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 12#if defined(__BORLANDC__) || defined(_MSC_VER) 13# define snprintf _snprintf // C compilers have this in stdio.h 14#endif 15 16// some old perls do not have this, try to make it work, no 17// guarantees, though. if it breaks, you get to keep the pieces. 18#ifndef UTF8_MAXBYTES 19# define UTF8_MAXBYTES 13 20#endif 21 22// three extra for rounding, sign, and end of string 23#define IVUV_MAXCHARS (sizeof (UV) * CHAR_BIT * 28 / 93 + 3) 24 25#define F_ASCII 0x00000001UL 26#define F_LATIN1 0x00000002UL 27#define F_UTF8 0x00000004UL 28#define F_INDENT 0x00000008UL 29#define F_CANONICAL 0x00000010UL 30#define F_SPACE_BEFORE 0x00000020UL 31#define F_SPACE_AFTER 0x00000040UL 32#define F_ALLOW_NONREF 0x00000100UL 33#define F_SHRINK 0x00000200UL 34#define F_ALLOW_BLESSED 0x00000400UL 35#define F_CONV_BLESSED 0x00000800UL 36#define F_RELAXED 0x00001000UL 37#define F_ALLOW_UNKNOWN 0x00002000UL 38#define F_HOOK 0x00080000UL // some hooks exist, so slow-path processing 39 40#define F_PRETTY F_INDENT | F_SPACE_BEFORE | F_SPACE_AFTER 41 42#define INIT_SIZE 32 // initial scalar size to be allocated 43#define INDENT_STEP 3 // spaces per indentation level 44 45#define SHORT_STRING_LEN 16384 // special-case strings of up to this size 46 47#define DECODE_WANTS_OCTETS(json) ((json)->flags & F_UTF8) 48 49#define SB do { 50#define SE } while (0) 51 52#if __GNUC__ >= 3 53# define expect(expr,value) __builtin_expect ((expr), (value)) 54# define INLINE static 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#define IN_RANGE_INC(type,val,beg,end) \ 64 ((unsigned type)((unsigned type)(val) - (unsigned type)(beg)) \ 65 <= (unsigned type)((unsigned type)(end) - (unsigned type)(beg))) 66 67#define ERR_NESTING_EXCEEDED "json text or perl structure exceeds maximum nesting level (max_depth set too low?)" 68 69#ifdef USE_ITHREADS 70# define JSON_SLOW 1 71# define JSON_STASH (json_stash ? json_stash : gv_stashpv ("JSON::XS", 1)) 72#else 73# define JSON_SLOW 0 74# define JSON_STASH json_stash 75#endif 76 77static HV *json_stash, *json_boolean_stash; // JSON::XS:: 78static SV *json_true, *json_false; 79 80enum { 81 INCR_M_WS = 0, // initial whitespace skipping, must be 0 82 INCR_M_STR, // inside string 83 INCR_M_BS, // inside backslash 84 INCR_M_C0, // inside comment in initial whitespace sequence 85 INCR_M_C1, // inside comment in other places 86 INCR_M_JSON // outside anything, count nesting 87}; 88 89#define INCR_DONE(json) ((json)->incr_nest <= 0 && (json)->incr_mode == INCR_M_JSON) 90 91typedef struct { 92 U32 flags; 93 U32 max_depth; 94 STRLEN max_size; 95 96 SV *cb_object; 97 HV *cb_sk_object; 98 99 // for the incremental parser 100 SV *incr_text; // the source text so far 101 STRLEN incr_pos; // the current offset into the text 102 int incr_nest; // {[]}-nesting level 103 unsigned char incr_mode; 104} JSON; 105 106INLINE void 107json_init (JSON *json) 108{ 109 Zero (json, 1, JSON); 110 json->max_depth = 512; 111} 112 113///////////////////////////////////////////////////////////////////////////// 114// utility functions 115 116INLINE SV * 117get_bool (const char *name) 118{ 119 SV *sv = get_sv (name, 1); 120 121 SvREADONLY_on (sv); 122 SvREADONLY_on (SvRV (sv)); 123 124 return sv; 125} 126 127INLINE void 128shrink (SV *sv) 129{ 130 sv_utf8_downgrade (sv, 1); 131 132 if (SvLEN (sv) > SvCUR (sv) + 1) 133 { 134#ifdef SvPV_shrink_to_cur 135 SvPV_shrink_to_cur (sv); 136#elif defined (SvPV_renew) 137 SvPV_renew (sv, SvCUR (sv) + 1); 138#endif 139 } 140} 141 142// decode an utf-8 character and return it, or (UV)-1 in 143// case of an error. 144// we special-case "safe" characters from U+80 .. U+7FF, 145// but use the very good perl function to parse anything else. 146// note that we never call this function for a ascii codepoints 147INLINE UV 148decode_utf8 (unsigned char *s, STRLEN len, STRLEN *clen) 149{ 150 if (expect_true (len >= 2 151 && IN_RANGE_INC (char, s[0], 0xc2, 0xdf) 152 && IN_RANGE_INC (char, s[1], 0x80, 0xbf))) 153 { 154 *clen = 2; 155 return ((s[0] & 0x1f) << 6) | (s[1] & 0x3f); 156 } 157 else 158 return utf8n_to_uvuni (s, len, clen, UTF8_CHECK_ONLY); 159} 160 161// likewise for encoding, also never called for ascii codepoints 162// this function takes advantage of this fact, although current gccs 163// seem to optimise the check for >= 0x80 away anyways 164INLINE unsigned char * 165encode_utf8 (unsigned char *s, UV ch) 166{ 167 if (expect_false (ch < 0x000080)) 168 *s++ = ch; 169 else if (expect_true (ch < 0x000800)) 170 *s++ = 0xc0 | ( ch >> 6), 171 *s++ = 0x80 | ( ch & 0x3f); 172 else if ( ch < 0x010000) 173 *s++ = 0xe0 | ( ch >> 12), 174 *s++ = 0x80 | ((ch >> 6) & 0x3f), 175 *s++ = 0x80 | ( ch & 0x3f); 176 else if ( ch < 0x110000) 177 *s++ = 0xf0 | ( ch >> 18), 178 *s++ = 0x80 | ((ch >> 12) & 0x3f), 179 *s++ = 0x80 | ((ch >> 6) & 0x3f), 180 *s++ = 0x80 | ( ch & 0x3f); 181 182 return s; 183} 184 185// convert offset pointer to character index, sv must be string 186static STRLEN 187ptr_to_index (SV *sv, char *offset) 188{ 189 return SvUTF8 (sv) 190 ? utf8_distance (offset, SvPVX (sv)) 191 : offset - SvPVX (sv); 192} 193 194///////////////////////////////////////////////////////////////////////////// 195// fp hell 196 197// scan a group of digits, and a trailing exponent 198static void 199json_atof_scan1 (const char *s, NV *accum, int *expo, int postdp, int maxdepth) 200{ 201 UV uaccum = 0; 202 int eaccum = 0; 203 204 // if we recurse too deep, skip all remaining digits 205 // to avoid a stack overflow attack 206 if (expect_false (--maxdepth <= 0)) 207 while (((U8)*s - '0') < 10) 208 ++s; 209 210 for (;;) 211 { 212 U8 dig = (U8)*s - '0'; 213 214 if (expect_false (dig >= 10)) 215 { 216 if (dig == (U8)((U8)'.' - (U8)'0')) 217 { 218 ++s; 219 json_atof_scan1 (s, accum, expo, 1, maxdepth); 220 } 221 else if ((dig | ' ') == 'e' - '0') 222 { 223 int exp2 = 0; 224 int neg = 0; 225 226 ++s; 227 228 if (*s == '-') 229 { 230 ++s; 231 neg = 1; 232 } 233 else if (*s == '+') 234 ++s; 235 236 while ((dig = (U8)*s - '0') < 10) 237 exp2 = exp2 * 10 + *s++ - '0'; 238 239 *expo += neg ? -exp2 : exp2; 240 } 241 242 break; 243 } 244 245 ++s; 246 247 uaccum = uaccum * 10 + dig; 248 ++eaccum; 249 250 // if we have too many digits, then recurse for more 251 // we actually do this for rather few digits 252 if (uaccum >= (UV_MAX - 9) / 10) 253 { 254 if (postdp) *expo -= eaccum; 255 json_atof_scan1 (s, accum, expo, postdp, maxdepth); 256 if (postdp) *expo += eaccum; 257 258 break; 259 } 260 } 261 262 // this relies greatly on the quality of the pow () 263 // implementation of the platform, but a good 264 // implementation is hard to beat. 265 // (IEEE 754 conformant ones are required to be exact) 266 if (postdp) *expo -= eaccum; 267 *accum += uaccum * Perl_pow (10., *expo); 268 *expo += eaccum; 269} 270 271static NV 272json_atof (const char *s) 273{ 274 NV accum = 0.; 275 int expo = 0; 276 int neg = 0; 277 278 if (*s == '-') 279 { 280 ++s; 281 neg = 1; 282 } 283 284 // a recursion depth of ten gives us >>500 bits 285 json_atof_scan1 (s, &accum, &expo, 0, 10); 286 287 return neg ? -accum : accum; 288} 289///////////////////////////////////////////////////////////////////////////// 290// encoder 291 292// structure used for encoding JSON 293typedef struct 294{ 295 char *cur; // SvPVX (sv) + current output position 296 char *end; // SvEND (sv) 297 SV *sv; // result scalar 298 JSON json; 299 U32 indent; // indentation level 300 UV limit; // escape character values >= this value when encoding 301} enc_t; 302 303INLINE void 304need (enc_t *enc, STRLEN len) 305{ 306 if (expect_false (enc->cur + len >= enc->end)) 307 { 308 STRLEN cur = enc->cur - (char *)SvPVX (enc->sv); 309 SvGROW (enc->sv, cur + (len < (cur >> 2) ? cur >> 2 : len) + 1); 310 enc->cur = SvPVX (enc->sv) + cur; 311 enc->end = SvPVX (enc->sv) + SvLEN (enc->sv) - 1; 312 } 313} 314 315INLINE void 316encode_ch (enc_t *enc, char ch) 317{ 318 need (enc, 1); 319 *enc->cur++ = ch; 320} 321 322static void 323encode_str (enc_t *enc, char *str, STRLEN len, int is_utf8) 324{ 325 char *end = str + len; 326 327 need (enc, len); 328 329 while (str < end) 330 { 331 unsigned char ch = *(unsigned char *)str; 332 333 if (expect_true (ch >= 0x20 && ch < 0x80)) // most common case 334 { 335 if (expect_false (ch == '"')) // but with slow exceptions 336 { 337 need (enc, len += 1); 338 *enc->cur++ = '\\'; 339 *enc->cur++ = '"'; 340 } 341 else if (expect_false (ch == '\\')) 342 { 343 need (enc, len += 1); 344 *enc->cur++ = '\\'; 345 *enc->cur++ = '\\'; 346 } 347 else 348 *enc->cur++ = ch; 349 350 ++str; 351 } 352 else 353 { 354 switch (ch) 355 { 356 case '\010': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'b'; ++str; break; 357 case '\011': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 't'; ++str; break; 358 case '\012': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'n'; ++str; break; 359 case '\014': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'f'; ++str; break; 360 case '\015': need (enc, len += 1); *enc->cur++ = '\\'; *enc->cur++ = 'r'; ++str; break; 361 362 default: 363 { 364 STRLEN clen; 365 UV uch; 366 367 if (is_utf8) 368 { 369 uch = decode_utf8 (str, end - str, &clen); 370 if (clen == (STRLEN)-1) 371 croak ("malformed or illegal unicode character in string [%.11s], cannot convert to JSON", str); 372 } 373 else 374 { 375 uch = ch; 376 clen = 1; 377 } 378 379 if (uch < 0x80/*0x20*/ || uch >= enc->limit) 380 { 381 if (uch >= 0x10000UL) 382 { 383 if (uch >= 0x110000UL) 384 croak ("out of range codepoint (0x%lx) encountered, unrepresentable in JSON", (unsigned long)uch); 385 386 need (enc, len += 11); 387 sprintf (enc->cur, "\\u%04x\\u%04x", 388 (int)((uch - 0x10000) / 0x400 + 0xD800), 389 (int)((uch - 0x10000) % 0x400 + 0xDC00)); 390 enc->cur += 12; 391 } 392 else 393 { 394 need (enc, len += 5); 395 *enc->cur++ = '\\'; 396 *enc->cur++ = 'u'; 397 *enc->cur++ = PL_hexdigit [ uch >> 12 ]; 398 *enc->cur++ = PL_hexdigit [(uch >> 8) & 15]; 399 *enc->cur++ = PL_hexdigit [(uch >> 4) & 15]; 400 *enc->cur++ = PL_hexdigit [(uch >> 0) & 15]; 401 } 402 403 str += clen; 404 } 405 else if (enc->json.flags & F_LATIN1) 406 { 407 *enc->cur++ = uch; 408 str += clen; 409 } 410 else if (is_utf8) 411 { 412 need (enc, len += clen); 413 do 414 { 415 *enc->cur++ = *str++; 416 } 417 while (--clen); 418 } 419 else 420 { 421 need (enc, len += UTF8_MAXBYTES - 1); // never more than 11 bytes needed 422 enc->cur = encode_utf8 (enc->cur, uch); 423 ++str; 424 } 425 } 426 } 427 } 428 429 --len; 430 } 431} 432 433INLINE void 434encode_indent (enc_t *enc) 435{ 436 if (enc->json.flags & F_INDENT) 437 { 438 int spaces = enc->indent * INDENT_STEP; 439 440 need (enc, spaces); 441 memset (enc->cur, ' ', spaces); 442 enc->cur += spaces; 443 } 444} 445 446INLINE void 447encode_space (enc_t *enc) 448{ 449 need (enc, 1); 450 encode_ch (enc, ' '); 451} 452 453INLINE void 454encode_nl (enc_t *enc) 455{ 456 if (enc->json.flags & F_INDENT) 457 { 458 need (enc, 1); 459 encode_ch (enc, '\n'); 460 } 461} 462 463INLINE void 464encode_comma (enc_t *enc) 465{ 466 encode_ch (enc, ','); 467 468 if (enc->json.flags & F_INDENT) 469 encode_nl (enc); 470 else if (enc->json.flags & F_SPACE_AFTER) 471 encode_space (enc); 472} 473 474static void encode_sv (enc_t *enc, SV *sv); 475 476static void 477encode_av (enc_t *enc, AV *av) 478{ 479 int i, len = av_len (av); 480 481 if (enc->indent >= enc->json.max_depth) 482 croak (ERR_NESTING_EXCEEDED); 483 484 encode_ch (enc, '['); 485 486 if (len >= 0) 487 { 488 encode_nl (enc); ++enc->indent; 489 490 for (i = 0; i <= len; ++i) 491 { 492 SV **svp = av_fetch (av, i, 0); 493 494 encode_indent (enc); 495 496 if (svp) 497 encode_sv (enc, *svp); 498 else 499 encode_str (enc, "null", 4, 0); 500 501 if (i < len) 502 encode_comma (enc); 503 } 504 505 encode_nl (enc); --enc->indent; encode_indent (enc); 506 } 507 508 encode_ch (enc, ']'); 509} 510 511static void 512encode_hk (enc_t *enc, HE *he) 513{ 514 encode_ch (enc, '"'); 515 516 if (HeKLEN (he) == HEf_SVKEY) 517 { 518 SV *sv = HeSVKEY (he); 519 STRLEN len; 520 char *str; 521 522 SvGETMAGIC (sv); 523 str = SvPV (sv, len); 524 525 encode_str (enc, str, len, SvUTF8 (sv)); 526 } 527 else 528 encode_str (enc, HeKEY (he), HeKLEN (he), HeKUTF8 (he)); 529 530 encode_ch (enc, '"'); 531 532 if (enc->json.flags & F_SPACE_BEFORE) encode_space (enc); 533 encode_ch (enc, ':'); 534 if (enc->json.flags & F_SPACE_AFTER ) encode_space (enc); 535} 536 537// compare hash entries, used when all keys are bytestrings 538static int 539he_cmp_fast (const void *a_, const void *b_) 540{ 541 int cmp; 542 543 HE *a = *(HE **)a_; 544 HE *b = *(HE **)b_; 545 546 STRLEN la = HeKLEN (a); 547 STRLEN lb = HeKLEN (b); 548 549 if (!(cmp = memcmp (HeKEY (b), HeKEY (a), lb < la ? lb : la))) 550 cmp = lb - la; 551 552 return cmp; 553} 554 555// compare hash entries, used when some keys are sv's or utf-x 556static int 557he_cmp_slow (const void *a, const void *b) 558{ 559 return sv_cmp (HeSVKEY_force (*(HE **)b), HeSVKEY_force (*(HE **)a)); 560} 561 562static void 563encode_hv (enc_t *enc, HV *hv) 564{ 565 HE *he; 566 567 if (enc->indent >= enc->json.max_depth) 568 croak (ERR_NESTING_EXCEEDED); 569 570 encode_ch (enc, '{'); 571 572 // for canonical output we have to sort by keys first 573 // actually, this is mostly due to the stupid so-called 574 // security workaround added somewhere in 5.8.x 575 // that randomises hash orderings 576 if (enc->json.flags & F_CANONICAL && !SvRMAGICAL (hv)) 577 { 578 int count = hv_iterinit (hv); 579 580 if (SvMAGICAL (hv)) 581 { 582 // need to count by iterating. could improve by dynamically building the vector below 583 // but I don't care for the speed of this special case. 584 // note also that we will run into undefined behaviour when the two iterations 585 // do not result in the same count, something I might care for in some later release. 586 587 count = 0; 588 while (hv_iternext (hv)) 589 ++count; 590 591 hv_iterinit (hv); 592 } 593 594 if (count) 595 { 596 int i, fast = 1; 597#if defined(__BORLANDC__) || defined(_MSC_VER) 598 HE **hes = _alloca (count * sizeof (HE)); 599#else 600 HE *hes [count]; // if your compiler dies here, you need to enable C99 mode 601#endif 602 603 i = 0; 604 while ((he = hv_iternext (hv))) 605 { 606 hes [i++] = he; 607 if (HeKLEN (he) < 0 || HeKUTF8 (he)) 608 fast = 0; 609 } 610 611 assert (i == count); 612 613 if (fast) 614 qsort (hes, count, sizeof (HE *), he_cmp_fast); 615 else 616 { 617 // hack to forcefully disable "use bytes" 618 COP cop = *PL_curcop; 619 cop.op_private = 0; 620 621 ENTER; 622 SAVETMPS; 623 624 SAVEVPTR (PL_curcop); 625 PL_curcop = &cop; 626 627 qsort (hes, count, sizeof (HE *), he_cmp_slow); 628 629 FREETMPS; 630 LEAVE; 631 } 632 633 encode_nl (enc); ++enc->indent; 634 635 while (count--) 636 { 637 encode_indent (enc); 638 he = hes [count]; 639 encode_hk (enc, he); 640 encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he)); 641 642 if (count) 643 encode_comma (enc); 644 } 645 646 encode_nl (enc); --enc->indent; encode_indent (enc); 647 } 648 } 649 else 650 { 651 if (hv_iterinit (hv) || SvMAGICAL (hv)) 652 if ((he = hv_iternext (hv))) 653 { 654 encode_nl (enc); ++enc->indent; 655 656 for (;;) 657 { 658 encode_indent (enc); 659 encode_hk (enc, he); 660 encode_sv (enc, expect_false (SvMAGICAL (hv)) ? hv_iterval (hv, he) : HeVAL (he)); 661 662 if (!(he = hv_iternext (hv))) 663 break; 664 665 encode_comma (enc); 666 } 667 668 encode_nl (enc); --enc->indent; encode_indent (enc); 669 } 670 } 671 672 encode_ch (enc, '}'); 673} 674 675// encode objects, arrays and special \0=false and \1=true values. 676static void 677encode_rv (enc_t *enc, SV *sv) 678{ 679 svtype svt; 680 681 SvGETMAGIC (sv); 682 svt = SvTYPE (sv); 683 684 if (expect_false (SvOBJECT (sv))) 685 { 686 HV *stash = !JSON_SLOW || json_boolean_stash 687 ? json_boolean_stash 688 : gv_stashpv ("JSON::XS::Boolean", 1); 689 690 if (SvSTASH (sv) == stash) 691 { 692 if (SvIV (sv)) 693 encode_str (enc, "true", 4, 0); 694 else 695 encode_str (enc, "false", 5, 0); 696 } 697 else 698 { 699#if 0 700 if (0 && sv_derived_from (rv, "JSON::Literal")) 701 { 702 // not yet 703 } 704#endif 705 if (enc->json.flags & F_CONV_BLESSED) 706 { 707 // we re-bless the reference to get overload and other niceties right 708 GV *to_json = gv_fetchmethod_autoload (SvSTASH (sv), "TO_JSON", 0); 709 710 if (to_json) 711 { 712 dSP; 713 714 ENTER; SAVETMPS; PUSHMARK (SP); 715 XPUSHs (sv_bless (sv_2mortal (newRV_inc (sv)), SvSTASH (sv))); 716 717 // calling with G_SCALAR ensures that we always get a 1 return value 718 PUTBACK; 719 call_sv ((SV *)GvCV (to_json), G_SCALAR); 720 SPAGAIN; 721 722 // catch this surprisingly common error 723 if (SvROK (TOPs) && SvRV (TOPs) == sv) 724 croak ("%s::TO_JSON method returned same object as was passed instead of a new one", HvNAME (SvSTASH (sv))); 725 726 sv = POPs; 727 PUTBACK; 728 729 encode_sv (enc, sv); 730 731 FREETMPS; LEAVE; 732 } 733 else if (enc->json.flags & F_ALLOW_BLESSED) 734 encode_str (enc, "null", 4, 0); 735 else 736 croak ("encountered object '%s', but neither allow_blessed enabled nor TO_JSON method available on it", 737 SvPV_nolen (sv_2mortal (newRV_inc (sv)))); 738 } 739 else if (enc->json.flags & F_ALLOW_BLESSED) 740 encode_str (enc, "null", 4, 0); 741 else 742 croak ("encountered object '%s', but neither allow_blessed nor convert_blessed settings are enabled", 743 SvPV_nolen (sv_2mortal (newRV_inc (sv)))); 744 } 745 } 746 else if (svt == SVt_PVHV) 747 encode_hv (enc, (HV *)sv); 748 else if (svt == SVt_PVAV) 749 encode_av (enc, (AV *)sv); 750 else if (svt < SVt_PVAV) 751 { 752 STRLEN len = 0; 753 char *pv = svt ? SvPV (sv, len) : 0; 754 755 if (len == 1 && *pv == '1') 756 encode_str (enc, "true", 4, 0); 757 else if (len == 1 && *pv == '0') 758 encode_str (enc, "false", 5, 0); 759 else if (enc->json.flags & F_ALLOW_UNKNOWN) 760 encode_str (enc, "null", 4, 0); 761 else 762 croak ("cannot encode reference to scalar '%s' unless the scalar is 0 or 1", 763 SvPV_nolen (sv_2mortal (newRV_inc (sv)))); 764 } 765 else if (enc->json.flags & F_ALLOW_UNKNOWN) 766 encode_str (enc, "null", 4, 0); 767 else 768 croak ("encountered %s, but JSON can only represent references to arrays or hashes", 769 SvPV_nolen (sv_2mortal (newRV_inc (sv)))); 770} 771 772static void 773encode_sv (enc_t *enc, SV *sv) 774{ 775 SvGETMAGIC (sv); 776 777 if (SvPOKp (sv)) 778 { 779 STRLEN len; 780 char *str = SvPV (sv, len); 781 encode_ch (enc, '"'); 782 encode_str (enc, str, len, SvUTF8 (sv)); 783 encode_ch (enc, '"'); 784 } 785 else if (SvNOKp (sv)) 786 { 787 // trust that perl will do the right thing w.r.t. JSON syntax. 788 need (enc, NV_DIG + 32); 789 Gconvert (SvNVX (sv), NV_DIG, 0, enc->cur); 790 enc->cur += strlen (enc->cur); 791 } 792 else if (SvIOKp (sv)) 793 { 794 // we assume we can always read an IV as a UV and vice versa 795 // we assume two's complement 796 // we assume no aliasing issues in the union 797 if (SvIsUV (sv) ? SvUVX (sv) <= 59000 798 : SvIVX (sv) <= 59000 && SvIVX (sv) >= -59000) 799 { 800 // optimise the "small number case" 801 // code will likely be branchless and use only a single multiplication 802 // works for numbers up to 59074 803 I32 i = SvIVX (sv); 804 U32 u; 805 char digit, nz = 0; 806 807 need (enc, 6); 808 809 *enc->cur = '-'; enc->cur += i < 0 ? 1 : 0; 810 u = i < 0 ? -i : i; 811 812 // convert to 4.28 fixed-point representation 813 u = u * ((0xfffffff + 10000) / 10000); // 10**5, 5 fractional digits 814 815 // now output digit by digit, each time masking out the integer part 816 // and multiplying by 5 while moving the decimal point one to the right, 817 // resulting in a net multiplication by 10. 818 // we always write the digit to memory but conditionally increment 819 // the pointer, to enable the use of conditional move instructions. 820 digit = u >> 28; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0xfffffffUL) * 5; 821 digit = u >> 27; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x7ffffffUL) * 5; 822 digit = u >> 26; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x3ffffffUL) * 5; 823 digit = u >> 25; *enc->cur = digit + '0'; enc->cur += (nz = nz || digit); u = (u & 0x1ffffffUL) * 5; 824 digit = u >> 24; *enc->cur = digit + '0'; enc->cur += 1; // correctly generate '0' 825 } 826 else 827 { 828 // large integer, use the (rather slow) snprintf way. 829 need (enc, IVUV_MAXCHARS); 830 enc->cur += 831 SvIsUV(sv) 832 ? snprintf (enc->cur, IVUV_MAXCHARS, "%"UVuf, (UV)SvUVX (sv)) 833 : snprintf (enc->cur, IVUV_MAXCHARS, "%"IVdf, (IV)SvIVX (sv)); 834 } 835 } 836 else if (SvROK (sv)) 837 encode_rv (enc, SvRV (sv)); 838 else if (!SvOK (sv) || enc->json.flags & F_ALLOW_UNKNOWN) 839 encode_str (enc, "null", 4, 0); 840 else 841 croak ("encountered perl type (%s,0x%x) that JSON cannot handle, you might want to report this", 842 SvPV_nolen (sv), (unsigned int)SvFLAGS (sv)); 843} 844 845static SV * 846encode_json (SV *scalar, JSON *json) 847{ 848 enc_t enc; 849 850 if (!(json->flags & F_ALLOW_NONREF) && !SvROK (scalar)) 851 croak ("hash- or arrayref expected (not a simple scalar, use allow_nonref to allow this)"); 852 853 enc.json = *json; 854 enc.sv = sv_2mortal (NEWSV (0, INIT_SIZE)); 855 enc.cur = SvPVX (enc.sv); 856 enc.end = SvEND (enc.sv); 857 enc.indent = 0; 858 enc.limit = enc.json.flags & F_ASCII ? 0x000080UL 859 : enc.json.flags & F_LATIN1 ? 0x000100UL 860 : 0x110000UL; 861 862 SvPOK_only (enc.sv); 863 encode_sv (&enc, scalar); 864 encode_nl (&enc); 865 866 SvCUR_set (enc.sv, enc.cur - SvPVX (enc.sv)); 867 *SvEND (enc.sv) = 0; // many xs functions expect a trailing 0 for text strings 868 869 if (!(enc.json.flags & (F_ASCII | F_LATIN1 | F_UTF8))) 870 SvUTF8_on (enc.sv); 871 872 if (enc.json.flags & F_SHRINK) 873 shrink (enc.sv); 874 875 return enc.sv; 876} 877 878///////////////////////////////////////////////////////////////////////////// 879// decoder 880 881// structure used for decoding JSON 882typedef struct 883{ 884 char *cur; // current parser pointer 885 char *end; // end of input string 886 const char *err; // parse error, if != 0 887 JSON json; 888 U32 depth; // recursion depth 889 U32 maxdepth; // recursion depth limit 890} dec_t; 891 892INLINE void 893decode_comment (dec_t *dec) 894{ 895 // only '#'-style comments allowed a.t.m. 896 897 while (*dec->cur && *dec->cur != 0x0a && *dec->cur != 0x0d) 898 ++dec->cur; 899} 900 901INLINE void 902decode_ws (dec_t *dec) 903{ 904 for (;;) 905 { 906 char ch = *dec->cur; 907 908 if (ch > 0x20) 909 { 910 if (expect_false (ch == '#')) 911 { 912 if (dec->json.flags & F_RELAXED) 913 decode_comment (dec); 914 else 915 break; 916 } 917 else 918 break; 919 } 920 else if (ch != 0x20 && ch != 0x0a && ch != 0x0d && ch != 0x09) 921 break; // parse error, but let higher level handle it, gives better error messages 922 923 ++dec->cur; 924 } 925} 926 927#define ERR(reason) SB dec->err = reason; goto fail; SE 928 929#define EXPECT_CH(ch) SB \ 930 if (*dec->cur != ch) \ 931 ERR (# ch " expected"); \ 932 ++dec->cur; \ 933 SE 934 935#define DEC_INC_DEPTH if (++dec->depth > dec->json.max_depth) ERR (ERR_NESTING_EXCEEDED) 936#define DEC_DEC_DEPTH --dec->depth 937 938static SV *decode_sv (dec_t *dec); 939 940static signed char decode_hexdigit[256]; 941 942static UV 943decode_4hex (dec_t *dec) 944{ 945 signed char d1, d2, d3, d4; 946 unsigned char *cur = (unsigned char *)dec->cur; 947 948 d1 = decode_hexdigit [cur [0]]; if (expect_false (d1 < 0)) ERR ("exactly four hexadecimal digits expected"); 949 d2 = decode_hexdigit [cur [1]]; if (expect_false (d2 < 0)) ERR ("exactly four hexadecimal digits expected"); 950 d3 = decode_hexdigit [cur [2]]; if (expect_false (d3 < 0)) ERR ("exactly four hexadecimal digits expected"); 951 d4 = decode_hexdigit [cur [3]]; if (expect_false (d4 < 0)) ERR ("exactly four hexadecimal digits expected"); 952 953 dec->cur += 4; 954 955 return ((UV)d1) << 12 956 | ((UV)d2) << 8 957 | ((UV)d3) << 4 958 | ((UV)d4); 959 960fail: 961 return (UV)-1; 962} 963 964static SV * 965decode_str (dec_t *dec) 966{ 967 SV *sv = 0; 968 int utf8 = 0; 969 char *dec_cur = dec->cur; 970 971 do 972 { 973 char buf [SHORT_STRING_LEN + UTF8_MAXBYTES]; 974 char *cur = buf; 975 976 do 977 { 978 unsigned char ch = *(unsigned char *)dec_cur++; 979 980 if (expect_false (ch == '"')) 981 { 982 --dec_cur; 983 break; 984 } 985 else if (expect_false (ch == '\\')) 986 { 987 switch (*dec_cur) 988 { 989 case '\\': 990 case '/': 991 case '"': *cur++ = *dec_cur++; break; 992 993 case 'b': ++dec_cur; *cur++ = '\010'; break; 994 case 't': ++dec_cur; *cur++ = '\011'; break; 995 case 'n': ++dec_cur; *cur++ = '\012'; break; 996 case 'f': ++dec_cur; *cur++ = '\014'; break; 997 case 'r': ++dec_cur; *cur++ = '\015'; break; 998 999 case 'u': 1000 { 1001 UV lo, hi; 1002 ++dec_cur; 1003 1004 dec->cur = dec_cur; 1005 hi = decode_4hex (dec); 1006 dec_cur = dec->cur; 1007 if (hi == (UV)-1) 1008 goto fail; 1009 1010 // possibly a surrogate pair 1011 if (hi >= 0xd800) 1012 if (hi < 0xdc00) 1013 { 1014 if (dec_cur [0] != '\\' || dec_cur [1] != 'u') 1015 ERR ("missing low surrogate character in surrogate pair"); 1016 1017 dec_cur += 2; 1018 1019 dec->cur = dec_cur; 1020 lo = decode_4hex (dec); 1021 dec_cur = dec->cur; 1022 if (lo == (UV)-1) 1023 goto fail; 1024 1025 if (lo < 0xdc00 || lo >= 0xe000) 1026 ERR ("surrogate pair expected"); 1027 1028 hi = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000; 1029 } 1030 else if (hi < 0xe000) 1031 ERR ("missing high surrogate character in surrogate pair"); 1032 1033 if (hi >= 0x80) 1034 { 1035 utf8 = 1; 1036 1037 cur = encode_utf8 (cur, hi); 1038 } 1039 else 1040 *cur++ = hi; 1041 } 1042 break; 1043 1044 default: 1045 --dec_cur; 1046 ERR ("illegal backslash escape sequence in string"); 1047 } 1048 } 1049 else if (expect_true (ch >= 0x20 && ch < 0x80)) 1050 *cur++ = ch; 1051 else if (ch >= 0x80) 1052 { 1053 STRLEN clen; 1054 1055 --dec_cur; 1056 1057 decode_utf8 (dec_cur, dec->end - dec_cur, &clen); 1058 if (clen == (STRLEN)-1) 1059 ERR ("malformed UTF-8 character in JSON string"); 1060 1061 do 1062 *cur++ = *dec_cur++; 1063 while (--clen); 1064 1065 utf8 = 1; 1066 } 1067 else 1068 { 1069 --dec_cur; 1070 1071 if (!ch) 1072 ERR ("unexpected end of string while parsing JSON string"); 1073 else 1074 ERR ("invalid character encountered while parsing JSON string"); 1075 } 1076 } 1077 while (cur < buf + SHORT_STRING_LEN); 1078 1079 { 1080 STRLEN len = cur - buf; 1081 1082 if (sv) 1083 { 1084 STRLEN cur = SvCUR (sv); 1085 1086 if (SvLEN (sv) <= cur + len) 1087 SvGROW (sv, cur + (len < (cur >> 2) ? cur >> 2 : len) + 1); 1088 1089 memcpy (SvPVX (sv) + SvCUR (sv), buf, len); 1090 SvCUR_set (sv, SvCUR (sv) + len); 1091 } 1092 else 1093 sv = newSVpvn (buf, len); 1094 } 1095 } 1096 while (*dec_cur != '"'); 1097 1098 ++dec_cur; 1099 1100 if (sv) 1101 { 1102 SvPOK_only (sv); 1103 *SvEND (sv) = 0; 1104 1105 if (utf8) 1106 SvUTF8_on (sv); 1107 } 1108 else 1109 sv = newSVpvn ("", 0); 1110 1111 dec->cur = dec_cur; 1112 return sv; 1113 1114fail: 1115 dec->cur = dec_cur; 1116 return 0; 1117} 1118 1119static SV * 1120decode_num (dec_t *dec) 1121{ 1122 int is_nv = 0; 1123 char *start = dec->cur; 1124 1125 // [minus] 1126 if (*dec->cur == '-') 1127 ++dec->cur; 1128 1129 if (*dec->cur == '0') 1130 { 1131 ++dec->cur; 1132 if (*dec->cur >= '0' && *dec->cur <= '9') 1133 ERR ("malformed number (leading zero must not be followed by another digit)"); 1134 } 1135 else if (*dec->cur < '0' || *dec->cur > '9') 1136 ERR ("malformed number (no digits after initial minus)"); 1137 else 1138 do 1139 { 1140 ++dec->cur; 1141 } 1142 while (*dec->cur >= '0' && *dec->cur <= '9'); 1143 1144 // [frac] 1145 if (*dec->cur == '.') 1146 { 1147 ++dec->cur; 1148 1149 if (*dec->cur < '0' || *dec->cur > '9') 1150 ERR ("malformed number (no digits after decimal point)"); 1151 1152 do 1153 { 1154 ++dec->cur; 1155 } 1156 while (*dec->cur >= '0' && *dec->cur <= '9'); 1157 1158 is_nv = 1; 1159 } 1160 1161 // [exp] 1162 if (*dec->cur == 'e' || *dec->cur == 'E') 1163 { 1164 ++dec->cur; 1165 1166 if (*dec->cur == '-' || *dec->cur == '+') 1167 ++dec->cur; 1168 1169 if (*dec->cur < '0' || *dec->cur > '9') 1170 ERR ("malformed number (no digits after exp sign)"); 1171 1172 do 1173 { 1174 ++dec->cur; 1175 } 1176 while (*dec->cur >= '0' && *dec->cur <= '9'); 1177 1178 is_nv = 1; 1179 } 1180 1181 if (!is_nv) 1182 { 1183 int len = dec->cur - start; 1184 1185 // special case the rather common 1..5-digit-int case 1186 if (*start == '-') 1187 switch (len) 1188 { 1189 case 2: return newSViv (-(IV)( start [1] - '0' * 1)); 1190 case 3: return newSViv (-(IV)( start [1] * 10 + start [2] - '0' * 11)); 1191 case 4: return newSViv (-(IV)( start [1] * 100 + start [2] * 10 + start [3] - '0' * 111)); 1192 case 5: return newSViv (-(IV)( start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 1111)); 1193 case 6: return newSViv (-(IV)(start [1] * 10000 + start [2] * 1000 + start [3] * 100 + start [4] * 10 + start [5] - '0' * 11111)); 1194 } 1195 else 1196 switch (len) 1197 { 1198 case 1: return newSViv ( start [0] - '0' * 1); 1199 case 2: return newSViv ( start [0] * 10 + start [1] - '0' * 11); 1200 case 3: return newSViv ( start [0] * 100 + start [1] * 10 + start [2] - '0' * 111); 1201 case 4: return newSViv ( start [0] * 1000 + start [1] * 100 + start [2] * 10 + start [3] - '0' * 1111); 1202 case 5: return newSViv ( start [0] * 10000 + start [1] * 1000 + start [2] * 100 + start [3] * 10 + start [4] - '0' * 11111); 1203 } 1204 1205 { 1206 UV uv; 1207 int numtype = grok_number (start, len, &uv); 1208 if (numtype & IS_NUMBER_IN_UV) 1209 if (numtype & IS_NUMBER_NEG) 1210 { 1211 if (uv < (UV)IV_MIN) 1212 return newSViv (-(IV)uv); 1213 } 1214 else 1215 return newSVuv (uv); 1216 } 1217 1218 len -= *start == '-' ? 1 : 0; 1219 1220 // does not fit into IV or UV, try NV 1221 if (len <= NV_DIG) 1222 // fits into NV without loss of precision 1223 return newSVnv (json_atof (start)); 1224 1225 // everything else fails, convert it to a string 1226 return newSVpvn (start, dec->cur - start); 1227 } 1228 1229 // loss of precision here 1230 return newSVnv (json_atof (start)); 1231 1232fail: 1233 return 0; 1234} 1235 1236static SV * 1237decode_av (dec_t *dec) 1238{ 1239 AV *av = newAV (); 1240 1241 DEC_INC_DEPTH; 1242 decode_ws (dec); 1243 1244 if (*dec->cur == ']') 1245 ++dec->cur; 1246 else 1247 for (;;) 1248 { 1249 SV *value; 1250 1251 value = decode_sv (dec); 1252 if (!value) 1253 goto fail; 1254 1255 av_push (av, value); 1256 1257 decode_ws (dec); 1258 1259 if (*dec->cur == ']') 1260 { 1261 ++dec->cur; 1262 break; 1263 } 1264 1265 if (*dec->cur != ',') 1266 ERR (", or ] expected while parsing array"); 1267 1268 ++dec->cur; 1269 1270 decode_ws (dec); 1271 1272 if (*dec->cur == ']' && dec->json.flags & F_RELAXED) 1273 { 1274 ++dec->cur; 1275 break; 1276 } 1277 } 1278 1279 DEC_DEC_DEPTH; 1280 return newRV_noinc ((SV *)av); 1281 1282fail: 1283 SvREFCNT_dec (av); 1284 DEC_DEC_DEPTH; 1285 return 0; 1286} 1287 1288static SV * 1289decode_hv (dec_t *dec) 1290{ 1291 SV *sv; 1292 HV *hv = newHV (); 1293 1294 DEC_INC_DEPTH; 1295 decode_ws (dec); 1296 1297 if (*dec->cur == '}') 1298 ++dec->cur; 1299 else 1300 for (;;) 1301 { 1302 EXPECT_CH ('"'); 1303 1304 // heuristic: assume that 1305 // a) decode_str + hv_store_ent are abysmally slow. 1306 // b) most hash keys are short, simple ascii text. 1307 // => try to "fast-match" such strings to avoid 1308 // the overhead of decode_str + hv_store_ent. 1309 { 1310 SV *value; 1311 char *p = dec->cur; 1312 char *e = p + 24; // only try up to 24 bytes 1313 1314 for (;;) 1315 { 1316 // the >= 0x80 is false on most architectures 1317 if (p == e || *p < 0x20 || *p >= 0x80 || *p == '\\') 1318 { 1319 // slow path, back up and use decode_str 1320 SV *key = decode_str (dec); 1321 if (!key) 1322 goto fail; 1323 1324 decode_ws (dec); EXPECT_CH (':'); 1325 1326 decode_ws (dec); 1327 value = decode_sv (dec); 1328 if (!value) 1329 { 1330 SvREFCNT_dec (key); 1331 goto fail; 1332 } 1333 1334 hv_store_ent (hv, key, value, 0); 1335 SvREFCNT_dec (key); 1336 1337 break; 1338 } 1339 else if (*p == '"') 1340 { 1341 // fast path, got a simple key 1342 char *key = dec->cur; 1343 int len = p - key; 1344 dec->cur = p + 1; 1345 1346 decode_ws (dec); EXPECT_CH (':'); 1347 1348 decode_ws (dec); 1349 value = decode_sv (dec); 1350 if (!value) 1351 goto fail; 1352 1353 hv_store (hv, key, len, value, 0); 1354 1355 break; 1356 } 1357 1358 ++p; 1359 } 1360 } 1361 1362 decode_ws (dec); 1363 1364 if (*dec->cur == '}') 1365 { 1366 ++dec->cur; 1367 break; 1368 } 1369 1370 if (*dec->cur != ',') 1371 ERR (", or } expected while parsing object/hash"); 1372 1373 ++dec->cur; 1374 1375 decode_ws (dec); 1376 1377 if (*dec->cur == '}' && dec->json.flags & F_RELAXED) 1378 { 1379 ++dec->cur; 1380 break; 1381 } 1382 } 1383 1384 DEC_DEC_DEPTH; 1385 sv = newRV_noinc ((SV *)hv); 1386 1387 // check filter callbacks 1388 if (dec->json.flags & F_HOOK) 1389 { 1390 if (dec->json.cb_sk_object && HvKEYS (hv) == 1) 1391 { 1392 HE *cb, *he; 1393 1394 hv_iterinit (hv); 1395 he = hv_iternext (hv); 1396 hv_iterinit (hv); 1397 1398 // the next line creates a mortal sv each time its called. 1399 // might want to optimise this for common cases. 1400 cb = hv_fetch_ent (dec->json.cb_sk_object, hv_iterkeysv (he), 0, 0); 1401 1402 if (cb) 1403 { 1404 dSP; 1405 int count; 1406 1407 ENTER; SAVETMPS; PUSHMARK (SP); 1408 XPUSHs (HeVAL (he)); 1409 sv_2mortal (sv); 1410 1411 PUTBACK; count = call_sv (HeVAL (cb), G_ARRAY); SPAGAIN; 1412 1413 if (count == 1) 1414 { 1415 sv = newSVsv (POPs); 1416 FREETMPS; LEAVE; 1417 return sv; 1418 } 1419 1420 SvREFCNT_inc (sv); 1421 FREETMPS; LEAVE; 1422 } 1423 } 1424 1425 if (dec->json.cb_object) 1426 { 1427 dSP; 1428 int count; 1429 1430 ENTER; SAVETMPS; PUSHMARK (SP); 1431 XPUSHs (sv_2mortal (sv)); 1432 1433 PUTBACK; count = call_sv (dec->json.cb_object, G_ARRAY); SPAGAIN; 1434 1435 if (count == 1) 1436 { 1437 sv = newSVsv (POPs); 1438 FREETMPS; LEAVE; 1439 return sv; 1440 } 1441 1442 SvREFCNT_inc (sv); 1443 FREETMPS; LEAVE; 1444 } 1445 } 1446 1447 return sv; 1448 1449fail: 1450 SvREFCNT_dec (hv); 1451 DEC_DEC_DEPTH; 1452 return 0; 1453} 1454 1455static SV * 1456decode_sv (dec_t *dec) 1457{ 1458 // the beauty of JSON: you need exactly one character lookahead 1459 // to parse everything. 1460 switch (*dec->cur) 1461 { 1462 case '"': ++dec->cur; return decode_str (dec); 1463 case '[': ++dec->cur; return decode_av (dec); 1464 case '{': ++dec->cur; return decode_hv (dec); 1465 1466 case '-': 1467 case '0': case '1': case '2': case '3': case '4': 1468 case '5': case '6': case '7': case '8': case '9': 1469 return decode_num (dec); 1470 1471 case 't': 1472 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "true", 4)) 1473 { 1474 dec->cur += 4; 1475#if JSON_SLOW 1476 json_true = get_bool ("JSON::XS::true"); 1477#endif 1478 return newSVsv (json_true); 1479 } 1480 else 1481 ERR ("'true' expected"); 1482 1483 break; 1484 1485 case 'f': 1486 if (dec->end - dec->cur >= 5 && !memcmp (dec->cur, "false", 5)) 1487 { 1488 dec->cur += 5; 1489#if JSON_SLOW 1490 json_false = get_bool ("JSON::XS::false"); 1491#endif 1492 return newSVsv (json_false); 1493 } 1494 else 1495 ERR ("'false' expected"); 1496 1497 break; 1498 1499 case 'n': 1500 if (dec->end - dec->cur >= 4 && !memcmp (dec->cur, "null", 4)) 1501 { 1502 dec->cur += 4; 1503 return newSVsv (&PL_sv_undef); 1504 } 1505 else 1506 ERR ("'null' expected"); 1507 1508 break; 1509 1510 default: 1511 ERR ("malformed JSON string, neither array, object, number, string or atom"); 1512 break; 1513 } 1514 1515fail: 1516 return 0; 1517} 1518 1519static SV * 1520decode_json (SV *string, JSON *json, char **offset_return) 1521{ 1522 dec_t dec; 1523 SV *sv; 1524 1525 /* work around bugs in 5.10 where manipulating magic values 1526 * will perl ignore the magic in subsequent accesses. 1527 * also make a copy of non-PV values, to get them into a clean 1528 * state (SvPV should do that, but it's buggy, see below). 1529 */ 1530 /*SvGETMAGIC (string);*/ 1531 if (SvMAGICAL (string) || !SvPOK (string)) 1532 string = sv_2mortal (newSVsv (string)); 1533 1534 SvUPGRADE (string, SVt_PV); 1535 1536 /* work around a bug in perl 5.10, which causes SvCUR to fail an 1537 * assertion with -DDEBUGGING, although SvCUR is documented to 1538 * return the xpv_cur field which certainly exists after upgrading. 1539 * according to nicholas clark, calling SvPOK fixes this. 1540 * But it doesn't fix it, so try another workaround, call SvPV_nolen 1541 * and hope for the best. 1542 * Damnit, SvPV_nolen still trips over yet another assertion. This 1543 * assertion business is seriously broken, try yet another workaround 1544 * for the broken -DDEBUGGING. 1545 */ 1546 { 1547#ifdef DEBUGGING 1548 STRLEN offset = SvOK (string) ? sv_len (string) : 0; 1549#else 1550 STRLEN offset = SvCUR (string); 1551#endif 1552 1553 if (offset > json->max_size && json->max_size) 1554 croak ("attempted decode of JSON text of %lu bytes size, but max_size is set to %lu", 1555 (unsigned long)SvCUR (string), (unsigned long)json->max_size); 1556 } 1557 1558 if (DECODE_WANTS_OCTETS (json)) 1559 sv_utf8_downgrade (string, 0); 1560 else 1561 sv_utf8_upgrade (string); 1562 1563 SvGROW (string, SvCUR (string) + 1); // should basically be a NOP 1564 1565 dec.json = *json; 1566 dec.cur = SvPVX (string); 1567 dec.end = SvEND (string); 1568 dec.err = 0; 1569 dec.depth = 0; 1570 1571 if (dec.json.cb_object || dec.json.cb_sk_object) 1572 dec.json.flags |= F_HOOK; 1573 1574 *dec.end = 0; // this should basically be a nop, too, but make sure it's there 1575 1576 decode_ws (&dec); 1577 sv = decode_sv (&dec); 1578 1579 if (offset_return) 1580 *offset_return = dec.cur; 1581 1582 if (!(offset_return || !sv)) 1583 { 1584 // check for trailing garbage 1585 decode_ws (&dec); 1586 1587 if (*dec.cur) 1588 { 1589 dec.err = "garbage after JSON object"; 1590 SvREFCNT_dec (sv); 1591 sv = 0; 1592 } 1593 } 1594 1595 if (!sv) 1596 { 1597 SV *uni = sv_newmortal (); 1598 1599 // horrible hack to silence warning inside pv_uni_display 1600 COP cop = *PL_curcop; 1601 cop.cop_warnings = pWARN_NONE; 1602 ENTER; 1603 SAVEVPTR (PL_curcop); 1604 PL_curcop = &cop; 1605 pv_uni_display (uni, dec.cur, dec.end - dec.cur, 20, UNI_DISPLAY_QQ); 1606 LEAVE; 1607 1608 croak ("%s, at character offset %d (before \"%s\")", 1609 dec.err, 1610 (int)ptr_to_index (string, dec.cur), 1611 dec.cur != dec.end ? SvPV_nolen (uni) : "(end of string)"); 1612 } 1613 1614 sv = sv_2mortal (sv); 1615 1616 if (!(dec.json.flags & F_ALLOW_NONREF) && !SvROK (sv)) 1617 croak ("JSON text must be an object or array (but found number, string, true, false or null, use allow_nonref to allow this)"); 1618 1619 return sv; 1620} 1621 1622///////////////////////////////////////////////////////////////////////////// 1623// incremental parser 1624 1625static void 1626incr_parse (JSON *self) 1627{ 1628 const char *p = SvPVX (self->incr_text) + self->incr_pos; 1629 1630 // the state machine here is a bit convoluted and could be simplified a lot 1631 // but this would make it slower, so... 1632 1633 for (;;) 1634 { 1635 //printf ("loop pod %d *p<%c><%s>, mode %d nest %d\n", p - SvPVX (self->incr_text), *p, p, self->incr_mode, self->incr_nest);//D 1636 switch (self->incr_mode) 1637 { 1638 // only used for initial whitespace skipping 1639 case INCR_M_WS: 1640 for (;;) 1641 { 1642 if (*p > 0x20) 1643 { 1644 if (*p == '#') 1645 { 1646 self->incr_mode = INCR_M_C0; 1647 goto incr_m_c; 1648 } 1649 else 1650 { 1651 self->incr_mode = INCR_M_JSON; 1652 goto incr_m_json; 1653 } 1654 } 1655 else if (!*p) 1656 goto interrupt; 1657 1658 ++p; 1659 } 1660 1661 // skip a single char inside a string (for \\-processing) 1662 case INCR_M_BS: 1663 if (!*p) 1664 goto interrupt; 1665 1666 ++p; 1667 self->incr_mode = INCR_M_STR; 1668 goto incr_m_str; 1669 1670 // inside #-style comments 1671 case INCR_M_C0: 1672 case INCR_M_C1: 1673 incr_m_c: 1674 for (;;) 1675 { 1676 if (*p == '\n') 1677 { 1678 self->incr_mode = self->incr_mode == INCR_M_C0 ? INCR_M_WS : INCR_M_JSON; 1679 break; 1680 } 1681 else if (!*p) 1682 goto interrupt; 1683 1684 ++p; 1685 } 1686 1687 break; 1688 1689 // inside a string 1690 case INCR_M_STR: 1691 incr_m_str: 1692 for (;;) 1693 { 1694 if (*p == '"') 1695 { 1696 ++p; 1697 self->incr_mode = INCR_M_JSON; 1698 1699 if (!self->incr_nest) 1700 goto interrupt; 1701 1702 goto incr_m_json; 1703 } 1704 else if (*p == '\\') 1705 { 1706 ++p; // "virtually" consumes character after \ 1707 1708 if (!*p) // if at end of string we have to switch modes 1709 { 1710 self->incr_mode = INCR_M_BS; 1711 goto interrupt; 1712 } 1713 } 1714 else if (!*p) 1715 goto interrupt; 1716 1717 ++p; 1718 } 1719 1720 // after initial ws, outside string 1721 case INCR_M_JSON: 1722 incr_m_json: 1723 for (;;) 1724 { 1725 switch (*p++) 1726 { 1727 case 0: 1728 --p; 1729 goto interrupt; 1730 1731 case 0x09: 1732 case 0x0a: 1733 case 0x0d: 1734 case 0x20: 1735 if (!self->incr_nest) 1736 { 1737 --p; // do not eat the whitespace, let the next round do it 1738 goto interrupt; 1739 } 1740 break; 1741 1742 case '"': 1743 self->incr_mode = INCR_M_STR; 1744 goto incr_m_str; 1745 1746 case '[': 1747 case '{': 1748 if (++self->incr_nest > self->max_depth) 1749 croak (ERR_NESTING_EXCEEDED); 1750 break; 1751 1752 case ']': 1753 case '}': 1754 if (--self->incr_nest <= 0) 1755 goto interrupt; 1756 break; 1757 1758 case '#': 1759 self->incr_mode = INCR_M_C1; 1760 goto incr_m_c; 1761 } 1762 } 1763 } 1764 1765 modechange: 1766 ; 1767 } 1768 1769interrupt: 1770 self->incr_pos = p - SvPVX (self->incr_text); 1771 //printf ("interrupt<%.*s>\n", self->incr_pos, SvPVX(self->incr_text));//D 1772 //printf ("return pos %d mode %d nest %d\n", self->incr_pos, self->incr_mode, self->incr_nest);//D 1773} 1774 1775///////////////////////////////////////////////////////////////////////////// 1776// XS interface functions 1777 1778MODULE = JSON::XS PACKAGE = JSON::XS 1779 1780BOOT: 1781{ 1782 int i; 1783 1784 for (i = 0; i < 256; ++i) 1785 decode_hexdigit [i] = 1786 i >= '0' && i <= '9' ? i - '0' 1787 : i >= 'a' && i <= 'f' ? i - 'a' + 10 1788 : i >= 'A' && i <= 'F' ? i - 'A' + 10 1789 : -1; 1790 1791 json_stash = gv_stashpv ("JSON::XS" , 1); 1792 json_boolean_stash = gv_stashpv ("JSON::XS::Boolean", 1); 1793 1794 json_true = get_bool ("JSON::XS::true"); 1795 json_false = get_bool ("JSON::XS::false"); 1796 1797 CvNODEBUG_on (get_cv ("JSON::XS::incr_text", 0)); /* the debugger completely breaks lvalue subs */ 1798} 1799 1800PROTOTYPES: DISABLE 1801 1802void CLONE (...) 1803 CODE: 1804 json_stash = 0; 1805 json_boolean_stash = 0; 1806 1807void new (char *klass) 1808 PPCODE: 1809{ 1810 SV *pv = NEWSV (0, sizeof (JSON)); 1811 SvPOK_only (pv); 1812 json_init ((JSON *)SvPVX (pv)); 1813 XPUSHs (sv_2mortal (sv_bless ( 1814 newRV_noinc (pv), 1815 strEQ (klass, "JSON::XS") ? JSON_STASH : gv_stashpv (klass, 1) 1816 ))); 1817} 1818 1819void ascii (JSON *self, int enable = 1) 1820 ALIAS: 1821 ascii = F_ASCII 1822 latin1 = F_LATIN1 1823 utf8 = F_UTF8 1824 indent = F_INDENT 1825 canonical = F_CANONICAL 1826 space_before = F_SPACE_BEFORE 1827 space_after = F_SPACE_AFTER 1828 pretty = F_PRETTY 1829 allow_nonref = F_ALLOW_NONREF 1830 shrink = F_SHRINK 1831 allow_blessed = F_ALLOW_BLESSED 1832 convert_blessed = F_CONV_BLESSED 1833 relaxed = F_RELAXED 1834 allow_unknown = F_ALLOW_UNKNOWN 1835 PPCODE: 1836{ 1837 if (enable) 1838 self->flags |= ix; 1839 else 1840 self->flags &= ~ix; 1841 1842 XPUSHs (ST (0)); 1843} 1844 1845void get_ascii (JSON *self) 1846 ALIAS: 1847 get_ascii = F_ASCII 1848 get_latin1 = F_LATIN1 1849 get_utf8 = F_UTF8 1850 get_indent = F_INDENT 1851 get_canonical = F_CANONICAL 1852 get_space_before = F_SPACE_BEFORE 1853 get_space_after = F_SPACE_AFTER 1854 get_allow_nonref = F_ALLOW_NONREF 1855 get_shrink = F_SHRINK 1856 get_allow_blessed = F_ALLOW_BLESSED 1857 get_convert_blessed = F_CONV_BLESSED 1858 get_relaxed = F_RELAXED 1859 get_allow_unknown = F_ALLOW_UNKNOWN 1860 PPCODE: 1861 XPUSHs (boolSV (self->flags & ix)); 1862 1863void max_depth (JSON *self, U32 max_depth = 0x80000000UL) 1864 PPCODE: 1865 self->max_depth = max_depth; 1866 XPUSHs (ST (0)); 1867 1868U32 get_max_depth (JSON *self) 1869 CODE: 1870 RETVAL = self->max_depth; 1871 OUTPUT: 1872 RETVAL 1873 1874void max_size (JSON *self, U32 max_size = 0) 1875 PPCODE: 1876 self->max_size = max_size; 1877 XPUSHs (ST (0)); 1878 1879int get_max_size (JSON *self) 1880 CODE: 1881 RETVAL = self->max_size; 1882 OUTPUT: 1883 RETVAL 1884 1885void filter_json_object (JSON *self, SV *cb = &PL_sv_undef) 1886 PPCODE: 1887{ 1888 SvREFCNT_dec (self->cb_object); 1889 self->cb_object = SvOK (cb) ? newSVsv (cb) : 0; 1890 1891 XPUSHs (ST (0)); 1892} 1893 1894void filter_json_single_key_object (JSON *self, SV *key, SV *cb = &PL_sv_undef) 1895 PPCODE: 1896{ 1897 if (!self->cb_sk_object) 1898 self->cb_sk_object = newHV (); 1899 1900 if (SvOK (cb)) 1901 hv_store_ent (self->cb_sk_object, key, newSVsv (cb), 0); 1902 else 1903 { 1904 hv_delete_ent (self->cb_sk_object, key, G_DISCARD, 0); 1905 1906 if (!HvKEYS (self->cb_sk_object)) 1907 { 1908 SvREFCNT_dec (self->cb_sk_object); 1909 self->cb_sk_object = 0; 1910 } 1911 } 1912 1913 XPUSHs (ST (0)); 1914} 1915 1916void encode (JSON *self, SV *scalar) 1917 PPCODE: 1918 PUTBACK; scalar = encode_json (scalar, self); SPAGAIN; 1919 XPUSHs (scalar); 1920 1921void decode (JSON *self, SV *jsonstr) 1922 PPCODE: 1923 PUTBACK; jsonstr = decode_json (jsonstr, self, 0); SPAGAIN; 1924 XPUSHs (jsonstr); 1925 1926void decode_prefix (JSON *self, SV *jsonstr) 1927 PPCODE: 1928{ 1929 SV *sv; 1930 char *offset; 1931 PUTBACK; sv = decode_json (jsonstr, self, &offset); SPAGAIN; 1932 EXTEND (SP, 2); 1933 PUSHs (sv); 1934 PUSHs (sv_2mortal (newSVuv (ptr_to_index (jsonstr, offset)))); 1935} 1936 1937void incr_parse (JSON *self, SV *jsonstr = 0) 1938 PPCODE: 1939{ 1940 if (!self->incr_text) 1941 self->incr_text = newSVpvn ("", 0); 1942 1943 /* if utf8-ness doesn't match the decoder, need to upgrade/downgrade */ 1944 if (!DECODE_WANTS_OCTETS (self) == !SvUTF8 (self->incr_text)) 1945 if (DECODE_WANTS_OCTETS (self)) 1946 { 1947 if (self->incr_pos) 1948 self->incr_pos = utf8_length ((U8 *)SvPVX (self->incr_text), 1949 (U8 *)SvPVX (self->incr_text) + self->incr_pos); 1950 1951 sv_utf8_downgrade (self->incr_text, 0); 1952 } 1953 else 1954 { 1955 sv_utf8_upgrade (self->incr_text); 1956 1957 if (self->incr_pos) 1958 self->incr_pos = utf8_hop ((U8 *)SvPVX (self->incr_text), self->incr_pos) 1959 - (U8 *)SvPVX (self->incr_text); 1960 } 1961 1962 // append data, if any 1963 if (jsonstr) 1964 { 1965 /* make sure both strings have same encoding */ 1966 if (SvUTF8 (jsonstr) != SvUTF8 (self->incr_text)) 1967 if (SvUTF8 (jsonstr)) 1968 sv_utf8_downgrade (jsonstr, 0); 1969 else 1970 sv_utf8_upgrade (jsonstr); 1971 1972 /* and then just blindly append */ 1973 { 1974 STRLEN len; 1975 const char *str = SvPV (jsonstr, len); 1976 STRLEN cur = SvCUR (self->incr_text); 1977 1978 if (SvLEN (self->incr_text) <= cur + len) 1979 SvGROW (self->incr_text, cur + (len < (cur >> 2) ? cur >> 2 : len) + 1); 1980 1981 Move (str, SvEND (self->incr_text), len, char); 1982 SvCUR_set (self->incr_text, SvCUR (self->incr_text) + len); 1983 *SvEND (self->incr_text) = 0; // this should basically be a nop, too, but make sure it's there 1984 } 1985 } 1986 1987 if (GIMME_V != G_VOID) 1988 do 1989 { 1990 SV *sv; 1991 char *offset; 1992 1993 if (!INCR_DONE (self)) 1994 { 1995 incr_parse (self); 1996 1997 if (self->incr_pos > self->max_size && self->max_size) 1998 croak ("attempted decode of JSON text of %lu bytes size, but max_size is set to %lu", 1999 (unsigned long)self->incr_pos, (unsigned long)self->max_size); 2000 2001 if (!INCR_DONE (self)) 2002 { 2003 // as an optimisation, do not accumulate white space in the incr buffer 2004 if (self->incr_mode == INCR_M_WS && self->incr_pos) 2005 { 2006 self->incr_pos = 0; 2007 SvCUR_set (self->incr_text, 0); 2008 } 2009 2010 break; 2011 } 2012 } 2013 2014 PUTBACK; sv = decode_json (self->incr_text, self, &offset); SPAGAIN; 2015 XPUSHs (sv); 2016 2017 self->incr_pos -= offset - SvPVX (self->incr_text); 2018 self->incr_nest = 0; 2019 self->incr_mode = 0; 2020 2021 sv_chop (self->incr_text, offset); 2022 } 2023 while (GIMME_V == G_ARRAY); 2024} 2025 2026SV *incr_text (JSON *self) 2027 ATTRS: lvalue 2028 CODE: 2029{ 2030 if (self->incr_pos) 2031 croak ("incr_text can not be called when the incremental parser already started parsing"); 2032 2033 RETVAL = self->incr_text ? SvREFCNT_inc (self->incr_text) : &PL_sv_undef; 2034} 2035 OUTPUT: 2036 RETVAL 2037 2038void incr_skip (JSON *self) 2039 CODE: 2040{ 2041 if (self->incr_pos) 2042 { 2043 sv_chop (self->incr_text, SvPV_nolen (self->incr_text) + self->incr_pos); 2044 self->incr_pos = 0; 2045 self->incr_nest = 0; 2046 self->incr_mode = 0; 2047 } 2048} 2049 2050void incr_reset (JSON *self) 2051 CODE: 2052{ 2053 SvREFCNT_dec (self->incr_text); 2054 self->incr_text = 0; 2055 self->incr_pos = 0; 2056 self->incr_nest = 0; 2057 self->incr_mode = 0; 2058} 2059 2060void DESTROY (JSON *self) 2061 CODE: 2062 SvREFCNT_dec (self->cb_sk_object); 2063 SvREFCNT_dec (self->cb_object); 2064 SvREFCNT_dec (self->incr_text); 2065 2066PROTOTYPES: ENABLE 2067 2068void encode_json (SV *scalar) 2069 ALIAS: 2070 to_json_ = 0 2071 encode_json = F_UTF8 2072 PPCODE: 2073{ 2074 JSON json; 2075 json_init (&json); 2076 json.flags |= ix; 2077 PUTBACK; scalar = encode_json (scalar, &json); SPAGAIN; 2078 XPUSHs (scalar); 2079} 2080 2081void decode_json (SV *jsonstr) 2082 ALIAS: 2083 from_json_ = 0 2084 decode_json = F_UTF8 2085 PPCODE: 2086{ 2087 JSON json; 2088 json_init (&json); 2089 json.flags |= ix; 2090 PUTBACK; jsonstr = decode_json (jsonstr, &json, 0); SPAGAIN; 2091 XPUSHs (jsonstr); 2092} 2093 2094