {
1099 unsigned char ch = *(unsigned char *)dec_cur++;
1100
1101 if (expect_false (ch == '"'))
1102 {
1103 --dec_cur;
1104 break;
1105 }
1106 else if (expect_false (ch == '\\'))
1107 {
1108 switch (*dec_cur)
1109 {
1110 case '\\':
1111 case '/':
1112 case '"': *cur++ = *dec_cur++; break;
1113
1114 case 'b': ++dec_cur; *cur++ = '\010'; break;
1115 case 't': ++dec_cur; *cur++ = '\011'; break;
1116 case 'n': ++dec_cur; *cur++ = '\012'; break;
1117 case 'f': ++dec_cur; *cur++ = '\014'; break;
1118 case 'r': ++dec_cur; *cur++ = '\015'; break;
1119
1120 case 'u':
1121 {
1122 UV lo, hi;
1123 ++dec_cur;
1124
1125 dec->cur = dec_cur;
1126 hi = decode_4hex (dec);
1127 dec_cur = dec->cur;
1128 if (hi == (UV)-1)
1129 goto fail;
1130
1131 // possibly a surrogate pair
1132 if (hi >= 0xd800)
1133 if (hi < 0xdc00)
1134 {
1135 if (dec_cur [0] != '\\' || dec_cur [1] != 'u')
1136 ERR ("missing low surrogate character in surrogate pair");
1137
1138 dec_cur += 2;
1139
1140 dec->cur = dec_cur;
1141 lo = decode_4hex (dec);
1142 dec_cur = dec->cur;
1143 if (lo == (UV)-1)
1144 goto fail;
1145
1146 if (lo < 0xdc00 || lo >= 0xe000)
1147 ERR ("surrogate pair expected");
1148
1149 hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000;
1150 }
1151 else if (hi < 0xe000)
1152 ERR ("missing high surrogate character in surrogate pair");
1153
1154 if (hi >= 0x80)
1155 {
1156 utf8 = 1;
1157
1158 cur = encode_utf8 (cur, hi);
1159 }
1160 else
1161 *cur++ = hi;
1162 }
1163 break;
1164
1165 default:
1166 --dec_cur;
1167 ERR ("illegal backslash escape sequence in string");
1168 }
1169 }
1170 else if (expect_true (ch >= 0x20 && ch < 0x80))
1171 *cur++ = ch;
1172 else if (ch >= 0x80)
1173 {
1174 STRLEN clen;
1175
1176 --dec_cur;
1177
1178 decode_utf8 (dec_cur, dec->end - dec_cur, &clen);
1179 if (clen == (STRLEN)-1)
1180 ERR ("malformed UTF-8 character in JSON string");
1181
1182 do
1183 *cur++ = *dec_cur++;
1184 while (--clen);
1185
1186 utf8 = 1;
1187 }
1188 else if (ch == '\t' && dec->json.flags & F_RELAXED)
1189 *cur++ = ch;
1190 else
1191 {
1192 --dec_cur;
1193
1194 if (!ch)
1195 ERR ("unexpected end of string while parsing JSON string");
1196 else
1197 ERR ("invalid character encountered while parsing JSON string");
1198 }
1199 }
1200 while (cur < buf + SHORT_STRING_LEN);
1201
1202 {
1203 STRLEN len = cur - buf;
1204
1205 if (sv)
1206 {
1207 STRLEN cur = SvCUR (sv);
1208
1209 if (SvLEN (sv) - cur <= len)
1210 json_sv_grow (sv, cur, len);
1211
1212 memcpy (SvPVX (sv) + SvCUR (sv), buf, len);
1213 SvCUR_set (sv, SvCUR (sv) + len);
1214 }
1215 else
1216 sv = newSVpvn (buf, len);
1217 }
1218 }
1219 while (*dec_cur != '"');
1220
1221 ++dec_cur;
1222
1223 if (sv)
1224 {
1225 SvPOK_only (sv);
1226 *SvEND (sv) = 0;
1227
1228 if (utf8)
1229 SvUTF8_on (sv);
1230 }
1231 else
1232 sv = newSVpvn ("", 0);
1233
1234 dec->cur = dec_cur;
1235 return sv;
1236
1237 fail:
1238 dec->cur = dec_cur;
1239 return 0;
1240 }
1241
1242 static SV *
decode_num(dec_t * dec)1243 decode_num (dec_t *dec)
1244 {
1245 int is_nv = 0;
1246 char *start = dec->cur;
1247
1248 // [minus]
1249 if (*dec->cur == '-')
1250 ++dec->cur;
1251
1252 if (*dec->cur == '0')
1253 {
1254 ++dec->cur;
1255 if (*dec->cur >= '0' && *dec->cur <= '9')
1256 ERR ("malformed number (leading zero must not be followed by another digit)");
1257 }
1258 else if (*dec->cur < '0' || *dec->cur > '9')
1259 ERR ("malformed number (no digits after initial minus)");
1260 else
1261 do
1262 {
1263 ++dec->cur;
1264 }
1265 while (*dec->cur >= '0' && *dec->cur <= '9');
1266
1267 // [frac]
1268 if (*dec->cur == '.')
1269 {
1270 ++dec->cur;
1271
1272 if (*dec->cur < '0' || *dec->cur > '9')
1273 ERR ("malformed number (no digits after decimal point)");
1274
1275 do
1276 {
1277 ++dec->cur;
1278 }
1279 while (*dec->cur >= '0' && *dec->cur <= '9');
1280
1281 is_nv = 1;
1282 }
1283
1284 // [exp]
1285 if (*dec->cur == 'e' || *dec->cur == 'E')
1286 {
1287 ++dec->cur;
1288
1289 if (*dec->cur == '-' || *dec->cur == '+')
1290 ++dec->cur;
1291
1292 if (*dec->cur < '0' || *dec->cur > '9')
1293 ERR ("malformed number (no digits after exp sign)");
1294
1295 do
1296 {
1297 ++dec->cur;
1298 }
1299 while (*dec->cur >= '0' && *dec->cur <= '9');
1300
1301 is_nv = 1;
1302 }
1303
1304 if (!is_nv)
1305 {
1306 int len = dec->cur - start;
1307
1308 // special case the rather common 1..5-digit-int case
1309 if (*start == '-')
1310 switch (len)
1311 {
1312 case 2: return newSViv (-(IV)( start [1] - '0' * 1));
1313 case 3: return newSViv (-(IV)( start [1] * 10 + start [2] - '0' * 11));
1314 case 4: return newSViv (-(IV)( start [1] * 100 + start [2] * 10 + start [3] - '0' * 111));
1315 case 5: return newSViv (-(IV)( start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 1111));
1316 case 6: return newSViv (-(IV)(start [1] * 10000 + start [2] * 1000 + start [3] * 100 + start [4] * 10 + start [5] - '0' * 11111));
1317 }
1318 else
1319 switch (len)
1320 {
1321 case 1: return newSViv ( start [0] - '0' * 1);
1322 case 2: return newSViv ( start [0] * 10 + start [1] - '0' * 11);
1323 case 3: return newSViv ( start [0] * 100 + start [1] * 10 + start [2] - '0' * 111);
1324 case 4: return newSViv ( start [0] * 1000 + start [1] * 100 + start [2] * 10 + start [3] - '0' * 1111);
1325 case 5: return newSViv ( start [0] * 10000 + start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 11111);
1326 }
1327
1328 {
1329 UV uv;
1330 int numtype = grok_number (start, len, &uv);
1331 if (numtype & IS_NUMBER_IN_UV)
1332 if (numtype & IS_NUMBER_NEG)
1333 {
1334 if (uv < (UV)IV_MIN)
1335 return newSViv (-(IV)uv);
1336 }
1337 else
1338 return newSVuv (uv);
1339 }
1340
1341 len -= *start == '-' ? 1 : 0;
1342
1343 // does not fit into IV or UV, try NV
1344 if (len <= NV_DIG)
1345 // fits into NV without loss of precision
1346 return newSVnv (json_atof (start));
1347
1348 // everything else fails, convert it to a string
1349 return newSVpvn (start, dec->cur - start);
1350 }
1351
1352 // loss of precision here
1353 return newSVnv (json_atof (start));
1354
1355 fail:
1356 return 0;
1357 }
1358
1359 static SV *
decode_av(dec_t * dec)1360 decode_av (dec_t *dec)
1361 {
1362 AV *av = newAV ();
1363
1364 DEC_INC_DEPTH;
1365 decode_ws (dec);
1366
1367 if (*dec->cur == ']')
1368 ++dec->cur;
1369 else
1370 for (;;)
1371 {
1372 SV *value;
1373
1374 value = decode_sv (dec);
1375 if (!value)
1376 goto fail;
1377
1378 av_push (av, value);
1379
1380 decode_ws (dec);
1381
1382 if (*dec->cur == ']')
1383 {
1384 ++dec->cur;
1385 break;
1386 }
1387
1388 if (*dec->cur != ',')
1389 ERR (", or ] expected while parsing array");
1390
1391 ++dec->cur;
1392
1393 decode_ws (dec);
1394
1395 if (*dec->cur == ']' && dec->json.flags & F_RELAXED)
1396 {
1397 ++dec->cur;
1398 break;
1399 }
1400 }
1401
1402 DEC_DEC_DEPTH;
1403 return newRV_noinc ((SV *)av);
1404
1405 fail:
1406 SvREFCNT_dec (av);
1407 DEC_DEC_DEPTH;
1408 return 0;
1409 }
1410
1411 static SV *
decode_hv(dec_t * dec)1412 decode_hv (dec_t *dec)
1413 {
1414 SV *sv;
1415 HV *hv = newHV ();
1416
1417 DEC_INC_DEPTH;
1418 decode_ws (dec);
1419
1420 if (*dec->cur == '}')
1421 ++dec->cur;
1422 else
1423 for (;;)
1424 {
1425 EXPECT_CH ('"');
1426
1427 // heuristic: assume that
1428 // a) decode_str + hv_store_ent are abysmally slow.
1429 // b) most hash keys are short, simple ascii text.
1430 // => try to "fast-match" such strings to avoid
1431 // the overhead of decode_str + hv_store_ent.
1432 {
1433 SV *value;
1434 char *p = dec->cur;
1435 char *e = p + 24; // only try up to 24 bytes
1436
1437 for (;;)
1438 {
1439 // the >= 0x80 is false on most architectures
1440 if (p == e || *p < 0x20 || *p >= 0x80 || *p == '\\')
1441 {
1442 // slow path, back up and use decode_str
1443 SV *key = decode_str (dec);
1444 if (!key)
1445 goto fail;
1446
1447 decode_ws (dec); EXPECT_CH (':');
1448
1449 decode_ws (dec);
1450 value = decode_sv (dec);
1451 if (!value)
1452 {
1453 SvREFCNT_dec (key);
1454 goto fail;
1455 }
1456
1457 hv_store_ent (hv, key, value, 0);
1458 SvREFCNT_dec (key);
1459
1460 break;
1461 }
1462 else if (*p == '"')
1463 {
1464 // fast path, got a simple key
1465 char *key = dec->cur;
1466 int len = p - key;
1467 dec->cur = p + 1;
1468
1469 decode_ws (dec); EXPECT_CH (':');
1470
1471 decode_ws (dec);
1472 value = decode_sv (dec);
1473 if (!value)
1474 goto fail;
1475
1476 hv_store (hv, key, len, value, 0);
1477
1478 break;
1479 }
1480
1481 ++p;
1482 }
1483 }
1484
1485 decode_ws (dec);
1486
1487 if (*dec->cur == '}')
1488 {
1489 ++dec->cur;
1490 break;
1491 }
1492
1493 if (*dec->cur != ',')
1494 ERR (", or } expected while parsing object/hash");
1495
1496 ++dec->cur;
1497
1498 decode_ws (dec);
1499
1500 if (*dec->cur == '}' && dec->json.flags & F_RELAXED)
1501 {
1502 ++dec->cur;
1503 break;
1504 }
1505 }
1506
1507 DEC_DEC_DEPTH;
1508 sv = newRV_noinc ((SV *)hv);
1509
1510 // check filter callbacks
1511 if (expect_false (dec->json.flags & F_HOOK))
1512 {
1513 if (dec->json.cb_sk_object && HvKEYS (hv) == 1)
1514 {
1515 HE *cb, *he;
1516
1517 hv_iterinit (hv);
1518 he = hv_iternext (hv);
1519 hv_iterinit (hv);
1520
1521 // the next line creates a mortal sv each time it's called.
1522 // might want to optimise this for common cases.
1523 cb = hv_fetch_ent (dec->json.cb_sk_object, hv_iterkeysv (he), 0, 0);
1524
1525 if (cb)
1526 {
1527 dSP;
1528 int count;
1529
1530 ENTER; SAVETMPS;
1531 PUSHMARK (SP);
1532 XPUSHs (HeVAL (he));
1533 sv_2mortal (sv);
1534
1535 PUTBACK; count = call_sv (HeVAL (cb), G_ARRAY); SPAGAIN;
1536
1537 if (count == 1)
1538 {
1539 sv = newSVsv (POPs);
1540 PUTBACK;
1541 FREETMPS; LEAVE;
1542 return sv;
1543 }
1544 else if (count)
1545 croak ("filter_json_single_key_object callbacks must not return more than one scalar");
1546
1547 PUTBACK;
1548
1549 SvREFCNT_inc (sv);
1550
1551 FREETMPS; LEAVE;
1552 }
1553 }
1554
1555 if (dec->json.cb_object)
1556 {
1557 dSP;
1558 int count;
1559
1560 ENTER; SAVETMPS;
1561 PUSHMARK (SP);
1562 XPUSHs (sv_2mortal (sv));
1563
1564 PUTBACK; count = call_sv (dec->json.cb_object, G_ARRAY); SPAGAIN;
1565
1566 if (count == 1)
1567 sv = newSVsv (POPs);
1568 else if (count == 0)
1569 SvREFCNT_inc (sv);
1570 else
1571 croak ("filter_json_object callbacks must not return more than one scalar");
1572
1573 PUTBACK;
1574
1575 FREETMPS; LEAVE;
1576 }
1577 }
1578
1579 return sv;
1580
1581 fail:
1582 SvREFCNT_dec (hv);
1583 DEC_DEC_DEPTH;
1584 return 0;
1585 }
1586
1587 static SV *
decode_tag(dec_t * dec)1588 decode_tag (dec_t *dec)
1589 {
1590 SV *tag = 0;
1591 SV *val = 0;
1592
1593 if (!(dec->json.flags & F_ALLOW_TAGS))
1594 ERR ("malformed JSON string, neither array, object, number, string or atom");
1595
1596 ++dec->cur;
1597
1598 decode_ws (dec);
1599
1600 tag = decode_sv (dec);
1601 if (!tag)
1602 goto fail;
1603
1604 if (!SvPOK (tag))
1605 ERR ("malformed JSON string, (tag) must be a string");
1606
1607 decode_ws (dec);
1608
1609 if (*dec->cur != ')')
1610 ERR (") expected after tag");
1611
1612 ++dec->cur;
1613
1614 decode_ws (dec);
1615
1616 val = decode_sv (dec);
1617 if (!val)
1618 goto fail;
1619
1620 if (!SvROK (val) || SvTYPE (SvRV (val)) != SVt_PVAV)
1621 ERR ("malformed JSON string, tag value must be an array");
1622
1623 {
1624 AV *av = (AV *)SvRV (val);
1625 int i, len = av_len (av) + 1;
1626 HV *stash = gv_stashsv (tag, 0);
1627 SV *sv;
1628
1629 if (!stash)
1630 ERR ("cannot decode perl-object (package does not exist)");
1631
1632 GV *method = gv_fetchmethod_autoload (stash, "THAW", 0);
1633
1634 if (!method)
1635 ERR ("cannot decode perl-object (package does not have a THAW method)");
1636
1637 dSP;
1638
1639 ENTER; SAVETMPS;
1640 PUSHMARK (SP);
1641 EXTEND (SP, len + 2);
1642 // we re-bless the reference to get overload and other niceties right
1643 PUSHs (tag);
1644 PUSHs (sv_json);
1645
1646 for (i = 0; i < len; ++i)
1647 PUSHs (*av_fetch (av, i, 1));
1648
1649 PUTBACK;
1650 call_sv ((SV *)GvCV (method), G_SCALAR);
1651 SPAGAIN;
1652
1653 SvREFCNT_dec (tag);
1654 SvREFCNT_dec (val);
1655 sv = SvREFCNT_inc (POPs);
1656
1657 PUTBACK;
1658
1659 FREETMPS; LEAVE;
1660
1661 return sv;
1662 }
1663
1664 fail:
1665 SvREFCNT_dec (tag);
1666 SvREFCNT_dec (val);
1667 return 0;
1668 }
1669
1670 static SV *
decode_sv(dec_t * dec)1671 decode_sv (dec_t *dec)
1672 {
1673 // the beauty of JSON: you need exactly one character lookahead
1674 // to parse everything.
1675 switch (*dec->cur)
1676 {
1677 case '"': ++dec->cur; return decode_str (dec);
1678 case '[': ++dec->cur; return decode_av (dec);
1679 case '{': ++dec->cur; return decode_hv (dec);
1680 case '(': return decode_tag (dec);
1681
1682 case '-':
1683 case '0': case '1': case '2': case '3': case '4':
1684 case '5': case '6': case '7': case '8': case '9':
1685 return decode_num (dec);
1686
1687 case 'f':
1688 if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5))
1689 {
1690 dec->cur += 5;
1691
1692 if (expect_false (!dec->json.v_false))
1693 dec->json.v_false = GET_BOOL (false);
1694
1695 return newSVsv (dec->json.v_false);
1696 }
1697 else
1698 ERR ("'false' expected");
1699
1700 break;
1701
1702 case 't':
1703 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4))
1704 {
1705 dec->cur += 4;
1706
1707 if (expect_false (!dec->json.v_true))
1708 dec->json.v_true = GET_BOOL (true);
1709
1710 return newSVsv (dec->json.v_true);
1711 }
1712 else
1713 ERR ("'true' expected");
1714
1715 break;
1716
1717 case 'n':
1718 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4))
1719 {
1720 dec->cur += 4;
1721 return newSVsv (&PL_sv_undef);
1722 }
1723 else
1724 ERR ("'null' expected");
1725
1726 break;
1727
1728 default:
1729 ERR ("malformed JSON string, neither tag, array, object, number, string or atom");
1730 break;
1731 }
1732
1733 fail:
1734 return 0;
1735 }
1736
1737 static SV *
decode_json(SV * string,JSON * json,STRLEN * offset_return)1738 decode_json (SV *string, JSON *json, STRLEN *offset_return)
1739 {
1740 dec_t dec;
1741 SV *sv;
1742
1743 /* work around bugs in 5.10 where manipulating magic values
1744 * makes perl ignore the magic in subsequent accesses.
1745 * also make a copy of non-PV values, to get them into a clean
1746 * state (SvPV should do that, but it's buggy, see below).
1747 *
1748 * SvIsCOW_shared_hash works around a bug in perl (possibly 5.16),
1749 * as reported by Reini Urban.
1750 */
1751 /*SvGETMAGIC (string);*/
1752 if (SvMAGICAL (string) || !SvPOK (string) || SvIsCOW_shared_hash (string))
1753 string = sv_2mortal (newSVsv (string));
1754
1755 SvUPGRADE (string, SVt_PV);
1756
1757 /* work around a bug in perl 5.10, which causes SvCUR to fail an
1758 * assertion with -DDEBUGGING, although SvCUR is documented to
1759 * return the xpv_cur field which certainly exists after upgrading.
1760 * according to nicholas clark, calling SvPOK fixes this.
1761 * But it doesn't fix it, so try another workaround, call SvPV_nolen
1762 * and hope for the best.
1763 * Damnit, SvPV_nolen still trips over yet another assertion. This
1764 * assertion business is seriously broken, try yet another workaround
1765 * for the broken -DDEBUGGING.
1766 */
1767 {
1768 #ifdef DEBUGGING
1769 STRLEN offset = SvOK (string) ? sv_len (string) : 0;
1770 #else
1771 STRLEN offset = SvCUR (string);
1772 #endif
1773
1774 if (offset > json->max_size && json->max_size)
1775 croak ("attempted decode of JSON text of %lu bytes size, but max_size is set to %lu",
1776 (unsigned long)SvCUR (string), (unsigned long)json->max_size);
1777 }
1778
1779 if (DECODE_WANTS_OCTETS (json))
1780 sv_utf8_downgrade (string, 0);
1781 else
1782 sv_utf8_upgrade (string);
1783
1784 SvGROW (string, SvCUR (string) + 1); // should basically be a NOP
1785
1786 dec.json = *json;
1787 dec.cur = SvPVX (string);
1788 dec.end = SvEND (string);
1789 dec.err = 0;
1790 dec.depth = 0;
1791
1792 if (dec.json.cb_object || dec.json.cb_sk_object)
1793 dec.json.flags |= F_HOOK;
1794
1795 *dec.end = 0; // this should basically be a nop, too, but make sure it's there
1796
1797 decode_ws (&dec);
1798 sv = decode_sv (&dec);
1799
1800 if (offset_return)
1801 *offset_return = dec.cur - SvPVX (string);
1802 else if (sv)
1803 {
1804 // check for trailing garbage
1805 decode_ws (&dec);
1806
1807 if (dec.cur != dec.end)
1808 {
1809 dec.err = "garbage after JSON object";
1810 SvREFCNT_dec (sv);
1811 sv = 0;
1812 }
1813 }
1814
1815 if (!sv)
1816 {
1817 SV *uni = sv_newmortal ();
1818
1819 // horrible hack to silence warning inside pv_uni_display
1820 COP cop = *PL_curcop;
1821 cop.cop_warnings = pWARN_NONE;
1822 ENTER;
1823 SAVEVPTR (PL_curcop);
1824 PL_curcop = &cop;
1825 pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ);
1826 LEAVE;
1827
1828 croak ("%s, at character offset %d (before \"%s\")",
1829 dec.err,
1830 (int)ptr_to_index (string, dec.cur),
1831 dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)");
1832 }
1833
1834 sv = sv_2mortal (sv);
1835
1836 if (!(dec.json.flags & F_ALLOW_NONREF) && json_nonref (sv))
1837 croak ("JSON text must be an object or array (but found number, string, true, false or null, use allow_nonref to allow this)");
1838
1839 return sv;
1840 }
1841
1842 /////////////////////////////////////////////////////////////////////////////
1843 // incremental parser
1844
1845 static void
incr_parse(JSON * self)1846 incr_parse (JSON *self)
1847 {
1848 const char *p = SvPVX (self->incr_text) + self->incr_pos;
1849
1850 // the state machine here is a bit convoluted and could be simplified a lot
1851 // but this would make it slower, so...
1852
1853 for (;;)
1854 {
1855 switch (self->incr_mode)
1856 {
1857 // reached end of a scalar, see if we are inside a nested structure or not
1858 end_of_scalar:
1859 self->incr_mode = INCR_M_JSON;
1860
1861 if (self->incr_nest) // end of a scalar inside array, object or tag
1862 goto incr_m_json;
1863 else // end of scalar outside structure, json text ends here
1864 goto interrupt;
1865
1866 // only used for initial whitespace skipping
1867 case INCR_M_WS:
1868 for (;;)
1869 {
1870 if (*p > 0x20)
1871 {
1872 if (*p == '#')
1873 {
1874 self->incr_mode = INCR_M_C0;
1875 goto incr_m_c;
1876 }
1877 else
1878 {
1879 self->incr_mode = INCR_M_JSON;
1880 goto incr_m_json;
1881 }
1882 }
1883 else if (!*p)
1884 goto interrupt;
1885
1886 ++p;
1887 }
1888
1889 // skip a single char inside a string (for \\-processing)
1890 case INCR_M_BS:
1891 if (!*p)
1892 goto interrupt;
1893
1894 ++p;
1895 self->incr_mode = INCR_M_STR;
1896 goto incr_m_str;
1897
1898 // inside #-style comments
1899 case INCR_M_C0:
1900 case INCR_M_C1:
1901 incr_m_c:
1902 for (;;)
1903 {
1904 if (*p == '\n')
1905 {
1906 self->incr_mode = self->incr_mode == INCR_M_C0 ? INCR_M_WS : INCR_M_JSON;
1907 break;
1908 }
1909 else if (!*p)
1910 goto interrupt;
1911
1912 ++p;
1913 }
1914
1915 break;
1916
1917 // inside true/false/null
1918 case INCR_M_TFN:
1919 incr_m_tfn:
1920 for (;;)
1921 switch (*p++)
1922 {
1923 case 'r': case 'u': case 'e': // tRUE, falsE, nUll
1924 case 'a': case 'l': case 's': // fALSe, nuLL
1925 // allowed
1926 break;
1927
1928 default:
1929 --p;
1930 goto end_of_scalar;
1931 }
1932
1933 // inside a number
1934 case INCR_M_NUM:
1935 incr_m_num:
1936 for (;;)
1937 switch (*p++)
1938 {
1939 case 'e': case 'E': case '.': case '+':
1940 case '-':
1941 case '0': case '1': case '2': case '3': case '4':
1942 case '5': case '6': case '7': case '8': case '9':
1943 // allowed
1944 break;
1945
1946 default:
1947 --p;
1948 goto end_of_scalar;
1949 }
1950
1951 // inside a string
1952 case INCR_M_STR:
1953 incr_m_str:
1954 for (;;)
1955 {
1956 if (*p == '"')
1957 {
1958 ++p;
1959 goto end_of_scalar;
1960 }
1961 else if (*p == '\\')
1962 {
1963 ++p; // "virtually" consumes character after \
1964
1965 if (!*p) // if at end of string we have to switch modes
1966 {
1967 self->incr_mode = INCR_M_BS;
1968 goto interrupt;
1969 }
1970 }
1971 else if (!*p)
1972 goto interrupt;
1973
1974 ++p;
1975 }
1976
1977 // after initial ws, outside string
1978 case INCR_M_JSON:
1979 incr_m_json:
1980 for (;;)
1981 {
1982 switch (*p++)
1983 {
1984 case 0:
1985 --p;
1986 goto interrupt;
1987
1988 case 0x09:
1989 case 0x0a:
1990 case 0x0d:
1991 case 0x20:
1992 if (!self->incr_nest)
1993 {
1994 --p; // do not eat the whitespace, let the next round do it
1995 goto interrupt;
1996 }
1997 break;
1998
1999 // the following three blocks handle scalars. this makes the parser
2000 // more strict than required inside arrays or objects, and could
2001 // be moved to a special case on the toplevel (except strings)
2002 case 't':
2003 case 'f':
2004 case 'n':
2005 self->incr_mode = INCR_M_TFN;
2006 goto incr_m_tfn;
2007
2008 case '-':
2009 case '0': case '1': case '2': case '3': case '4':
2010 case '5': case '6': case '7': case '8': case '9':
2011 self->incr_mode = INCR_M_NUM;
2012 goto incr_m_num;
2013
2014 case '"':
2015 self->incr_mode = INCR_M_STR;
2016 goto incr_m_str;
2017
2018 case '[':
2019 case '{':
2020 case '(':
2021 if (++self->incr_nest > self->max_depth)
2022 croak (ERR_NESTING_EXCEEDED);
2023 break;
2024
2025 case ']':
2026 case '}':
2027 if (--self->incr_nest <= 0)
2028 goto interrupt;
2029 break;
2030
2031 case ')':
2032 --self->incr_nest;
2033 break;
2034
2035 case '#':
2036 self->incr_mode = INCR_M_C1;
2037 goto incr_m_c;
2038 }
2039 }
2040 }
2041
2042 modechange:
2043 ;
2044 }
2045
2046 interrupt:
2047 self->incr_pos = p - SvPVX (self->incr_text);
2048 //printf ("interrupt<%.*s>\n", self->incr_pos, SvPVX(self->incr_text));//D
2049 //printf ("return pos %d mode %d nest %d\n", self->incr_pos, self->incr_mode, self->incr_nest);//D
2050 }
2051
2052 /////////////////////////////////////////////////////////////////////////////
2053 // XS interface functions
2054
2055 MODULE = JSON::XS PACKAGE = JSON::XS
2056
2057 BOOT:
2058 {
2059 int i;
2060
2061 for (i = 0; i < 256; ++i)
2062 decode_hexdigit [i] =
2063 i >= '0' && i <= '9' ? i - '0'
2064 : i >= 'a' && i <= 'f' ? i - 'a' + 10
2065 : i >= 'A' && i <= 'F' ? i - 'A' + 10
2066 : -1;
2067
2068 json_stash = gv_stashpv ("JSON::XS" , 1);
2069 bool_stash = gv_stashpv ("Types::Serialiser::Boolean", 1);
2070 bool_false = get_bool ("Types::Serialiser::false");
2071 bool_true = get_bool ("Types::Serialiser::true");
2072
2073 sv_json = newSVpv ("JSON", 0);
2074 SvREADONLY_on (sv_json);
2075
2076 CvNODEBUG_on (get_cv ("JSON::XS::incr_text", 0)); /* the debugger completely breaks lvalue subs */
2077 }
2078
2079 PROTOTYPES: DISABLE
2080
2081 void CLONE (...)
2082 CODE:
2083 // as long as these writes are atomic, the race should not matter
2084 // as existing threads either already use 0, or use the old value,
2085 // which is sitll correct for the initial thread.
2086 json_stash = 0;
2087 bool_stash = 0;
2088 bool_false = 0;
2089 bool_true = 0;
2090
new(char * klass)2091 void new (char *klass)
2092 PPCODE:
2093 {
2094 SV *pv = NEWSV (0, sizeof (JSON));
2095 SvPOK_only (pv);
2096 json_init ((JSON *)SvPVX (pv));
2097 XPUSHs (sv_2mortal (sv_bless (
2098 newRV_noinc (pv),
2099 strEQ (klass, "JSON::XS") ? JSON_STASH : gv_stashpv (klass, 1)
2100 )));
2101 }
2102
2103 void boolean_values (JSON *self, SV *v_false = 0, SV *v_true = 0)
2104 PPCODE:
2105 self->v_false = newSVsv (v_false);
2106 self->v_true = newSVsv (v_true);
2107 XPUSHs (ST (0));
2108
2109 void get_boolean_values (JSON *self)
2110 PPCODE:
2111 if (self->v_false && self->v_true)
2112 {
2113 EXTEND (SP, 2);
2114 PUSHs (self->v_false);
2115 PUSHs (self->v_true);
2116 }
2117
2118 void ascii (JSON *self, int enable = 1)
2119 ALIAS:
2120 ascii = F_ASCII
2121 latin1 = F_LATIN1
2122 utf8 = F_UTF8
2123 indent = F_INDENT
2124 canonical = F_CANONICAL
2125 space_before = F_SPACE_BEFORE
2126 space_after = F_SPACE_AFTER
2127 pretty = F_PRETTY
2128 allow_nonref = F_ALLOW_NONREF
2129 shrink = F_SHRINK
2130 allow_blessed = F_ALLOW_BLESSED
2131 convert_blessed = F_CONV_BLESSED
2132 relaxed = F_RELAXED
2133 allow_unknown = F_ALLOW_UNKNOWN
2134 allow_tags = F_ALLOW_TAGS
2135 PPCODE:
2136 {
2137 if (enable)
2138 self->flags |= ix;
2139 else
2140 self->flags &= ~ix;
2141
2142 XPUSHs (ST (0));
2143 }
2144
2145 void get_ascii (JSON *self)
2146 ALIAS:
2147 get_ascii = F_ASCII
2148 get_latin1 = F_LATIN1
2149 get_utf8 = F_UTF8
2150 get_indent = F_INDENT
2151 get_canonical = F_CANONICAL
2152 get_space_before = F_SPACE_BEFORE
2153 get_space_after = F_SPACE_AFTER
2154 get_allow_nonref = F_ALLOW_NONREF
2155 get_shrink = F_SHRINK
2156 get_allow_blessed = F_ALLOW_BLESSED
2157 get_convert_blessed = F_CONV_BLESSED
2158 get_relaxed = F_RELAXED
2159 get_allow_unknown = F_ALLOW_UNKNOWN
2160 get_allow_tags = F_ALLOW_TAGS
2161 PPCODE:
2162 XPUSHs (boolSV (self->flags & ix));
2163
2164 void max_depth (JSON *self, U32 max_depth = 0x80000000UL)
2165 PPCODE:
2166 self->max_depth = max_depth;
2167 XPUSHs (ST (0));
2168
2169 U32 get_max_depth (JSON *self)
2170 CODE:
2171 RETVAL = self->max_depth;
2172 OUTPUT:
2173 RETVAL
2174
2175 void max_size (JSON *self, U32 max_size = 0)
2176 PPCODE:
2177 self->max_size = max_size;
2178 XPUSHs (ST (0));
2179
2180 int get_max_size (JSON *self)
2181 CODE:
2182 RETVAL = self->max_size;
2183 OUTPUT:
2184 RETVAL
2185
2186 void filter_json_object (JSON *self, SV *cb = &PL_sv_undef)
2187 PPCODE:
2188 {
2189 SvREFCNT_dec (self->cb_object);
2190 self->cb_object = SvOK (cb) ? newSVsv (cb) : 0;
2191
2192 XPUSHs (ST (0));
2193 }
2194
2195 void filter_json_single_key_object (JSON *self, SV *key, SV *cb = &PL_sv_undef)
2196 PPCODE:
2197 {
2198 if (!self->cb_sk_object)
2199 self->cb_sk_object = newHV ();
2200
2201 if (SvOK (cb))
2202 hv_store_ent (self->cb_sk_object, key, newSVsv (cb), 0);
2203 else
2204 {
2205 hv_delete_ent (self->cb_sk_object, key, G_DISCARD, 0);
2206
2207 if (!HvKEYS (self->cb_sk_object))
2208 {
2209 SvREFCNT_dec (self->cb_sk_object);
2210 self->cb_sk_object = 0;
2211 }
2212 }
2213
2214 XPUSHs (ST (0));
2215 }
2216
2217 void encode (JSON *self, SV *scalar)
2218 PPCODE:
2219 PUTBACK; scalar = encode_json (scalar, self); SPAGAIN;
2220 XPUSHs (scalar);
2221
2222 void decode (JSON *self, SV *jsonstr)
2223 PPCODE:
2224 PUTBACK; jsonstr = decode_json (jsonstr, self, 0); SPAGAIN;
2225 XPUSHs (jsonstr);
2226
decode_prefix(JSON * self,SV * jsonstr)2227 void decode_prefix (JSON *self, SV *jsonstr)
2228 PPCODE:
2229 {
2230 SV *sv;
2231 STRLEN offset;
2232 PUTBACK; sv = decode_json (jsonstr, self, &offset); SPAGAIN;
2233 EXTEND (SP, 2);
2234 PUSHs (sv);
2235 PUSHs (sv_2mortal (newSVuv (ptr_to_index (jsonstr, SvPV_nolen (jsonstr) + offset))));
2236 }
2237
2238 void incr_parse (JSON *self, SV *jsonstr = 0)
2239 PPCODE:
2240 {
2241 if (!self->incr_text)
2242 self->incr_text = newSVpvn ("", 0);
2243
2244 /* if utf8-ness doesn't match the decoder, need to upgrade/downgrade */
2245 if (!DECODE_WANTS_OCTETS (self) == !SvUTF8 (self->incr_text))
2246 if (DECODE_WANTS_OCTETS (self))
2247 {
2248 if (self->incr_pos)
2249 self->incr_pos = utf8_length ((U8 *)SvPVX (self->incr_text),
2250 (U8 *)SvPVX (self->incr_text) + self->incr_pos);
2251
2252 sv_utf8_downgrade (self->incr_text, 0);
2253 }
2254 else
2255 {
2256 sv_utf8_upgrade (self->incr_text);
2257
2258 if (self->incr_pos)
2259 self->incr_pos = utf8_hop ((U8 *)SvPVX (self->incr_text), self->incr_pos)
2260 - (U8 *)SvPVX (self->incr_text);
2261 }
2262
2263 // append data, if any
2264 if (jsonstr)
2265 {
2266 /* make sure both strings have same encoding */
2267 if (SvUTF8 (jsonstr) != SvUTF8 (self->incr_text))
2268 if (SvUTF8 (jsonstr))
2269 sv_utf8_downgrade (jsonstr, 0);
2270 else
2271 sv_utf8_upgrade (jsonstr);
2272
2273 /* and then just blindly append */
2274 {
2275 STRLEN len;
2276 const char *str = SvPV (jsonstr, len);
2277 STRLEN cur = SvCUR (self->incr_text);
2278
2279 if (SvLEN (self->incr_text) - cur <= len)
2280 json_sv_grow (self->incr_text, cur, len);
2281
2282 Move (str, SvEND (self->incr_text), len, char);
2283 SvCUR_set (self->incr_text, SvCUR (self->incr_text) + len);
2284 *SvEND (self->incr_text) = 0; // this should basically be a nop, too, but make sure it's there
2285 }
2286 }
2287
2288 if (GIMME_V != G_VOID)
2289 do
2290 {
2291 SV *sv;
2292 STRLEN offset;
2293
2294 if (!INCR_DONE (self))
2295 {
2296 incr_parse (self);
2297
2298 if (self->incr_pos > self->max_size && self->max_size)
2299 croak ("attempted decode of JSON text of %lu bytes size, but max_size is set to %lu",
2300 (unsigned long)self->incr_pos, (unsigned long)self->max_size);
2301
2302 if (!INCR_DONE (self))
2303 {
2304 // as an optimisation, do not accumulate white space in the incr buffer
2305 if (self->incr_mode == INCR_M_WS && self->incr_pos)
2306 {
2307 self->incr_pos = 0;
2308 SvCUR_set (self->incr_text, 0);
2309 }
2310
2311 break;
2312 }
2313 }
2314
2315 PUTBACK; sv = decode_json (self->incr_text, self, &offset); SPAGAIN;
2316 XPUSHs (sv);
2317
2318 self->incr_pos -= offset;
2319 self->incr_nest = 0;
2320 self->incr_mode = 0;
2321
2322 sv_chop (self->incr_text, SvPVX (self->incr_text) + offset);
2323 }
2324 while (GIMME_V == G_ARRAY);
2325 }
2326
incr_text(JSON * self)2327 SV *incr_text (JSON *self)
2328 ATTRS: lvalue
2329 CODE:
2330 {
2331 if (self->incr_pos)
2332 croak ("incr_text can not be called when the incremental parser already started parsing");
2333
2334 RETVAL = self->incr_text ? SvREFCNT_inc (self->incr_text) : &PL_sv_undef;
2335 }
2336 OUTPUT:
2337 RETVAL
2338
incr_skip(JSON * self)2339 void incr_skip (JSON *self)
2340 CODE:
2341 {
2342 if (self->incr_pos)
2343 {
2344 sv_chop (self->incr_text, SvPV_nolen (self->incr_text) + self->incr_pos);
2345 self->incr_pos = 0;
2346 self->incr_nest = 0;
2347 self->incr_mode = 0;
2348 }
2349 }
2350
incr_reset(JSON * self)2351 void incr_reset (JSON *self)
2352 CODE:
2353 {
2354 SvREFCNT_dec (self->incr_text);
2355 self->incr_text = 0;
2356 self->incr_pos = 0;
2357 self->incr_nest = 0;
2358 self->incr_mode = 0;
2359 }
2360
2361 void DESTROY (JSON *self)
2362 CODE:
2363 SvREFCNT_dec (self->v_false);
2364 SvREFCNT_dec (self->v_true);
2365 SvREFCNT_dec (self->cb_sk_object);
2366 SvREFCNT_dec (self->cb_object);
2367 SvREFCNT_dec (self->incr_text);
2368
2369 PROTOTYPES: ENABLE
2370
encode_json(SV * scalar)2371 void encode_json (SV *scalar)
2372 PPCODE:
2373 {
2374 JSON json;
2375 json_init (&json);
2376 json.flags |= F_UTF8;
2377 PUTBACK; scalar = encode_json (scalar, &json); SPAGAIN;
2378 XPUSHs (scalar);
2379 }
2380
decode_json(SV * jsonstr)2381 void decode_json (SV *jsonstr)
2382 PPCODE:
2383 {
2384 JSON json;
2385 json_init (&json);
2386 json.flags |= F_UTF8;
2387 PUTBACK; jsonstr = decode_json (jsonstr, &json, 0); SPAGAIN;
2388 XPUSHs (jsonstr);
2389 }
2390
2391